This study examines the pervasive nature of trauma exposure in journalism and its impact across all stages of the news reporting cycle. Drawing from a comprehensive literature review, we highlight the significant gap in research on the neuroscience of journalism and journalist PTSD, despite the high prevalence of PTSD among war correspondents and photojournalists. We present a conceptual framework illustrating how primary and secondary trauma exposures occur during news gathering, reporting, and consumption phases. The study also reviews empirical findings on PTSD's effects on brain structure and function, identifying symptom subclusters and their associated brain regions. This knowledge informs the development of biotypes for more personalized and effective treatment strategies for journalists with PTSD while also emphasizing an urgent need for comprehensive support systems for journalists. This interdisciplinary approach uniting research in journalism, neuroscience, psychology, and organizational management, also advances the understanding of trauma in journalism and its broader implications for news consumption and societal trust in media institutions. Interventions that can support journalists’ mental well-being can also enhance the quality of news reporting and contribute to a more resilient and informed society.

Introduction

A journalist’s job is to tell stories, whether good or bad. Trauma is an underappreciated aspect of journalists reporting on tragic events and relaying this news to audiences (Hill, 2018). Trauma that originates from a direct source is called primary trauma, while “the spread of negative emotional and cognitive states from those who are traumatized to those who have close contact with these individuals” is known as secondary trauma and is much less appreciated, both in the journalism mental health literature as well as the PTSD literature more generally (Motta, 2023). It is important to examine the role of both primary and secondary trauma in journalism, not only in journalists but at all steps of the news cycle. Understanding PTSD in journalists and general consumers of news therefore not only requires understanding where trauma exposure occurs across the news ecosystem but also a dedicated effort to better understand causes and effects of PTSD in this population to help journalists, especially war journalists and war photojournalists, but also the audiences they serve, to have better mental health outcomes and to improve journalism as a profession.

Trauma is involved in all steps in the news cycle

Table 1 conceptualizes the news reporting cycle in the context of primary and secondary trauma exposure. As Table 1 demonstrates, trauma exposure can occur at all three phases of this news reporting cycle, making the case for investigating the traumatic exposure of not only journalists in reporting news but potentially, audiences consuming news content.

The management structure for journalists, which typically has a hierarchical structure with journalists reporting to various levels of editors, creates opportunities for all involved in news production to develop secondary or even primary PTSD. A study of local journalists reporting on the 1995 Oklahoma City Bombing, Hill (2018) found that organizational leaders can identify trauma reactions of journalists, and may even become victims of secondary traumatic stress when journalists are assigned to cover a primary trauma such as war or terrorism. Even newsroom journalists who are not out in the field experience repeated exposure to traumatic news (Shah et al., 2020). While in the aftermath of the Oklahoma City Bombing, counseling, leadership, and social support increased for journalists, training has not changed in the nearly 30 years since the tragic event (Hill, 2018). Studying PTSD in journalism is a small community of practice, collecting more and more evidence that journalists and newsrooms face unique challenges when reporting on difficult events such as war, natural disasters, or terrorism. Furthermore, the trauma they are exposed to and report on can be further passed down the line beyond journalists and newsroom management to the communities served, as is discussed next.

Studies on the neuroscience of journalism and journalist PTSD are scarce

While a vast literature has documented the effects of stress and anxiety on the brain, few studies have examined the neuroscience of journalism consumption, let alone journalist PTSD, specifically. Studies on the effect of news consumption on the brain are scant but reveal news consumption’s clear effects on behavior, cognition, and decision-making and help explain why people may “doomscroll” or read “clickbait” articles (Dan et al., 2020; Li et al., 2021; Wei et al., 2022). Researchers have also examined the role of stress in news consumption in a very limited context. A study of doomscrolling of COVID-19 news in the pandemic found that greater exposure to COVID-19 news was associated with more stress about COVID, greater depression and anxiety, and increased drug use as a coping mechanism (Dyar et al., 2024). Another study found that women exhibit greater signs of stress reactivity than men after reading bad news (Marin et al., 2012). More recently, Shabahang and colleagues (2024) discovered that doomscrolling evokes existential anxiety in both the United States and Iran.

Further work is needed to determine in what ways the neuroscience of news gathering and reporting is similar to or different from that of news consumption. Only a handful of studies have examined journalism’s effect on journalists, though the existing evidence suggests that PTSD is an underreported problem among practitioners. Jukes (2017) terms the phrase “affective journalism,” rooted in neuroscience and referring to “affective processes, behaviours and practices that lie at the heart of journalists’ work when covering traumatic news stories.” Jukes (2017) learns that wartime journalists must work to actively distance themselves from the traumatic events they are covering as a buffer from ill effects.

Feinstein et al. (2002) note that war journalism is “a hazardous profession,” with PTSD rates matching those of combat veterans, adding, “despite the risks inherent in reporting war, we could find no research on the psychological health of war reporters.” This is echoed by Marais and Stewart (2005), who suggest that journalist temperament contributes to development of PTSD. A 2023 study of East African journalists (Radoli, 2023) found that trauma visuals can contribute to escalation of journalist PTSD.

War photojournalists are particularly vulnerable to PTSD. A study of photojournalists assigned to cover traumatic events by Newman et al. (2003) revealed that those assigned to photograph a greater number of traumatic events experienced greater distress. Jonisová (2022) writes about the negative impact of war photography on photojournalists’ mental health, as it brings them closer to the primary trauma source, “risking their lives to bring us a visual testimony directly from the battlefront.”

A small, dedicated community of practice, both dedicated to training journalists better and advancing the research on trauma and journalism, is emerging to help journalists better cover such events while advancing pedagogical approaches that protect journalist mental health. This highly interdisciplinary work unites scholars in journalism training and education, neuroscience, psychology, organizational management, and more. The Journalism Education and Trauma Research Group - North America (n.d.), a group of researchers dedicated to understanding the role of trauma in journalists’ lives, held its first symposium, titled “On Sacred Ground,” in 2024. The Columbia University Dart Center for Journalism and Trauma (n.d.) unites journalists and trauma researchers, including neuroscientists studying post-traumatic stress disorder, to help journalists grapple with reporting on tragic events.

This work may also provide insights into the neural underpinnings of primary and secondary trauma; empirically validate non-stereotypical traumatic experiences; and improve PTSD treatments by considering lived experiences of people not stereotypically associated with PTSD, given that trauma occurs on a spectrum, and its neurobiological effects can be observed on a spectrum as well (Stark, 2015). Studying journalist PTSD and the downstream effects of traumatic news reporting for consumers of news could also have broad social implications for our interactions with news institutions and the work of building trust with news audiences, given that links are emerging between PTSD and social dysfunction and trust (Winkeler, 2023).

Empirical findings of PTSD effects on brain structure

About 6-7% of the population will meet the criteria for PTSD in their lifetime, with those experiencing traumatic events most likely to develop PTSD. Individual, trauma-focused therapies are most effective to treat the disorder; those who are unable to access trauma-focused therapies can also benefit from “symptom reduction” using selective serotonin reuptake inhibitors (SSRIs) and serotonin-norepinephrine reuptake inhibitors (SNRIs) (Schrader and Ross, 2021). Different types of PTSD may benefit from different therapies. For example, individuals with comorbid PTSD and major depressive disorder (MDD), in a randomized clinical trial of exposure therapy, benefitted more from virtual reality exposure therapy than prolonged imagined exposure when either was combined with d-cycloserine, an antibiotic used as a partial NMDA agonist to boost the effects of exposure therapy (Difede et al., 2022).

Understanding the neurobiology of PTSD and its many different combinations can help answer empirical questions and create more effective treatments, especially given the heterogeneity of PTSD and may have implications for persistence of various symptoms post treatment (Ringwald, 2024; Ringwald et al., 2024). There are 20 symptoms defined in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM V) which, when combined, create 636,120 unique symptom combinations (Galatzer and Levy, 2013). Common comorbidities of PTSD include anxiety and depression (Spinhoven et al., 2014) as well as a higher risk of being diagnosed with a substance use disorder (Goldstein et al., 2016). A large study in the US found that 44.6% of people living with PTSD also struggled with alcohol use disorder (AUD) or substance use disorder (Simpson et al., 2019). AUD and SUD have also been linked to specific features of the disorder, such as dissociation (Najavits and Walsh, 2012) and dysphoria, which can include hyperarousal and emotional numbing symptoms (Biehn et al., 2015). PTSD has demonstrated effects on brain function, structure, and biochemistry (Harnett et al., 2020) which are increasingly well-understood.

Trauma exposure can also create neurobiological signatures without a diagnosis of PTSD, and these signatures can vary. Table 2 (appendix—see PDF) lists symptom subclusters derived from the underlying structure of PTSD (e.g., Elklit and Shevlin, 2010; Krause, et al., 2007) which are differentially associated with different brain regions, not only in PTSD but also in cases in which trauma exposure has happened but symptoms are below the threshold for a PTSD diagnosis. This is consistent with the idea of a trauma spectrum (Stark, 2015) on which both journalists and consumers of news experience primary and secondary trauma at varying levels. While the exact functions and roles of each brain region are outside the scope of this article, it is notable that differences in brain structure exist between different subclusters of symptoms, lending evidence to the idea that symptoms of PTSD, which often overlap with those in anxiety and depression, may lead to unique neural signatures (Tozzi et al., 2024).

The case for PTSD biotypes in war journalism

Reflecting on the diversity of experiences in living with PTSD, and their potential neurological effects, can lead to the establishment of biotypes in which PTSD patients differ on symptoms, behavioral and cognitive performance, and response to medications and treatments (Tozzi et al., 2024). Such biotypes can help those living with PTSD obtain more tailored treatments and help them live better lives, as well as advance the science of trauma and potentially, better illuminate its role in our daily lives as consumers of news.

Journalists are trained to tell stories, so talking to journalists about their traumatic experiences may also support their posttraumatic growth by helping with the establishment of a trauma narrative (van Der Kolk, 2022). Talking to war journalists about their PTSD symptoms, treatments, and other aspects of living with PTSD can help define biotypes that can be further examined using neuroimaging to understand the full range of PTSD lived experiences with implications for treatment and recovery. War journalists are the population of journalists that are most likely to develop PTSD. Feinstein et al. (2002) found not only that war journalists and combat veterans had similar levels of PTSD, but that war journalists scored higher on depression and PTSD measures, and self-reported drinking more alcohol, compared to other journalists. Despite the risks, journalists may not be informed about emotional impacts of this work, and may have inadequate access to mental health resources such as counseling (Newman et al., 2003).

Furthermore, biotypes of PTSD may differ between people actively fighting in a combat zone and those there to report on it. For example, the locus of control may differ between combat veterans and war journalists, and research shows that those with a higher external locus of control are more likely to develop PTSD (Güzel et al., 2024). A combat veteran may have a high internal locus of control, believing that their work can improve the lived experiences of local people, as Marcus Luttrell did as a Navy Seal, as detailed in his memoir Service: A Navy Seal at War (Luttrell and Hornfischer, 2014). Journalists, as observers, may have different loci of control. A former Iraq war journalist I interviewed stated that he did not feel his objectives as a war journalist, on the other hand, were accomplished, and that he felt shame for being among people who did not have a choice to be in a war zone.

Lastly, biotypes may differ between war journalists themselves. In the next section, I detail the PTSD symptoms of two war journalists – both managers of a large journalistic organization in Iraq during the height of the insurgency in the Iraq War – who live with vastly different PTSD symptoms. One has greater re-experiencing symptoms, while the other has greater hyperarousal symptoms. Research shows that hyperarousal and re-experiencing have opposite effects on affective habituation, which is a phenomenon that occurs when one is repeatedly exposed to a traumatic stimulus, as measured by functional magnetic resonance imaging. McCurry et al. (2020) looked at affective habituation in combat veterans and found that hyperarousal and re-experiencing symptoms engaged different brain circuits in affective habituation. In the study, hyperarousal symptoms linked to decreased habituation to negative stimuli in areas of frontal and temporal gyri, ventromedial prefrontal cortex and anterior insula. On the other hand, re-experiencing symptoms were linked to increased affective habituation towards negative stimuli in a similar set of brain regions. More research is needed to better understand PTSD biotypes, and talking to war journalists can be an important step forward towards a better understanding of the disorder.

Looking to war journalists’ lived experiences for clues about PTSD biotypes

“Post-traumatic stress disorder isn’t just for soldiers. After years of covering war and tragedy in the Middle East and Southeast Asia for Reuters, it happened to me,” says Dean Yates, a journalist turned workplace mental health expert and public speaker who was diagnosed with PTSD in 2016 (Yates, 2016). He describes his diagnosis as “plunging into an abyss” (Yates, 2016). “It’s hard to stop falling when you don’t understand why you fell in the first place,” he writes in his memoir, Line in the Sand. “After my diagnosis, I felt no emotion, nothing. I had a name for my condition, but no relief” (Yates, 2023, p. 28).

Yates spent 26 years at Reuters, serving as bureau chief in Iraq. He later served as their head of mental health strategy from 2017 to 2020. Yates is a workplace mental health expert, public speaker, podcast host and journalist. He is an outspoken advocate on mental health, press freedom and government accountability. In his book, Line in the Sand, he details his experiences working as Bureau chief in Iraq during the war. “As bureau chief of the largest foreign news organisation in Baghdad, I fear for the safety of my staff more than anything” (Yates, 2023, p. 4), Yates writes. Journalism staff not only died as a consequence of the wartime military operations but were also deliberately targeted by terrorists during his tenure.

All of it is just another day at work for Yates, who says of the journalists’ deaths, which he wrote a story about for Reuters: “Of course, this is news. More emails arrive asking the same question. All my foreign colleagues wait until I write a five-paragraph story before reporting the event. This makes it real, not a nightmare. Doing something I’ve done thousands of times for Reuters across Asia and the Middle East for fifteen years” (Yates, 2023, p. 11). Doing something he loves allows him to say “focused,” as he writes in his book: “More precisely, the work, the responsibility I carry, stops me falling apart” (Yates, 2023, p. 11).

Work became a survival mechanism and way of life for Yates, but the trauma eventually started to catch up with him. At one point, he asked his editors if he could leave the reporting zones to work remotely, which they accepted. Then, he began having intrusive reminders of the various events he’d experienced, forcing him to work less. “Dozens of sights and sounds began intruding into my head,” says Yates of his PTSD symptoms: “intermittent at first, frequent now” (Yates, 2023, p. 24). He also has nightmares: “Even when I’m not being pursued in my sleep, the scenes are always violent. [...] I’m in Iraq or elsewhere in the Middle East. People I know sometimes appear, usually journalists” (Yates, 2023, p. 41). Yates, who also reported from Banda Aceh, Indonesia, during its tragic 9.5-magnitude earthquake, continues: “I also show up in Indonesia or Australia. Then there is lots of water, flooding. I must flee or drown” (Yates, 2023, p. 41).

To deal with his symptoms, Yates says he self medicated codeine, Xanax, and alcohol, though his psychiatrist urged him to rest. “Problem is – on sick leave and therefore free of editing copy, writing emails and taking part in daily news-planning calls – my mind has declared war on itself” (Yates, 2023, p. 41). Yates describes a lack of support from Reuters, such as no “internal peer network of journalists who support distressed colleagues” (Yates, 2023, p. 42). “It’s like I don’t exist,” he writes (Yates, 2023, p. 42).

Even prior to his diagnosis, spending years covering wars and catastrophes around the world, Yates struggled with normal life. His book portrays him as sensitized to a life of chaos and trauma, drawn to “newsroom noise, the pull of a big story,” even on the occasion of the birth of his child with his partner, a fellow journalist (Yates, 2023, p. 30).

“I don’t know anyone who has worked in war zones as a journalist who has not been damaged by it,” says a war journalist to whom I spoke that preceded Yates as Reuters Baghdad Bureau chief. However, his symptoms, as he describes them in his own words, are a bit different. While Yates struggles with reexperiencing symptoms, his anonymous colleague was diagnosed with both PTSD and attention-deficit hyperactivity disorder or ADHD. Luderer et al. (2020) found that ADHD was associated with a higher risk for traumatic events, self-reported PTSD, and higher severity of PTSD patients with alcohol dependence.

The anonymous bureau chief developed substance abuse problems, especially with alcohol, when he began self-medicating to help his brain stop “racing around” and “as a way to focus.”

“In a way, it worked, at first, because I was able to calm my brain and write much better for a while, but because I was using increasing amounts of alcohol, it escalated, and then I was going through withdrawal. I wasn’t drinking to write, I was drinking to not go through withdrawal,” he says. He was able to stop drinking, and realized that he could never drink again because it would put him back into the same cycle.

His PTSD is, in his words, “a little bit complex because I’ve been in personal danger a whole bunch of times. I’ve been in a helicopter that was shot down, I’ve been bombed, I’ve been shot at, but as far as I know, none of those incidents really traumatized me, and I don’t think they caused me any harm. What I think caused me the harm was that…I was in charge of all journalists in the Middle East which included Iraq, Iran, Sudan, Yemen, all these war zones and the constant stress of that, I think, kind of wore me down.”

Dangers journalists face in warzones both relate to their own survival and that of their colleagues. While the unnamed war journalist woke up to the sound of bombs daily, he “just kind of got used to it…you just kind of learn to ignore it.” What he cites as contributing to his PTSD is having to account for others in the warzone. Because terrorist groups in the region targeted foreign journalists, the Bureau had to work with local Iraqi journalists to do reporting.

“All day I'd be just on edge just worrying about them and worrying about their safety, and I only ever really relaxed about 11 pm at night, when I knew all the cars were back safely. And I could kinda relax. And this went on for two years. And so I think part of you know, the damage that was done to me is mainly just this attritional tension and stress, day after day, not really worrying about myself, but worrying about other people. So I think that kind of wore me down,” he says. He and Yates were also affected when two of their colleagues were killed by a United States military helicopter, the details of which were released by Wikileaks. Eventually, the anonymous journalist says, he got to a point where, when he would hear loud bombs in the middle of the night, he would just go back to sleep. He notes that his colleagues did, too. “That was kind of a sign for me that I’d crossed a line into a different kind of thinking, where I can be bombed and just go back to bed and think it’s normal,” he said.

A few aspects of PTSD in journalism are “a bit different and interesting” to the anonymous war journalist:

(1) Most journalists will never want to say they have PTSD and are ashamed about it because they volunteered to go to the war zone. When war journalists go into a war zone, it is by choice, and often they have fought hard to get there “because it’s good for your career,” he says.

(2) “You’re reporting on people who don’t have a choice. They are stuck there.” The people living in warzones did not get to choose to be there or not, says the war journalist. However, journalists can call their editor at any time and quit. “That creates a sense of shame, I think, for journalists when they are affected by PTSD. You feel like, ‘Oh, what right do I have to have become ill from this?’ Other people who have no choices are having to go on with their lives. So that’s one reason journalists deny [that they have PTSD].”

(3) Vicarious trauma can play a large role. The unnamed war journalist said that the most damaging thing for his mental health was seeing his colleagues go into danger. He was not as worried about himself and didn’t himself face any real danger, but found “just seeing the awful consequences for others” damaging.

(4) Related to PTSD is the idea of “moral injury” which the National Center for PTSD (n.d.) describes as “distressing psychological, behavioral, social, and sometimes spiritual aftermath of exposure to such events.” The unnamed war journalist describes it as “a disease of your soul” in which the things you see that are “morally indefensible … can make you a bit unwell. It’s part of the whole story,” he says. He relates a quote from Dispatches by war journalist Michael Herr: “What I didn't know, and it took the war to teach me, was that you're not only responsible for the things that you do. You're also responsible for the things that you see.’” When you witness events in a war zone, he says, even if “you didn't cause it, and it wasn't your fault, you do feel a sense of moral responsibility, and it does harm your mental health.”

Conclusion

Journalism exposes its practitioners and consumers alike – especially on issues such as wars, disasters, and terrorist events – to trauma. Research shows that journalists, especially wartime journalists, including photographers, face PTSD rates similar to combat veterans. Despite this, there is a critical lack of mental health support and resources for wartime journalists and their newsrooms. Research on the neuroscience of journalism and journalist PTSD is limited, despite extensive literature documenting stress and anxiety’s effects on the brain in other contexts. The few available studies paint a picture of traumatic news coverage that leads to adverse behavioral and cognitive outcomes not only in journalists but in consumers of news as well.

Because journalists are natural storytellers, looking at their experiences can also help develop biotypes of PTSD which may differ in terms of symptoms, treatments, and neurobiological substrates. Empirical findings on the neurobiology of PTSD provide a basis for developing better treatments based on such biotypes. One step forward towards a biotype may involve identifying symptom subclusters and looking at the brain regions associated with them to gain insight into PTSD’s underlying neural mechanisms. This knowledge, combined with individual symptom profiles and treatment responses, can inform the development of biotypes for more personalized interventions, helping not only journalists but all people living with PTSD.

Understanding the neurobiology of PTSD in journalists is crucial for developing tailored treatments and support systems, ultimately improving both journalist well-being and the quality of news reporting. War journalists’ experiences underscore the urgent need for comprehensive support systems throughout the news ecosystem. By recognizing journalists' unique challenges and leveraging neuroscience and psychology insights, scientists and clinicians can develop more effective interventions, and the pedagogy of journalism can be improved to support mental health capacity-building in the newsroom. This approach can enhance news reporting quality and contribute to a more resilient society.

Gardener Comments

Andrew Neff (Ph.D. Neuroscience):
I can see value in replacing DSM diagnostic categories & generic symptoms with more specific symptoms & symptom-clusters that were defined with feedback from people with lived experience. New sub-categories of PTSD could inspire the development of new psychotherapies or social interventions, and incorporating new symptoms checklists could be built into machine learning models to better predict individualized treatment response (likely alongside fMRI and other measures). To be convinced of the value of this approach, you need to accurately predict treatment response, but also need to create a system that increases the total number of people treated.

The one thing I don’t think will work out is using neuroimaging to identify biological intervention targets, insofar as 1) fMRI voxels are ambiguous summaries of ~10,000+ neurons, and 2) non-invasive stimulation technology is even less precise and generally unable to target more than one region at a time. Even if we had better interventional tech (like surgically implanted electrodes), of the brain’s infinite properties, I am doubtful that manipulating fMRI voxels is a powerful and specific lever, and historically we haven’t had many (...any) successes with this approach.

Ben Curtis (Photojournalist, Nieman Fellow):

This article provides a broad look at the latest research into the role of traumatization in news reporting and photojournalism and aspects of the neuroscience involved, making the case for treating trauma experienced through such work, especially conflict reporting, as a specific biotype of symptoms and causation.

Despite the high prevalence of PTSD in these individuals there is a relative paucity (with some notable exceptions) of research specific to them, and so this article is very welcome.

George (psychologist):
This is perhaps the most tentative yes I've given to one of these and I think I'm very much on the fence with whether or not I actually like this paper, especially in its current form.

I believe this is a novel area (to my limited knowledge of clinical psych) and I think its focus on PTSD in war journalists is super interesting. However, I am very sceptical of the theory behind this paper and use of neuroscience to justify further categorisation of maladpative behaviour even more than the DSM-5 already does.

I do not know anything about biotypes and I have never heard the term before, but from the current writing I struggle to believe in their worth in this analysis. The neuroscience of PTSD is mentioned briefly and clusters of behaviour displayed by different people with PTSD. But I feel like the analysis only really describes individual experiences rather than something like a factor/ cluster analysis of PTSD symptoms. The analysis which is presented is a bit like a thematic analysis in the presentation and summary of your data. If this is to be published, I think a mixed-media thematic analysis of war journalists' shared experience would be a very cool way to go with the data you've presented. I struggle with the neuroscience part of the PTSD biotypes as without brain scans etc., I really don't know what lived experiences can really tell you about the brain, so a behavioural idiographic approach to this analysis not only makes more sense but seems a bit more theoretically coherent. I would prefer it greatly if you moved theory onto something like process-based therapy which focuses on the behaviour symptoms and lived experiences of people with PTSD, but I'm sure any clincal theory of behaviour will add some structure and merit to your discussion.

Allen Arthur (Director of engagement at Solutions Journalism Network, freelance journalist):
I believe this is an important tangle of issues to pursue for the health of media, democracy, and our communities. While potential PTSD in journalists may seem niche, the reality is that the seemingly endless cycle of reporting primarily on the world's worst and most devastating incidents is profoundly deteriorating the mental health and well-being of journalists. This, in turn, has led them to be less curious and more burnt out, along with producing an unfortunately (though understandably) cynical and hopeless worldview that filters into their reporting. That has negative effects on the fabric of our communities in a variety of ways we are just starting to understand, including perpetuating that cynicism and hopelessness among news consumers. The far-reaching effects of trauma on those tasked with helping others understand the world shouldn't be underestimated and deserve much more study.

Anonymous 1:
The writing is quite good. The text was easy to read and the topic was presented in a straightforward way with organized evidence. I did not feel there was a compelling argument for why war journalists' PTSD subtypes need to be or should be studied rather than studying PTSD subtypes generally. However, I recommend publication because it was intriguing to see these ideas combined this way, and a compelling case is made that the study of war journalists' PTSD *could* provide insight and better outcomes.

Oswell Moyo (Media and Communication Researcher, MPhil Media and Society Studies, Bsc Journalism and Media Studies):

Does the article contain novel ideas that have the potential to advance science?

This article brings into fore novel ideas about important but neglected Biotypes of PTSD in War Journalists. The article comes at a time when the world is getting polarized and rooted in economic wars fronted by the largest economies, namely the United States of America and China. At the time of writing, Russian- Ukraine war, Israel and Palestine Conflict, unending African conflicts in the SAHEL Region, Democratically Republic of Congo and Mozambique conflicts are at the centre stage with the need for research on the neuroscience of journalism and journalist PTSD. This article has potential to advance interdisciplinary discipline of science within the field of journalism studies, psychology, medicine and war studies.

Comments

Contextualization (background to the problem, rationale for the research, problem identification, objectives of the study)

• The title of the article can be expanded into more detail. It is too narrow. The reader must be given a glimpse of the geographical location of the study. I Am sure you are aware that experiences of war journalists in Ukraine can be different from those in DRC hence I suggest a slight change of title. The researcher can rewrite the title like this “Examining Biotypes of PTSD amongst journalists covering war in Global North.”

• Kindly expand on the following point “the management infrastructure for journalists is lacking” What exactly?

• Kindly include scholarly reference to buttress this point, “While in the aftermath of the Oklahoma City Bombing, counseling, leadership, and social support increased for journalists, training has not changed in the nearly 30 years since the tragic event.”

• Is it possible to give a brief background in the introductory section? What motivated you to undertake the study? How many journalists have been affected? Kindly touch on those.

Discussion (contribution of scholarship- the core of study)

The study definitely contributes to scholarship in both the area of journalism, psychology and neuroscience. The discussion can be thickened by bringing in theory and comparing it with similar studies. I am impressed by thick qualitative raw data that is informative.

References

1. Biehn, TL, Contractor, AA, Elhai, JD, Tamburrino, M, Fine, TH, Cohen, G, Shirley, E, Chan, PK, Liberzon, I, Calabrese, JR, & Galea, S. (2016). Latent dimensions of posttraumatic stress disorder and their relations with alcohol use disorder. Social Psychiatry and Psychiatric Epidemiology, 51(3), 421–429. https://doi.org/10.1007/s00127-015-1135-x

2. Brenner LA. (2011). Neuropsychological and neuroimaging findings in traumatic brain injury and post-traumatic stress disorder. Dialogues Clin Neurosci, 13(3):311-23. https://doi.org/10.31887/DCNS.2011.13.3/lbrenner

3. Casteen, EJ, Nielsen, SR, Olson, EA, Frederiks, K, & Rosso, IM. (2022). Reexperiencing and anxious arousal symptoms in relation to volumes of thalamus nuclei in posttraumatic stress spectrum adults. Brain and Behavior, 12, e2639. https://doi.org/10.1002/brb3.2639

4. Columbia University Dart Center for Journalism and Trauma. (n.d.).

https://dartcenter.org/

1. Dan, O, Leshkowitz, M, and Hassin, RR. (2020). On clickbaits and evolution: curiosity from urge and interest. Current Opinion in Behavioral Sciences, 35, 150-156. https://doi.org/10.1016/j.cobeha.2020.09.009

2. Difede, J, Rothbaum, BO, Rizzo, AA, Wyka, K, Spielman, L, Reist, C, Roy, MJ, Jovanovic, T, Norrholm, SD, Cukor, J, Olden, M, Glatt, CE, & Lee, FS. (2022). Enhancing exposure therapy for posttraumatic stress disorder (PTSD): a randomized clinical trial of virtual reality and imaginal exposure with a cognitive enhancer. Translational Psychiatry, 12(1), 299. https://doi.org/10.1038/s41398-022-02066-x

3. Dubberley, S. (2020). Finally recognizing secondary trauma as a primary issue. Columbia Journalism Review. https://www.cjr.org/analysis/finally-recognizing-secondary-trauma-as-a-primary-issue.php

4. Dyar, C., Crosby, S., Newcomb, M. E., Mustanski, B., & Kaysen, D. (2024). Doomscrolling: Prospective associations between daily COVID news exposure, internalizing symptoms, and substance use among sexual and gender minority individuals assigned female at birth. Psychology of Sexual Orientation and Gender Diversity, 11(1), 139–152. https://doi.org/10.1037/sgd0000585

5. Elklit, A., & Shevlin, M. (2007). The structure of PTSD symptoms: a test of alternative models using confirmatory factor analysis. The British journal of clinical psychology, 46(Pt 3), 299–313. https://doi.org/10.1348/014466506X171540

6. Galatzer-Levy, IR and Bryant, RA. (2013). 636,120 Ways to Have Posttraumatic Stress Disorder. Perspect Psychol Sci, 8(6):651-62. https://doi.org/10.1177/1745691613504115 PMID: 26173229.

7. Hill, D. (2018). Timeline of trauma: A case study of newsroom management and support during and after the Oklahoma City bombing. [Doctoral dissertation, University of Oklahoma]. https://shareok.org/handle/11244/54336

8. Feinstein, A, Owen, J, and Blair, N. (2002). A Hazardous Profession: War, Journalists, and Psychopathology. American Journal of Psychiatry, 159(9). https://doi.org/10.1176/appi.ajp.159.9.1570

9. Goldstein, RB, Smith, SM, Chou, SP, Saha, TD, Jung, J, Zhang, H, Pickering, RP, Ruan, WJ, Huang, B, & Grant, BF. (2016). The epidemiology of DSM-5 posttraumatic stress disorder in the United States: Results from the National Epidemiologic Survey on Alcohol and Related Conditions-III. Social Psychiatry and Psychiatric Epidemiology, 51, 1137-1148. https://doi.org/10.1007/s00127-016-1208-5

10. Güzel, A, Tekin, CS, Yamaç, SU. (2024). Exploring the impacts of perceived locus of control on post-traumatic stress disorder among disaster survivors: A systematic review. Journal of Psychiatric and Mental Health Nursing. https://www.doi.org/10.1111/jpm.13030

11. Harnett, NG, Goodman, AM, Knight, DC. (2020). PTSD-related neuroimaging abnormalities in brain function, structure, and biochemistry. Experimental Neurology, 330: 113331. https://doi.org/10.1016/j.expneurol.2020.113331

12. Henigsberg, N, Kalember, P, Petrović, ZK, & Šečić, A. (2019). Neuroimaging research in posttraumatic stress disorder - Focus on amygdala, hippocampus and prefrontal cortex. Progress in Neuro-Psychopharmacology & Biological Psychiatry, 90, 37–42. https://doi.org/10.1016/j.pnpbp.2018.11.003

13. Hopper, JW, Frewen, PA, van der Kolk, BA, & Lanius, RA. (2007). Neural correlates of reexperiencing, avoidance, and dissociation in PTSD: Symptom dimensions and emotion dysregulation in responses to script‐driven trauma imagery. Journal of Traumatic Stress, 20(5), 713-725. https://doi.org/10.1002/jts.20284

14. Jeong, H, Chung, YA, Ma, J, Kim, J, Hong, G, Oh, JK, Kim, M, Ha, E, Hong, H, Yoon, S, & Lyoo, IK. (2019). Diverging roles of the anterior insula in trauma-exposed individuals vulnerable or resilient to posttraumatic stress disorder. Scientific Reports, 9(1), 15539. https://doi.org/10.1038/s41598-019-51727-3

15. Jonisová, E. (2022). The Importance and Consequences of War Photography. Editura Casa Cărții de Știință. https://www.ceeol.com/search/article-detail?id=1081632

16. Jukes, S. (2017). Affective journalism – uncovering the affective dimension of practice in the coverage of traumatic news. [Doctoral thesis, Goldsmiths, University of London]. https://research.gold.ac.uk/id/eprint/20635/

17. Krause ED, Kaltman S, Goodman LA, Dutton MA. (2007). Longitudinal Factor Structure of Posttraumatic Stress Symptoms Related to Intimate Partner Violence. Psychological Assessment, Vol. 19, No. 2, 165-175. https://doi.org/10.1037/1040-3590.19.2.165

18. Kroes, MC, Whalley, MG, Rugg, MD, & Brewin, CR. (2011). Association between flashbacks and structural brain abnormalities in posttraumatic stress disorder. European Psychiatry : The Journal of the Association of European Psychiatrists, 26(8), 525–531. https://doi.org/10.1016/j.eurpsy.2011.03.002

19. Li, X, Zhou, J, Xiang, H, and Cao, J. (2022). Attention Grabbing through Forward Reference: An ERP Study on Clickbait and Top News Stories. International Journal of Human-Computer Interaction, 40(11), 3014-3029. https://doi.org/10.1080/10447318.2022.2158262

20. Luderer, M, Reinhard, I, Richter, A, Kiefer, F, and Weber, T. (2020). ADHD Is Associated with a Higher Risk for Traumatic Events, Self-Reported PTSD, and a Higher Severity of PTSD Symptoms in Alcohol-Dependent Patients. Eur Addict Res, 26 (4-5): 245–253. https://doi.org/10.1159/000508918

21. Luttrell, M, and Hornfischer, JD. (2014). Service: A Navy SEAL at War. Back Bay Books.

22. Marais A and Stuart AD. (2005). The Role of Temperament in the Development of Post-Traumatic Stress Disorder Amongst Journalists. South African Journal of Psychology, 35(1):89-105. https://doi.org/10.1177/008124630503500106

23. Marin, M-F, Morin-Major, J-K, Schramek, TE, Beaupré, A, Perna, A, Juster, R-P, and Lupien, SJ. (2012). There Is No News Like Bad News: Women Are More Remembering and Stress Reactive after Reading Real Negative News than Men. PLoS ONE 7(10): e47189. https://doi.org/10.1371/journal.pone.0047189

24. Michaels TI, Stone E, Singal S, Novakovic V, Barkin RL, Barkin S. (2021) Brain reward circuitry: The overlapping neurobiology of trauma and substance use disorders. World J Psychiatry. 2021 Jun 19;11(6):222-231. https://doi.org/10.5498/wjp.v11.i6.222

25. Milad, MR, Quirk, GJ, Pitman, RK, Orr, SP, Fischl, B, and Rauch, SL. (2007) A role for the human dorsal anterior cingulate cortex in fear expression. Biol Psychiatry. 2007;62:1191–4. https://doi.org/10.1016/j.biopsych.2007.04.032

26. McCurry, KL, Frueh, BC, Chiu, PH, & King-Casas, B. (2020). Opponent Effects of Hyperarousal and Re-experiencing on Affective Habituation in Posttraumatic Stress Disorder. Biological psychiatry. Cognitive neuroscience and neuroimaging, 5(2), 203–212. https://doi.org/10.1016/j.bpsc.2019.09.006

27. Motta, RW. (2023). Secondary Trauma: Silent Suffering and Its Treatment. Springer Cham. https://doi.org/10.1007/978-3-031-44308-4

28. Najavits, L. M., & Walsh, M. (2012). Dissociation, PTSD, and Substance Abuse: An Empirical Study. Journal of Trauma & Dissociation, 13(1), 115–126. https://doi.org/10.1080/15299732.2011.608781

29. National Center for PTSD. (n.d.) Moral Injury. https://www.ptsd.va.gov/professional/treat/cooccurring/moral_injury.asp

30. Newman, E., Simpson, R., & Handschuh, D. (2003). Trauma exposure and post-traumatic stress disorder among photojournalists. Visual Communication Quarterly, 58(1), 4–13. https://doi.org/10.1080/15551390309363497

31. Nutt, D. J., & Malizia, A. L. (2004). Structural and functional brain changes in posttraumatic stress disorder. The Journal of Clinical Psychiatry, 65 Suppl 1, 11–17. https://pubmed.ncbi.nlm.nih.gov/14728092/

32. Radoli, LO. (2023). Courting Trauma: An Unspoken Mental Health Crisis Among Journalists in East Africa. Research Association for Interdisciplinary Studies Conference 2023. https://rais.education/wp-content/uploads/2023/09/0292.pdf

33. Ringwald, W. R., Feltman, S., Clouston, S., Mann, F. D., Ruggero, C., & Kotov, R. (2024, May 24). Mechanisms of stress that maintain posttraumatic stress disorder across 20 years. https://doi.org/10.31234/osf.io/egaxv

34. Ringwald, W. [@WhitneyRingwald]. (2024, May 24). Symptom persistence was sig for all dimensions. The difference between the hyperarousal effect estimate was outside the 95% CI for the effect in other dimensions, but it was a very small difference in absolute effect size. [X Post]. X. https://x.com/WhitneyRingwald/status/1794143843870290347.

35. Rosso, IM, Crowley, DJ, Silveri, MM, Rauch, SL, and Jensen, JE. (2017). Hippocampus Glutamate and N-Acetyl Aspartate Markers of Excitotoxic Neuronal Compromise in Posttraumatic Stress Disorder. Neuropsychopharmacology, 42(8), 1698–1705. https://doi.org/10.1038/npp.2017.32

36. Schrader, C, & Ross, A. (2021). A Review of PTSD and Current Treatment Strategies. Missouri Medicine, 118(6), 546–551.

37. Shabahang, R, Hwang, H, Thomas, EF, Aruguete, MS, McCutcheon, LE, Orosz, G, Khanzadeh, AAH, Chirani, BM, and Zsila, Á. (2024). Doomscrolling evokes existential anxiety and fosters pessimism about human nature? Evidence from Iran and the United States. Computers in Human Behavior Reports, 100438.

38. Shah, SF, Jan, F, Ginossar, T, McGrail, JP, Baber, D, & Ullah, R. (2020). Trauma exposure and posttraumatic stress disorder among regional journalists in Pakistan. Journalism, 23(2), 391-408.

39. Simmons, A. N., Paulus, M. P., Thorp, S. R., Matthews, S. C., Norman, S. B., & Stein, M. B. (2008). Functional activation and neural networks in women with posttraumatic stress disorder related to intimate partner violence. Biological Psychiatry, 64(8), 681–690. https://doi.org/10.1016/j.biopsych.2008.05.027

40. Simpson, T. L., Rise, P., Browne, K. C., Lehavot, K., & Kaysen, D. (2019). Clinical presentations, social functioning, and treatment receipt among individuals with comorbid life-time PTSD and alcohol use disorders versus drug use disorders: Findings from NESARC-III. Addiction, 114(6), 983-993. https://doi.org/10.1111/add.14565

41. Spinhoven, P, Penninx, BW, van Hemert, AM, de Rooij, M, & Elzinga, BM. (2014). Comorbidity of PTSD in anxiety and depressive disorders: prevalence and shared risk factors. Child Abuse & Neglect, 38(8), 1320–1330. https://doi.org/10.1016/j.chiabu.2014.01.017

42. Stark, EA, Parsons, CE, Van Hartevelt, TJ, Charquero-Ballester, M, McManners, H, Ehlers, A, Stein, A, and Kringelbach, ML. (2015). Post-traumatic stress influences the brain even in the absence of symptoms: A systematic, quantitative meta-analysis of neuroimaging studies. Neuroscience and Biobehavioral Reviews, 56, 207–221. https://doi.org/10.1016/j.neubiorev.2015.07.007

43. Tozzi, L, Zhang, X, Pines, A, Olmsted, AM, Zhai, ES, Anene, ET, Chesnut, M, Holt-Gosselin, B, Chang, S, Stetz, PC, Ramirez, CA, Hack, LM, Korgaonkar, MS, Wintermark, M, Gotlib, IH, Ma, J, and Williamset LM. (2024). Personalized brain circuit scores identify clinically distinct biotypes in depression and anxiety. Nature Medicine, 30, 2076–2087. https://doi.org/10.1038/s41591-024-03057-9

44. van der Kolk, B. (2022). The Body Keeps the Score. Penguin Books.

45. van der Kruijs, SJ, Bodde, NM, Carrette, E, Lazeron, RH, Vonck, KE, Boon, PA, Langereis, GR, Cluitmans, PJ, Feijs, LM, Hofman, PA, Backes, WH, Jansen, JF, & Aldenkamp, AP. (2014). Neurophysiological correlates of dissociative symptoms. Journal of Neurology, Neurosurgery, and Psychiatry, 85(2), 174–179. https://doi.org/10.1136/jnnp-2012-302905

46. van Rooij, SJ, Kennis, M, Vink, M, & Geuze, E. (2016). Predicting Treatment Outcome in PTSD: A Longitudinal Functional MRI Study on Trauma-Unrelated Emotional Processing. Neuropsychopharmacology, 41(4), 1156–1165. https://doi.org/10.1038/npp.2015.257

47. Wei F and Nguyen, UT. (2022). An Attention-Based Neural Network Using Human Semantic Knowledge and Its Application to Clickbait Detection. IEEE Open Journal of the Computer Society, 3, 217-232. https://www.doi.org/10.1109/OJCS.2022.3213791

48. Winkeler, K. (2023). Mechanisms of social dysfunction and treatment-related change in Veterans with posttraumatic stress disorder (PTSD). [Master’s thesis, Virginia Tech]. https://vtechworks.lib.vt.edu/items/3165eb30-9eac-451c-8adb-a9d786e935f2

49. Yates, D. (2016). The road to Ward 17: A reporter’s battle with PTSD. Reuters Investigates. https://www.reuters.com/investigates/special-report/ptsd-witness-yates/

50. Yates, D. (2023). Line in the Sand: A Life-Changing Journey Through a Body and a Mind After Trauma. Pan McMillan Australia.

51. Journalism Education and Trauma Research Group - North America. (n.d.) https://www.facebook.com/profile.php?id=61560521622594

Ungrading is the practice of removing traditional grading systems, often based on the belief that grades do not adequately reflect student knowledge or that they undermine deeper learning. The value of ungrading, particularly considering major assignments like exams, is supported mostly by theoretical scholarship and qualitative studies. This article describes a controlled experiment testing whether it is detrimental to give students an unconditional 100% on a test, in terms of future performance on that same test. These experiments took place over three semesters, at two universities, in two different undergraduate classes (neuroscience and psychology), including a total of 409 students. Results were mixed: during the first two semesters, when comparing students who were originally graded to those who were not, there was no difference in performance on that same test 2-3 months later. However, in the final semester, students who were traditionally graded scored 4-5% better on the same test one week to six weeks later. Consequently, this data supports the broader testing of diverse grading practices, including transcripts that do not contain grades.

Introduction

Theoretically, an instructor's recommendation to schools and employers, in the form of transcript grades, could act as carrots and sticks to promote greater learning. One way to evaluate this prospect is to randomly assign some students to complete individual assignments for a grade, while merely encouraging other students to complete the assignment without any tangible consequences.

When homework is voluntary, unsurprisingly, college students rarely do it, yet those who are required to do homework exhibit either little (2-4%) or no performance increases on later exams (Trost & Salehi-Isfahani, 2012; Pozo et al., 2006; Grodner & Rupp, 2013). Similarly mixed results have been observed when grading quizzes. For example, in a psychology class, a study found that during units in which quizzes were graded, students were more likely to earn A’s on a unit exam (Dalfen et al., 2018). However, another study showed that students who were graded on quizzes actually performed 4% worse on a final exam, compared to those that were only given practice quizzes (Wickline & Spektor, 2011). Similarly, another study found that students who were assigned graded quizzes scored worse on the final exam (75%) compared to those who were given no quizzes (78%) and those who were given practice quizzes (82%) (Khanna, 2015).

One reason why grades may not motivate long-term learning is that they incentivize cramming (Fergus, 2022; Rodriguez et al., 2018) and cheating (Vandehey et al., 2007). In contrast, spreading study sessions over time enhances long-term learning, though it may reduce the efficiency of test performance relative to the time spent studying. Similarly, activities like creatively exploring material beyond the course, comparing new concepts with idiosyncratic prior knowledge, or critiquing a professor’s ideas are rarely efficient for exam preparation. It takes less time to accept that "the amygdala is the fear center of the brain" than to ponder what "is the" means or how amygdala research impacts clinical care for anxiety disorders. Without the pressure of grades, students will have more time to think creatively, critically, and personally, in a way that could foster lasting understanding.

Several studies, conducted both in K-12 and college classrooms, have demonstrated that providing comments on assignments or quizzes, without grades, increases student interest in class material (Butler & Nisan, 1986; Butler, 1988), increases the chance that students will voluntarily choose a challenging assignment in the future (Harter, 1978; Deci, 1999), and may even increase performance on later tests (Wickline & Spektor, 2011; Khanna, 2015). In universities that rely on narrative transcripts (which do contain instructor evaluations but do not contain letter grades), students are more likely to feel like they are using their time valuably (Chamberlin et al., 2023), which could perhaps lead to greater retention rates.

The study attempts to build upon prior studies in two ways. First, this study provides enhanced methodological rigor, for example, by assigning students to conditions based on their last name (as opposed to prior quasi-experimental studies that have assigned students to conditions based on course section, as in Wickline and Skeptor [2011] and Khanna [2015]), and testing performance on the same exam (as opposed to comparing across different exams, as in Dalfen et al. [2018]). Moreover, the experiments presented here appear to be the first to test a high-stakes assessment such as an exam (20-25% of one’s course grade).

Methods

Procedure and Primary Outcome Measures

Each semester included either one or two midterm exams (see Table 1 for a description of each study). For each exam, some students were traditionally graded, and others were given 100% based solely on completion (yet all students saw their actual grades until the end of the semester as a way to provide feedback). 

The primary outcome was the percentage score on a surprise re-test of each midterm exam, administered remotely (on Canvas) during a final exam timeslot, 1 week to 3 months after completing the original midterm exam. This measure was selected as an attempt to simulate students' retention after completing a class (and perhaps therefore, preparedness for further classes or jobs). To maintain the deception, all students were told they would be tested on content that was not explicitly included in the midterm exams, albeit the exact instructions differed by semester (see Table 1).

In semester 1, group assignment was based on course section. In semesters 2 and 3, group assignment was based on last name (rather than true randomization, in an attempt to minimize this experiment’s intrusiveness into the student's learning experience). Students could not be blinded to their experimental condition, and I did not blind myself during the analysis regarding which student was in which group.

All midterm exams (including the introductory psychology course) were closed-note, multiple choice, administered in class on the learning management system Canvas, and all covered similar topics in biological psychology and were drawn from the same textbook (forthcoming with Cambridge University Press). Cronbach’s alpha was calculated from semester 3’s exams, including .85 and .84 for the exam 1 and exam 2 midterm, and .76 and .57 for the exam 1 and 2 retention test. Semester 3’s exam questions are included on OSF, the same questions were used in prior experiments with slight modifications to wording and answer choices. 

Pre-semester Attitudes Toward Grades Survey

Surveys were administered at the beginning of each semester as a means for gauging prior knowledge, interests, and soliciting preliminary attitudes toward grades. A summary of relevant questions is presented in Figure 2 and the full surveys are included on OSF. These data were collected from students who took part in the experiments cited in the main text, as well as 29 additional students from the professor’s other introductory psychology class at Emory University who did not take part in the ungrading experiments.

Midterm Exam Survey and Planned Data Collection

During semester 3, surveys were administered at the end of each exam asking students about how they studied, how much they studied, and whether they attributed any mental health challenges to the exam (defined broadly to include stress, depression, anxiety, or other mental health conditions). Although this data was collected, it was only available for a limited analysis due to unforeseen limitations in the course management software (in summary, in the course management platform Canvas, the “New Quizzes” feature only provides summary statistics for each question, it does not allow responses to be broken down by individual student - if this feature becomes available, this data may be uploaded to the OSF portal). Table 3 presents this data in full, in the order that the survey was administered at the end of the exam, without breaking this data down by group.

Additional planned data collection, initially outlined in the OSF pre-registration, did not occur. The collection of these data was omitted due to concerns over the potential undue time burden on participants and my belief that the data described in the last paragraph would more effectively capture student experiences.

Participant Population

There were no a priori criteria for determining sample size. Using an 80% power threshold, semester 1’s study was able to detect a standardized mean difference of 0.62, or 6.8% on the exam. In semester 2, the exam 1 and 2 retention tests could detect a standardized mean difference of 0.61 (8.4%) and 0.60 (5.6%). In semester 3, the exam 1 and 2 retention tests could detect a standardized mean difference of 0.22 (3%) and 0.24 (4.2%).

The sample size was limited by the number of participants in each class, and the total number of experiments (3) was determined based on the goal of improving internal validity (semi-random assignment in experiments 2 and 3), conceptually replicating original findings (experiment 2 and 3), and generalizing to a different class and university (experiment 3). 

In all semesters, students were asked to voluntarily disclose race, ethnicity, and gender (Table 3). At Emory University, due to inconsistencies in the survey format, this data can not reliably be attributed to specific students who were in this experiment (as distinguished from other classes), and therefore data are presented in aggregate.

Rationale for Excluding Subjects

I attempted to exclude all test scores in which I had reason to believe that scores did not reflect a student's knowledge, such as when I suspected that a student cheated or did not try on the exam. These rationales evolved each semester and became formalized by the final experiment. In all semesters, there were no procedures for imputing missing data for students who did not complete their retention test (final exam). More detail in pre-registrations for semester’s two and three at OSF (https://osf.io/6d4c7/).

Note that a very large number of students were excluded (6 of 47, 5 of 47, and 141 of 455 in semesters 1-3 respectively) largely explained by the absence of a concrete incentive for effort during completion-graded exams (note that during in-person mid-term exams at Indiana University, several students brazenly walked out after completing the exam in only a few minutes). However, when conducting the same analyses without excluding subjects, only one test changed from significant to non-significant (semester 3, unit 1 retention test, as will be further described in the results section).

Statistics

Exam scores were evaluated with a Wilcoxon Rank Sum test between all study groups for each midterm and retention test, leading to multiple statistical tests per exam during semesters 2 and 3. Each comparison helps answer a unique question, and therefore, no corrections for multiple comparisons are reported.

Transparency and Openness

I reported how I determined the sample size, all data exclusions, all manipulations, and all measures in the study, and the study is reported following applicable Journal Article Reporting Standards (Applebaum, et al., 2018) with small exceptions (e.g. inclusion of a flowchart). The experiments in semesters 2 and 3 were pre-registered before data collection, and data from all experiments are available at https://osf.io/6d4c7 (along with course syllabi and surveys). Data were analyzed using JMP Pro 17.

IRB Approval and Ethics

The experiments conducted here were deemed exempt from review by Institutional Review Boards at Emory University and Indiana University. Nonetheless, care was taken not to burden students with extra surveys or activities beyond the strict requirements of the class. Participants were not compensated for participation.

Results

In all experiments, some students were given an automatic 100% on a midterm exam, while others were graded traditionally. Student performance was evaluated on the midterm exam, as well as 7-86 days later to test their retention of the material.

The experiments in semesters 1 and 2 were both conducted at Emory University in a 200-level elective course called Clinical Neuroscience. In semester 1, there were no significant differences between traditionally-graded and completion-graded students at all, either for the midterm exam (z = 1.87, p = 0.06, 95% CI: -2.3% to 17.7%) or the retention exam (z = 0.66, p = 0.51, 95% CI: -7.6% to 5.7%) administered 67 days later (Figure 1).

In semester 2, two midterm exams were included, and students crossed over (such that they were traditionally-graded on one exam and completion-graded on the other), and at the end of the semester, both tests were re-administered. Traditionally graded students scored 15.8% higher on the unit 1 mid-term (z = 3.40, p < 0.01, 95% CI: 7.1% to 24.6%) and 14.4% higher unit 2 mid-term (z = 4.00, p < 0.01, 95% CI: 8.2% to 20.7%). However, there were no significant differences on the unit 1 retention test (z = 0.99, p = 0.32, 95% CI: -4.5 to 13.2%) nor on the unit 2 retention test (z = 1.07, p = 0.28, 95% CI: -3.3% to 8.5%), administered 86 and 65 days later.

Semester 3’s experiments in Fall 2023 were conducted at Indiana University in an introductory psychology class (emphasizing research methods, neuroscience, and cognition), which commonly fulfills a general education requirement. This semester, two midterm exams were included alongside two sets of quizzes. Again, students crossed over, such that everyone was traditionally-graded on one exam and one quiz set. These two independent variables were fully crossed, such that some students were traditionally graded on both quizzes and exams in the same unit, while others were mixed. Again, at the end of the semester, both tests were re-administered. This paper’s main text does not break down data by quiz-condition because no significant differences were observed (the full dataset is available on OSF).

During semester 3, on the unit 1 midterm, traditionally-graded students scored 12.5% higher (z = 8.6, p < 0.01, 95% CI: 11.0% to 13.9%), and for the unit 2 midterm, traditionally graded students scored 11.0% higher (z = 6.4, p < 0.01, 95% CI: 9.5% to 12.4%). In unit 1's retention test, traditionally graded students scored 4.8% higher (z = 3.1, p < 0.01, 95% CI: 2.8% to 6.9%), while in unit 2’s retention test, traditionally graded students scored 3.8% higher (z = 3.0, p < 0.01, 95% CI: 2.1% to 5.6%). This semester’s retention exams were administered 48 and 9 days after the midterm. As noted in the methods, the only instance in which excluding subjects affected decisions of statistical significance was during semester 3’s unit 1 retention exam: When all students were included, there no longer was any significant difference observed in exam scores between groups.

Depending on the semester, grading an exam has mixed effects on performance on a retention test administered 9-86 days after a midterm. Top left and bottom two rows: In each graph, group means & standard errors reflect the statistics cited in the main text which excluded participants who were suspected of cheating or not trying. However, for semester 3, all individual data points are included in the background image for clarity since a large number of students were excluded. Top right: Mean difference in exam scores between traditionally-graded and completion-graded students, as a function of retention-test interval time, including 95% confidence interval. * indicates p < 0.05.

Discussion

Using null-hypothesis significance testing with a conventional threshold of p < 0.05, this study observed mixed results. The final semester’s experiments supported the hypothesis that grading exams improves long-term learning (by 4% to 5%), or consequently, that midterm grades reflect some durable knowledge. However, the other two semesters' data do not support the hypothesis.

One explanation for the discrepant results is the sample size. Both significant findings were observed at Indiana University in which data were collected from 278-324 students, in contrast to the experiments at Emory, which included 41-44 students each. Supporting this hypothesis, in two of three experiments at Emory, traditionally-graded students scored 2% to 4% higher on average. This observation illustrates that although studies differed in terms of reaching a conventional threshold for rejecting the null hypothesis, all studies observed similar average differences in retention-grades. If it is true and generalizable that grading exams leads to up to 5% greater retention, it could represent a benchmark for comparison: people who advocate for ungrading would have to justify that the benefits of ungrading exceed this number. 

A second possibility is that the effect of grading depends on the class and student populations. The first two experiments were conducted at a selective private college, in a 200-level elective neuroscience course. Many of these students were contemplating scientific careers, and therefore, perhaps were motivated to study regardless of grades. Moreover, these classes included 25 students, allowing me to frequently interact with individual students and to assign student presentations at the end of the semester. If the experiments at Emory are generalizable to other elective courses, other small classes, or other selective universities, this result could undermine a major rationale underlying a foundational educational practice.

One final explanation is the timing of the retention exam. Due to idiosyncrasies in each class (based on distinct curricular requirements), the retention interval in both Indiana University experiments was shorter than all studies conducted at Emory (9-48 days vs 65-86 days). Supporting this hypothesis, when comparing the rate of performance decay from midterm to the retention test, all five studies observed that student performance on traditionally-graded exams decayed faster than performance on completion-graded exams (which often did not decay at all). Again, if true, this result could undermine a major rationale underlying a foundational educational practice.

These findings align with a meta-analysis showing that undergraduate grades account for only 4-6% of the variance in supervisor-rated job performance across various industries (Roth et al., 1996). If students who perform well on midterm exams do not sustain this performance even before the semester ends, then in some respects, they may be no better prepared for future education or careers. Grades today are compromised by factors such as cheating (Vandehey et al., 2007), cramming (Fergus, 2022; Rodriguez et al., 2018), knowledge of one’s instructor biases (Becker et al., 1968), and test-taking skills (Towns & Robinson, 1993; Rogers & Harley, 1999). This suggests that if colleges were to limit third-party access to student performance metrics, they might not be withholding highly valuable information.

The simplest alternative to a traditional grading system is one in which grades are kept confidential between a student and their instructor. The value of confidentiality in grading may be seen by analogy to other professions: doctors don’t share health information (like a patient’s drug use) because patients would stop sharing medically-relevant behaviors, lawyers don’t share a defendant's entire informal testimony because it could harm the defendant's case, financial advisors don’t share their client's assets because it could lead to exploitation. Perhaps something similar would be valuable in education. If student performance was confidential, at least some students would engage with the material more critically, creatively, and personally. Moreover, perhaps instructors could more accurately assess durable learning, as assignment scores may be less tainted by cheating and memorization. Moreover, perhaps this would force employers and graduate schools to invest more time or money into developing procedures for hiring and admissions that better predict future performance, such as work samples, personality tests, or integrity tests (Schmidt & Hunter, 1998).

Overall, the number of stakeholders to grades is vast and diverse: it is worthwhile to consider the value of grades both in the classroom as well as to third parties. It is likely that a professor’s evaluation of their student’s class performance is useful for some purposes but not for others, and for this reason alone, it may be useful for universities to continue providing grades to third parties. However, the value provided to third parties should be weighed against the costs and benefits of this practice to students in terms of learning and well-being.

Gardener Comments

Greg Baker:
While it's possible to nitpick the methodology or the analysis, it's valuable research overall. It is a bold attempt at answering a question that few are brave enough even to ask.

Richard Sprague:
Excellent example of SoS paper: yes, it's okay that the data collection wasn't completely consistent across the universities. The point is to get the data out there, note the problems, and let others potentially learn something.

Anonymous 1:
The problem with the study is the lack of meta-analysis of the overall results. With small sample sizes, varying levels of 'significance' (p values won't be consistently below 5%) are expected solely due to sampling error (because of low power), and this doesn't mean the results are "mixed". That's the mistake that Schmidt and Hunter wrote about for many decades. To avoid this error is easy, convert the results to Cohen's d for each study, and do a meta-analysis. It's easy with R. Looking at your figure 1, it's easy to see that in 9 or 10 out of the 10 comparisons, traditional grading did better. Thus, we can be pretty confident the traditional grading effect is positive, not zero or negative. Note: since you have multiple outcomes for some samples, you probably have to use multi-level meta-analysis to be entirely rigorous, though I doubt this would affect things much. Read this: Doing_Meta_Analysis_in_R

I don't see where it is written what software the analyses were done with. Ideally, you will add the output from this to the OSF. I looked but didn't see any code output (R markdown, Python notebook, SPSS output, or something else).

DK (PhD in psychology):
There are several problems I see in this research. Most notably, too few universities and participants were used to get reliable insights into the impact of exam scores on knowledge retention. Nevertheless, there are some positive aspects, such as pre-registration, and I am generally against the "file-drawer problem", given that many studies that are conducted are eventually not published, so I would recommend publishing this and making sure it can inform other similar studies. 

Malmesbury:
I really liked this article and the general approach of doing an experiment to check whether the foundational things we take for granted actually work as intended. The author uses a cross-over design where university students are promised an unconditional 100% grade, and measures the effect on exam grades, as well as a later retention test.

The paper is well-written; it explains clearly how the experimental process was altered a bit over time. A few points weren't very clear to me:

• For semester 3, one exclusion criterion is an overall score <50%. However, the plots show a lot of points below that grade. Is it because it's the overall score over multiple exams, and these points are compensated by a better grade on another exam? In the latter case, this could cause problems with Berkson's paradox.

• Given a large number of students have grades near or below the "no effort" exclusion criterion, I'm worried that it could distort the results (for example, an increase in variance could lead to an artifactual increase in mean).

• I haven't checked the numbers myself, but according to the author, the results wouldn't be different without the ad-hoc exclusion criteria. In that case, I think it'd be fine to just show the data without excluding students. Then of course it means deviating from pre-registration for the later studies, but as long as things are straightforward and transparent I don't think it's a problem. Ideally, it should be easy to check whether the exclusion of students is creating statistical artifacts, and if that's the case that's a good enough justification to deviate from pre-registration.

Heidi Zamzow (PhD student, Psychological and Behavioural Science):
This study had some interesting ideas which would be useful to develop further. However, I did not see how the research question, methods, findings, and particularly the conclusion link up. 

I was unclear on whether the issue was more about the practice of assigning grades or rather the practice of sharing grades with third parties, which to me seem quite different. 

The evidence cited didn't always seem to support the argument being made. For instance, showing that providing comments on assignments can have beneficial outcomes does not imply that grades are useless or harmful, but rather that comments are helpful.

I find the author's claim that the study 'provides enhanced methodological rigour' a bit dubious. I found the justification for the exclusion criteria to be weak. If the results are to be useful for real-world application, a more transparent approach would be to report the outcome for the whole sample and then perhaps a subgroup with these exclusions. 

I did not see any mention of how the data from the surveys administered at the beginning of the semester were used; it seems it would be necessary to control for prior knowledge, at a minimum, and report findings in the text with and without covariates. Determining the influence of demographic characteristics such as age, gender, race also would be important, but unfortunately due to the design of the study these could not be tested as covariates.

Though the author did note the differences between the two universities/classes could account for the 'mixed findings', I would say they are TOO different to be compared in the same study at all. 

The author touches on some relevant theory (i.e., self-determination theory) but the article could be improved by developing this further to show how theory informed the hypothesis being tested (which actually wasn't clearly stated).

I was also a bit unclear on the ethics, as it seemed the experiment had the potential to impact students' grades at the end of the year?

On a broader note, to address the issue of graduate schools and employers using GPA to predict future performance, the problem seems to be more one of changing the criteria for acceptance/entrance rather than changing the grading system per se. This is already being done, for instance, in more 'elite' universities in the US, where GPA counts for only a small portion of the admittance criteria -- just one factor amongst many.

Phil Filippak:
The main thing I want to add is that I think preserving grading in universities is common sense, but it's good to see at least some statistics supporting that notion. I suspect that the exams might have gotten too easy in general, since the distributions of grades are too close to the top for most students (but maybe I'm mistaken, and it's the students who have gotten really smart in the last few decades).

Billy Bob:
I found this article to be interesting and engaging. I believe that the article sets up an interesting and relevant premise and that the experiments do generally address the primary question. Given the circumstances under which these data were collected, I think that the experimental design and data collection processes are adequate. My primary issue is in data analysis, particularly Fig. 1 Semester 3. It seems to me that the statistical tests for the Retention data of both Units 1 and 2 are primarily reporting on the size of the datasets as opposed to actual differences between the means or the distributions. With larger sample sizes, common statistical tests that are meant to test the equivalence of means between two datasets tend to fail because the exact means are not equal. If the author has the time and bandwidth, I would recommend that they consider performing permutation tests for all of the Semester 3 data. Not only can these tests potentially deal with this issue of large sample size, but these tests can also be used to make stronger statements between datasets, such as that two distributions are equivalent. 

Joe R:
This looks like a reasonable question to ask and a decent, if possibly flawed, attempt to answer it. I felt confused by the exclusion criteria. In particular, Semester 3 excluded a lot of students, and the explanation listed three criteria, said that only one was excluded, but also said that only one was used. What of the third? Which contributed the most to exclusions? I worry that excluding that many students (particularly ones who "completed assignments toward the end of the semester that essentially constituted studying for the exam") might confound the results. I lacked the time for a deep dive into the data; these and other potential issues should probably be checked more thoroughly before publishing. But I didn't notice any glaring flaws in the writeup itself. I think studying retention specifically, rather than initial test performance, was a good decision and produced interesting results. 

References

1. Appelbaum, M., Cooper, H., Kline, R. B., Mayo-Wilson, E., Nezu, A. M., & Rao, S. M. (2018). Journal article reporting standards for quantitative research in psychology: The APA Publications and Communications Board task force report. American Psychologist, 73(1), 3-25. https:/doi.org/10.1037/amp0000191

2. Barbayannis, G., Bandari, M., Zheng, X., Baquerizo, H., Pecor, K., & Ming, X. (2022). Academic stress and mental well-being in college students: Correlations, affected groups, and COVID-19. Frontiers in Psychology, 13. https:/doi.org/10.3389/fpsyg.2022.886344

3. Becker, H. S., Geer, B., & Hughes, E. C. (1968). Making the Grade: The Academic Side of College Life. Wiley.

4. Blum, S. (2020). Ungrading: Why rating students undermines learning (and what to do instead). West Virginia University Press.

5. Butler, R., & Nisan, M. (1986). Effects of no feedback, task-related comments, and grades on intrinsic motivation and performance. Journal of Educational Psychology, 78(3), 210-216. https://doi.org/10.1037/0022-0663.78.3.210

6. Butler, R. (1988). Enhancing and undermining intrinsic motivation: the effects of task-involving and ego-involving evaluation on interest and performance. British Journal of Educational Psychology, 58(1), 1–14. https://doi.org/10.1111/j.2044-8279.1988.tb00874.x

7. Chamberlin, K., Yasué, M., & Chiang, I. C. A. (2023). The impact of grades on student motivation. Active Learning in Higher Education, 24(2), 109-124. https://doi.org/10.1177/1469787418819728

8. Costanzo, M., & Philpott, J. (1986). Predictors of therapeutic talent in aspiring clinicians: A multivariate analysis. Psychotherapy: Theory, Research, Practice, Training, 23(3), 363-369. https://doi.org/10.1037/h0085624

9. Dalfen, S. R., Fienup, D. M., & Sturmey, P. (2018). Effects of a contingency for quiz accuracy on exam scores. Behavior Analysis in Practice, 11, 106-113. https://doi.org/10.1007/s40617-018-0226-z

10. Deci, E. L., Vallerand, R. J., Pelletier, L. G., & Ryan, R. M. (1991). Motivation and education: The self-determination perspective. Educational Psychologist, 26(3-4), 325-346. https://doi.org/10.1207/s15326985ep2603&4_6

11. Deci, E. L., Koestner, R., & Ryan, R. M. (1999). A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychological bulletin, 125(6), 627. https://doi.org/0.1037/0033-2909.125.6.627

12. Fergus, S. (2022). Are undergraduate students studying smart? Insights into study strategies and habits across a programme of study. Journal of University Teaching and Learning Practice (JUTLP). 19(2), 110-127. https://doi.org/10.53761/1.19.2.8 

13. Greene, E. B. (1931). The retention of information learned in college courses. The Journal of Educational Research, 24(4), 262-273. https://doi.org/10.1080/00220671.1931.10880208

14. Grodner, A., and Rupp, N. G. (2013). The role of homework in student learning outcomes: Evidence from a field experiment. The Journal of Economic Education, 44, 93–109. https://doi.org/10.1080/00220485.2013.770334

15. Gustav, A. (1969). Retention of course material after varying intervals of time. Psychological Reports, 25(3), 727-730. https://doi.org/10.2466/pr0.1969.25.3.727

16. Harter, S. (1978). Pleasure derived from challenge and the effects of receiving grades on children's difficulty level choices. Child Development, 49(3), 788–799. https://doi.org/10.2307/1128249 

17. Healthy Minds Study among Colleges and Universities (2022-2023) [Data set]. Healthy Minds Network, University of Michigan, University of California Los Angeles, Boston University, and Wayne State University. Retrieved March 20, 2024, from https://healthymindsnetwork.org/research/data-for-researchers.

18. Inoue, A. B. (2019). Labor-based grading contracts: Building equity and inclusion in the compassionate writing classroom, 2nd ed. The WAC Clearinghouse; University Press of Colorado. https://doi.org/10.37514/PER-B.2022.1824 

19. Khanna, M. M. (2015). Ungraded pop quizzes: Test-enhanced learning without all the anxiety. Teaching of Psychology, 42(2), 174-178. https://doi.org/10.1177/0098628315573144

20. Kohn, A. (2011). The case against grades. Educational Leadership, 69(3), 28-33.

21. Landrum, R. E., & Gurung, R. A. (2013). The memorability of introductory psychology revisited. Teaching of Psychology, 40(3), 222–227. https://doi.org/10.1177/0098628313487417

22. Mena, J. A., & Stevenson, J. R. (2022). The promise of labor-based grading contracts for the teaching of psychology and neuroscience. Teaching of Psychology. https://doi.org/10.1177/00986283221119783

23. Pozo, S., & Stull, C. (2006). Requiring a math skills unit: Results of a randomized experiment. The American Economic Review, 96, 437–41. https://doi.org/10.1257/000282806777212486

24. Rodriguez, F., Rivas, M. J., Matsumura, L. H., Warschauer, M., & Sato, B. K. (2018). How do students study in STEM courses? Findings from a light-touch intervention and its relevance for underrepresented students. PloS One, 13(7), e0200767. https://doi.org/10.1371/journal.pone.0200767

25. Rogers, W. T., & Harley, D. (1999). An empirical comparison of three-and four-choice items and tests: Susceptibility to testwiseness and internal consistency reliability. Educational Psychology Measurement, 59(2), 234–247. https://doi.org/10.1177/00131649921969820

26. Roth, P. L., BeVier, C. A., Switzer III, F. S., & Schippmann, J. S. (1996). Meta-analyzing the relationship between grades and job performance. Journal of Applied Psychology, 81(5), 548-556. https://doi.org/10.1037/0021-9010.81.5.548

27. Schinske, J., & Tanner, K. (2014). Teaching more by grading less (or differently). CBE—Life Sciences Education, 13(2), 159-166.  https://doi.org/10.1187/cbe.CBE-14-03-0054

28. Schmidt, F. L., & Hunter, J. E. (1998). The validity and utility of selection methods in personnel psychology: Practical and theoretical implications of 85 years of research findings. Psychological Bulletin, 124(2), 262-274. https://doi.org/10.1037/0033-2909.124.2.262

29. Stommel, J. (2023). Do we need the word “ungrading”? Zeal: A Journal for the Liberal Arts, 1(2), 82-87.

30. Towns, M. H., & Robinson, W. R. (1993). Student use of test-wiseness strategies in solving multiple-choice chemistry examinations. Journal of Research in Science Teaching, 30, 709–722. https://doi.org/10.1002/tea.3660300709

31. Trost, S., & Salehi-Isfahani, D. (2012). The effect of homework on exam performance: Experimental results from principles of economics. Southern Economic Journal, 79(1), 224-242. https://doi.org/10.4284/0038-4038-79.1.224

32. Vandehey, M., Diekhoff, G., & LaBeff, E. (2007). College cheating: A twenty-year follow-up and the addition of an honor code. Journal of College Student Development, 48(4), 468-480. https://doi.org/10.1353/csd.2007.0043

33. Wickline, V. B., & Spektor, V. G. (2011). Practice (rather than graded) quizzes, with answers, may increase introductory psychology exam performance. Teaching of Psychology, 38(2), 98-101. https://doi.org/10.1177/0098628311401580

34. Yang, C., Luo, L., Vadillo, M. A., Yu, R., & Shanks, D. R. (2021). Testing (quizzing) boosts classroom learning: A systematic and meta-analytic review. Psychological bulletin, 147(4), 399. https://doi.org/10.1037/bul0000309

How can we create a container for knowledge about AI consciousness? This work introduces a new framework based on physicalism, decoherence, and symmetry. Major arguments include (1) atoms are a more sturdy ontology for grounding consciousness than bits, (2) Wolfram’s ‘branchial space’ is where an object’s true shape lives, (3) electromagnetism is a good proxy for branchial shape, (4) brains and computers have significantly different shapes in branchial space, (5) symmetry considerations will strongly inform a future science of consciousness, and (6) computational efficiency considerations may broadly hedge against “s-risk”.

I. AI consciousness is in the wind

AI is the most rapidly transformative technology ever developed. Consciousness is what gives life meaning. How should we think about the intersection? A large part of humanity’s future may involve figuring this out. But there are three questions that are actually quite pressing, and we may want to push for answers on:

1. What is the default fate of the universe if a technological singularity happens and breakthroughs in consciousness research don’t?

2. What interesting qualia-related capacities does humanity have that synthetic superintelligences might not get by default?

3. What should CEOs of leading AI companies know about consciousness?

The following is a wide-ranging safari through various ideas and what they imply about these questions. Some items are offered as arguments, others as hypotheses or observations; I’ve tried to link to the core sources in each section. In the interests of exposition, I’ve often erred on the side of being opinionated. But first — some preliminaries about why AI consciousness is difficult to study.

Key references:

• Faggella, D. (2023). A Worthy Successor — The Purpose of AGI

II. The social puzzle of AI consciousness

“AI consciousness” is an overloaded term, spanning at least three dimensions:

1. Human-like responsive sophistication: does this AI have a sense of self? Is it able to understand and contextualize intentions, capabilities, hidden state, and vibes, both in itself and others? The better AI is at playing the games we think of as characteristically human (those which are intuitive to us, and those we ascribe status to), the more “consciousness” it has.

2. In-group vs Big Other: is the AI part of our team? Our Team has interiority worth connecting with and caring about (“moral patienthood”). The Other does not.

3. Formal phenomenology: in a narrow and entirely technical and scientifically predictive sense, if you had the equations for qualia (the elements and composition of subjective experience) in-hand, would this AI have qualia?

It’s surprisingly difficult to talk about technical details of AI consciousness for at least three reasons. First, the other non-technical considerations are more accessible and act as conversational attractors. Second, AI consciousness is at the top of an impressive pyramid of perhaps 10-20 semi-independent open problems in metaphysics, and being “right” essentially relies on making the correct assumption for each while having no clean experimental paradigm nor historical tradition to fall back on — in some ways AI consciousness is the final boss of philosophy. 

Third, humans are coalitional creatures — before we judge the truth of a statement, we instinctually evaluate its implications for our alliances. To take an opinionated position on AI consciousness is to risk offending our coalition members, ranging from colleagues, tenure committees, donors, & AI labs, each with their own forms of veto power. This pushes theorists toward big-tent, play-it-safe, defer-to-experts positions.

But in truth, there are no experts in AI consciousness, and it’s exactly in weird positions that may offend intuitive and coalitional sensibilities where the truth is most likely to be found. As Eric Schwitzgebel puts it, “Common sense is incoherent in matters of metaphysics. There’s no way to develop an ambitious, broad-ranging, self-consistent metaphysical system without doing serious violence to common sense somewhere. It’s just impossible. Since common sense is an inconsistent system, you can’t respect it all. Every metaphysician will have to violate it somewhere.”

Key references:

• Hoel, E.P. (2024). Neuroscience is pre-paradigmatic. Consciousness is why

• Schwitzgebel, E. (2024). The Weirdness of the World

• Johnson, M.E. (2022). It From Bit, Revisited

III. Will future synthetic intelligences be conscious?

The question of machine consciousness rests on ‘what kind of thing’ consciousness is. If consciousness is a lossy compression of complex biological processes, similar to “metabolism” or “mood”, asking whether non-biological systems are conscious is a Wittgensteinian type error — i.e. a nonsensical move, similar to asking “what time is it on the sun?” or trying to formalize élan vital. When we run into such category errors, our task is to stop philosophically hitting ourselves; i.e. to debug and dissolve the confusion that led us to apply some category in a context where it’s intrinsically ill-defined. 

On the other hand, if consciousness is a “first-class citizen of reality” that’s definable everywhere, like electric current or gravity, machine consciousness is a serious technical question that merits a serious technical approach. I believe consciousness is such a first-class citizen of reality. Moreover,

I believe synthetic intelligences will be conscious, albeit with a huge caveat.

AIs will be conscious (because most complex heterogenous things probably are): Just as we’re made of the same stuff as rocks, trees, and stars, it’s difficult to formalize a theory of consciousness where most compound physical objects don’t have roughly the same ontological status when it comes to qualia. I.e. I take it as reasonable that humans are less ‘a lone flickering candle of consciousness’ and more a particularly intelligent, cohesive, and agentic “qualiafauna” that has emerged from the endemic qualiaflora. We are special — but for quantitative, not qualitative, reasons. Synthetic intelligences will have qualia, because the universe is conscious by default. We don’t have to worry about the light of consciousness going out — though we can still worry about it going weird.

Caveat: Only real things can be conscious. 

There’s a common theme of attributing consciousness to the highest-status primitive. Theology, Psychology, and Physics have each had their time in the sun as “the most real way of parametrizing reality” and thus the ‘home domain’ of consciousness. Now that software is eating the world, computation is king — and consciousness joins its court. In other words, ‘what kind of thing consciousness is’ is implicitly not just an assertion of metaphysics but of status.

Although software is ascendant, computational theory is still in search of an overarching framework. The story so far is that there are different classes of computation, and problems and computational systems within each class are equivalent in fundamental ways. Quantum computers aside, all modern systems are equivalent to what we call a “Turing machine” — essentially a simple machine that has (1) a tape with symbols on it, (2) an ‘action head’ that can read and write symbols, and (3) rules for what to do when it encounters each symbol. All our software, from Microsoft Excel to GPT4, is built from such Turing-level computations.

Although computational theory in general may prove to intersect with physics (e.g. digital physics, cellular automatons), Turing-level computations in particular seem formally distinct from anything happening in physics. We speak of a computer as “implementing” a computation — but if we dig at this, precisely which Turing-level computations are happening in a physical system is defined by convention and intention, not objective fact. 

• In mathematical terms, there exists no 1-to-1 and onto mapping between the set of Turing-level computations and the set of physical microstates (broadly speaking, this is a version of the Newman Problem).

• In colloquial terms, bits and atoms are differently shaped domains and it doesn’t look like they can be reimagined to cleanly fit together. 

• In metaphysical terms, computations have to be physically implemented in order to be real. However, there are multiple ways to physically realize any (Turing-level) computation, and multiple ways to interpret a physical realization as computation, and no privileged way to choose between them. Hence it can be reasonably argued that computations are never “actually” physically implemented.

To illustrate this point, imagine drawing some boundary in spacetime, e.g. a cube of 1mm^3. Can we list which Turing-level computations are occurring in this volume? My claim is we can’t, because whatever mapping we use will be arbitrary — there is no objective fact of the matter (see Anderson & Piccinini 2024).

And so, because these domains are not equivalent, we’re forced to choose one (or neither) as the natural home of consciousness; it cannot be both. I propose we choose the one that is more real — and while computational theory is beautiful, it’s also a “mere” tautological construct whereas physics is predictive. I.e. electrons are real in more ways than Turing-level bits are, and so if consciousness is real, it must be made out of physics. If it’s possible to describe consciousness as (hyper)computation, it’ll be described in a future computational framework that is isomorphic to physics anyway. Only hardware can be conscious, not software.

This may sound like “mere metaphysics” but whether physical configurations or computational processes are the seat of value is likely the fault-line in some future holy war.*

*I think the best way to adjudicate this is predictiveness and elegance. Maxwell and Faraday assumed that electromagnetism had deep structure and this led to novel predictions, elegant simplifications, and eventually, the iPhone. Assuming qualia has deep structure should lead to something analogous.

Core reference for my argument: 

• Anderson, N., & Piccinini, G. (2024). The Physical Signature of Computation: A Robust Mapping Account

Key references supporting consciousness as computational:

• Safron, A. (2021). IWMT and the physical and computational substrates of consciousness

• Rouleau, N., & Levin, M. (2023). The Multiple Realizability of Sentience in Living Systems and Beyond

• Bach, J. (2024). Cyber Animism

• Butlin, P., & Long, R., et al. (2023). Consciousness in Artificial Intelligence: Insights from the Science of Consciousness

• Levin, M. (2022). Technological Approach to Mind Everywhere: An Experimentally-Grounded Framework for Understanding Diverse Bodies and Minds

• Levin, M. (2024). The Space Of Possible Minds

Key references supporting consciousness as physical, or not Turing-level computational:

• Piccinini, G. (2015). Physical Computation: A Mechanistic Account 

• Johnson, M.E. (2017). Against functionalism

• Aaronson, S. (2014). “Could a Quantum Computer Have Subjective Experience?”

• Johnson, M.E. (2016). Principia Qualia

• Johnson, M.E. (2019). Taking monism seriously

• Kleiner, J. (2024). Consciousness qua Mortal Computation

• Kleiner, J. (2024). The Newman Problem of Consciousness Science

• Hales, C.G., & Ericson, M. (2022). Electromagnetism's Bridge Across the Explanatory Gap: How a Neuroscience/Physics Collaboration Delivers Explanation Into All Theories of Consciousness

• Johnson, M.E. (2022). AIs aren’t conscious; computers are

• McCabe, G. (2004). Universe creation on a computer

• Schiller, D. (2024). Functionalism, integrity, and digital consciousness

• Tononi, G., & Koch, C. (2014). Consciousness: Here, There but Not Everywhere

• Pachniewski, P. (2022). Not artificially conscious

See also:

• Lee, A.Y. (2024). Objective Phenomenology

• Kleiner, J. (2024). Towards a structural turn in consciousness science

• Johnson, M.E. (2022). It From Bit, Revisited

• Ladyman, J. (2023). Structural Realism

• Kanai, R., & Fujisawa, I. (2023). Towards a Universal Theory of Consciousness

• Forthcoming from Dalrymple and from Gorard

IV. We should not rely on AIs or brain emulations to accurately self-report qualia

Many of the most effortlessly intuitive human capacities have proven the most difficult to replicate in artificial systems. Accurately reporting phenomenology may be a particularly thorny problem.

I suggested in Principia Qualia that our capacity to accurately report our phenomenology rests on a laboriously evolved system of correlations that’s very particular to our substrate:

I.e. we can talk “about” our qualia because qualia-language is an efficient compression of our internal logical state, which evolution has beaten into systematic correlation with our actual qualia. This is a contingent correlation, not an intrinsic feature of reality.

If we transfer an organism’s computational signature to a new substrate, the new substrate it’s running on will have some qualia (because ~everything physical has qualia), but porting a computational signature, no matter how well it replicates behavior, will not necessarily replicate the qualia traditionally associated with the signature or behavior. By shifting the physical basis of the system, the link between “physical microstate” and “logical state of the brain’s self-model” breaks and would need to be re-evolved.

Over the long term, most classes of adaptive systems are in fact likely to (re)develop such language games that are coupled to their substrate qualia, for the same reasons our words became systematically coupled to our brain qualia — but the shape of their concepts and dimensions of normative loading may be very different. Language’s structure comes from its usefulness, and if we were to design a reporting language for “functionally important things about nervous systems” vs a reporting language for “functionally important things about computer state,” we’d track very different classes of system & substrate dynamics.

Don’t trust what brain uploads or synthetic intelligences say about their qualia — though by all means be kind to them.[1]

Key references:

• Johnson, M.E. (2016). Principia Qualia

• Kleiner, J., & Hoel, E.P. (2021). Falsification and consciousness

• Hoel, E.P. (2024). AI Keeps Getting Better at Talking About Consciousness

• Johnson, M.E. (2019). Taking monism seriously

Exploring the nature of systematic correlations between reality, brain, and language:

• Safron, A. (2021). IWMT and the physical and computational substrates of consciousness

• Ramstead, M., et al. (2023). On Bayesian mechanics: a physics of and by beliefs

• Long, R. (2023). What to think when a language model tells you it's sentient

• Quine, W.V.O. (1960). Word and Object

V. Technological artifacts will have significantly different qualia signatures & boundaries than evolved systems

In “What’s out there?” I suggested that:

A key lens I would offer is that the functional boundary of our brain and the phenomenological boundary of our mind overlap fairly tightly, and this may not be the case with artificial technological artifacts. And so artifacts created for functional purposes seem likely to result in unstable phenomenological boundaries, unpredictable qualia dynamics and likely no intentional content or phenomenology of agency, but also ‘flashes’ or ‘peaks’ of high order, unlike primordial qualia. We might think of these as producing ‘qualia gravel’ of very uneven size (mostly small, sometimes large, odd contents very unlike human qualia).

Our intuitions have evolved to infer the internal state of other creatures on our tree of life; they’re likely to return nonsense values when applied to technological artifacts, especially those utilizing crystallized intelligence. 

There’s lively discussion around whether Anthropic’s “Claude” chatbot is conscious (and Claude does nothing to deflate this). But if consciousness requires something to be physically instantiated, every ‘chunk’ of consciousness must have extension in space and time. Where is Claude’s consciousness? Is it associated with a portion of the GPU doing inference in some distant datacenter, or a portion of the CPU and I/O bus on your computer, or in the past humans that generated Claude’s training data, or the datacenter which originally trained the model? Is there a singular “Claude consciousness” or are there thousands of small shards of experience in a computer? What we speak of as “Claude” may not have a clean referent in the domain of consciousness, and in general we should expect most technological artifacts to have non-intuitive projections into consciousness.

This observation, although important, is also somewhat shallow — of course computers will exhibit different consciousness patterns than brains. To go deeper, we need to look at the details of our substrate.

Key references:

• Johnson, M.E. (2019). What’s out there?

• Wollberg, E. (2024). Qualia Takeoff in The Age of Spiritual Machines

• Johnson, M.E. (2022). Qualia Astronomy & Proof of Qualia

VI. Branchial space is where true shape lives

A strange but absolutely central concept in modern physics is that quantum particles naturally exist in an ambivalent state — a “multiple positions true at the same time” superposition. Decoherence is when interaction with the environment forces a particle to commit to a specific position, and this (wave-like) superposition collapses into one of its (particle-like) component values. The Copenhagen interpretation suggested decoherence is random, but over the past ~2 decades Hugh Everett’s many-worlds interpretation (MWI) has been gaining favor. MWI frames decoherence as a sort of “branching”: instead of the universe randomly choosing which value to collapse into, all values still exist but in different branches of reality.

E.g. let’s say we’re observing a cesium-137 atom. This atom is unstable and can spontaneously decay (a form of decoherence) into either barium-137 or barium-137m. It decays into barium-137m. The many-worlds interpretation (MWI) claims that it did both — i.e. there’s a branch of reality where it decayed into barium-137, and another branch where it decayed into barium-137m, and we as observers just happen to be in the latter branch. MWI may sound like a very odd theory, but it collects and simplifies an enormous amount of observations and confusions about what happens at the quantum level.

Stephen Wolfram suggests understanding the MWI in terms of branchial graphs where each interaction which can cause decoherence creates a new branch. This sort of graph gets Vast very quickly, but in principle each branch is perfectly describable. Wolfram’s new physics proposes the universe can be thought of as the aggregate of all such graphs, which he calls “branchial space”:

Tracing through the connections of a branchial graph gives rise to the notion of a kind of space in which states on different branches of history are laid out. In particular, branchial space is defined by the pattern of entanglements between different branches of history in possible branchial graphs.

Different branches of reality split off and can diverge — but they can also interact, and merge (recohere). Many “weird quantum effects” such as the double-slit experiment can be formally reframed as arising from interactions between branches (this is the core thesis behind quantum computing), and time itself may be understood as emergent from branchial structure.

The “branchial view” suggests that different types of objects are different types of knots in branchial space, as defined by (1) how their particles are connected, (2) what patterns of coherence and decoherence this allows, and (3) the branches that form & interact due to this decoherence. A wooden chair, for instance, is a relatively static ‘knot’ (though there’s always froth at the quantum level); a squirrel is a finely complex process of knotting; the sun is a 4.5 billion-years-long nuclear Gordian weave.

The “branchial view” matters for consciousness because decoherence may be necessary for, if not identical to, consciousness.

Decoherence is often seen as an impediment to consciousness; e.g. Max Tegmark argues that predictive systems must minimize it as a source of uncertainty. On the other hand, Scott Aaronson argues decoherence is instead a necessary condition for consciousness:

[Y]es, consciousness is a property of any suitably-organized chunk of matter. But, in addition to performing complex computations, or passing the Turing Test, or other information-theoretic conditions that I don’t know (and don’t claim to know), there’s at least one crucial further thing that a chunk of matter has to do before we should consider it conscious. Namely, it has to participate fully in the Arrow of Time. More specifically, it has to produce irreversible decoherence as an intrinsic part of its operation. It has to be continually taking microscopic fluctuations, and irreversibly amplifying them into stable, copyable, macroscopic classical records.

[...]

So, why might one conjecture that decoherence, and participation in the arrow of time, were necessary conditions for consciousness? I suppose I could offer some argument about our subjective experience of the passage of time being a crucial component of our consciousness, and the passage of time being bound up with the Second Law. Truthfully, though, I don’t have any a-priori argument that I find convincing. All I can do is show you how many apparent paradoxes get resolved if you make this one speculative leap. 

[…]

There’s this old chestnut, what if each person on earth simulated one neuron of your brain, by passing pieces of paper around.  It took them several years just to simulate a single second of your thought processes.  Would that bring your subjectivity into being?  Would you accept it as a replacement for your current body?  If so, then what if your brain were simulated, not neuron-by-neuron, but by a gigantic lookup table?  That is, what if there were a huge database, much larger than the observable universe (but let’s not worry about that), that hardwired what your brain’s response was to every sequence of stimuli that your sense-organs could possibly receive.  Would that bring about your consciousness?  Let’s keep pushing: if it would, would it make a difference if anyone actually consulted the lookup table?  Why can’t it bring about your consciousness just by sitting there doing nothing?

Aaronson goes on to list some paradoxes and puzzling edge-cases that resolve if ‘full participation in the Arrow of Time’ is a necessary condition for a system being consciousness: e.g., whether brains which have undergone Fully Homomorphic Encryption (FHE) could still be conscious (no – Aaronson suggests that nothing with a clean digital abstraction layer could be) or whether a fully-reversible quantum computer could exhibit consciousness (no – Aaronson argues that no fully-reversible process could be). (Paragraph from Johnson 2016)

I agree with Aaronson and propose going further (“MBP Hypothesis” in Appendix E, Johnson 2016; see also significant work in Chalmers & McQueen 2021). Briefly, my updated 2024 position is “an experience is an object in branchial space, and the magnitude of its consciousness is the size of its branchial graph.” Pick a formal specification of branchial space, add a boundary condition for delineating subgraphs (e.g. ontological, topological, amalgamative-majority, dispersive/statistical, compositional), and we have a proper theory.

Slightly rephrased: the thesis of “Qualia Formalism” (QF) or “Information Geometry of Mind” (IGM) is that a proper formalism for consciousness exists:

An information geometry of mind (IGM) is a mathematical representation of an experience whose internal relationships between components mirror the internal relationships between the elements of the subjective experience it represents. A correct information geometry of mind is an exact representation of an experience. More formally, an IGM is a mathematical object such that there exists an isomorphism (a one-to-one and onto mapping) between this mathematical object and the experience it represents. Any question about the contents, texture, or affective state of an experience can be answered in terms of this geometry. (Johnson 2023)

If such a formalism exists, a core question is how to derive it. I’m speculating here that perhaps (a) what Wolfram calls “branchial space” is the native domain of this formalism, (b) an IGM/QF will be isomorphic to a bounded branchial graph, and (c) solving the binding/boundary problem is identical with determining the signature for where one bounded graph ends and another begins.

However, to recap — this section’s thesis is that branchial space is where true shape lives. This has three elements:

1. Decoherence is a crucial part of reality and can be understood in terms of branchial space;

2. Decoherence may be necessary for consciousness;

3. If we wish to understand the “true shape” of something in a way that may reflect its qualia, we should try to infer its shape in branchial space.

Key references:

• Aaronson, S. (2014). “Could a Quantum Computer Have Subjective Experience?”

• Sandberg, A., et al. (2017). That is not dead which can eternal lie: the aestivation hypothesis for resolving Fermi’s paradox

• Tegmark, M. (1999). The importance of quantum decoherence in brain processes

• Tegmark, M. (2014). Consciousness as a State of Matter

• Johnson, M.E. (2016). Principia Qualia

• Johnson, M.E. (2023). Qualia Formalism and a Symmetry Theory of Valence

• Wikipedia, accessed 29 April 2024. Quantum decoherence

• Wolfram, S., et al. (2020). The Wolfram Physics Project; example of branchial expansion

• Barrett, A. (2014). An integration of integrated information theory with fundamental physics

• Albantakis, L., et al. (2023). Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms

• Chalmers, D.J., & McQueen, K.J. (2021). Consciousness and the Collapse of the Wave Function

VII. Brains and computers have vastly different shapes in branchial space

A defining feature of brains is self-organized criticality (SOC). “Critical” systems are organized in such a way that a small push can change their attractor; “self-organized” means assembled by intrinsic and stochastic factors, not top-down design. A property of SOC systems is they stay SOC systems over time — the system evolves criticality itself as one of its attractors. In other words, the brain is highly sensitive to even small inputs, and will eventually regenerate this sensitivity almost regardless of what the input is.

Brains at the edge of criticality can be thought of as ‘perching’ on their symmetries/ambivalences/sensory superpositions: multiple interpretations for inputs can be considered, and as they get ruled out the system can follow the energy gradient downwards (“symmetry breaking”). Later as the situation is metabolized the system recovers its perch, ready for the next input.[2] Given that these ‘perch positions’ are local optimums of systemic potential energy and *sensory* superpositions, and given evolution’s tendency towards scale-free motifs, I suspect these perches might also be local statistical optimums for quantum superpositions (and thus ambient decoherence).

Relatedly, self-organized criticality makes brains very good at amplifying decoherence. Tiny decoherence events can make neurons near their thresholds activate, which can snowball and influence the path of the whole system. Such decoherence events are bidirectionally coupled to the brain’s information processing: local quantum noise influences neural activity, and neural activity influences local quantum noise.

My conclusion is that the brain is an extremely dynamic branchial knotting process, with each moment of experience as a huge, scale-free knot in branchial space.[3]

Meanwhile, modern computers minimize the amplification of decoherence. A close signature of decoherence is heat, which computers make a lot of — but work very hard to buffer the system against in order to maintain what Aaronson calls a “clean digital abstraction layer”. Every parameter of a circuit is tuned to prevent heat and quantum fluctuations from touching and changing its intended computation (Kleiner & Ludwig 2023).

This makes computers rather odd objects in branchial space. They have noise/decoherence in proportion to their temperature (this is called “Johnson noise” after John B. Johnson; no relation) and occasionally they’ll sample it for random numbers, but mostly computers are built to be deterministic — the macroscopic behavior of circuits implementing a typical computation end up exactly the same in almost all branches. This makes the computation very differently represented in branchial space compared to the Johnson noise of the circuits implementing it — and compared to how a brain’s computations are represented in branchial space.

Clarifying what this means for computer consciousness essentially involves three questions:

1. What are the major classes of objects in branchial space?[4][5][6]

2. What are the branchial-relevant differences between brains and computers?

3. How do we construct a “qualia-weighted” branchial space such that the size of the subgraph corresponds with the amount of consciousness?[7]

These questions are challenging to address properly given current theory. As an initial thesis, I’ll suggest that a reasonable shortcut to profiling an object’s branchial shape is to evaluate it for criticality and electromagnetic flows. I discussed the former above; the next section discusses the latter.

Key references:

• Kleiner, J., & Ludwig, T. (2023). If consciousness is dynamically relevant, artificial intelligence isn't conscious

• Wikipedia, accessed 26 April 2024. Johnson-Nyquist noise

• Hoel, E.P., et al. (2013). Quantifying causal emergence shows that macro can beat micro; primer

• Johnson, M.E. (2019). Neural Annealing: Toward a Neural Theory of Everything

• Johnson, M.E. (2024). Minds as Hyperspheres

• Zurek, W.H. (2009). Quantum Darwinism

• Tegmark, M. (2014). Consciousness as a State of Matter

• Olah, C. (2024). Distributed Representations: Composition & Superposition

VIII. A cautious focus on electromagnetism

A growing trend in contemporary consciousness research is to focus on the electromagnetic field. Adam Barrett describes the basic rationale for ‘EM field primacy in consciousness research’ in An integration of integrated information theory with fundamental physics:

1. Quantum fields are fundamental entities in physics, and all particles can be understood as

ripples in their specific type of field.

2. Since they’re so fundamental, it seems plausible that these fields could be carriers for consciousness.

3. The gravity, strong, and weak nuclear fields probably can’t support the complexity required for human consciousness: gravity’s field is too simple to support structure since it only attracts, and disturbances in the other two don’t propagate much further than the width of an atom’s nucleus.

4. However, we know the brain’s neurons generate extensive, complex, and rapidly changing patterns in the electromagnetic field.

5. Thus, we should look to the electromagnetic field as a possible ‘carrier’ to consciousness (Summary of Barrett 2014, quoted from Johnson 2016)

W. H. Zurek makes a complimentary point in his famous essay Quantum Darwinism:

Suitability of the environment as a channel [for information propagation] depends on whether it provides a direct and easy access to the records of the system. This depends on the structure and evolution of [the environment] E. Photons are ideal in this respect: They interact with various systems, but, in effect, do not interact with each other. This is why light delivers most of our information. Moreover, photons emitted by the usual sources (e.g., sun) are far from equilibrium with our surroundings. Thus, even when decoherence is dominated by other environments (e.g., air) photons are much better in passing on information they acquire while “monitoring the system of interest”: Air molecules scatter from one another, so that whatever record they may have gathered becomes effectively undecipherable.

These are substantial arguments and strongly suggest that electromagnetism plays a dominant role in binding human-scale experience. However, I would suggest three significant caveats:

1. A dominant (statistical) role is not necessarily an exclusive (ontological) role;

2. A force being necessary for binding does not imply that it’s sufficient for describing or instantiating all experiential elements / records;

3. Human-scale experiences happen on a certain energy scale, and dynamics that hold at this energy scale may not hold at other scales. I.e., it seems plausible that other forces play a more significant role in binding at quantum scales, or in cosmological megastructures (e.g. black holes).

I take the branchial view of consciousness as more philosophically/ontologically precise than the electromagnetic view. However, the EM view is generally a useful compression of the branchial view, since most of the branchial dynamics associated with variance in human-scale experience are mediated by the electromagnetic field.

This shortcut is likely similarly relevant for computer consciousness. So — what do we know about brain EMF vs computer EMF?

Brain EMF profile: The brain uses voltage potentials extensively across organs, various classes of chemical gradients, cellular membranes (~−50 mV, inside negative) and axons (~-70mV), as well as within cells to assemble structures (DNA, proteins, etc). Brains also use chemical reactions (“metabolism”) extensively during normal operation and these reactions are primarily mediated through valence shells, an electromagnetic phenomenon.

As a speculative sweeping characterization, I suspect the overall configuration resembles a highly layered configuration of nested & partially-overlapping electromagnetic shells (cf. unreleased summer 2021 talk on the binding/boundary problem). The overall electrical polarity of the shells may be a simple proxy for both decoherence and consciousness — e.g. the voltage potential between brainstem and brain surface should increase during psychedelics, meditation, and arousal, and decrease with age (Michael Johnson & Max Hodak in conversation, 2024). 

Computer EMF profile: Computer substrates are mostly non-reactive semiconductors which prioritize conditional ease of electron flow and adjustment of phase; voltage flows through gates based on simple logic, and the configuration of these gates changes each clock cycle (typically in the GHz range). Voltage in computer logic gates approximates square waves (whereas bioelectric voltage is sinusoidal) and the sharper these transition are, the greater ‘splash’ in the EM field. The magnitude of this ‘splash’ may or may not track branchial expansion / strength of consciousness. Computers can be turned off (thus adding another category of object); brains cannot be.

Computer chips are designed to maintain neutral overall polarity, but the voltage channels (which can be thought of as directed graphs for current, essentially functioning as ‘waveguides’ for EM waves) that instantiate a computational state are relatively high voltage (~600mV+) compared to biological voltages (although getting smaller each chip generation).[8] I’ll propose characterizing the electromagnetic profile of a modern processor as a strobing electromagnetic lattice.

Ultimately, these sorts of characterizations are data-starved, and one of the big ‘intuition unlocks’ for consciousness research in general — and the branchial view in particular — will be the capacity to visualize the EM field in realtime & high resolution. What stories might we tell if we had better tools?

Key references:

• Zurek, W.H. (2009). Quantum Darwinism

• Barrett, A. (2014). An integration of integrated information theory with fundamental physics

• Gomez-Emilsson, A., & Percy, C. (2023). Don’t forget the boundary problem! How EM field topology can address the overlooked cousin to the binding problem for consciousness

• Hales, C.G., & Ericson, M. (2022). Electromagnetism's Bridge Across the Explanatory Gap: How a Neuroscience/Physics Collaboration Delivers Explanation Into All Theories of Consciousness

(See also An introduction to Susan Pockett: An electromagnetic theory of consciousness)

IX. The Symmetry Theory of Valence is a Rosetta Stone

In 2016 I offered the Symmetry Theory of Valence: the symmetry of an information geometry of mind corresponds with how pleasant it is to be that experience. I.e. the valence of an experience is due entirely to its structure, and symmetry in this structure intrinsically feels good (Johnson 2016, Johnson 2021, Johnson 2023).

Rephrased: if the proper goal of consciousness research is to construct a mathematical formalism for an experience (essentially, a high-dimensional shape that exactly mirrors the structure of an experience), STV predicts that the symmetry of this shape corresponds with the pleasantness of the experience. “Symmetry” is a technical term, but Frank Wilczek suggests “change without change” as a useful shorthand: for each symmetry something has, there exists a mathematical operation (e.g. a flip or rotation) that leaves it unchanged.

The fundamental question in phenomenology research is where to start — what are the natural kinds of qualia? STV has three answers to this:

1. Valence is a natural kind within experience;

2. Symmetry is a natural kind within any formalism which can represent experiential structure;

3. These are the same natural kind, just in different domains.

Explicitly, the Symmetry Theory of Valence is a theory of valence and is testable as such; I offer some routes in my 2023 summary paper. Tacitly, STV is also a collection of implications about “what kind of thing” consciousness research is. Just like the first line in the Rosetta Stone offered a wide range of structural constraints on Ancient Egyptian, if we can say ‘one true thing’ about qualia this may inform a great deal about potential approaches.

Perhaps the most significant tacit implication is importing physics’ symmetry aesthetic into consciousness research. Nobel Laureate P. W. Anderson famously remarked “It is only slightly overstating the case to say that physics is the study of symmetry”; Nobel Laureate Frank Wilczek likewise describes symmetry as a core search criterion:

[T]he idea that there is symmetry at the root of Nature has come to dominate our understanding of physical reality. We are led to a small number of special structures from purely mathematical considerations—considerations of symmetry—and put them forward to Nature, as candidate elements for her design. [...] In modern physics we have taken this lesson to heart. We have learned to work from symmetry toward truth. Instead of using experiments to infer equations, and then finding (to our delight and astonishment) that the equations have a lot of symmetry, we propose equations with enormous symmetry and then check to see whether Nature uses them. It has been an amazingly successful strategy. (Wilczek 2016)

If STV is true, Wilczek’s observation about the centrality of symmetry likely also applies to consciousness. Initial results seem promising; a former colleague (A.G.E.) once suggested that “Nothing in psychedelics makes sense except in light of the Symmetry Theory of Valence” — and I tend to agree.

Concretely, importing physics’ symmetry aesthetic predicts there will be phenomenological conservation laws (similar to conservation of energy, charge, momentum, etc), and suggests a phenomenological analogue to Noether’s theorem. The larger point here is that consciousness research need not start from scratch, and just as two points define a line, and three lines define a plane, it may not take too many dualities such as STV to uniquely identify the mapping between the domains of consciousness & physics. Optimistically, “solving consciousness” could take years, not centuries.

STV and the “branchial view” described in Sections VI-VIII are separate hypotheses, but looking at them side-by-side elicits certain questions. How should we think about “branchial symmetry” and evaluating valence of knots-as-moments-of-experience in branchial space? Should we look for metrics of graph uniformity, recoherence/unity, non-interference? Physics has perhaps half a dozen formally equivalent interpretations of reality; I expect STV will too.[9]

Key references:

• Johnson, M.E. (2016). Principia Qualia

• Johnson, M.E. (2023). Qualia Formalism and a Symmetry Theory of Valence

• Johnson, M.E. (2021). A Primer on the Symmetry Theory of Valence

• Wolfram, S. (2021). The Concept of the Ruliad

• Johnson, M.E. (2019). Taking monism seriously

• Safron, A., et al. (2023). Making and breaking symmetries in mind and life

• Wilczek, F. (2016). A Beautiful Question: Finding Nature’s Deep Design

• Gross, D.J. (1996). The role of symmetry in fundamental physics

• Brading, K., & Castelanni, E. (Ed). (2003). Symmetries in physics: philosophical reflections

(Thanks also to David Pearce for his steadfast belief in valence (or “hedonic tone”) as a natural kind.)

X. Will it be pleasant to be a future superintelligence?

The human experience ranges from intense ecstasy to horrible suffering, with a wide middle. However, we also have what I would call the “Buddhist endowment” — that if and when we remove expectation/prediction/tension from our nervous system, it generally self-organizes into a high-harmony state as the default.

If we continue building smarter computers out of Von Neumann architecture GPUs, or eventually switch to e.g. quantum, asynchronous, or thermodynamic processors, what valence of qualia is this likely to produce? Will these systems have any similar endowments? Are there considerations around what design choices we should avoid?

I’ll offer five hypothesis about AI/computer valence:

Hypothesis #1: Architecture drives valence in top-down systems, data drives valence in bottom-up systems. If the valence of an experience derives from its structure, we should evaluate where the structure of systems comes from. The structure of self-organized systems (like brains) varies primarily due to the data flowing through them — evolutionarily, historically, and presently. The structure of top-down systems (like modern computers) varies primarily due to fixed architectural commitments.

Hypothesis #2: Computer valence is dominated by chip layout, waveguide interference, waveform shape, and Johnson noise. I argue above that decoherence is necessary and perhaps sufficient for consciousness. Computers create lots of decoherence along their electrified circuits; however, this decoherence is treated as “waste heat” (essentially the Johnson noise of the circuit) and is largely isolated from influencing the computation. I suspect a future science of machine qualia will formalize how a circuit’s voltage pattern, physical microstructure, computational architecture, nominal computation, and Johnson noise interact to project into branchial space. 

Whether there is a systematic connection between high-level computational constructs (e.g. virtual characters in a video game) and qualia is extremely muddy at this point and likely highly dependent on hardware and software implementation; potentially true if the system is designed to make it true, but likely not the case by default. I.e., neither brain emulations nor virtual characters in video games will be “conscious” in any real sense by default, although we could design a hardware+software environment where they would be.

Hypothesis #3: Valence shells influence phenomenological valence: variance in the chemical structures that comprise a system’s substrate contributes to stochastic variance in its phenomenal binding motifs. These factors will influence phenomenological structure, and thus valence.[10]

Hypothesis #4: Machine consciousness has a quadrimodal possibility distribution: instead of biology’s continuous and dynamic range of valence, I expect the substrates of synthetic intelligences to reliably lead to experiences which have either (a) extreme negative valence, (b) extreme positive valence, (c) extremely neutral valence, or (d) swings between extremely positive and negative valence. Instead of a responsive & continuous dynamism as in biology, whatever physical substrate & computational architecture the computer chip’s designers originally chose will likely lock in one of these four valence scenarios regardless of what is being computed (see hypothesis #1).

Hypothesis #5: Trend toward positive valence in optimal systems: reducing energy loss is a primary optimization target for microprocessor design, and energy loss is minimized when all forms of dissonance are minimized. In theory this should lead to a trend away from the production of negative valence as computers get more energy efficient. To paraphrase Carl Shulman, pain and pleasure may not be equally energy-efficient — and this bodes well for future computers.

Key references:

• Johnson, M.E. (2016). Principia Qualia

• Johnson, M.E. (2023). Qualia Formalism and a Symmetry Theory of Valence

• Extropic (2024). Ushering in the Thermodynamic Future; thread on thermodynamic processors

On energy loss & architectural imperatives:

• Friston, K. (2010). The free-energy principle: a unified brain theory?

• Ramstead, M., et al. (2023). On Bayesian mechanics: a physics of and by beliefs

• Safron, A. (2020). An Integrated World Modeling Theory (IWMT) of Consciousness

XI. Will artificial superintelligences be interested in qualia?

We care about our qualia, and sometimes the qualia of those around us. Why?

I’d offer three reasons:

1. As noted above, humans developed increasingly sophisticated and coherent terminology about “consciousness” because it was a great compression schema for evaluating, communicating, and coordinating internal state;

2. Caring about the qualia of ourselves and others has substantial instrumental value — happy creatures produce better results across many tasks. This led us to consider qualia as a domain of optimization;

3. We are porous harmonic computers that catch feels from those around us, incentivizing us to care about nearby harmonic computers, even if it’s not in our narrow-sense self-interest.

These factors, iterated across hundreds of thousands of years and countless social, political, and intellectual contexts, produced a “language game” which approximates a universal theory of interiority.

Humanity is now in the process of determining whether this “as-if” universal theory of internal state can be properly systematized into an actual universal theory of internal state — whether qualia is the sort of thing that can have its own alchemy-to-chemistry transition, or its Maxwell’s Laws moment, where a loose and haphazard language-game can find deep fit with the structure of reality and “snap into place”. I’m optimistic the answer is yes, and not only because it’s a Pascalian wager.

Humans care about qualia. Will Artificial Superintelligences (ASIs)? 

I’d suggest framing this in terms of sensitivity. Humans are sensitive to qualia — we have a map of what’s happening in the qualia domain, and we treat it as a domain of optimization. We are Qualia Sensitive Processes (QSPs). Most of the universe is not sensitive to qualia — it is made up of Qualia Insensitive Processes (QIPs), which do not treat consciousness as either an explicit or implicit domain of optimization.

This distinction suggests reframing our question: is the modal synthetic superintelligence a QSP? Similarly — is QSP status a convergent capacity that all sufficiently advanced civilizations develop (like calculus), or is it a rare find, and something that could be lost during a discontinuous break in our lineage? What parts of the qualia domain do QSPs tend to optimize for — is it usually valence or are there other common axes to be sensitive to? Can we determine a typology of cosmological (physical) megastructures which optimize for each common qualia optimization target?

As a starting assumption, I suggest we can view the “qualia language game” as a particularly useful compression of reality. If this compression is useful or incentivized for future superintelligences, it will get used, and the normative loading we’ve baked into it will persist. If not, it won’t. There are no laws of the universe forcing ASIs to intrinsically care about consciousness and valence, but they won’t intrinsically disregard it either. 

Qualia has a variable causal density, which informs the usefulness of modeling it

In Taking Monism Seriously, I suggested:

There may be many possible chemical foundations for life (carbon, silicon, etc), but there will tend to be path-dependent lock-in, as biological systems essentially terraform their environments to better support biology. Terran biology can be thought of as a coordination regime that muscled out real or hypothetical competition to become the dominant paradigm on Earth. Perhaps we may find analogues here in past, present, and future phenomenology.

The more of a certain kind of structure present in the environment, the easier it is to model, remix, use as infrastructure, and in general invest in similar structure — e.g. the more DNA-based organisms in an ecosystem, the easier it is for DNA-based organisms to thrive there. The animal kingdom seems to be providing a similar service for consciousness, essentially “qualiaforming” reality such that bound (macroscopically aggregated) phenomenological experience is increasingly useful as both capacity and model. Rephrased: minds are points of high causal density in qualiaspace; the more minds present in an ecosystem, the more valuable it is to understand the laws of qualia.

I think causal density is a particularly useful lens by which to analyze both systems and ontologies. Erik P. Hoel has written about causal emergence and how certain levels of abstraction are more “causally dense” or efficacious to model than others; I suspect we can take Hoel’s hypothesis and evaluate causal density between dual-aspect monism’s different projections of reality. I.e. the equations of qualia may not be particularly useful for modeling stellar fusion, but they seem relatively more useful for predicting biological behavior since the causal locus of many decisions is concentrated within clean phenomenological boundaries. 

The equations of qualia don’t seem particularly useful for understanding the functional properties of modern computers. Whether studying phenomenology will stay useful for humans, and become useful for modeling AI behavior, is really up to us and our relationships with AI, neurotechnology, and our egregores. “Who colonizes whom?” may revolve around “who is legible to whom?”.

Finally — to reiterate a point, all physical processes have projections into the qualia domain. Whatever ASIs are doing will still have this projection! I.e. the risk is not that consciousness gets wiped out, it’s that whatever optimization target ASI settles on has a “bad” projection into the qualia domain, while at the same time shifting the local environment away from the capacity or interest to self-correct. But there are reasons to believe suffering is energetically inefficient and will get optimized away. So even if we don’t make ASIs explicitly care about consciousness, the process that created them may still implicitly turn out to be a QSP.

Key references:

• Wittgenstein, L. (1953). Philosophical Investigations

• Quine, W.V.O. (1960). Word and Object

• Hoel, E.P. (2017). When the Map Is Better Than the Territory (see also Erik’s primer)

• Johnson, M.E. (2019). What’s out there?

XII. Where is consciousness going?

I can say AI consciousness is a wild topic not just because it crosses the two most important topics of the day but also because there’s a drought of formal models and intuitions diverge profoundly on how to even start. Here’s a recap of what I believe are the most important landmarks for navigation:

1. AI consciousness is as much a social puzzle as a technical one;

2. We should distinguish “software consciousness” from “hardware consciousness”; only the latter can be a well-formed concept;

3. We should carefully trace through where humans’ ability to accurately report qualia comes from, and shouldn’t assume artificial systems will get this ‘for free’;

4. Artificial systems will likely have significantly different classes (& boundaries) of qualia than evolved systems;

5. Decoherence seems necessary for consciousness, and patterns of decoherence (formalized as “branchial space”) encode the true shape of a system;

6. Brains and computers have vastly different shapes in branchial space;

7. The Symmetry Theory of Valence is a central landmark for navigating valence, and qualia in general;

8. Hardware qualia spans several considerations, which may draw from similar considerations as materials science & system architecture design;

9. Future AIs may or may not be interested in qualia, depending on whether modeling qualia structure has instrumental value to them; 

10. Today, the qualia domain has points of high causal density, which we call “minds”. Modern computers are an example of how the locus of causality can be different.

Armed with this list, we can circle back and try to say a few things to our original questions:

1. What is the default fate of the universe if the singularity happens and breakthroughs in consciousness research don’t?

There’s a common trope that an ASI left to its own devices would turn the universe into “computronium” — a term for “the arrangement of matter that is the best possible form of computing device”. I believe that energy efficiency considerations weigh heavily against this being an “s-risk”, although hard physical optimizations would have a significantly elevated chance of such compared to the status quo. My concerns are more social, e.g. morality inversions and conflating value and virtue.

2. What interesting qualia-related capacities does humanity have that synthetic superintelligences might not get by default?

Our ability to accurately report our qualia, and that we care about qualia, are actually fairly unique and something that AIs and even ASIs will not get by default. If we want to give them these capacities, we should understand how evolution gave them to us. A unified phenomenological experience that feels like it has causal efficacy (the qualia of “Free Will”) may have similar status.

3. What should CEOs of leading AI companies know about consciousness?

Distinguishing “hardware qualia” vs “software qualia” is crucial; the former exists, the latter does not. “CEOs of the singularity” should expect that consciousness will develop as a full scientific field in the future, likely borrowing heavily from physics, and that this may be a once-in-a-civilization chance to design AIs that can deeply participate in founding a new scientific discipline. Finally, I’d (somewhat self-interestedly) suggest being aware of the Symmetry Theory of Valence; it’ll be important.

In the longer term, the larger question seems to be: “What endowments has creation, evolution, and cosmic chance bequeathed upon humanity and upon consciousness itself? Of these, which are contingent (and could be lost) and which are eternal?” — and if some have grand visions to aim at the very heavens and change the laws of physics… what should we change them to?

Acknowledgements:

Thank you Dan Faggella for his “A Worthy Successor” essay, which inspired me to write; Radhika Dirks for past discussion about boundary conditions; Justin Mares and Janine Leger for their tireless encouragement; David Pearce & Giulio Tononi for their 2000s-era philosophical trailblazing; and my parents. Thanks also to Pasha Kamyshev, Roger’s Bacon, Romeo Stevens, George Walker, Pawel Pachniewski, and Leopold Haller for offering feedback on drafts, and Seeds of Science reviewers for their comments.

Article sent for review 15 May 2024 and published June 2024.

Notes:

[1] Although a “naive brain upload” may not replicate the original’s qualia, I anticipate the eventual development of a more sophisticated brain uploading paradigm that would. This would involve specialized hardware, perhaps focused on shaping the electromagnetic field using brain-like motifs.

[2] Thanks to Romeo Stevens for the metaphor.

[3] If something affects consciousness it will affect the shape of the brain in branchial space. As an example from my research — vasomuscular clamps reduce local neural dynamism, temporarily locking nearby neurons into more static “computer-like” patterns. This introduces fragments of hard structure into cognition & phenomenology, which breaks symmetries and forces the rest of the knot to form around this structure.

• Johnson, M.E. (2023). Principles of Vasocomputation: A Unification of Buddhist Phenomenology, Active Inference, and Physical Reflex (Part I)

• Moore, C., Cao, R. (2008). The hemo-neural hypothesis: on the role of blood flow in information processing

• Jacob, M., et al. (2023) Cognition is entangled with metabolism: relevance for resting-state EEG-fMRI

[4] One of the most important physics themes of the last 20 years is W. H. Zurek’s Quantum Darwinism. Zurek’s basic project has been to rescue normality: for almost a century physicists had bifurcated their study of reality into the quantum and the macro, with no clean bridge to connect the two. The quantum realm is characterized by fragile, conditional, and non-local superpositions; the “classical” realm is decidedly localized, objective, and durable. Somehow, quantum mechanics naturally adds up to everyday normality — but physicists were a little evasive on exactly how.

Zurek’s big idea was positing a darwinian ecology at the quantum level. The randomness of decoherence generates a wide range of quantum configurations; most of these configurations are destroyed by interaction with other systems, but a few are able to not only survive interactions with its environment but reproduce it. These winners become consensus across both systems and branches, which grants them attributes we think of as “classical” or “objective”:

Only states that produce multiple informational offspring – multiple imprints on the environment – can be found out from small fragments of E. The origin of the emergent classicality is then not just survival of the fittest states (the idea already captured by einselection), but their ability to “procreate”, to deposit multiple records – copies of themselves – throughout E.

Proliferation of records allows information about S to be extracted from many fragments of E … Thus, E acquires redundant records of S. Now, many observers can find out the state of S independently, and without perturbing it. This is how preferred states of S become objective. Objective existence – hallmark of classicality – emerges from the quantum substrate as a consequence of redundancy.

…

Consensus between records deposited in fragments of E looks like “collapse”.

…

Quantum Darwinism – upgrade of E to a communication channel from a mundane role it played in [the way physics has historically talked about] decoherence[.]

Zurek’s basic thesis is that physicists tend to think about decoherence in isolation, whereas they should also consider it as a universal selection pressure — one which has preferentially populated our world (Wolfram would say ‘branchial space’) with certain classes of systems, and has thus put broad-ranging constraints on what exists.

[5] Joscha Bach observes that we are actively making new classes of objects in branchial space:

The particle universe is a naturally occurring error correcting code on the quantum universe. Particles are stable enough to carry information across the junctures in the branching substrate universe, which makes control structures (atoms, cells, minds) possible.

If humans successfully build quantum computers, they impose new error correcting codes on the quantum substrate and are effectively creating a new type of particle, but one that has an evolved quantum technological agency as its precondition.

[6] With a nod to Frank Wilczek we can reasonably expect that the most mathematical beautiful formulation of branchial space will be the most qualia-accurate. This may be helpful or not, depending on priors.

[7] As a case study on what sorts of “branchially active” substances are possible, this passage from Maheffey 2021 is striking:

Plutonium is a very strange element, and some of its characteristics are not understood. It has seven allotropes, each with a different crystal structure, density, and internal energy, and it can switch from one state to another very quickly, depending on temperature, pressure, or surrounding chemistry. This makes a billet of plutonium difficult to machine, as the simple act of peeling off shavings in a lathe can cause an allotropic change as it sits clamped in the chuck. Its machining characteristic can shift from that of cast iron to that of polyethylene, and at the same time its size can change.

You can safely hold a billet in the palm of your hand, but only if its mass and even more importantly its shape does not encourage it to start fissioning at an exponentially increasing rate. The inert blob of metal can become deadly just because you picked it up, using the hydrogen in the structure of your hand as a moderator and reflecting thermalized neutrons back into it and making it go supercritical. The ignition temperature of plutonium has never been established. In some form, it can burst into white-hot flame sitting in a freezer.

[8] The NES’s core voltage was 5V; The ENIAC had a plate voltage of 200V-300V; myocardiocytes (heart muscle cells) have action potentials of around 90mV; the Large Hadron Collider (LHC) creates a voltage potential of ~6.5 trillion volts; the voltage potential between the earth and the ionosphere is around 300kV.

[9] My intuition is that combining the “symmetry view” and “branchial view” could offer heuristics for addressing the binding/boundary problem: how to determine the boundary of a conscious experience. E.g.,

1. We can interpret a moment of experience as a specific subset of branchial space;

2. The information content of this subset (i.e. the composition of the experience) can be phrased as a set of symmetries and broken symmetries;

3. The universe has intrinsic compositional logic (vaguely, whatever the Standard Model’s gauge group SU(3)×SU(2)×U(1) is a projection of; speculatively, located in Wolfram’s “Rulial space”), which can be defined as which symmetries and broken symmetries can be locally combined;

4. This compositional/perspectival limit may in turn determine a natural limit for the set of local nodes that can be combined into a ‘unified’ subgraph, vs when a new unified subgraph must be started.

I.e. just as the compositional logic of the universe doesn’t allow particles to have spin or electrical charge values of 5/7ths, it’s possible that some combinations of phenomenal information can’t exist in a unified experience — and this may uniquely determine a hard boundary for each experience. Reaching a little further in search of concrete elegance, perhaps the limit of an experience, the limit of a branchial subgraph, and the limit of a particular type/superset of local gauge equivalence are all the same limit. I hope to discuss this further in an upcoming essay.

[10] Beata Grobenski also noted the connection between valence shells & phenomenological valence in a recent piece.

Which kind of evidence would we need to see to believe that artificial neural networks can suffer? We review neuroscience literature, investigate behavioral arguments and propose high-level considerations that could shift our beliefs. Of these three approaches, we believe that high-level considerations, i.e. understanding under which circumstances suffering arises as an optimal training strategy, is the most promising. Our main finding, however, is that the understanding of artificial suffering is very limited and should likely get more attention.

Gardener Comments

Greg Baker (Lecturer in AI at Macquarie University):

This is like an inverse survey paper: it summarizes a wide variety of deeper topics, except that some of those topics haven't been written up elsewhere.

Arturo Macías:
I am a classical epiphenomenalist,  so I support the publication from a discrepant viewpoint. I think that while there is not a clear theory on consciousness in the paper, all issues commented are relevant, and the selected literature is very useful; the branchial space part looks especially relevant, given the relation between time and consciousness. 

Michael Bukatin (PhD in computer science):
This is an interesting position paper by a well-known consciousness researcher. I hope it gets published.

It addresses the most glaring defects present in almost all approaches to the "hard problem of consciousness" and should serve as a starting point for subsequent fruitful discussions.

I'll touch upon some possible follow-up points in these comments, including some additions and also including some places where I have significant disagreements with the author.

## Fundamentals

### Difficulties of talking about consciousness in general and AI consciousness in particular

The author notes that it is particularly difficult to talk about AI consciousness because people's views tend to be distorted by them taking into account implications of those views for their alliances and also by entanglements between "is X conscious" and status of that X.

Another difficulty seems to come from the fact that **"core intuitions about consciousness tend to cluster into two camps"** ["Why it's so hard to talk about Consciousness"] with Camp 2 being convinced that subjective experience and qualia exist in a fundamental way and Camp 1 having the intuition that "consciousness as a non-special high-level phenomenon". People belonging to different camps usually don't have productive conversations about consciousness.

I firmly belong to Camp 2, and so does the author, but we have to accept that the paper in question is unlikely to find acceptance within Camp 1, at least as things stand today (I'll touch upon how this might eventually change in the next subsection).

### Fundamental role of qualia and of practices of predictive science

One core defect of a typical approach to the "hard problem of consciousness" is that a typical approach focuses on the question "what makes something conscious or not conscious" ignoring the issues related to the nature of qualia and to the structure of spaces of qualia.

However, I think that the **"hard problem of qualia"** is the difficult core of the "hard problem of consciousness". If the "hard problem of qualia" is solved the rest is likely to go easier (how and why the qualia cluster into subjective entities, how the symmetry breaks result in a particular subjective entity being "me" versus all other subjective entities being "not me", and so on). So those approaches to the "hard problem of consciousness" which are content to sidestep understanding the issues related to qualia seem to be mostly missing the point. In this sense, it is very good that the paper in question emphasizes the central role of qualia and of our understanding of the nature and structure of qualia.

Another core defect of a typical approach to the "hard problem of consciousness" is that a typical approach is mostly speculative, saying plausible things, but not being rich on any empirical predictions. So we have literally hundreds of different approaches purporting to be the candidate solutions for the "hard problem of consciousness" without much to guide our choice between those candidate solutions.

But if one believes together with the author and myself that consciousness and qualia are "first-class citizens of reality", fundamental aspects of the reality we are in, then it is reasonable to demand that the novel theories of consciousness meet the same criteria as novel theories of physics (the same criteria which general relativity and quantum theory have met before).

Namely, a good theory of consciousness should make **novel unexpected empirical predictions** (including, ideally,  **new methods to achieve novel subjective phenomenology**). The author emphasizes this position saying:

>I think the best way to adjudicate this is predictiveness and elegance. Maxwell and Faraday assumed that electromagnetism had deep structure and this led to novel predictions, elegant simplifications, and eventually, the iPhone. Assuming qualia has deep structure should lead to something analogous.

If this level of understanding of consciousness is achieved, this would go a long way convincing people from Camp 1 as well (similarly to most scientists believing these days  (but not a hundred years ago) that general relativity and quantum theory make sense).

## Further remarks

I have listed the reasons for the paper in question to be a good, fruitful, methodologically sound starting point for subsequent discussions.

Now it's time to look at the details including disagreements.

### A nitpick: attention

The paper in question says:

>If consciousness is a lossy compression of complex biological properties, similar to “attention” or “mood”, asking whether non-biological systems are conscious is a Wittgensteinian type error

Well, no, "artificial attention" is the core innovation responsible for the spectacular

success of the modern cutting edge AI systems. This is a relatively recent innovation which is not trying to model the detailed biological mechanisms of "attention", but is doing a very good job modelling the **effects** of "attention".

So "attention" is certainly not a good example in this sense; non-biological AI systems are highly capable of exhibiting "attention" and many of them are using "artificial attention" as the core of their cognitive engines.

### Central disagreement: on the correspondence between physical and conscious entities

I share the author's assumption that we are dealing with "physics-like fields of qualia", or something in this spirit. This is precisely why I can't agree with the oversimplified correspondence between physical entities and conscious entities.

The objections come from two directions:

  * a physical entity is likely to contain multiple consciousnesses

  * a consciousness might extend beyond a single physical entity

#### 1) a physical entity is likely to contain multiple consciousnesses

If we think that all physical processes are conscious, this means that there are many consciousnesses within a human brain. "Subconscious processes" are likely to have their own consciousness, but that consciousness and those qualia are not included into the consciousness of a person. "Thermal noise" and such in the brain might also have associated qualia, but they are not included into the consciousness of a person either.

So a brain of an unconscious human has processes which have consciousness, but those processes are not included into the (temporarily absent) consciousness of the person in question. And even a dead body might have some qualia associated with it (because any piece of physical reality is conjectured to have some), but there is no reason to believe that those qualia have anything to do with that person.

So, turning to the author's analysis of qualia associated with an inference in an ML model, if the key qualia are indeed "thermal noise", then those qualia don't have anything to do with the essence of the particular inference in the particular ML model and, just like for a human, would not be part of the consciousness of the **"temporary virtual character"** simulated by an LLM at the moment. If the "thermal noise" is the only type of qualia present here, then the process corresponding to the **"temporary virtual character"** is a P-zombie.

Now, I have to say, that I find it highly unlikely that in a world where almost everything is conscious to some extent, a coherent "temporary virtual character" would be a zombie, especially given that we know that autoregressive Transformers are "less feed-forward than they seem" and that they, in fact, do emulate recurrent machines when used in the autoregressive mode (see e.g. "Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention").

A more promising approach would be to try to understand the nature of "purely linguistic qualia" those inference might or might not possess. And, of course, with the emerging multimodal models, one should revisit the issues like "do all human linguistic qualia have audio-visual nature or not", and 'do "linguistic qualia" (if any) in multimodal models have audio-visual nature or not', and so on.

In any case, "the LLM inferences are conscious, but only feel low-level noise" conjecture seems like a cop out and is almost certainly wrong. "Thermal noise"-associated qualia might be present, but they are unlikely to be part of the LLM consciousness. If that's the only qualia present in that context,  then LLM inferences themselves are likely to be P-zombies.

However, what is important to keep in mind is that the property of being conscious or not is not a property of a given LLM, but a property of a given inference, of a given **simulation** or of entities within that simulation [See the **Simulator Theory** sequence by Janus and the paper in Nature, “Role play with large language models”, co-authored by Murray Shanahan and Janus team]. It might be that the same LLM produces smart and very conscious sessions and dumb sessions devoid of meaningful subjectivity. LLMs are **generators of "virtual realities"**, and some of those realities might have sentiences associated with them, while others might lack sentiences (or, at least, the sentiences might be of wildly different natures and complexities for different sessions of the same LLM).

#### 2) a consciousness might extend beyond a single physical entity

The second class of objections comes from the fact that for a human brain we actually don't know if the engaged qualia fields are physically confined within the head and body or extend beyond it (assuming that those qualia are in the same physical space at all). Perhaps, the fact that we see the "world out there" actually corresponds to the engaged visual qualia fields being "out there" and not within the brain. We just don't know (because our progress on the "hard problem of qualia" is next to non-existent).

What does this have to do with AI consciousness? When one asks what class of software processes corresponds to a physical computer, the "Turing machines" answer is wrong on many levels. Right now, the important aspect for us is that the computer is not isolated, it interacts with the external world. Technically speaking, if one wants to talk in terms of Turing machines, one would need to talk about "Turing machines with oracles", with the whole world being the oracle in question. This, by the way, invalidates the famous "Goedel argument" by Roger Penrose, see, for example, my essay "Reading Roger Penrose".

It is known that when people interact, various synchronization effects are observed. In particular, those synchronization effects are often observed introspectively by at least one of the participants. The nature of those synchronization effects is not well-understood.  But we need to ponder whether some synchronization effects between a person and an LLM talking to each other might be present. We really don't know much about the nature of reality we inhabit, and should keep a good deal of open mind about any particular property of that reality which might seem obvious to us. A given property might seem obvious, but might, nevertheless, be an incorrect approximation, screening important aspects of reality from us. In the last section of this comment I'll talk about the possibilities to amplify those synchronization effects, and what those possibilities might imply.

**Summarizing my main objections**: 

  * a physical entity is likely to contain multiple consciousnesses,

  * a consciousness might extend beyond a single physical entity,

  * and these properties have all kinds of non-trivial implications. 

### Merging human and AI consciousness

For an actually working theory of qualia and of consciousness we hope to discover in the future, it's not enough to just have a strong predictive power.

We want to be able to experience "what is it like to be a bat" or "what is it like to be a particular LLM inference in progress", and so on.

One conscious entity wants to be able to experience what it is like to be another potentially conscious entity (this is never an entirely safe endeavor, so all kinds of ethical and safety caveats apply).

This is the **Holy Grail** of consciousness science, the vital empirical counterpart to our theoretical work.

We are seeing rapid progress in high-quality **non-invasive brain-computer interfaces** and in our ability to create tightly closed loops between humans and electronic systems using those interfaces (the risks are potentially quite formidable and one needs to keep those risks in mind while engaging in these kinds of activities).

So it is possible that we might be able to investigate the degree of consciousness of various electronic devices empirically.

If humans don't do that, many of the advanced AI systems will be curious enough about human consciousness and human qualia and will likely initiate merging experiments from their side in order to satisfy their curiosity and to experience "what is it like to be a human".

Nicholas Craycraft (ex-google full stack engineer):
I like this article's overview, but as with everything qualia, the hypotheses come off as scientifically untestable. There is no clear way to do science that dissociates form from function. I would like there to be more of a focus on tackling the testability of this domain.

There are already countless philosophers doing the same thing this essay does, summarizing other philosophies and coming up with one's own paradigm.

This article is strong where it is arguing for models that function for engineering purposes. It is weak wherever it makes metaphysical claims ("Consciousness is what gives life meaning" is already contestable. I would argue that structural correlations within a context gives things meaning, in that this is literally what meanings functionally are.) It is strong when it is summarizing a wide range of philosophical positions on these topics, and it is weak where it takes its own positions without a clear testing plan. It just reads as cherry picking the worldviews that the author favors.

If the article must make such confident assertions in its conclusion -- it would be nice if it could first retread the philosophical heuristics/priors one last time before saying things like:

Q "What should CEOs of leading AI companies know about consciousness?"
A "Among other things they should understand that software qualia isn't real."

This wasn't argued for and is a serious S-risk if the author is wrong! The 'argument' given is: axiom : consciousness is a singular thing that can only have one natural home.

aesthetic preference??: I propose we choose the one that is more real.

but "real" by what metric? I have to assume the logic goes:

axiom : descriptive reductionism
axiom : descriptive reductionism -> mereological nihilism
therefore : mereological nihilism

I should note that my biases differ sharply from the author's. I am more of a functional holist and simulationist regarding phenomenology. I find it inconceivable that we will discover a source of atomic-level phenomenology that cannot be replicated using larger-scale structures to produce functionally identical claims of phenomenology. Moreover, I cannot envision any test that would discern 'functional' from 'actual' phenomenology that I would agree upon.

Richard Sprague:
A clear rejection for SoS in its current form, but contains numerous ideas that I wish were better articulated. As written currently, it's just a jumble of ideas badly in need of a hard-nosed editor. Singularity? AI company CEOs? a lot of random topics -- pick one.

Worthiness for SoS: while I don't necessarily think the bullet-point style of writing disqualifies it *per se*, in this case it looks more like a Powerpoint than a real paper.

That said, there are many interesting ideas in this paper. Although I've read much of the related popular literature about consciousness (Goff, Hoffman, Seth, etc.) I've not run into some of what seem to be unique insights. If nothing else, his general direction seems original in the sense of synthesizing ideas from multiple sources and adding many new of his own.

He takes panpsychism as truth, without explanation. At minimum he must realize this puts him in a very controversial part of the room, and he should defend himself accordingly. Since his entire argument depends on it, this is a pretty important part to clear up.

Interesting idea (I bet it's been well-explored, but I haven't seen this) about Turing machines' computations not mapping into their physical substrate. That flickering pattern of bits going on and off, instantiated as zillions of transistor voltages going up and down, bears no Turing equivalence?! Now that I think about it, that's an impressive finding.

Another interesting side-point about Wolfram's Branchial space and how maybe decoherence is some kind of interaction between unrelated branches. Especially interesting since, as he notes, current computers, unlike brains, are specifically designed to minimize decoherence. 

Jack Arcalon:
An array of highly speculative but long-term efforts to extend consciousness research. Right now the mystery of awareness is so great that it may require intensive brainstorming and overview impressions in the hope this may lead to one great insight.

A compact, inexpensive repeated survey on American adults’ attitudes toward Artificial General Intelligence (AGI) revealed a stable ordering but changing magnitudes of agreement toward three statements. Contrasting 2021 to 2023, American adults increasingly agreed AGI was possible to build. Respondents agreed more weakly that AGI should be built. Finally, American adults mostly disagree that an AGI should have the same rights as a human being; disagreeing more strongly in 2023 than in 2021.

Introduction

Are we ready for in silico equals? Depending on who you choose to listen to, you will find arguments that this question is premature (Marcus, 2022) timely (Bubeck et al., 2023) or already too late (Yudkowsky, 2023). Here we present the results of a short, repeated survey on attitudes toward Artificial General Intelligence (AGI). We can say with confidence that (in our samples) American adults believe building an AGI is possible. Many (but not the majority) agree that AGI should be created. Most disagree with the idea that an AGI should have the same rights as a human being.

Prior Work

This study is not the only recent assessment of attitudes toward artificial intelligence. Sindermann et al. (2021) recently contributed the Attitude Towards Artificial Intelligence (ATAI) scale – a five-item scale measuring two factors: Acceptance and Fear. The ATAI comprised concise prompts such as “I trust artificial intelligence” and “Artificial intelligence will destroy mankind.” Across Chinese and German respondents, those who scored high in Acceptance expressed high willingness to use artificial intelligence systems (e.g. self-driving cars). Those who scored high on the Fear factor expressed lower willingness. Overall, on the 11-point Likert range, mean responses for both factors sat near the midpoint.

Similarly, Schepman and Rodway (2020) created a 32-item General Attitudes towards Artificial Intelligence Scale with two equally-sized subscales for positive and negative attitudes. In contrast to Sindermann et al., Schepman and Rodway emphasized the difference between attitudes and technological readiness (Lam et al., 2008). Technological readiness, they argued, predicts consumer decisions – e.g. will I buy and use a home smart speaker? However, artificial intelligence systems are more frequently deployed by organizations, not individuals, and the decision is made for the individual, not by the individual. Therefore, they argue it is important to measure general attitudes toward artificial intelligence and comfort with proposed applications (e.g. using facial recognition to automatically detect, identify and fine jaywalkers) to inform legislators and organizations. Overall, Schepman and Rodway found that positive attitudes were associated with respondents’ perceived utility of potential applications and negative attitudes with perceived malevolence (“I find Artificial Intelligence sinister”) or dystopian potential (“Artificial Intelligence is used to spy on people.”)

Neither of the above, and no other work we are aware of has specifically measured attitudes toward AGI specifically. Additionally, repeated surveys such as this one are rare. The current study contributes results on a novel facet of artificial intelligence attitudes and provides evidence toward temporal trends.

Methods

The data we collected and the analysis code we wrote to support this article are publicly available at https://osf.io/df2yx/. In March 2021 and again in April 2023, we gathered responses to three survey items regarding AGI. In the prompt, we defined AGI this way: “Artificial General Intelligence (AGI) refers to a computer system that could learn to complete any intellectual task that a human being could.” The survey items listed in Table 1 followed.

2021 Survey

We sought low-cost, high-volume respondent recruitment. After evaluating several alternatives, we selected Google Surveys (Sostek & Slatkin, 2017). At the time, Google Surveys provided a service to distribute one-item surveys to a representative sample of American adults (Keeter & Christian, 2012) for $0.10 per respondent. Self-report age and sex demographics were automatically appended to each response. The platform claimed to deliver samples representative across age, sex and geographic region, with the Census Bureau’s American Community Survey as the reference compared against.

In March 2021, we distributed three separate one-item surveys using the AGI definition from above and the items from Table 1. The platform selected respondents, sent a push notification inviting them to the survey and collected the response. We requested 300 respondents to each item and paid $90 total: $0.10 per item * 300 respondents * 3 items. In Table 2, one can see the Google Surveys platform over-delivered on quantity of respondents in each case; we believe this was due to the platform attempting to fulfill its promise to automatically adjust sampling to deliver a representative sample. Figure 1 depicts what a respondent would have seen.

2023 Survey

The 2023 survey was distributed on the Prolific Academic platform. We were unable to use Google Surveys, because the product was abandoned and shut down in November 2022. Prolific (2014) also provides distribution of online surveys to samples of American adults. (See Adams et al., 2020 for a recent evaluation of the platform.) It is important to note that on Prolific a premium is charged to target a representative sample, and we paid the premium to do so. The total cost for 459 completed responses was $988, or $2.15 each.

Unlike the 2021 survey, the 2023 participants responded to all three items. The three items consisted of the same statements from Table 1. A seven-point Likert scale was again the set of possible responses, however, each point received its own verbal label. Figure 2 depicts what a respondent would have seen. The survey was created as a Qualtrics XM “Matrix table.” (In 2021, we had separated each item into its own administration only because that delivered the best cost-to-response ratio due to the affordances of Google Surveys). In 2023, because the entire sample responded to all three items, we were able to perform a correlational analysis (see Discussion) that was not possible for the 2021 survey.

Sample Demographics

In Table 2, we present the total count of respondents per sample, and percentages per Sex and Age category (On both platforms, only values of Female and Male were provided for Sex). Recall that each item was fielded to a separate sample in 2021, while in 2023 all participants responded to all three items simultaneously.Table 2. Demographics of each sample.

Results

We began with the simplest analysis. Using only the most recent data (viz. the 2023 Prolific sample), we contrasted the average response for all three items. We make the typical assumption of a Likert scale, and assign each of the seven responses a number from -3 to +3. Mean responses for each item are plotted in Figure 3.

Averages can sometimes distort the truth, and some scholars would object to mapping categorical responses to numerical values. For those reasons, we also present histograms containing the raw frequency of each categorical response here as Figure 4.

Next, we turn to a temporal analysis; we will contrast the magnitude of agreement for each item between the 2021 and 2023 samples. We make some assumptions to do so. First, we assume the samples are comparable. Second, we assume the Likert-scale responses can be mapped to equally-spaced numeric intervals of -3 for Strongly disagree through +3 for Strongly agree. Finally, we assume the platform effects (i.e. Google Surveys versus Prolific) are negligible. With these assumptions, we plot the change over years in Figure 5.

Welch two sample t-tests (which adjust degrees of freedom to account for unequal variance) indicated that the 2021 to 2023 differences for Possible to build and Same rights as a human were statistically significant with p < 0.001. The difference for Should be built (p = 0.10) did not reach the conventional criterion for statistical significance.

Finally, we explore potential differences across age and sex demographics. Figure 6 plots mean responses per age category. (In the Google Surveys data, only a bin is reported for each respondent. We mapped the age in years variable available in the Prolific sample to the same bins.) Statistical analysis (generalized linear models estimating Response from Age * Year) confirms visual inspection of the plots: agreement differs by year of sample for Possible to build and Same rights as a human as discussed above, and the strongest association with age is more disagreement from older adults for Same rights as a human in 2023.

Figure 7 plots 2021 and 2023 responses by sex. Both groups’ results conform to the patterns in Figure 5: increasing agreement for Possible to build; little change for Should be built and more severe disagreement for Same rights as a human. In a linear regression, the interaction between Sex and Year was statistically significant for Same rights as a human.

Discussion

Artificial General Intelligence – by our chosen definition – is not here yet. But American adults increasingly believe it is possible to build. Their agreement that AGI should be built is not increasing at the same rate. Their willingness to endorse equal rights for an AGI is trending in the opposite direction. As artificial intelligence systems become ubiquitous (as they are currently poised to) persistent, consistent measurement of these attitudes bears consideration.

We examined the relationship between attitudes and age (and attitudes and sex) not because we had theory or even a hypothesis to test. Instead, we explored the data for the same reason George Mallory gave for climbing Mt. Everest: “Because it’s there.” We urge readers to not overinterpret the results. Instead, note that across every demographic category, the trajectory of attitudes over time is consistent in direction. This is not a landscape of crossover interactions. There is a hint that males and older adults are souring to the idea of AGI peers faster than their (human) counterparts. Future scholars could follow up on that idea. However, we urge future surveys to attempt to discover correlates more predictive of AGI attitudes than mere demographics.

More generally, we argue that in the current moment, temporal trends at the aggregate level are more likely to be interesting than static subsample analyses. The capabilities of AI are developing rapidly, and public opinion toward its applications is likely also in flux. We understand the convenience of demographic subsample analyses – the marginal cost is only one more line of R code. Repeating a survey, on the other hand, roughly doubles the amount of time and effort of the research project. It is our hope that more researchers accept that burden, however. Compact, inexpensive, repeated surveys such as this one are an example we hope others follow. We will repeat this survey again in the spring of 2024 using the same methods as the 2023 administration.

For the 2023 survey only, every respondent encountered all three items. Thus, we can explore the associations between attitudes. Figure 8 contains boxplots for each pair of items. The thick black horizontal line in each marks the median response; blue diamonds mark the mean response. Responses for all items were positively correlated. Respondents who were more skeptical about the possibility of AGI disagreed that it should be built (r = 0.37, CI = [0.29, 0.45], p < 0.001). This makes intuitive sense. If you believe AGI is not possible, presumably you also believe it would be a waste of time to try to build one. Those who were more skeptical about the possibility of AGI also more strongly disagreed that it should have the same rights as a human (r = 0.13, CI = [0.03, 0.21], p = 0.007). Note, however, that human rights for AGI is unpopular across the spectrum of belief in its possibility.

Agreement that AGI should be built is associated with less disagreement that AGI should have the same rights as a human (r = 0.30, CI = [0.21, 0.38], p < 0.001).  On average, those who support building AGI are against granting them rights, but less so than those against building AGI.  Indeed, among those disagreeing that AGI should be built, lack of support for rights was nearly unanimous.

Agreement that AGI should be built is associated with less disagreement that AGI should have the same rights as a human (r = 0.30, CI = [0.21, 0.38], p < 0.001).  On average, those who support building AGI are against granting them rights, but less so than those against building AGI.  Indeed, among those disagreeing that AGI should be built, lack of support for rights was nearly unanimous.

Limitations

We present these results based on data we have instead of the ideal data we might wish we had.  Ideal data collection would certainly have used the same platform.  It was not our choice for Google to abandon their survey product, but perhaps we should have foreseen such.  In any case, respondent recruitment method and survey platform are unavoidably confounded with time in our data.

It might have been useful to collect data at finer temporal resolution.  One strength of the current interval is that it straddles an extremely salient news event in artificial intelligence – namely, OpenAI’s public release of ChatGPT.  Given the investment and interest sparked by the recent generative AI wave, we expect AGI speculation and discussion will be increasingly salient to the general public.  Future data will provide more clarity on the possible trends and associations explored here.

Conclusions

Like all responsible scholars, we disclaim the idea that these results are definitive or comprehensive.  Americans’ attitudes toward AGI and artificial intelligence more generally deserve and are gaining increasing attention (O’Shaughnessy et al., 2023; Zhang & Dafoe, 2019).  However, to our knowledge, these are the first results from a repeated survey regarding AGI fielded to national, representative samples.  Thus, they can provide a reference point for further investigation.  We take the liberty here to make the following recommendations: Large, carefully sampled, well-resourced surveys such as the General Social Survey, American National Election Studies and the World Values Survey should add and continue to field an item or scale specifically regarding AGI.  Independent of these projects, scholars can and should make use of inexpensive, representative online survey platforms to consistently and persistently measure attitudes toward technology.

Gardener Comments

Andrew Neff (PhD in neuroscience):
Really interesting, I recommend publication, I just came here to leave a small recommendation about Figure 5-7 related to presentation. As it is, the Y-axis range does not capture the entire range, which could lead readers to overestimate the differences that were observed.

Joe R:
Overall this struck me as an excellent paper. The (mundane, not statistical) significance of the results seems dubious, but I'm tempted to recommend this paper for its impressive dedication to accuracy, brevity, and clarity. 

Pros: 

• A good introduction succinctly summarizing results.

• Publicly accessible data. 

• A clear and legible explanation of methods. 

• Simple, legible, appropriate graphics. 

• A commitment to repeat the survey in 2024. 

• Open acknowledgement of flaws. 

Cons: 

• Simple, general questions that could be measuring lots of different things (like a general halo or horns effect of AI news). 

• A relatively small sample for a survey of national relevance (~300). But still respectable. 

• The inherent unreliability of survey data. (Acknowledged by the authors). 

I'm honestly not quite clear on how useful these data will be to researchers. I agree with the authors that we shouldn't read too much into the results. But for ~$1k/year it is probably a cheap test worth doing. And having been done, it should probably be shared. Kudos to the authors for one of the best write-ups I've seen so far.

C:
I like the blunt honesty that the authors use with regard to their non-existing hypotheses. I also greatly value that they made their assumptions clear. However, I am hesitant for the following reasons:

• Unclear how large the 2021 sample was. Unclear how sample size was determined

• 2021 was on a star-rating, in 2023 all items had labels. Why did you choose this different scale structure when implementing it in Qualtrics? And how did this potentially affect the results?

• I agree that longitudinal studies are necessary, but I hope that more researchers follow panel-designs, in which they survey the same people repeatedly. This controls for a lot of unobserved variables and maybe you could consider a panel design for your study as well

Dr. Payal B. Joshi:
The article is relevant to the present trends on artificial intelligence. While there is a plethora of information pertaining to the topic that ranges from technological applications to mystical uses, the present study attempts to explore a lesser-known terrain on attitudes towards AI. The survey results are critically analyzed and limitations of the study are presented. Maybe one aspect that seems skipped by the author is - the study is based on American people's attitudes only: Can the results be entirely extrapolated to other parts of the world? Not quite. Also, I found a few polarized responses a little obvious that respondents provided therein - hence a detailed survey (as already highlighted to be conducted by author is mentioned) shall serve the purpose. 

Overall, the paper is well-presented and thus I highly recommend publishing the paper as it is. 

Thomas Gladwin (PhD):
I enjoyed reading this paper, on a clearly topical issue. It should help inspire future research - for instance, would opinions change from before to after watching the Star Trek trial of Commander Data? I appreciated the transparency on methods and the research process, as well as the creative and clear visualizations.

A minor comment: "We make the typical assumption of a Likert scale, and assign each of the seven responses a number from -3 to +3." and "some scholars would object to mapping categorical responses to numerical values." It's great that this is touched on, but maybe it would be helpful to make explicit what the "typical assumption" is and provide a reference to a paper on the issue? (E.g., for a counterargument to the "against" position", https://link.springer.com/article/10.1007/s10459-010-9222-y.)

References

1. Adams, T., Li, Y., & Liu, H. (2020). A Replication of Beyond the Turk: Alternative Platforms for Crowdsourcing Behavioral Research – Sometimes Preferable to Student Groups. AIS Transactions on Replication Research, 6(1). https://doi.org/10.17705/1atrr.00058

2. Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., Nori, H., Palangi, H., Ribeiro, M. T., & Zhang, Y. (2023). Sparks of Artificial General Intelligence: Early experiments with GPT-4 (arXiv:2303.12712). arXiv. https://doi.org/10.48550/arXiv.2303.12712

3. Keeter, S., & Christian, L. (2012). A Comparison of Results from Surveys by the Pew Research Center and Google Consumer Surveys. https://www.pewresearch.org/politics/2012/11/07/a-comparison-of-results-from-surveys-by-the-pew-research-center-and-google-consumer-surveys/

4. Lam, S. Y., Chiang, J., & Parasuraman, A. (2008). The effects of the dimensions of technology readiness on technology acceptance: An empirical analysis. Journal of Interactive Marketing, 22(4), 19–39.

5. Marcus, G. (2022, July 1). Artificial General Intelligence Is Not as Imminent as You Might Think. Scientific American. https://www.scientificamerican.com/article/artificial-general-intelligence-is-not-as-imminent-as-you-might-think1/

6. O’Shaughnessy, M. R., Schiff, D. S., Varshney, L. R., Rozell, C. J., & Davenport, M. A. (2023). What governs attitudes toward artificial intelligence adoption and governance? Science and Public Policy, 50(2), 161–176.

7. Prolific. (2014). https://www.prolific.co/academic-researchers

8. Schepman, A., & Rodway, P. (2020). Initial validation of the general attitudes towards Artificial Intelligence Scale. Computers in Human Behavior Reports, 1, 100014.

9. Sindermann, C., Sha, P., Zhou, M., Wernicke, J., Schmitt, H. S., Li, M., Sariyska, R., Stavrou, M., Becker, B., & Montag, C. (2021). Assessing the attitude towards artificial intelligence: Introduction of a short measure in German, Chinese, and English language. KI-Künstliche Intelligenz, 35, 109–118.

10. Sostek, K., & Slatkin, B. (2017). How Google Surveys Works. http://g.co/surveyswhitepaper

11. Yudkowsky, E. (2023, March 29). Pausing AI Developments Isn’t Enough. We Need to Shut it All Down. Time. https://time.com/6266923/ai-eliezer-yudkowsky-open-letter-not-enough/

12. Zhang, B., & Dafoe, A. (2019). Artificial intelligence: American attitudes and trends. Available at SSRN 3312874.

Abstract

That policy makers will ever rationally respond to scientific warnings about the ecological crisis should be treated as a falsifiable hypothesis. After more than five decades of such warnings, there is a strong case for skepticism. Climate and other ecological tipping points constitute the quantitative thresholds beyond which current political systems can definitively be said to have failed. This presents a mandate to generate broad consensus on where tipping points lie, and at what proximity to them new strategies should be pursued. Central to any new strategy should be an understanding of why the old one failed—an understanding of why those in power almost exclusively derive from academic backgrounds other than physical science, and the psychological differences between those who issued or received so many warnings of collapse. To that end, a psychological trait syndrome relevant to political power is proposed, based on correlations between academic specialization, psychometric results, and the behavior of powerful people across a wide range of societies. This proposed syndrome consists of four covarying dimensions of individual difference. These are perceptions of hierarchy vs. egalitarianism, established knowledge vs. open inquiry, physical vs. symbolic action, and schematic vs. particular knowledge.      

Introduction

In December of 2017, BioScience published an article with 15,364 scientist signatories, from  184 countries—the most to ever cosign and formally support a journal article—stating that “we have unleashed a mass extinction event” and that the “widespread misery and catastrophic biodiversity loss” this entailed would soon be inevitable, absent a massive shift (Ripple et al. 2017). Or rather, they issued the successor to a similar warning made twenty five years previous. Exercising a considerable capacity for understatement, the authors of “World Scientists' Warning to Humanity: A Second Notice” said that “in most respects, we have not heeded” the first one, issued by the Union of Concerned Scientists and ~1700 other scientists.

A quarter century is such a significant interval, the global crisis we are in the midst of so profound, and the behavior of political and economic institutions so utterly disconnected from anything that could plausibly be described as a meaningful response, that a wide range of scientific questions emerge about the core assumptions underlying communications such as this. Why continue this strategy? Because if over 1,700 scientific signatories isn't sufficient, 15,364 ought to do the trick? At what point would it be true that such communications have empirically failed? And at what point does this failure indicate a fundamental inability of the extant system to acknowledge our crisis? 

If science is to be defined, to a significant extent, by its responsiveness to empirical reality, this point must exist. But we have pushed the global system into a state where outcomes have become very uncertain—far beyond what anyone can plausibly describe as safe (Hansen et al. 2013; Kemp et al. 2022)—and so far, this scientific dialogue has not begun. It is difficult to think of a more concretely useful and theoretically intriguing scientific question than what social structures are compatible with ecological survival, and if we are, by any chance, living in one that isn't. But to begin a scientific assessment of whether the world's political systems are capable of decisive ecological action requires no longer simply assuming that they are. 

For instance, the second “World Scientists' Warning to Humanity,” says “As most political leaders respond to pressure…dogged opposition can be overcome and political leaders compelled to do the right thing.” But how do we know? Such a significant assertion obviously warrants the same scientific scrutiny we would apply to statements like “the horse was domesticated on the Pontic Steppe” or “spadefoot toad cannibalism is mediated by corticotropin releasing hormone.” We have no case studies in fundamental socioeconomic transformations of the kind recommended by climate and ecological scientists. A significant curtailment of fossil fuel extraction, deforestation, and animal product consumption—fairly standard items in the policy prescriptions of scientific warnings to policy makers—would constitute one of the major economic reorganizations of human history, roughly akin in scale (but very different in nature) to the industrial revolution or the post-World War Two “great acceleration,” as described by (Steffen et al. 2015).   

Thus it is an open scientific question whether or not current political systems are capable of affecting the major, unprecedented shifts necessary to avoid catastrophe. This scientific question is probably best framed by physical thresholds of runaway climate change—so-called tipping points. Scientific entities such as the IPCC have made the not-exactly-apolitical decision to simply disregard self-perpetuating climate processes, despite the likely central role such processes played in past mass extinctions (Rogelj et al. 2018). However, we know that the climate system contains feedbacks—such as the release of methane from melting permafrost, the total loss of marine ice, and the conversion of forests to lower biomass ecosystems—which are associated with the earth's past mass extinctions. Political systems can be said to have decisively failed to avert global catastrophe when these processes have been triggered. And strong arguments can be made that it is no longer warranted to place any faith in these systems once there is a significant possibility these processes have been triggered. 

We can use the question of whether the world's forests are a net carbon source or sink as an example of establishing a “significant possibility” that these processes have been initiated. If a number of years transpire in which forests lose more carbon to fire and severe weather than they gain in biomass (Baccini et al. 2017; Harris et al.), the possibility that forests are no longer viable in the current climate can be placed in a confidence interval. Obviously, the loss of forests exacerbates the climate conditions causing forest loss, creating a feedback loop. Overall climate system tipping points can be identified by estimating the triggers for multiple feedbacks of this nature. 

The question of exactly where to place such a threshold, and say that beyond it, a political system has failed to avert catastrophe is of course partially a matter of judgement. But for our purposes, the point is that the threshold certainly exists somewhere—it is obviously no longer reasonable to write earnest warnings to policy makers towards the tail end of, say, an end-Permian magnitude extinction—and it is essential that we try to identify it. Otherwise, we have no way to evaluate the question of how to interact with current political systems—and what options exist other than the strategies which have been tried unsuccessfully thus far—in a scientific manner. 

What do we learn about science from its perennial debates?

Kemp et al. (2022) explore the strong statistical bias of climate research to ignore the possibility that “abrupt and/or irreversible changes may be triggered at a temperature threshold”. If we are rapidly approaching, have arrived at, or are beyond a catastrophic threshold, what does this imply for scientists' strategies for ecological survival? That they should give up on life itself? Or that they should give up on issuing warnings to those in power, and instead attempt to directly intervene in the dynamics of power, to assert new decision making processes wherein scientific comprehension plays a larger role? To rephrase this in more typically political language: when is the science clear that collapse can only be averted by sociopolitical revolution? And if that time has come, or when it does, what will scientific deliberation about how to achieve that revolution look like, and what conclusions will it produce?      

Of course, anyone remotely familiar with science as it is actually practiced, here in corporeal reality, is well aware that such a process would be sharply constrained by worldviews. For these deliberations to proceed, it would be necessary for a specific subset of scientists to identify one another. It would be necessary for scientists to apply to themselves the very methods they use to study the nature and corollaries of the belief systems of others. 

There are numerous indications of an underlying psychological framework involved in various scientific outlooks. For instance, there are statistical predictors of academic specializations lacking an inherent logical relationship with the field of study. Students with majors such as business, economics, and business administration have higher scores on psychological survey instruments like the Social Dominance Orientation (SDO) and symbolic racism scales than those with majors such as anthropology or sociology (Sidanius et al. 2003). Math and physical science graduates form a distinct cluster in their scores on psychometrics such as Need for Closure and Disgust, while graduates in applied technical fields like engineering form another distinct cluster (Gambetta and Hertog 2016). And academic training itself appears to enhance the psychological tendencies which cause individuals to self-select into a given field—engineering students, for instance, begin their studies with a more hierarchical worldview than physical or social scientists, and that hierarchical outlook intensifies during their training (Haley and Sidanius 2005). Despite intriguing indications of the presence of distinct psychologies in the academy, the corollaries of scientific beliefs have not been given the rigorous, systematic attention the beliefs of non-scientists have received. We do not know what predictors might exist, for instance, of preferences for scientific positions within disciplines.

I believe a more self-referential epistemology, involving concerted inquiry into the kinds of minds that favor various scientific paradigms, is necessary for science to progress. The generality of this point is illustrated by perennial debates within various disciplines, as pointlessly repetitive as dire warnings about the ecological crisis. Such debates often feature obsessively stylized reasoning, in service of some unarticulated (but often very apparent) worldview. 

There are many examples of this phenomenon, such as the perennially divergent interpretations of the same anthropological data found in discourses such as the Kalahari debate (Solway and Lee 1990), and its more recent counterpart concerning east African foragers (Porter and Marlowe 2007), or perennially divergent interpretations of the same archaeological findings, in cases like the Indus Valley civilization (Green 2020). To examine one example in some detail, reflecting on the “morality play, seemingly bound forever to the wheel of intellectual life,” of the previous century's worth of nature-nurture debates, Tooby and Cosmides (1995) describe the “innumerable incarnations” of this perennial rupture: “rationalism versus empiricism, heredity versus environment, instinct versus learning, nature versus nurture, human universals versus cultural relativism, human nature versus human culture, innate behavior versus acquired behavior, Chomsky versus Piaget, biological determinism versus social determinism, essentialism versus social construction, modularity versus domain-generality” (Tooby and Cosmides 1995). While some echelons of social science have made significant progress toward biologically integrated explanatory frameworks in the thirty years subsequent to this writing, it is hardly the case that these incommensurate modes of reasoning have been truly reconciled. “A rift runs through anthropology” say Camilla, Power, and Callan (2017), in their introduction to a book on human origins, sounding little different in its account of “radically opposed” epistemologies than Tooby and Cosmides in 1995, or for that matter Konrad Lorenz in the 1950s.

Politically predictive psychological traits

My perspective on this particular schism, derived from three decades in ecological politics, illustrates the potential for the kind of inquiry I am proposing. It has long been clear to me that my own and many other social movements are hindered by the nature-nurture thinking which still—after all the exhausting dialogues—animates some social theorists's refusal to reconcile with biology. This is hardly surprising, as political tendencies and social science share a frequent aversion to human nature (Schroder 2023). But what is equally clear to me is that, in movements and the academy alike, there is a distinct underlying psychology involved in this rejection of human nature, which reveals itself through a number of covariates. 

The most immediately apparent is that biophobia, as it has been termed, expresses a more general orientation toward the symbolic, at the expense of the physical. This manifests as a more assiduous concern with symbolic harm, e.g. through language, on the part of social constructivists, and a more pronounced inclination toward concrete, physical action on the part of those who favor biologically integrated explanations. Chomsky and Foucault's famous 1971 debate is an example, with Chomsky speaking about the biology of language acquisition and advocating for disruptive civil disobedience, while Foucault expresses skepticism about human nature and insists we should engage in critique of institutions, such as the academy, for subtly perpetuating relations of domination (Chomsky and Foucault 1971). Some version of this psychological difference is abundantly clear in any number of movement contexts where strategy is being debated, engendering considerable misery for all involved. But I think we could uncover other psychological corollaries of nature-nurture perceptions if we used the wide range of techniques scientists have used to understand, for instance, left-right political differences.     

Political psychologists and others have illuminated a complex, idiosyncratic landscape of traits that correlate with this enduring, fundamental form of political division. A great deal of the results concern fearfulness, with right-wing political perceptions correlating with greater activity of the amygdala during risk-taking activities (Schreiber et al. 2013), stronger tendency to interpret ambiguous facial expressions as threatening (Vigil et al. 2010), greater startle responsiveness with exposure to loud noises (Oxley et al. 2008), heightened general physiological arousal in response to threatening imagery (Dodd et al. 2012), and a greater self-reported fear of death—with this being the strongest correlation of any psychological variable in a large meta-analysis (Jost et al. 2003 ). Then again, a number of the correlated traits have nothing to do with fear, such as use of body language in conversation (Carney, Jost, and Gosling 2008) and IQ scores (Hodson and Busseri 2012; Kemmelmeir 2008; Stankov 2009). 

It is beyond our current scope to place these myriad results into a unitary explanatory framework. Hibbing, Smith, and Alford (2014) provide a more comprehensive review of politically correlated traits, and the thematic convergences of numerous studies. I only reference these diverse findings to indicate the ambitious theorizing they could inform (e.g. Schroder 2022),  and thus the possibility of applying these wide-ranging techniques—such as brain imaging during experimental game play, psychological self-report, and physiological measures of stress responses, to name a few—to partisans of scientific debates, such as the interminable nature-nurture dialogue. Does the anterior cingulate cortex activity levels of people who find absolute social constructivist claims convincing—for instance, that Hopi people do not possess a sense of linear time (Brown 1991)—differ from those who find such notions ridiculous?   

The above example is an instance of left-right psychological difference; ACC volume (Kanai et al. 2012) and activity (Amodio et al. 2007) is greater among left-leaning study participants. Such findings are, of course, guided by hypotheses, and I'm not certain there's any particular reason to speculate that the ACC—which is involved in everything from modulating aggression to disengaging from habitual behavioral responses—varies with one's perspective on nature-nurture dialogues. However, I would like to advance a tentative hypothesis of temperamental divisions within what we broadly describe as science, guided by existing findings.    

A psychological trait syndrome apparent in the exercise of political power 

This hypothesis emerges from what has always struck me as an unfortunate gap in the empirical foundations of science: there is no discipline which requires direct experience with attempting to achieve political outcomes in order to claim expertise. Scientific proclamations about the social leveling mechanisms of hunter-gatherers or the intergroup aggression of chimpanzees require somebody to actually go spend time around hunter-gatherers or chimpanzees, but statements about the amenability of political systems to vaguely specified forms of “pressure”, such as the one found in the scientists' second warning to humanity, do not require any such fieldwork. I do not think I would've formulated the hypothesis I am presenting here absent direct experience with political processes. 

This hypothesis was formulated by observing the personalities involved in political conflicts. I was attempting to understand why people with certain temperaments and worldviews consistently seemed to inhabit certain specialized social roles. It consists of a schema of four covarying dimensions of perceptual difference. These four perceptual oppositions are hierarchical vs. egalitarian, established vs. innovative, corporeal vs. symbolic, and definitive/schematic vs. ambiguous/particular.   

The first is perhaps hardly surprising, as this is the most fundamental and enduring form of political division across human contexts. Left and right have no particular established meaning, even among—perhaps particularly among—specialists (Jost et al. 2003; What Is Politics 2020) but the most recurrent definition is that leftism is a politics of egalitarianism and rightism a politics of hierarchy (Jost et al. 2003; Tuschman 2013 ).

Politically, the established vs. innovative opposition is frequently expressed as tradition vs. progress. The terms conservative and progressive are used interchangeably, for better or worse, with right and left. Robinson et al. (2015) show significant differences in references to the past and future based on political outlook.  

The third perceptual binary I hypothesized was in mode of instantiation of a given worldview, a corporeal vs. symbolic axis. As numerous studies reviewed by Haley and Sidanius (2005) describe, police and military employment strongly predicts a hierarchical worldview. The metrics of worldview used in these studies, such as the Social Dominance Orientation Scale and the Right Wing Authoritarianism Scale, also negatively predict an environmental worldview (Altemeyer 1998). Likewise, professional specializations concerned with the administration of economic activity, involving the direct manipulation of the physical world tend to have more hierarchical and extractive/utilitarian worldviews (Gambetta and Hertog 2016). 

Correspondingly, those who specialize in the production of knowledge and culture tend to have more egalitarian and ecological worldviews. My experience of conflict around environmental issues was thus of one side contending in the symbolic world while the other contended in the physical world. Those advocating for ecological protection might write books, give speeches, hold protests, and issue dire warnings. Those seeking maximum resource extraction, on the other hands, tended to be the ones in a position to arrest protestors or to actually occupy the industries (and their regulatory agencies) which environmentalists wrote to or protested. Naturally, this correlation between worldview and social role had a great deal to do with the final outcomes of these conflicts.          

Our final opposition is between the definitive and schematic vs. the ambiguous and particular.  I cannot overstate the extent to which a certain epistemic paralysis, born of incessant reference to reality's inherent ambiguities and local vicissitudes, is profoundly disruptive of political organizing efforts. Bizarre as it may sound, contemporary political groups fail most often through self-reflection of a decidedly academic nature. Do we really know if we should try to stop this coal export facility from being built? Or is our conception of climate change really just a reductionist way of seeing the world that fails to account for equally valid indigenous cosmologies which transcend western notions of causality? Should we really try to implement a local network of food distribution production and distribution? Or is what we think of as food an inherently colonial construct rooted in soil fertility metaphors designed to justify male domination? Of course, asking questions of this broad nature can be productive, but refusing to act because one can never fully answer them guarantees passivity. Think of Foucault repudiating Chomsky's organizing, citing our inability to ever really disentangle our perceptions from their unconscious frames of reference and socially conditioned schematics. If real estate developers and police departments experience a similar paralysis, I am unaware of it. 

An inherent relationship between the third and fourth of these opposing perceptual modes should be noted. I believe a tendency to physically manipulate reality correlates with abstract, schematic perceptions of reality, because “Legibility is a condition of manipulation” (Scott 1998). Some level of reductionism is required for any instrumental action, and excessive reductionism likewise seems to correlate with the kind of excessive, catastrophic manipulations of the physical world undertaken by those in power. 

Temperament, worldview, academic specialization, and power 

Gambetta and Hertog (2016) document correlations between academic specialization and affiliation with different types of terrorist organizations, over a number of decades and a range of political contexts, involving hundreds of participants in groups ranging from the Red Army Faction to Islamic Jihad. The correlation is particularly striking among engineers, who are vastly overrepresented in right-wing terrorist groups of the islamic and western varieties. Engineers are seventeen times more likely to be found in their sample of islamic terrorists than would be expected if they were found in proportion to their presence in the adult male population. 

Gambetta and Hertog utilize data from the European Social Survey (ESS) on 11,183 male graduates, from 17 western European countries, with results that indicate the existence of three statistically robust, temperamentally distinct populations in the academy. In addition to presence in terrorist groups, they report correlations between academic specialization and ESS data on three traits strongly associated with political perception: disgust, need for closure (NFC), and in-group/out-group distinction. Higher measures of any of these three traits is predictive of right-wing political perception. 

The first of these populations practices engineering, medicine, law, and business, economics, and administration. These disciplines are overrepresented in right-wing terrorist groups and, as we would expect, their practitioners have higher than average scores on the three traits. The next of the three populations practice math and physical science: this population is not significantly present in terrorist groups and has societally average scores on these three traits. Finally, we find particularly low scores on these traits—predicting left-wing political perception—among practitioners of social and psychological sciences and the humanities, as we would expect from their disproportionate representation in left-wing terrorist groups.         

These findings militate for a program of concerted inquiry into the kinds of minds found in various academic departments. I relate them to my hypothesized perceptual differences to indicate potential paths of inquiry.  

The egalitarianism vs. hierarchy component of my psychological schema is explicitly present in the left-right aspect of Gambetta and Hertog's findings. However, a motivation for my hypothesis was to account for recurrent uniformities between political systems with divergent ideologies. Something to note about the disciplines with a right-wing bias is that they are heavily overrepresented in institutions of power in modern societies, regardless of official ideology. Business and law are the most common degrees among US policy makers, while Stalin complained that all the socialists had been replaced by engineers (Curtis 1992). Scott  (1998) calls the worldview of those who orchestrated “the most tragic episodes” of late 19^th and 20^th century state power Authoritarian High Modernism, noting this worldview unites Lenin, Robert McNamara, and the Shah of Iran. “As a faith, it was shared by many across a wide spectrum of political ideologies. Its main carriers and exponents were the avant-garde among engineers, planners, technocrats, high-level administrators, architects, scientists, and visionaries.”

I believe Scott is describing one extreme of my hypothesized psychological trait syndrome—a correlated preference for hierarchy, established knowledge, and corporeal action based on schematic abstraction—in positions of power in technological mass societies, while Gambetta and Hertog are identifying this same trait syndrome in the academy. Asking what kinds of people exercise what kinds of power allows us to transcend the chaos inherent in taking overtly stated ideologies at face value. The same people, for instance, comprised the security services of both the USSR and Russia in its current hyper-capitalist phase. Surely attributing some significance to this fact makes more sense than describing their system of mass coercion as a fundamentally left-wing phenomenon until 1991, and right-wing thereafter. It would make a great deal of sense if, for instance, personnel of the FSB scored higher on the Right-Wing Authoritarianism scale than the average Russian, even when it was called the KGB. This prediction emerges from the fact that high RWA predicts, for instance, hyper-violent partisanship of both Israeli and Palestinian nationalism (Tuschman 2013). In other words, it predicts a fundamental temperament that consistently manifests in the killing and caging of others, while contextually varying in the rationale for doing so.    

Our three populations exhibit continuous variation with respect to the established vs. innovative component of the hypothesized trait syndrome. The four disciplines associated with institutional power (engineering, medicine, law, and business, economics, and administration) concern the application of existing knowledge, and so fall on the established end of the continuum, as expected. Likewise, in Gambetta and Hertog's sample, this population scores high on a subcomponent of the Need for Closure scale called Traditionalism, which measures preferences for social arrangements based on traditional family and religious authority. Math and science are concerned with innovations and undiscovered terrain, but they build on previous work, meaning knowledge can, in fact, become established. A recurrent frustration with many permutations of social science is a lack of truly definitive conclusions, and thus of progress of any kind (e.g. Tooby and Cosmides 1995).    

The corporeal vs. symbolic axis is clearly present along our proposed disciplinary-perceptual continuum, again displaying continuous variation. The institutional power-associated disciplines are fundamentally concerned with the manipulation of corporeal reality. Science and mathematics is a dynamic integration of the accumulation of pure knowledge and its physical application. The only physical manifestation of many varieties of social science and humanities is the publication of books.

Likewise, the presence of the definitive/schematic vs. ambiguous/particular axis in academic specializations is conspicuous. As we have already noted, certain elements of social science and humanities are characterized by a sort of general epistemic fatalism, while physical science and mathematics comprise a dynamic synthesis, attempts at reductionism and systematic knowledge conducted with a measure of epistemic humility. Engineering and other technocratic disciplines are notorious for the cheerful certainty of their predictions, sometimes in the presence of screamingly apparent contradictory evidence.

Scott calls Authoritarian High Modernism “a faith that borrowed, as it were, the legitimacy of science and technology,” characterized by “unscientifically optimistic” assessments of its own technical mastery. Anyone familiar with the rosy forecasts found in Environmental Impact Statements for new fossil fuel infrastructure, or US military strategies for increased troop commitments, knows the distinctly schismatic quality this “synoptic view of a selective reality” possesses. Towering infernos and enraged violence are disregarded in favor of confidence intervals and first derivatives—a model is asserted to possess greater validity than reality itself. Gambetta and Hertog note that “engineers strongly believe that science can solve environmental problems” to a much greater degree than scientists themselves, and echo Scott's description of their worldview, noting that “engineering students, like followers of text-based religions, rely …  on answers that have already been given.” Likewise, Scott notes  Lenin's belief in the “objective and logically inevitable” judgements of the Bolsheviks, while Gambetta and Hertog describe a jihadist engineering student's YouTube videos, where he explains how the curves produced by two simple formulas represent the fight between al-Qaeda and its enemies, offering unequivocal mathematical proof that al-Qaida will prevail.

These correlated psychological tendencies indicate that the broad construct of science  encompasses distinct epistemologies and patterns of conduct. Understanding these distinctions is central to understanding the nature of power in the societies in which science is conducted, and thus its ability to solve problems. 

Conclusions

Two concrete scientific initiatives have been proposed in the course of this discussion, the first of which informs the second. The first is the identification of climate and ecological tipping points, and their articulation as thresholds for the failure of the current scientific and political approach. The literature on tipping points is currently quite diffuse; an integrated assessment of the many potential runaway processes which have been identified, and an effort to establish a broad consensus on how to determine when they have been triggered, would be a major development. This would require political will on the part of scientists: as Kemp (2022) and others have noted, the IPCC simply refrains from modeling feedbacks, and this has significantly influenced scientists' perceptions of what constitutes “mainstream” climate analysis. 

The establishment of quantitative physical thresholds would provide a metric for evaluating the strategies thus far employed in fields like climate science and conservation biology for averting catastrophe. The core premise of these strategies—that policymakers embedded in the current systems of power will someday rationally respond to the warnings and recommendations of scientists—would then gain the status of a falsifiable hypothesis. Ideally—although this obviously involves some level of discretion—the threshold for declaring the current paradigm a failure would exist when tripping points are imminent, but somewhere before their actual triggering. 

There are certainly examples of the scientific acknowledgement of the intransigence of extant political systems. For instance, Gardner and Bullock (2021) state that, because of the climate crisis, the traditional goals of conservation science are no longer physically attainable. Noting that the field relies on “false assumptions of how to catalyze transformative change,” the authors suggest their scientific discipline transform into one they call survival ecology. 

Like Gardner and Bullock, here I am advocating that the question of how to avert global catastrophe be treated as an open-ended scientific one. A meaningful scientific inquiry into why the system has thus far been unresponsive scientists' warnings would, of course, be a vast and multifaceted one. We could note any number of conceivable starting points for this journey, such as the long history of those in power exhibiting non-responsiveness to other existential crises (Tuchman 1984), and the tendency toward collapse of hierarchical mass societies throughout the archaeological record (Tainter 1988; Scott 2009; 2017). 

However, the second concrete scientific initiative presented here concerns a particularly central aspect of this inquiry: the underlying cause of the landscape of variable perceptions climate and ecological scientists find themselves in. In particular, it proposes to search for answers to two very fundamental questions. One, why is it that when physical scientists are communicating their results to policymakers, they are always communicating with people from different disciplinary backgrounds? Two, why do people from different disciplinary backgrounds—even if those backgrounds involve a great deal of technical reasoning—vary in the ways they do in their perceptions of environmental issues? 

The proposal is that the methods of political psychology, which have produced a body of robust results on populations outside the academy—with findings from brain imaging, psychological questionnaire, physiological, and behavioral studies corroborating one another—be applied within it. A number of studies could be directly replicated with academics as their subjects. In addition to the dimensions of individual variation found within the existing political psychology literature, I have proposed a syndrome of four correlated psychological traits, extending current findings, which I believe warrants study. 

It seems likely that some people will have a broad aversion to the study of any psychological differences in the academy. This is perhaps why there aren't a large number of such studies, despite that each one I am aware of has, in fact, found interesting and meaningful differences. The existence of a psychological corollary to, for instance, a particular divide within a discipline, or an entire discipline, might be perceived as diminishing the credibility of that discipline or partisans of that divide. But of course, the fact that certain kinds of minds are attuned to certain aspects of reality, or certain kinds of explanation, doesn't invalidate them. Certain personalities are more or less attuned to threat—obviously this does not invalidate the objective reality of both threatening and benevolent circumstances. I believe that scientific knowledge would be significantly advanced by a better understanding of the different kinds of minds that produce it.     

According to West (2017), the pace at which complex systems generate new crises is commensurate with their scale. As global human society generates new crises ever faster, it appears to be outpacing the ability of science to adapt and respond (e.g. Kirchner 2022). The need to adapt to—and meaningfully influence—these changes has engendered numerous meta-scientific proposals, to which these two are added. One, that we treat dominant assumptions about how to address the ecological crisis as falsifiable hypotheses. Two, that science applies its methods for understanding belief systems, and their underlying temperaments, to itself. 

Gardener Comments

Josh Randall:
This paper presents two central ideas, that certain disciplines of academia either produce or attract people with specific psychological traits along four axes and that this idea should belong to a line of inquiry that focuses on learning about the politics of the world using empirical data of immense social change. I think this first idea could benefit from an explicit comparison to other social tests' categories in addition to their results as described in the manuscript. The second idea is given almost no additional information and is possibly a more important question, especially as it critiques several currently existing disciplines without fully explaining where this empirical study of collapse could come from.

Ted Wade:
The article first says that we need a scientific basis for deciding when current strategies to mobilize against ecological collapse have failed and we must try some other, unspecified course of action. It brings up tipping points, but maybe we have already passed one or more. At any rate, changing strategy after a tipping point seems futile. The author then hypothesizes 4 psychological trait dimensions, and offers a combination of existing research and speculation in applying these to 3 populations: entrenched authorities, violent extremists, and scientists. Along the way, the writing is erudite and interesting, with some very quotable lines, but I found it hard to track where it was going. The Conclusion section mostly does not follow from the body, but does offer that the "trait syndrome" discussed might help with finding a path to survival. That's all the cash value that I see. What I would like to see would be a reason to pursue this research instead of the many other explanations of our societal paralysis, because we don't have much bandwidth or time remaining to get this figured out. The three populations mentioned all have a role in what might happen, but if we did the research and nailed down their positions on the trait dimensions, what would we do with that? We would still need to learn how those groups can be made to agree on actions with huge and uncertain costs and benefits. I realize that is an impossible ask for a single call-to-action paper. 

Anonymous1:
The article is vague, and it can be summarized into a complaint: our ecological concerns do not translate into political action. It transpires some resentment against the technical professions, and in fact is a clear example of why people in government and the corporate world, even if sympathetic to the concerns of the author, do not act on their advice: the advice itself is not produced in actionable form! Shall we apply a carbon tax? To which countries? How do we avoid the free rider problem? These are the problems to be addressed. And they shall be addressed with both scientific AND strategic soundness. No hint of understanding of particular interests and adversarial relations or real world politics is found in the entire article. No references to serious political science literature, institutional mechanisms, or any other relevant social science is offered. Even the most important issue (the tragedy of commons in a world with multiple and adversarial countries) is not discussed. 

Mark:
Overall, I think this is well situated and thoughtful. With absolutely no pressure to cite this work, the author might be interested in Almaatouq et al. 2022 and many of the papers it discusses which have a strong thematic overlap to this work.

Dr. Payal B. Joshi:
The article is well-presented and the topic is relevant and timely. I recommend publishing the article as it is.

Joe R:
1. Does the article contain novel ideas that have the potential to advance science? 

Asking what kinds of biases inform our thinking is an important part of advancing human knowledge. Noticing said biases in oneself is a rare skill, and often crucial to avoiding mistakes in science. Studying those biases scientifically isn't exactly a novel idea, but there's definitely room for a lot more growth in that arena. Philosopher, know thyself; scientist, study thyself. "We must apply our methods for understanding belief systems, and their underlying temperaments, to ourselves" (p. 11) is, out of context, an excellent piece of advice for anyone, especially those dedicated to the systematic, evidence-based study of the ground truths of reality. 

Unfortunately, the author had to go and add context. 

2. Does the article include adequate justification for its ideas and how they could advance science? 

It's hard to find an actual fully-stated hypothesis in any of the sections except maybe the one starting on page 6. But based on past replication failures, when someone says "Aha! I have found the four fundamental axes on which political opinions vary!" they tend to be overhyping at best and tragically mistaken at worst. And that's for n=10,000 giant personality test studies, compared to this paper's n=0 personal experience. It is especially suspicious when the author might have an incentive to label the side of the axis they're on as the Good and Righteous side. Consequently, I am highly skeptical of these claims. 

3. Does the article contain high-quality writing?

It is hard to extract a unified or coherent claim from this paper. I scoured the document for a core claim or concrete proposal, but failed to find one I could clearly restate. It reads more like a passive-aggressive manifesto than a scientific paper, and I am getting a hard-to-describe suspicion that the author is trying to say "academics are biased and can't agree, THEREFORE, my intellectual opponents disagree with me because of their evil angry fearful monkey brains" in scientific-sounding language, hoping we won't notice the shift from a claim with decent evidence to a claim with basically none. And I have absolutely no idea where they're going with the claims about "powerful" disciplines sharing psychological traits. Are they looking to overthrow the math majors?

The best I can manage to extract from this paper is their original point about climate studies, which seems to go "nobody in power listened to us, it is time for REVOLUTION!" I see precisely zero actionable or realistic proposals for such, let alone any assurance that trying won't result in a giant flaming disaster. I expect that if this article were to be published and, against all odds, taken seriously by mainstream science, it would only precipitate a massive flamewar in which academics accuse one another of hierarchical/established/corporeal/definitive/schematic bias or their counterparts, declare one another shills for the powerful, and somehow manage to blame it all on the engineers and business majors. (Okay, so not that different from present dialogues). 

There might be some kernels of truth buried in this paper - and it cites some extant studies that sorta point in the same direction - but accusing entire disciplines as being inhabited by hidebound loons is the sort of thing that ought to be handled carefully. I'm seeing a lot more heat than light, and I can't recommend a paper this bloated and incendiary as kindling for a bright new paradigm. 

Enrique Muñoz:
The article is well structured and easy to follow for non-specialists. I merit the self-reflection of the author which, undoubtedly, we all should do more often.

References

1. Amodio D. M., Jost J. T., Master S. L. & Yee, C. M. (2007) Neurocognitive correlates of liberalism and conservatism. Nature Neuroscience 10:1246–47.

2. Baccini, A., Walker, W., Carvalho, L., Farina, M., Sulla-Menashe, D., & Houghton, R. A. (2017). Tropical forests are a net carbon source based on aboveground measurements of gain and loss. Science, 358(6360), 230–234. doi:10.1126/science.aam5962

3. Brown, D. E. (1991) Human Universals. McGraw-Hill.

4. Carney, D. R., Jost J.T. & Gosling, S. D. (2008) The secret lives of liberals and conservatives: Personality profiles, interaction styles, and the things they leave behind. Political Psychology 29: 807– 40. 

5. Chomsky, N. & Foucault, M. (1971) Transcript of debate on human nature: Justice versus power. https://chomsky.info/1971xxxx/

6. Curtis, A. (1992) Pandora's Box: A Fable from the Age of Science. Pt. 1: The Engineers Plot.

7. Dodd, M. D., Balzer, A., Jacobs, C. M., Gruszczynski, M. W., Smith, K. B. & Hibbing, J. R. (2012) The political left rolls with the good and the political right confronts the bad: connecting physiology and cognition to preferences. Philosophical Transactions: Biological Sciences 367(1589):640-49. 

8. Gambetta, D. & Hertog, S. (2016) Engineers of Jihad: The Curious Connection between Violent Extremism and Education. Princeton University Press.

9. Gardner, C. J. and Bullock, J. M. (2021) In the climate emergency, conservation must become survival ecology. Front. Conserv. Sci. 2:659912. doi: 10.3389/fcosc.2021.659912 

10. Green, A. S. (2020). Killing the Priest-King: Addressing Egalitarianism in the Indus Civilization. Journal of Archaeological Research, 29(2), 153–202. doi:10.1007/s10814-020-09147-9

11. Haley, H. & Sidanius, J. (2005) Person-organization congruence and the maintenance of group-based social hierarchy: A social dominance perspective. Group Processes and Intergroup Relations 8(2):187-203.  

12. Hansen, J., et al. (2008) Atmospheric CO2: Where should humanity aim? The Open Atmospheric Science Journal 2:217-231. DOI 10.2174/1874282300802010217

13. Harris, N. L., Gibbs, D. A., Baccini, A., Birdsey, R. A., de Bruin, S., Farina, M., … Tyukavina, A. (2021). Global maps of twenty-first century forest carbon fluxes. Nature Climate Change, 11(3), 234–240. doi:10.1038/s41558-020-00976-6

14. Hibbing, J. R., Smith, K. B., & Alford, J. R. (2014) Differences in negativity bias underlie variations in political ideology. Behavioral and Brain Sciences 37:297–307.

15. Hodson, G. & Busseri, M. A. (2012) Bright minds and dark attitudes: Lower cognitive ability predicts greater prejudice through right-wing attitudes and low intergroup contact. Psychological Science 23(2):187-95. 

The essay advocates for diverse approaches in presenting a researcher's scientific contributions in a project. Taking inspiration from sports journalism and its visualization of football players' data, the essay suggests that a radar plot, incorporating CRediT contributor role data, enables multiple authors of a scientific paper to illustrate their contributions in a more specific manner. The suggested method, though subject to bias reporting, pays credit to different aspects of a research project, from conceptualization, analysis, administration, to writing and revising. It not only enables both academics and lay readers to better understand the considerable amount of work required in every project but also calls for the need to employ diverse viewpoints in science.

Introduction

Scientific profiles of researchers use single-number metrics, such as publications, citations, or h-index, to present an overview of the authors’ impacts. Online scientific profiles, such as the Web of Science or ORCID, primarily utilize single-number metrics for visualization. For example, the Web of Science provides a plot showing times cited and publications over time on a researcher’s profile. While it can be argued that single-number metrics offer a simple yet elegant portrayal of impacts, the glaring problems of metric manipulation and substituting metrics for actual quality are damaging to the scientific community [1,2]. Drawing inspiration from football analytics and data visualization, this essay aims to explore a different method to visualize a researcher’s scientific contributions.

Why football analytics? In recent years, statistical analysis has indeed become an integral part of performance analysis in various sports, such as basketball and baseball. The book Moneyball: The Art of Winning an Unfair Game, and its movie Moneyball (2011), has popularized the concept of using statistical analysis to gain an advantage in sports, particularly for teams with lesser financial power. However, the success of statistics in basketball or baseball does not translate well to football due to the low-scoring nature of the game [3]. Hence, the scoreline does not necessarily reflect the performance of the team or a player. Moreover, as a data point often registers an on-ball action, off-ball actions are often neglected [4]. Despite the shortcomings of football data, football writers have employed different methods and visualization techniques to demonstrate a player’s contribution. The visualization used in this article is the radar chart.

Plotting the radar chart with CRediT data

The radar chart, more commonly referred to as a pizza chart, is extensively utilized in football analytics [5]. This chart offers a comprehensive way to summarize various statistics of a player during different phases of actions in football—be it attacking, defending, or possession. Here is an example of a pizza chart:

In science, while there is no consensus regarding research phases, the CRediT – Contributor Roles Taxonomy — is a common method for researchers to declare their contributions in a paper [7]. Presently, numerous publishers have adopted CRediT, including Cell Press, De Gruyter Open, Elsevier, SAGE, Springer, and others [7]. Additionally, free services such as Rescognito (https://rescognito.com/) also utilize CRediT to enable researchers to acknowledge their contributions and those of their peers. Rescognito is also using CRediT data for visualization, though  the visualization tools are still in the beta stage. CRediT’s 14 contributor roles are clearly defined and easy to understand. However, beyond a declaration in the manuscript, there seems to be no apparent utilization of CRediT.

Hence, in lieu of football statistics, this essay attempts to visualize researcher statistics using the 14 contributor roles from CRediT. These roles were grouped into 4 categories representing phases of research: Conceptualization, Analysis, Administration, and Writing.

Figure 2 provides a visualization of CRediT roles for an early career social scientist who published 31 articles from 2018 to 2023 (please note that Writing – original draft, Writing – review & editing, Conceptualization were shortened respectively changed to Writing, Revising, Concept for clarity). 

As depicted in Figure 2, Researcher A's primary contributions lie in writing and revising manuscripts, as well as responsibilities in data curation and investigation. To further illustrate the potential of the pizza chart, I have visualized a comparison between Researcher A and a hypothetical senior Researcher B in In Figure 3 (for visual clarity, I shorten Project Administration to Admin, and Funding Acquisition to Funding). As you can see, Researcher B assumes the role of the senior researcher and makes significant contributions to supervision and conceptualization in the studies.

A pizza chart can enhance the presentation of several aspects. Firstly, it effectively illustrates an individual scientist's contribution to a paper. This feature is particularly valuable for researchers in large research groups or fields where collaborative, multi-authored papers are common. Additionally, the chart serves as a useful tool for visualizing how scientific contributions evolve over time. For example, as a young social scientist, the focus was primarily on manuscripts and data (like Researcher A). In contrast, senior researchers tend to shift their contributions towards the administrative phase or key aspects of the study, such as the main concept or methodology (like Researcher B). Variations of the pizza chart allow for comparisons between researchers and, with enough data, between a researcher and the field's average. Moreover, the chart can also represent author positions (first, corresponding, or co-author) and replace the researchers' images in the inner circle.

Limitations and Closing Remarks

Indeed, there is no one-size-fits-all method to fully capture the beauty and complexity of research or a person’s career [8]. Therefore, we shouldn’t limit our presentations of science to single-number metrics or only a handful of visualizations based on those metrics. I argue for diversity in presenting scientific contributions, providing examples like the radar chart and CRediT contributor roles data.

For a complex activity like science, it is quite peculiar to observe the limited ways scientists publicly present their work. The development of scientific journals and their ecosystem can be attributed to both historical and practical reasons. However, as Adam Mastroianni argued, the current ecosystem has its limitations and, in a certain sense, has failed us [9]. Even though the journal impact factor was not initially intended for rating science, the precise number has failed to facilitate a healthy system, despite its mathematical appeal [10]. In “The Seduction of Clarity,” philosopher C. Thi Nguyen argued against the illusion of mathematical beauty. Using the standardized value systems of bureaucracies as an example of how clarity can exploit our cognitive vulnerabilities, he pointed out: “Quantified systems are, by design, highly usable and easily manipulable. They provide a powerful experience of cognitive facilities. It is much easier to do things with grades and rubrics than it is with qualitative descriptions” [11]. Similarly, single, easy-to-use, easy-to-understand metrics like journal impact factors, citations, or publication numbers make things simpler for everyone. However, in our pursuit of the clarity these numbers provide, the scientific community often sacrifices a qualitative, nuanced, and complex understanding of science.

In football, there exists a clear cognitive dissonance between what one observes on the field and the actual results. When Brazil experienced a humiliating defeat on their home turf during the 2014 World Cup against Germany, the underlying data presented a different picture, with Brazil actually superior in several aspects (shots, shots on target, possession, dangerous attacks, to name a few) [12]. Yet, Brazil lost 1 – 7 to Germany in that semifinal. This disparity in football necessitates that data analysts exercise creativity and caution when interpreting football data. Likewise, we can adopt diverse perspectives to discuss and understand science.

This essay advocates for diversity in the presentation of scientific contributions. The once favored single, easy-to-use, and easy-to-understand metrics have had their moments. It's time to move away from relying solely on these metrics and begin developing new languages to represent ourselves. My suggestion is to visualize the CRediT data. However, this is just one approach; there are other possibilities. Currently, the availability of CRediT data poses a significant challenge in implementing this method. Publishers like Scopus, Web of Science, and others may need to consider providing it to the public [13]. Additionally, we should be mindful of potential biases arising from the self-reported nature of the CRediT data when discussing the results.

The scientific community has been calling for changes in how we use metrics. However, finding a suitable replacement remains a challenge. This article presents an attempt to adopt a type of chart from football analytics for research purposes. This marks an early effort, and there are still many shortcomings. Nonetheless, as science promotes increased openness and transparency, a new strategy is imperative for a cost-effective and more informative outcome [14,15].

Acknowledgement: The authors would like to express their gratitude for colleagues who have shown tremendous support: Manh-Tung Ho, Hong-Kong T. Nguyen, Thanh-Huyen T. Nguyen, Ngoc-Thang B. Le, Hung-Hiep Pham, Alex Thao P. Luu.

Supplementary Information: R code used to generate figures is available in SI. 

Gardener Comments

Note: The author’s responses are interleaved throughout this section. 

Anonymous 1 (9 years of experience as journal editor):
I like the approach. However, I would like to see more examples from other researchers. I suggest you pick a top researcher with a large number of publications, so one can draw a plot based on many data points. If you picked two researchers, you can compute a contrast version of the plot where you subtract one's scores from the other, showing their relative preferences for roles. Ideally, there would be a website where users can look up any researcher (e.g. by ORCID id) and see their plot, as well as the contrast with another researcher.

Manh-Toan Ho: Thank you for your comment. I have provided a comparison plot between two researchers. 

Andrew Neff (Psychology PhD):
I agree that this visualization and the CRediT system in general provide a more accurate metric of experience than authorship alone, I think it could improve hiring and retention. I also agree a lot depends on the availability of CRediT data, and wonder if a platform like Google scholar would allow scientists to add this info post-publication if a journal doesn’t naturally list it. I also am starting to wonder whether quantifying each CRediT category could also be helpful, and possibly even necessary, if some (lazier) people are going to use this info in hiring.

Ahmad Ozair (MD, clinical researcher with experience in bibliometrics and scientometrics research):
The article presents an interesting method of visualizing scientific contribution of authors to peer-reviewed publications. Certain considerations, however, limit the rigor of the work and of its external generalizability: 

1. The author's own radar plot is drawn from 15 papers - which is a very small sample. They should strongly consider expanding this to papers beyond 2020, as this is not too challenging. 

Manh-Toan Ho: I have updated the data, as well as adding one comparison plot. 

2. Author says that "For a complex activity such as science, it is rather strange to see the limited ways scientists present their works." - saying this seemingly negates the entire field of bibliometrics and scientometrics which have extensive visualization options developed over decades.

Manh-Toan Ho: “For a complex activity like science, it is quite peculiar to observe the limited ways scientists publicly present their work.”

I have added the word publicly to clarify what I want to say in this sentence. My point is that the applications of “entire field of bibliometrics and scientometrics which have extensive visualization options developed over decades” are somewhat limited to a niche area rather than a widely adopted standard. When we describe someone’s research career in public, we typically only use numbers of publications, h-index or JIF to illustrate. That is something I find strange.

3. One limitation that may be added is the lack of rigor in the CREDIT listing itself by researchers. In many labs and groups, CREDIT is an afterthought and authorship order continues to be the primary point of discussion. Additionally, despite efforts, the majority of papers published lack complete CREDIT-appropriate listing, which limits automatic analyses. 

Manh-Toan Ho: I added this as a limitation to the discussion. However, this is a general problem when a new standard is being introduced, I think as many publishers are using CREDIT, it will soon be the norm and authors will have to take it more seriously. Moreover, I think the effort to use the CREDIT data will also improve the general perception of CREDIT as well.

4. Please consider adding to the discussion, the work being pioneered by the RESCOGNITO team. They have a great visualization tool using CREDIT already.

Manh-Toan Ho: I have mentioned RESCOGNITO in the article.

Josh Randall:
This manuscript describes an interesting means of visualizing researchers' contributions to the creation of publications. It takes advantage of the use of statistics in sports analysis to develop a system for showing how authors can contribute in different ways for the final publication. Two criticisms that I have are: a different form of visualization does not fundamentally move away from the publication focused description of a researcher's merit and so many of the components of constructing a specific paper such as administration in a specific lab group may not translate well in any other setting. The most useful aspect of this visualization would be seeing how one specific researcher changes through time, but I do not think this solves the fundamental problems with using H index to determine employability or success of a researcher.

Manh-Toan Ho: Thank you for your comments. 

I think your concerns are with the systematic ways of doing things in academia, and of course, that problems can’t be solved with just using a new way of visualization. Scientific communities are constantly finding new ways, such as Seed of Science, so I think the accumulation of all the efforts will eventually lead to meaningful changes.

Duy Nghia Pham:
1. I think the “Software” role should be placed in Analysis, not Administration.

Manh-Toan Ho: Thank you for your suggestion. I have switched Software to Analysis accordingly.

2. In figure 2, all 14 roles should be visualized (even if some are zeros) for a complete picture of one's contribution and easy comparison with other researchers.

Manh-Toan Ho: I have updated the data and visualized all roles.

3. While the Credit data might be missing, the author positions (first, last, co- added up to 100%) are available on Scopus or WoS, which provide similar information on the contribution. These 3 roles can be plotted in radar chart too. The inner circle (instead of showing the footballer/researcher face) may be used for the percentage of corresponding role (e.g, as a pie chart).

Manh-Toan Ho: I have included this point in suggestions for future directions. 

Vamsi Makineni:
Although the suggestion is interesting, I believe the author should expand upon the limitations and how we can solve them. The idea of a pizza chart for science can easily be ill-defined as the components are rough estimates of contribution. Also, one's role can be deduced in the acknowledgement section. It is apparent that early-stage researchers will mostly help with data collection and gradually transition to conceptualization as they gain experience so I do not see how the chart will provide very important information. Who should it benefit? At most, I see this proposal benefiting talented early-stage researchers who help in conceptualization. Of course, unlike football there is no easy point system to assign for this criteria. A single idea could be momentous to solve a problem the research team has had, so how many points should this idea receive? I believe that the author should take these comments into consideration and rewrite the paper.

Manh-Toan Ho: My initial idea and preliminary results are simple: visualizing a researcher, and comparing two researchers. However, with enough data, I think there are more ways to show how a researcher ‘performs’. For instance, comparing how a researcher contributes in a year to his team’s average, or to the average of the authors with the same age.  I think the whole community can benefit, with large enough data of course. Back to my comparison with football or sports, the rise of data analytics in sports eventually brought sports into the modern age with new technologies in recruitment or managing players’ performances. So while I understand your skepticism, I think it is better to look at the bright side.

Dr. Payal B. Joshi:
The idea of using pizza chart/radar plots and utilizing solitary publishing record that features CRediT information is flawed. It is good to note that the author realized that when concluding as its quoted, “one size-does not fit all” however, going ahead with the idea seems odd or depicts a dilemma for a researcher. Do we really need pictorial representation of each scientists/researchers’ output? Any metric or chart that is developed on a flawed premise is not useful. There is a constant effort that universities do not push metrics to quantify science and research integrity is already at stake. CRediT information is provided by the authors which itself is highly biased data. Author has completely missed the gift authorship and increasing disparities in today’s collaborative papers. Most of the authors are listed to act as funders for open access fee funding. 

Even if we keep research integrity aside, the idea is flawed. Why? - The answer is by the time you make a pizza chart based on CRediT from publishing record of past 10 years, and unfortunately few papers face retractions - what will be the case then? Is not the author’s record questionable in the first place. If author has manipulated data due to which he/she faced retraction, how credible is the CRediT information? Has the present author thought about that and thus can accordingly plan the R code run. I urge the author to read literature of Vuong QH whose work is already cited in the present article known for his research integrity work.

My final take is rather than ideating on novel author charts, we need to think on ways to curb any new formulation of metrics as that makes us accountants and not scientists. We need to perform qualitative analysis and enhance post-publication discourse to truly attract talent and not manipulators. I, thus, do not recommend publishing this article, however, can be considered as a blog on Seeds of Science. Pardon me as I cannot be more positive for the present article.

Manh-Toan Ho: CRediT information, despite its flaws, is still one of the most orthodox and systematic information that publishers have on author contributions. So, my point is to find a better use of the information that we have, and already gathered. Also, I agree that we need better qualitative analysis and enhance post-publication discourse, however, it doesn’t mean that we should only focus on this aspect. Improving the way we talk about ourselves is also a way to move forward. Moreover, I think a lot of your questions are technical in nature, like the gardens of forking paths. There will eventually be so many micro-decisions to make when dealing with data, and I think the best way to deal with that is be clear of how you use your data and your methods.

Dr. Christian Thurn:
I like the idea and there are many great things in the article. The sentence that I enjoyed most was “For a complex activity such as science, it is rather strange to see the limited ways scientists present their works.” 

Please make an R-package or share the R code for the radar chart, so that others can use it as well. 

Manh-Toan Ho: I have provided snippets of code, as well as making the R file available in the supplementary information. 

I don’t understand the second sentence: “The visualization of the single number metrics on online scientific profile like Web of Science is also minimal with citation or publications trends over time.” Something weird with this sentence: “The radar chart (or more commonly, pizza chart) that has been widely used in football analytics [5].” 

All else is good – short and clear

Manh-Toan Ho: Thank you for your comments, I have revised the sentences for clarity.

Jay Matthews:
In general I'm excited about the prospect of visually depicting skills, perhaps as an alternative to CVs, and finding alternative ways of showcasing competency—especially as careers become more diverse and interdisciplinary. Should this paper be published, I would hope that it could serve as a starting point for what I think might be a more important discussion, that is the validity of these measures. I believe you've covered this well in your Limitations and Conclusions section. In short, a visual representation that presents a broader view of contribution and participation could be beneficial and seems like it has a place in the future, however, this paper feels a bit more like a first draft. The 

Difficulty of Measurement:

I started working on a project to develop measures (for which you've used CRediT data) in order to build a ranking system, but it quickly went awry as human measures have been known to do. The actual measurement of skills, competencies, and experience is interesting to me and could perhaps add value to the recruitment industry, especially in light of the 'academic arms race'. Measures that will accurately reflect not only participation and completion, but be an indicator for quality and skill is perhaps somewhere this can go? Not to mention the kinds of patterns we might find when comparing the frequency of roles fulfilled and demographic factors like age, gender, field.

Comparisons to other work: 

Perhaps a more holistic picture of the person could build on Dan Robles's Curiosume? Perhaps the same skills can stand and a dimensionality of skill level could be added? (e.g. my level of experience (or skill) vs how interested I am in it). DataCamp has a similar visualisation for all the platform members along core skills related to different career choices (Data Analyst vs Data Scientist). The 80,000 hours site has a lot of preliminary work for competencies and skill outlines for different professions too, if this were to be compared or used elsewhere. If nothing else, this idea could benefit from comparison to more than just football, though it is a nice and easy starting point reference! 

Potential concerns:

- Why is the CRediT data so scarce? What might be preventing people from using this? 

Manh-Toan Ho: Thank you for your above discussion, they are insightful and valuable. Regarding the scarcity of CRediT data, even though declaration of contributions is a common practice for most publications nowadays, I do not know many services that publicly declare to use this data for further analysis, or any well-known publications/reports that focus especially on CRediT data. There is Rescognito (https://rescognito.com/), which makes use of CRediT data. However, their system is based on what authors claim, not an automatic data scraper. This makes it hard for any application of CRediT data on a mass scale.

- Might there be negative effects on researchers' careers for making this more explicit? (e.g. If I eagerly participate in project to grow my career but not given opportunity to do more than admin, might this exclude me from future non-administrative opportunities) 

Manh-Toan Ho: The roles in CRediT are industry standard. At a certain point in a researcher’s career, he/she will contribute in that role. So I do not think there will be any negative effects.

- Is this CRediT system useful and, if so, widely applicable across different types of research? Does standardisation increase commodification?

Manh-Toan Ho: So far, I decided to use CRediT because they are the standards for most publishers, as I have written in the article. My idea is when most publishers actively collect this information, it would be easier to aggregate, collect and apply the data on a large scale.

- Productivity is a difficult thing to measure, especially when producing knowledge work. In football (and perhaps even in other jobs that are more clerical or unit-based), there is a very clear measure or success with ROI attached to it. For research, it'd be difficult to think of a good comparison. Knowledge work doesn't always have a direct correlation between time or training and the quality of output. Perhaps this will change! I appreciate the note in the limitations about how individual stats don't necessarily correlate with winning, however, there are enough knowns to build a prediction market around football outcomes. Is the same true for research? And if so, what would the predictions be of? What is the ROI of a single research paper, and can it ever truly be counted by itself, and in how many years of increments? This is not meant as a criticism of this paper or idea, but rather as an encouragement — because researchers are dependent on jobs, organisational rewards systems and performance reviews to make a living, I believe there may be a lot of value in taking this implicit measures and modelling them more explicitly because it may enable us to interrogate the methodology and assumptions more incisively! 

If all goes well, this practice could create great opportunities to identify and take on career development projects in areas you'd like to present more strongly (for instance the author of this paper on visualisation has a 2 on the visualisation spoke of the graph).

Étienne FD:
Although it would be beneficial to refine the idea further, radar charts for scientists' contributions are a highly interesting idea to explore. It's valuable to spend some time researching and communicating how scientists actually perform their work and how that can affect the way science is conducted.

References

1. Tran T, et al. (2020). Scrambling for higher metrics in the Journal Impact Factor bubble period: a real-world problem in science management and its implications. Problems and Perspectives in Management, 18(1), 48-56.

2. Vuong QH. (2019). Breaking barriers in publishing demands a proactive attitude. Nature Human Behaviour, 3(10), 1034.

3. Stevenson T. (2020). The difficulty of statistically analyzing match performance. 

4. Muller J. (2021). Introducing the no-touch all-stars. The Athletic. Retrieved from https://theathletic.com/3028824/2021/12/22/introducing-the-no-touch-all-stars/

5. Knutson T. (2017). Revisiting radars. StatsBomb. URL: https://statsbomb.com/2017/05/revisiting-radars

6. Koetsier, R. (2021). Percentile Radars/Pizza's. Getting blue fingers. URL: https://www.gettingbluefingers.com/tutorials/RadarPizzaChart

7. CASRAI. (2021). CRediT – Contributor Roles Taxonomy. CASRAI. URL: https://casrai.org/credit/

8. Vuong QH. (2020). Reform retractions to make them more transparent. Nature, 582(7811), 149.

9. Mastroianni A. (2022). The rise and fall of peer review. Experimental History. 

10. Seglen PO. (1997). Why the impact factor of journals should not be used for evaluating research. BMJ, 314(7079), 497.

11. Nguyen CT. (2021). The Seductions of Clarity. Royal Institute of Philosophy Supplement, 89, 227–255. doi:10.1017/s1358246121000035

12. Biermann, C. (2019). Football hackers: The science and art of a data revolution. Kings Road Publishing.

13. Vuong QH (2017). Open data, open review and open dialogue in making social sciences plausible. Nature: Scientific Data Updates. 

14. Vuong QH. (2018). The (ir)rational consideration of the cost of science in transition economies. Nature Human Behaviour, 2(1), 5.

15. Ho MT, Ho MT, Vuong QH. (2021). Total SciComm: A Strategy for Communicating Open Science. Publications, 9(3), 31

Abstract

The lack of time travellers visiting us may be seen as evidence that time travel is not possible. In this article, I argue an alternative explanation is that we are not economically important enough to our descendants to justify the costs of time travel. Using a cost-benefit analysis, I elaborate on this argument. I suggest that the major cost of time travel is likely to be the energy cost, whilst the largest benefit of time travel is knowledge which the present possesses, but the future has lost. Focusing on this benefit, I argue it is extremely unlikely that we possess a piece of knowledge which is sufficiently important to a future civilisation (system critical), but also has been lost by said civilisation. This is to say, we may not have been visited by time travellers because we are not important enough.

Introduction

One experimental way to prove the feasibility of travel forwards and backwards in time (i.e., time travel) would be to stage a large, broadcasted event. The location for such an event should be temporally significant—say, Greenwich—and the advertising for the event should be substantial. By substantial, I mean significant enough to be remembered for all of human history (call this the chrono-permanency criterion). Assuming such a feat of marketing is accomplished, matters turn to the event itself. The event should be a welcoming party for our descendants. It should serve as a message that at the particular moment in history that the event takes place—perhaps New Year’s Eve, 2023?—our descendants are to send a chrononaut to make contact. If no one shows up, this suggests time travel is not possible. If someone shows up, time travel would be demonstrated.

The above experiment builds from a logical argument: that if time travel were possible, we would be awash with time travellers. As we are not, time travel is not possible. Yet, this argument assumes something quite significant: that we, or any civilisation prior to ours, are worth visiting.1 In this article, I propose an economic cost-benefit analysis of time travel. I argue the main economic benefit which our descendants may receive via time travel is knowledge which we currently possess, but they have lost. Furthermore, this knowledge must be sufficiently critical to our descendants to justify the costs of time travel, which are likely to be dominated by energy costs. I posit that even assuming the energy requirements for time travel are met by a human civilisation in the future, it is highly unlikely that that same civilisation will come to depend on a piece of knowledge which we currently possess, but they have lost and cannot rediscover by other means. In other words, I argue that even assuming time travel is possible, our epoch is unlikely to offer any economic benefit to a future, time travelling civilisation.

The Economics of Time Travel to Date

Various economic perspectives on time travel have been proposed. A notable flurry of commentary, albeit tongue-in-cheek, began in 2006 as Tyler Cowen (2006) began musing on the economics of relativity, drawing on Krugman’s (2010 [1978]) Theory of Interstellar Trade (also see Cowen, 2008a).2 Cowen (2008a, 2008b) and others (Gans, 2008; Morehouse, 2012; Whitman, 2008) would build on these ideas further to incorporate labour within ‘transtemporal markets.’ The basic argument of these perspectives is that future epochs would be incentivised to use time travel to exploit the cheaper labour of the past, select epoch-specific legal structures which were beneficial for business, and embrace arbitrage opportunities for resources across time-periods. Gans (2008) is an exception insofar as they are critical of these arguments, and suggest that the clear, overwhelming economic advantage for a time traveller comes from using knowledge of the future to play markets of the past. Indeed, in a recent paper, Swinton (2021) argues that the stock market proves time travel to be impossible, because markets are less rational than would be expected if a flurry of actors with enhanced foresight were actively trading.

What is surprising about this literature (besides its existence) are the dual assumptions that a) time travellers would broadly operate within the same or similar economic system to that of the present; and b) that the past offers efficiencies for this economic system which the present would not. These assumptions are not wholly unjustified. For instance, if time travel were invented tomorrow, any would-be time traveller would have globalised capitalism as their framework of economic reference, and with that in mind, might indeed travel to the past to purchase Apple shares.

But if time travel is invented many years from now, be it centuries or millennia, there is no compelling reason to believe a time traveller of, say, the 41st millennium would behave, economically, like an agent from the 21st century. Globalised capitalism is around 100 years’ old; industrial capitalism perhaps 200; and proto-liberal capitalism maybe only 300 to 400 (Wood, 1999). Human societies have, historically, demonstrated substantial capacity to change in (relatively) little time (Graeber and Wengrow, 2021). I will return to this broad criticism at the end of this article. Furthermore, none of the economic benefits of time travel offered in this literature are exclusive benefits of time travel. The notion of cheaper labour, arbitrage opportunities, or legal opportunities are really all advantages of expansion of the economic domain, much in the same way that colonial expansion, or the expansion of credit and financialisation, have been described (e.g., Arrighi, 2009). This is to say, for a time travelling civilisation, there is little reason to believe the advantages of trading with the past would be greater than trading with other planetary civilisations (or species).3 Even if the arguments stand, they are not really advantages of time travel per se.

A Cost-Benefit Analysis of Time Travel

These criticisms point towards a cost-benefit analysis approach. Time travel may only occur at a historical moment when the necessary components of time travel technology can be assembled. This is to say, only a civilisation that can support the costs of time travel could become time travellers. Furthermore, such a civilisation would only accept these costs for benefits which must be achieved through time travel, and not through alternative means, unless the alternative means were costlier than time travel (which, I would argue, largely depends on the nature of time travel technology itself).

Whilst one may consider a plethora of social and ethical costs associated with time travel, such as the grandfather paradox or post-colonial critiques applied trans-temporally, I argue the overwhelming economic cost of time travel is the energy cost. As others have acknowledged (Cowen, 2008a, 2006; Krugman, 2010 [1978]), near luminal or superluminal travel produces sufficient relativistic effects as to constitute time travel, at least into the future (relative to the slower observer). Achieving such speeds requires a tremendous amount of energy—for superluminal travel, an infinite amount of energy, based on our current understanding.4 Black holes—infamous galactic objects which warp spacetime—are formed by the deaths of the most massive of stars. These events involve enormous amounts of energy. Whilst physics has much less to say about the mechanisms for travelling backwards in time5, taking forward time travel as a rough guide, it is likely—should it be possible—that the primary constraining factor will also be energy.6 I take the notion of warping spacetime and superluminal travel as guides to time travel technology, and not as a presumptive means of time travelling. This is to say, because various theoretical ideas of time travel imply huge amounts of energy would be required, I am assuming a similar amount of energy would be required for time travel by any means.7

Energy is also an ideal variable to consider when speculating on the likely stage of civilisational development necessary for time travel. The Kardashev scale, developed by Nikolai Kardashev (1964) measures the development of (inter)planetary civilisations based on the amount of energy they are able to capture. A Type I civilisation is able to harness all the energy of its home star which reaches its planet. A Type II civilisation can capture all the energy of its home star, through innovations such as a Dyson Sphere (Dyson, 1960). A Type III civilisation can capture all the energy of its home galaxy. Current estimates suggest Earth has a Kardashev value of approximately 0.73, based on 2021 energy consumption figures (BP Statistic Review of World Energy, 2021).8 Carl Sagan adapted Kardashev’s ‘Types’ into a continuous scale, which is where this non-integer value comes from. Doing so in 1971, Sagan then estimated Earth’s Kardashev value at 0.70, implying human civilisation may reach Type I in around 500 years, assuming linear growth in energy consumption. Given the likely energy demands required for, say, the time-distorting effects of a black hole, time travel is unlikely to be achieved in the next few centuries without a historically monumental increase in energy production.

Turning to the question of benefits, I argue time travel is unlikely to be pursued unless the resource gained through the activity is extremely scarce or difficult to acquire in a future period, but is relatively less scarce (or even abundant) in a previous period. Again, given any time travelling civilisation has likely secured dominion over at least its home star, if not a sizeable proportion of its home galaxy, the notion that such scarce resources would be labour or raw materials seems unfounded. On a galactic scale, raw materials are likely to be abundant, especially given the incomprehensible energy resources available to a time travelling civilisation. Such resources could likely support substantially more labour, meaning labour is unlikely to be scarce either. Instead, the most feasible resource which one may consider abundant today, but scarce in the (far) future, is today’s knowledge.

By ‘today’s knowledge,’ I mean everything which is known, by at least someone or something, today. Whilst this definition is necessarily broad, I will generally focus on knowledge of processes (e.g., programming knowledge, manufacturing knowledge). This is because, given the likely large energy costs, knowledge in the sense of historic knowledge, or whatever might be gleamed through ‘chrono-tourism,’ is unlikely to be sufficiently important to a future society to warrant the use of time travel, at least economically speaking. Whilst I consider ‘today’s knowledge’ to encompass everything which is known today, I will generally focus on ‘system critical’ (SC) knowledge. By this, I mean knowledge which is irreplaceable within systems the future civilisation relies upon for maintaining its way of life. I focus on this type of knowledge for two reasons. Firstly, such SC knowledge is the only subset of ‘today’s knowledge’ which may overcome the likely immense energy costs of time travel, in terms of economic benefit. Secondly, because the recent examples of knowledge loss which exist tend to be SC knowledge.

Knowledge is always being lost. History provides a plethora of examples, such as Greek fire and Roman concrete. Several foundational principles of calculus have been found in works attributable to Archimedes, but which—for one reason or another—were lost for centuries (Flood, 2011). Newton’s work on gravity itself owes much to an almost lost set of astronomical observations made by Jeremiah Horrocks, whose early death meant the observations were only (and luckily) published as an appendix in someone else’s book (Ferguson, 2023). In past decades, one example of SC knowledge which was lost is fogbank, a highly classified ‘ingredient’ within the manufacturing process of the United States’ nuclear warheads. Fogbank was so classified, in fact, that no records of the production process were made, with all individuals formerly familiar with the process either dead, or unable to remember. It took US nuclear engineers over a decade to rediscover how to make fogbank (Lillard, 2009).

The idea that advanced civilisations could come to be reliant on knowledge that they no longer have knowledge of may seem silly, but it is both a complete possibility, and indeed, somewhat evidenced in our own society. Complex systems and processes almost never start off complex. Rather, complexity is almost always the result of additional processes being ‘tacked onto’ existing, simple processes (Frey, 2020). This is common in programming—a simple website may be further developed over time, with more code written to facilitate more features. As computers came to be adopted within the finance industry in the 1960s and 70s, foundational software was written in older programming languages to facilitate basic banking services. Finance, being a SC sector within our civilisation, broadly took the ‘if it ain’t broken, don’t fix it,’ approach to this foundational code, almost never rewriting, updating, or otherwise tweaking the code, but instead, tacking on new code to perform new services. This approach is somewhat understandable. Firstly, when launching a completely new version of a complex system, there is a substantial risk that something does not work, and that spectacular errors arise. Secondly, the costs of updating the whole system rarely outweigh the immediate benefits, meaning cost-constrained organisations, such as banks, may always find incentive to forestall major updates.

Of course, the fragility of old systems supporting new systems is not necessarily a problem if knowledge of the old system remains, even when a newer version would be less fragile, or otherwise better (e.g., more efficient). But as systems become more complex, individual competencies within an organisation tend to ‘specialise,’ with understanding of the whole fading from any one individual’s grasp (Beer, 1993, 1975, 1973; Braverman, 1974). As the workforce composition of an organisation changes, with new employees replacing the retiring, SC knowledge can become eroded through partial knowledge transfer and miscommunication.

It might be hard to imagine that a time travelling civilisation would find itself dependent upon today’s knowledge, and simultaneously ignorant of it. Whilst this is a core aspect of the argument I am to make; it is not wholly unheard of within time travel circles. One of the most infamous stories of time travel found online is that of John Titor, who (in 2001) claimed to be a US army soldier from the year 2036. Titor’s supposed mission was to retrieve an IBM 5100 computer from the 1970s, as various critical systems in 2036 used legacy software which was going to break, less compatible technology was acquired to fix the problem (Dodds, 2015). Whilst likely a fanciful tale concocted for an internet forum, the key premise is telling: time travel may occur when future civilisations lose access to SC knowledge.

The Inequality of Time Travel

Cost-benefit calculations are limited in all manner of ways. For instance, one may overestimate some variable within the analysis, or omit a variable entirely. It is my ambition with the arguments above, and the inequality below, to highlight worthwhile aspects of a cost-benefit analysis of time travel, rather than incorporate all possible considerations (of which there are too many).9

Equation (1) presents the basic inequality:

In Equation (1), Cj is the money cost of one Joule of energy, and Ej is the amount of energy, measured in Joules, required for time travel. The left side of the inequality thus captures the energy costs of time travel. On the right side, Pr F is the probability that knowledge F is lost; Pr(SCF) is the probability that knowledge F is system critical; and F is the money benefit of rediscovering knowledge F. A time-travelling civilisation is expected to time travel from time t+n to time t only when the energy costs of doing so are less than the expected payoffs of retrieving knowledge F. alpha is a money cost deflator so the value of money on each side of the inequality is comparable. Where money costs are estimated within the same time period, may be ignored. Likewise, alpha may appear on either side of the inequality, provided the correct adjustment to alpha is made.10

For discursive purposes, what is worthwhile focusing on are the two probabilities, Pr F and Pr(SCF) , on the right side of Equation (1). One way of modelling Pr F is via Equation (2):

where n is the number of years since knowledge F was discovered, r is the ‘reproduction rate’ of knowledge, or broadly, the number of years within a human generation. m is the number of ‘entities’ with knowledge of knowledge F at time t, per Equation (3):

where N (H, F,t) is the number of humans with knowledge of F at time t, and N(C,F,t) is the number of computers (as well as other records, such as books) with ‘knowledge’ of F at time t.11 Note that Equation (2) uses the reciprocal of m, which might be understood as the ‘rate of forgetting,’ compared to m, which is the ‘rate of remembering.’ Thus, (1-1/m)^n/r describes the probability of F being remembered over n years, hence why one less this is equal to Pr F.12

Per Equations (2) and (3), knowledge which is very niche or esoteric is known by few people and recorded in few places. Thus, m will tend to be small, and Pr F will rapidly move towards 1. Inversely, where knowledge is very broad or fundamental, many people will know of it, and there will also be many records of it. Thus, m will tend to be large, and Pr F will very slowly move towards 1.13

This is important when considering Pr(SCF) . By definition, Pr(SCF) concerns knowledge which is fundamental to a civilisation. As such, even though Pr(SCF) may be difficult to estimate, one can infer that when Pr(SCF) = 1, m will be very large and thus Pr F ≈ 0. Equally, when Pr F = 1 , Pr (SCF) ≈ 0. This is to say, these two multiplicative probabilities likely have a relationship such that they counteract one another. Only in very particular instances—say, the fogbank incident where fogbank was both critical to the US nuclear arsenal but also highly classified, so few knew of it—will Pr F Pr(SCF) ≈ 0 not be found.14 Therefore, even allowing F to be arbitrarily large, it is very unlikely that the conditions are met for the inequality to be satisfied.

If on the 31st of December 2023 an event is held to welcome our time travelling descendants, and no one shows up, we need not assume this is because time travel is impossible. Instead, the conditions for it to be economically beneficial for our descendants to visit us may just not be sufficient. Put less politely, we as a civilisation are unlikely to be important enough to warrant visiting.

Is the Future so Utilitarian?

As above, we cannot necessarily assume too much about the values of a future civilisation, or their political and economic arrangements. I have used this argument to critique previous economic musings on time travel. Yet, the same argument applies to this contribution. Indeed, perhaps more so. Whereas previous authors (e.g., Swinton, 2021) have generally assumed near-future time travel, I have made the case that time travel will only be possible to a civilisation in the far future, where more time has elapsed, giving way to more scope in the evolution of their socioeconomic system.

Thus, it is not necessarily reasonable to assume that a future civilisation will undertake the utilitarian calculus of a cost-benefit analysis. Indeed, much of my argument is contingent on only two civilisational developments, energy capture and time travel. Yet, in achieving these developments, others (technological, cultural, political, social, etc.) are almost certain to occur. The near-term likelihood of time travel will also certainly spark a vibrant ethical debate, which may include consideration of issues presented here. For instance, in our own epoch, climate change has led to initiatives like the Svalbard Global Seed Vault to preserve biodiversity in plant life. A civilisation which acquires time travel technology may, even before acquisition, recognise the dangers of losing knowledge and set up initiatives to preserve it, if not whole ethical systems and legislative approaches to regulate the emerging technology’s usage.

Equally, without making such an assumption, there is limited analysis which can be undertaken. Indeed, any speculation about the practices of the future must be caveated with all the problems of prediction or inference. Thus, whilst it is highly unlikely that the exact procedure outlined here will be the decision-making framework of a future civilisation, or that the philosophical basis of such a framework would hold either, it is nevertheless just as good a framework to adopt as any other for the purposes of postulation. Furthermore, the approach outlined here avoids some of the more questionable assumptions of previous authors, such as the assumptions of comparable trade arrangements or arrangements of production. By grounding the discussion only in discussions of energy costs and knowledge (information) benefits, little about, say, the economic system of the future, needs to be assumed.

As a final note, I have generally limited my analysis to questions of economic costs and benefits and have been cautious throughout this article to focus on these economic considerations, whilst offering less comment on—as above—alternative motives which may govern the development and use of time travel. One need only to ask oneself what one would do with the ability to time travel to realise that ‘acquiring lost knowledge,’ is a very niche motivation. Just as people travel the globe, and increasingly beyond, so too wanderlust is likely to extend to time travel. The prospect of visiting epochs radically different to one’s own represents an intrinsic motivation for many. This analysis would likely be incomplete without an acknowledgement of this.

And, perhaps someday, economists will have to calculate the various costs and benefits of chrono-tourism, such as those of cultural exchange, and supplement the model proposed here, assuming it is at all relevant. For instance, the (relatively) recent development of the atomic bomb, space travel, and the internet, may all point to the past century or so as an especially interesting time period for a future civilisation built on these (by their standards) primitive technologies. Thus, non-economic terms should be included in any cost-benefit analysis of time travel. Their omission here is wholly the result of practical limitations.

Conclusion

In this article, I argue that (accepting time travel is possible) time travellers may not have visited us because our present offers little economic benefit to them. I postulate that the major economic cost of time travel is likely to be the energy cost, and this cost is likely to be extremely large. Furthermore, I have argued the major economic benefit of time travel is likely to be today’s knowledge, and specifically, knowledge which is ‘system critical’ for a future civilisation, but which said civilisation has lost. Yet, the likelihood of knowledge being lost and for said knowledge to be system critical is extremely small. Thus, I conclude that there is only a very narrow set of circumstances where the present is a likely destination for any time traveller, economically speaking. Much like a parent or grandparent may lament that their immediate family do not visit often enough, present society may need to entertain the possibility that, from an economic perspective, our descendants do not find us especially important.

Acknowledgements

I am grateful to Kira Rhodes, Richard Whittle, and Adam Davidson for their kind comments and suggestions. All errors are my own.

Gardener Comments

Note: the author has responded to comments in the subsequent section.

Gadi Lifshitz:
While the article is well written, and I think has some interesting insights (and so, worth publishing), in my view focusing only on the economic utility is a bit problematic.

The economics of time travel (well, much like any travel - even travel for work purposes) will definitely be a consideration, but as time travel holds a vast potential for financial gain (According to Einstein, “Compound interest is the eighth wonder of the world"...), there's a good chance it will be secondary to other reasons (many of which were mentioned in the article itself, even if as side notes). It would have been interesting to address both the potential gains and other constraints in more details (maybe in follow up articles).

Arturo Macías (Economist):
The main issue to deal with regarding time travel is what do you expect to be the result of an intervention over the past. What happens when the time traveler goes back and kills baby Hitler? Does a new branch reality appear while the older stays? Does the old reality (including the time traveler and the suffering inflicted by WW2) Vanishes? (as you can imagine, this question suggests that my view is that time travel is not only physically, but ontologically impossible). The ontology of time travel and the consequences of intervention shall be discussed before anything else.

An important consequence of this perspective is that the main reason to travel back in time is to alter the past, a possibility not even discussed, and (depending on your ontology of time travel) with a high potential economic value.

A final remark: if energy costs are the limit to time travel, those costs would probably be proportional to the mass of the time vessel. Miniaturization could make time travel viable, but probably we shall look for ultra-small time travelers.

Srinivas Vamsi Parasa:
It is not clear how the ideas presented in this paper can be 'falsified'. The core idea/hypothesis of this paper is (1) the major economic cost of time travel is the extremely high energy cost (unverifiable claim) (2) major benefit of time travel is the lost 'system critical' knowledge which is of importance to a future civilization (unverifiable claim). Ironically, we would need time travel to the future or an encounter with a person who traveled from the future to make sure these assumptions are indeed correct. Therefore, while the speculations presented in the article sound to be plausible, it is not clear how we can 'falsify' the assumptions and claims made in the paper.

Antonio Hanna-Amodio:
In a world where humans frequently assume an exalted status, this article offers a refreshing alternative perspective: rooted in both empiricism and humility.

Ted Wade:
This is a well-posed, interesting, and novel take on a fascinating question. I can’t see any relevance to current science.

Chronos (Emeritus Professor of Temporal Paradoxes):
Quite right - I spent a fortune to travel back here, and now I'm stuck in 2023!

Joe R:
Suppose Elon Musk had a time machine that cost $10B per trip (~5% of his net worth). What would you have to say to him to convince him not to use it? Does anyone really, sincerely believe that "hey, that's kinda expensive and you probably won't make back costs" would do the trick? This is Elon Musk; he'd just respond with "so what, it's a TIME MACHINE!" And then he'd either make $100B anyway or get eaten by a dinosaur. But the point is, good luck stopping him.

This paper is well-written, but makes enormous assumptions and fails to justify them. I expect its value to science will in fact be net negative, distracting researchers from valuable but less cool-sounding topics and by upholding a subpar standard for argument.

Their main argument seems to be: assuming time travel is possible, the only thing that would justify the energy expenditure is system critical knowledge possessed by the past. We couldn't have any knowledge valuable enough, therefore time travelers wouldn't bother.

This argument makes no sense. For one thing, why is system-critical knowledge the only thing presumed valuable enough? What about untapped physical resources? Genetic samples of extinct species? Anthropological data? The ability to directly observe known supernovae? Status and bragging rights?

For another, why would a future society be concerned solely or even principally with economic benefits? If the future happens to be rich, then people will want to time travel even if it's expensive, the same way people today go on wildly expensive vacations. Suppose Cj and Ej are low enough relative to the energy output of a galaxy-spanning civilization that at least one wealthy or well-connected historian, writer, botanist, evangelist, philanthropist, or dedicated LARPer could afford a trip to the past. Why wouldn't they go? Current humans care about lots of things. Future humans (or whatever we become if we make it that far) probably will to. There could be many reasons to travel to the past. There might be plenty of reasons not to (e.g. paradox, which the author ignores), but I doubt "well, it costs a lot of money" will stop literally everyone from bothering.

The narrowest possible interpretation of this paper is "the past probably contains little to no system-critical knowledge worth time-traveling for". This seems...probably true? And the author makes a decent stab at demonstrating it. But it is a massive and unjustified leap to also claim, as they do in their abstract and conclusion, that such knowledge is the main reason anyone would bother. It's an even more unjustified leap to say "and maybe, instead of physics, economics is why we don't see any time travelers!" The narrow claim is trivial and the broader claims are wrong.

Phil Filippak:
First of all, to rephrase David Deutsch, every next century is the most important one in the history of our civilization. We may be deemed unworthy of visiting but at the same time, any century prior to the one that has time travel discovered, may be. I would assume that future generations may be interested in building a comprehensive understanding of history, and therefore will be inclined to visit most of the previous centuries, if not decades.

At the same time, if they are not only technically but also socially advanced, they may not want to ruin or pollute history with the notion of actual time travel, and so even if they will have traveled to the past, they would do that covertly, and no event will attract them.

Considering energy expenditure, we don't know how much energy will be available to our descendants even in 50 years, not to say anything about thousands of years.

Overall, I believe that the idea of time travel is so rarely used in modern scientific thought that it's worth publishing despite the possible flaws that the paper might contain (and I am definitely in no position to notice and point out those if they are present).

Author’s Response to Comments

One of the major contributions of this article, in my opinion, is to entertain. Whilst I have had discussions with people, many of whom think the argument I propose is strange, or perhaps cold, or perhaps completely wrong, I am pleased that in many instances my conversations about this idea have sparked interest, engagement, and discussion. Being wrong matters little if it results in good conversation, in my opinion.

I give this preface to support me in responding to the general sentiment of those who have responded negatively to this article. I have organised their points as: a) the core idea is poorly thought out, and contains many omissions; b) it is too assumptive; c) the approach cannot be falsified; and d) it makes no contribution to science (and perhaps even hinders it). I will try to take each point in turn.

In the introduction, I offer at least eight alternative explanations for why time travellers may not visit the party in my thought experiment. The article pursues only one. Yes, I concede, I have not gone into such inane detail outlining every possibility, but that is not the objective of the article. It is, as above, to entertain (hence why the eighth argument is that the party might not be that good) whilst, I hope, exploring interesting ideas such as measurement of civilisational development, the role of knowledge and lost knowledge, and the limits of cost/benefit analysis. I am not convinced that by exploring every possible twist and turn in the thought experiment that this article would be improved. I am convinced that it would be less entertaining, though I am of course biased.

It is! Whilst I do not deny that it is a bit of a weak argument, I would point critics to my discussion of the utilitarian assumptions of the article, wherein I note that there is almost no basis for many of the assumptions I am making, but that there is little basis for any possible assumptions. The article is speculating about the decision-making processes of a far future civilisation—yes, my assumptions are going to be on shaky ground. But I am sceptical any steady ground can be found.

This criticism I considered very surprising. If one examines the cost/benefit analysis literature, one finds an endless debate between those who support the approach from a pragmatic perspective, and those who are critical of it from (often) a social perspective. The debate is never that CBA is right, and almost always around the implications of thinking CBA is right. Convincing yourself that a poor model is correct is a dangerous deception, and to this end I have much sympathy with those who are critical of CBA. Equally, I am sympathetic to the argument that even formulating a bad model is worthwhile as an exploratory exercise. I think that is what I am trying to do here. Thus, it has never really occurred to me that some would want a ‘falsifiable’ model, because I have never for a moment conceived of the model as an attempt to get to the ‘right’ answer (or the wrong answer if falsified). I have always approached this article with the perspective that the exercise itself is what matters, and the criticisms of my assumptions is where the value lies. Hence my preface. Or, to put it another way, maybe the real insights were the friends we made along the way. If one wants a falsifiable model, then this contribution clearly fails.

Similarly, to the above point, I was quite put off by this comment, because I conceive of the whole scenario as a vehicle for a discussion of ideas, rather than the scenario itself being the main contribution. As I show in the paper, several authors in economics have engaged in similar, perhaps tongue-in-cheek approaches, to play around with economic ideas. I fail to see the harm in this—as an educator, I can only see the positives of wrapping up intellectual ideas in an accessible format, which I hope I am doing here (though perhaps I am failing). To be accused of making a negative contribution was particularly hurtful, though this comment must be wrong. To be pedantic, this contribution does not destroy knowledge. If it is a distraction, then it only makes the process of discovering new knowledge slower; it does not drive it into the negative.

I would like to thank the ‘gardeners’ for their comments, regardless of any disagreements I may have with them, as they have added to the further development of this article.

References

1. Arrighi, G. (2009) ‘The Long Twentieth Century: Money, Power and the Origins of our Times’ Verso Books: UK

2. Beer, S. (1993) ‘Designing Freedom’ Anansi: Canada

3. Beer, S. (1975) ‘Platform for Change’ Wiley: UK

4. Beer, S. (1973) ‘Decision and Control’ Wiley: UK

5. BP Statistical Review of World Energy (2021) ‘bp Statistical Review of World Energy’ [Online] [Date accessed: 30/03/2023]: https://www.bp.com/content/dam/bp/business-sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp-stats-review-2022-full-report.pdf

6. Braverman, H. (1974) ‘Labor and Monopoly Capital’ Monthly Review: USA

7. Cowen, T. (2006) ‘The economics of relativity’ Marginal Revolution. [Online] [Date accessed: 28/03/2023]: https://marginalrevolution.com/marginalrevolution/2006/01/the_economics_o.html

8. Cowen, T. (2008a) ‘The theory of interstellar trade’ Marginal Revolution. [Online] [Date accessed: 28/03/2023]: https://marginalrevolution.com/marginalrevolution/2008/03/the-theory-of-i.html

9. Cowen, T. (2008b) ‘Outsourcing, taken to extremes’ Marginal Revolution. [Online] [Date accessed: 28/03/2023]: https://marginalrevolution.com/marginalrevolution/2008/03/sentences-to-po.html

10. Dodds, L. (2015) ‘Who was John Titor, the ‘time traveller’ who came from 2036 to warn us of a nuclear war?’ The Telegraph. [Online] [Date accessed: 30/03/2023]: https://web.archive.org/web/20170622161305/http://www.telegraph.co.uk/news/science/11945420/Who-was-John-Titor-the-time-traveller-who-came-from-2036-to-warn-us-of-a-nuclear-war.html

11. Dyson, F. J. (1960) ‘Search for Artificial Stellar Sources of Infrared Radiation’ Science, 131(3414), pp. 1667-1668

12. Ferguson, D. (2023) ‘The forgotten maths genius who laid the foundations for Isaac Newton’ The Guardian. [Online] [Date accessed: 29/03/2023]: https://www.theguardian.com/science/2023/mar/19/jeremiah-horrocks-the-forgotten-maths-genius-who-laid-the-foundations-for-isaac-newton

13. Flood, A. (2011) ‘Archimedes Palimpsest reveals insights centuries ahead of its time’ The Guardian. [Online] [Date accessed: 29/03/2023]: https://www.theguardian.com/books/2011/oct/26/archimedes-palimpsest-ahead-of-time

14. Frey, C. B. (2020) ‘The Technology Trap: Capital, Labor, and Power in the Age of Automation’ Princeton University Press: UK

15. Gans, J. (2008) ‘The Real Economics of Time Travel’ [Online] [Date accessed: 28/03/2023]: https://economics.com.au/2008/03/29/the-real-economics-of-time-travel/

16. Graeber, D., Wengrow, D. (2021) ‘The Dawn of Everything’ Penguin Books: UK

17. Kardashev, N. S. (1964) ‘Transmission of Information by Extraterrestrial Civilizations’ Soviet Astronomy, 8(2), pp. 217-221

18. Krugman, P. (2010 [1978]) ‘The theory of interstellar trade’ Economic Inquiry, 48, pp. 1119-1123

19. Lillard, J. (2009) ‘Fogbank: Lost Knowledge Regained’ Nuclear Weapons Journal, 2, pp. 20-21

20. Morehouse, I. M. (2012) ‘The Economic Value of Time Travel’ The Foundation for Economic Education. [Online] [Date accessed: 28/03/2023]: https://fee.org/articles/the-economic-value-of-time-travel/

21. Swinton, J. R. (2021) ‘Economics and Time Travel v1’ [Online] [Date accessed: 28/03/2023]: https://www.researchgate.net/publication/351360747_Economics_and_Time_Travel_v1

22. Whitman, G. (2008) ‘Transtemporal Economics’ [Online] [Date accessed: 28/03/2023]: http://agoraphilia.blogspot.com/2008/03/transtemporal-economics.html

23. Wood, E. M. (1999) ‘The Origin of Capitalism’ Verso Books: UK

Abstract

Introduction: Motile and sensory (primary) cilia are organelles that are found on the surface of almost all cells. Defects in cilia cause a number of multi-organ diseases known as ciliopathies, which have clinically heterogeneous symptoms. This heterogeneity makes diagnosing cilia disorders challenging and clinicians often rely on genetic sequencing to delineate ciliopathies from other diseases. However, there is not a consensus on which sequencing tools are most optimal for ciliopathy diagnosis and research. 

Methods: Here I review the implications of next-generation sequencing tools for ciliopathy diagnostics. I describe landmark studies that showed ciliopathies as genetic conditions and transition to the advantages and challenges of using next-generation sequencing techniques. In particular, I compare studies that utilized targeted sequencing with those that used whole-exome and/or whole-genome sequencing.

Discussion: High throughput screens can identify novel cilia genes and show promise as a robust diagnostic tool in clinical settings. Moreover, I compare the effectiveness of whole-exome and whole-genome sequencing both for basic science research and clinical applications, arguing that whole-exome sequencing is a sufficient first pass in clinical settings. I also acknowledge that ciliopathies are associated with many, both significant and insignificant, genetic variants making interpreting next-generation sequencing data an ongoing challenge for scientists and clinicians.

Conclusion: This review demonstrates the increasing body of knowledge on cilia genomics and highlights that next-generation sequencing will be integral towards optimizing diagnostics for this heterogeneous and debilitating group of disorders.

Introduction

Cilia are present on almost every cell type in the human body. (Ishikawa and Marshall 2011) They are microtubule-based cellular projections that extend from the plasma membrane. Despite their prevalence, for many years, these hair-like projections, particularly non-motile cilia, were thought to be vestigial structures with no important function. (Wheway, Nazlamova, and Hancock 2018) However, after years of scientific disregard, “the cilium has emerged as a key organelle in numerous physiological and developmental processes”. (Ishikawa and Marshall 2011)

Cilia are assembled on basal bodies, which are derived from centrioles, and have a microtubule-based axoneme. (Horani et al. 2016) Cilia can be either motile, as seen in ependymal cells that line brain vesicles, or immotile/sensory, as seen in the photoreceptor cells of the retina or the collecting ducts of the kidney. (Reiter and Leroux 2017; Vladar and Stearns 2007) Motile cilia enable cell movement and the movement of fluids across the surface of cells. Sensory (primary) cilia are specialized for signal transduction and act as ‘antennae’, sensing extracellular signals like growth factors, odorants, and developmental morphogens. 

Research has revealed that defective primary and motile cilia can result in a number of human diseases, including retinal degeneration, polycystic kidney disease, and primary ciliary dyskinesia. (Hildebrandt, Benzing, and Katsanis 2011) Since cilia are ubiquitous, mutations in cilia-related genes can cause multi-organ disorders, which are called ciliopathies. (Hildebrandt, Benzing, and Katsanis 2011) Ciliopathies can be either autosomal dominant or recessive and diagnosing these disorders can be very challenging as clinical presentations are often very heterogeneous. (Braun and Hildebrandt 2017) This creates a clinical need to optimize genetics-based diagnostic tools for ciliopathies. Moreover, there is a knowledge gap in which genetic variations are causative for ciliopathies. Here, I review the implications of using next-generation sequencing tools to diagnose ciliopathies and to identify novel genetic variants that are associated with cilia defects. I will highlight the transition from landmark linkage analysis to high throughput screens, while also discussing the strengths of different approaches, which include targeted next-generation, whole-exome, and whole-genome sequencing. Based on this synthesis, I argue that whole-exome sequencing is a sufficient first pass in clinical settings, while whole-genome sequencing is more informative for basic science research. 

Discussion

i. Early Linkage Mapping Studies Associated Some Disorders with Variants in Cilia Genes

A landmark study in the field of cilia genetics was the identification of a candidate gene associated with monogenic polycystic kidney disease (PKD). (Reeders et al. 1985) PKD is a common and often lethal multi-organ disease with around 12.5 million people affected worldwide. (Chebib and Torres 2016) The authors of this influential paper identified four families that had PKD-like symptoms and relied on traditional genome mapping techniques to identify the location of a potential causative gene. They used genetic markers and looked for linkages with a potential disease allele. Ultimately, linkage analysis and lod score calculations allowed researchers to map the PKD-1 gene to the short arm of chromosome 16. However, at the time, researchers did not know that this gene was related to the primary cilium. The link was established about a decade later when another research group found a functional homologue of PKD-1 in a model organism, C. elegans. (Barr and Sternberg 1999) These authors generated PKD-1 knock-out organisms and saw that PKD-1 is part of the signaling cascade that is necessary for cilia formation. This was one of the first instances where a cilia-related gene was associated with a human disorder and undoubtedly generated a lot of interest in finding genes that encode for structural or functional cilia proteins.  

With an increase in genome mapping experiments, another group of researchers looked for genetic associations for a disease that is characterized by retinal dystrophy, obesity, renal malformations, and learning disabilities. (Forsythe and Beales 2013) This disease is now coined Bardet-Biedl syndrome (BBS) and for many years its molecular basis remained elusive. This was mostly due to the fact that different researchers were mapping it to different loci, with at least six loci being associated with phenotypic BBS. (Leppert et al. 1994) Different research groups focused on different genetic loci and translated them to model organism experiments to see where the encoded protein localized. The commonality was that these proteins localized to the base of the primary cilium, the basal body. (Katsanis et al. 2000; Mykytyn et al. 2002) We now know that the basal body is necessary for cilia formation and, in most cases cilia maintenance, in ciliated cells. (Magescas et al. 2021; Breslow and Holland 2019)  The fact that defects in the primary cilium can cause multi-organ disorders further fueled interest in dissecting the molecular mechanisms of ciliopathies.

ii. Sanger Sequencing to Diagnose Known Ciliopathies 

As scientists started to learn more about the genetic causes of ciliopathies, Sanger sequencing of exons in known cilia genes became the standard for clinical diagnostics. Researchers built on the aforementioned work on BBS, which in 2011 was associated with 15 cilia genes. In one study, researchers did Sanger Sequencing on 55 families with BBS. (Chen et al. 2011) Sequencing 142 exons showed that 84% of the patients had potentially pathogenic variations in one of the 12 cilia genes. The mutational analysis also revealed 21 novel mutations in these known genes, including 10 novel missense variations and 2 frameshift mutations. These results show the high genetic heterogeneity that can lead to BBS and ciliopathies at large. In addition, this heterogeneity can lead to the following two challenges. Firstly, it can make predictive genotyping challenging, meaning that it is difficult to know whether or not someone who does not show a ciliopathy phenotype, but has a variant in a ciliary gene, will develop clinical symptoms. Secondly, Sanger sequencing relies on previous knowledge of ciliopathy genes, but it is possible that known genes or exons of genes are not adequate to get a whole genetic picture of a ciliopathy patient. Given the large genetic heterogeneity, it is possible that exons that were not sequenced might have pathogenic variants or there might be mutations in other unidentified ciliary genes.

iii. Targeted Next-Generation Sequencing 

These challenges associated with Sanger sequencing coincided with a period where next-generation sequencing tools were becoming more and more prevalent in research and clinical settings. (Koboldt et al. 2013) After linking many renal disorders to cilia perturbations, scientists aimed to identify the mutation profile of ciliary genes in autosomal recessive PKD patients. Thus, targeted next-generation sequencing emerged as a cost-effective and rapid way to assess variants in cilia genes. An important study assessed allelic variants in 191 structural and functional genes of the primary cilium in tissue samples from 7 PKD patients. (Skalická et al. 2018) This was a significantly more high-throughput experiment compared to early studies that found one or two genes associated with familial PKD. Illumina sequencing revealed pathogenic variants in 39 genes encoding various structural components of the primary cilium. The most frequently mutated genes were those that encode centriolar and centrosomal proteins; other studies have also shown mutations in centriolar genes as the most common causes of ciliopathies. (Hildebrandt, Benzing, and Katsanis 2011) Since centrioles are essential for cilia formation it is likely that these patients do not form cilia and as such have severe disease phenotypes. 

Interestingly, none of the samples showed mutations in intraflagellar transport (IFT) proteins, which are a group of proteins that carry essential protein cargo to the tip of the cilium. This highlights the genetic heterogeneity of  PKD since the same phenotype can be traced back to an autosomal recessive mutation in an IFT protein in another study. (Qin, Rosenbaum, and Barr 2001) However, I believe that the lack of IFT proteins could also be due to the small sample size of this study (n=7 patients). Moreover, this also highlights a weakness of targeted sequencing. It is possible that IFT genes were not sequenced comprehensively, preventing the researchers from identifying causal variants in these highly redundant genes. Given a large number of studies that show IFT proteins as causal factors in renal ciliopathies, I believe that IFT proteins are important for proper cilia function and thus a genetic cause for ciliopathies. (Reiter and Leroux 2017) The authors of this study did not conduct extensive sequencing of all known IFT proteins. Overall, targeted next-generation sequencing has been key in highlighting genetic variations that lead to cilia defects, but can miss some pathogenic variants that are implicated in ciliopathies.

iv. Whole Exome Sequencing is an Effective Diagnostic Tool and Integral for Identifying Novel Ciliopathy Genes 

As the cost of per nucleotide next-generation sequencing continued to decrease, researchers started to complement targeted studies with exome-wide approaches. Whole-exome sequencing (WES) yields a more comprehensive analysis of genetic variations that lead to ciliopathies. Moreover, WES can reveal novel genes that are important for cilia function, while still generating rapid and high-depth results. WES was used on six families with members that had ambiguous ciliopathy-like symptoms. (Castro-Sánchez et al. 2017) It is important to highlight that these patients had not received an official ciliopathy diagnosis. Despite this ambiguous clinical presentation, all the patients had mutations in BBS genes, which are a group of genes that encode for basal body proteins. This group of BBS proteins, sometimes referred to as the ‘BBSome’ transport essential signal receptor proteins to the cilium and thus defects in the ‘BBSome’ impair the cilium’s ability to detect external signals. (Jin and Nachury 2009) BBSome defects result in many ciliopathy phenotypes, as mentioned previously. Moreover, WES revealed mutations in three other genes that were proposed to be “novel candidate cilia genes.” The functions of these genes are unknown, but they are all differentially expressed in dividing cells, suggesting that they might play a role in centriole function, which is essential for ciliogenesis. Examples like this show that WES is a useful strategy to diagnose patients with unclear phenotypes, while also revealing novel genes that might be related to cilia biology. 

WES can also allow clinicians to understand the underlying genetics of chronic kidney disorders, a common ciliopathy phenotype. Knowing that kidney failure is due to primary cilia defects can guide optimal treatment. In this pursuit, a research group did WES on around 100 patients that had chronic kidney disease before age 25 and found that 7 of them had mutations in cilia genes. (Mann et al. 2019) Interestingly, one patient with a ciliopathy variant did not have any known family history of renal disease, so the mode of transmission remains an open question. Regardless, WES is an informative tool for ciliopathy patients that are in need of kidney transplantation management.  

Moving onto disorders associated with defects in motile cilia, we can see that a large number of studies have also relied on WES. An important study analyzed primary ciliary dyskinesia (PCD), a defect of motile cilia that leads to chronic respiratory symptoms. This disease has an autosomal recessive mode of inheritance and is a known ciliopathy. (Horani et al. 2016) Researchers utilized whole-genome sequencing on one family and WES on another family and saw that all PCD patients had a homozygous loss-of-function in a ciliary axoneme gene. (Onoufriadis et al. 2014) However, this was not the only gene associated with PCD. Another group identified nine related subjects with PCD from geographically dispersed Amish communities and performed WES of two affected individuals and their unaffected parents. (Horani et al. 2012) WES confirmed the previously known autosomal-recessive mode of inheritance and identified a missense mutation in the HEATR2 gene as the causal variant in these families. (Lucas et al. 2020) This gene is essential for the stability of the inner microtubule structure that allows for the movement of motile cilia. (King 2016) This study shows that WES can be an effective diagnostic tool for PCD and also reveals that mutations in various ciliary genes can be sufficient for PCD. This begs the question of do all mutations that are linked to PCD generate the same phenotype? 

Next-generation sequencing has been key in linking changes in different ciliary genes to phenotype severity. As of 2020, more than 40 different genes have been reported that cause PCD. (Lucas et al. 2020) There have been attempts at linking certain genes to milder phenotypes, but I think it is important to exercise caution as these links are made based on small numbers of patients with variations in any one gene. An example is DNAH9 mutations, a protein that is essential for axoneme stability, being associated with a mild respiratory phenotype. (Postema et al. 2020) Overall, WES is shown to be effective at diagnosing ciliopathies associated with defects in motile and/or primary cilia.

v. Is Whole-Genome Sequencing More Robust at Diagnosing Ciliopathies?

A growing body of literature shows that WES is an effective way of diagnosing ciliopathies, and yet exome-sequencing does not take into account a very large portion of the DNA of patients. This begs the question of would whole-genome sequencing be more robust? While there is no literature that directly addresses this question, an important point to note is that almost all known causative variants that lead to ciliopathies are in the coding regions of the genome. Several studies have used WGS in families that have ciliopathies and only identified variants in coding regions. 

However, WGS can be more effective at diagnosing patients that have atypical symptoms. Many studies show that WGS is impactful in identifying novel developmental disorders, and ciliopathies should not be an exception. (Wright et al. 2015) While WES is more prevalent, WGS is a valuable tool when diagnosing patients that have a diverse spectrum of disease characteristics, particularly since it can reveal changes in non-coding regions of other genes that can be contributing to the ciliopathy-like phenotype. In one recent study, scientists did WGS on 4 children with suspected ciliopathies. (Strong et al. 2021) The authors identified that all 4 children had mutations in cilia genes, but two patients had additional pathogenic variants for non-coding regions that are associated with liver disease. 

In addition, WGS is important at identifying structural changes and copy number defects. One study that conducted WGS in 11 different families found a tandem duplication of 3 exons of an IFT gene in 8 families, which all had uncharacterized ciliopathies. (Geoffroy et al. 2018) Importantly, this IFT protein, which is involved with carrying the building blocks of cilia to the tip of the axoneme, had been missed by whole-exome sequencing. This study shows that ciliopathies are not just caused by nucleotide-level variants or deletions and that WGS is instrumental in detecting structural rearrangements, including copy number changes. Moreover, other studies have suggested that whole-genome sequencing is more powerful than whole-exome sequencing at detecting exome variants. (Belkadi et al. 2015) Lastly, it is known that transcription-level regulation is an important part of ciliogenesis. (Choksi et al. 2014) There is some evidence that mutations in ciliogenesis regulators would lead to mild to severe ciliopathies. For instance, proteins that regulate the expression levels of two cilia genes were shown to be associated with Joubert syndrome. (Lee et al. 2012) Mutations in this regulatory gene were shown to give rise to an ambiguous ciliopathy-like phenotype. Such genetic mechanisms will be revealed as WGS on ciliopathy patients get more prevalent.  

Overall, WES has been very effective at identifying known and find novel genetic variants that lead to ciliopathies. It is likely that it is sufficient as a first-pass in clinical settings. However, I do not think that a targeted-gene panel would be robust enough to rule out a cilia defect for a patient that has ciliopathy-like symptoms. WES would be an important follow-up for patients that had a ‘normal’ gene panel sequencing. In addition, WGS can play an important role in research by tackling a knowledge gap in cilia biology: can ciliopathies be associated with variations in non-coding regions of the genome?

vi. Challenges of Using Whole-Exome and Whole-Genome Sequencing as Diagnostics

We have demonstrated that whole-exome and/or whole-genome sequencing are important venues for ciliopathy detection and the identification of novel genes that are implicated in such diseases. Yet there are challenges associated with using WES and/or WGS in clinic for diagnostic purposes. Firstly, cilia genes tend to be highly variable and lack mutation hot spots. Many variants generate wild-type cilia and thus it can make diagnosing clinically ambiguous patients challenging. Moreover, different families tend to have different variations in the same gene, making developing precise diagnostic tools a complex task. Secondly, some cilia genes have duplicated regions in other parts of the chromosome. Take the PKD1 gene, for example. Several studies have shown that the high sequence similarity between the pseudogenes and their parent genes can obscure the detection of the pathogenic mutation. (Ali et al. 2019; Eisenberger et al. 2015) An added challenge is the fact that pseudogenes for original ciliopathy-associated genes are located proximal to the original gene. (Ali et al. 2019)

Moreover, mutations in the same ciliary gene can give rise to different symptoms, making it challenging for physicians to initiate genetic sequencing or interpret the results. For instance, mutations in the CEP290, a centrosomal protein, can lead to retina phenotypes and/or kidney problems. (Baker and Beales 2009) It is possible that it can cause other symptoms as well and if physicians are not aware of them it might be difficult for them to associate genetic variants with the phenotype. Since the boundaries that separate ciliopathy symptoms are very fluid wide-scale implementation can be a challenge.

Lastly, there are some medical ethics challenges associated with using WES in clinic. WES may identify mutations in genes unrelated to the studied disease, which has major implications for patients and their relatives. This would not be the case with targeted sequencing and yet we have talked about the limitations of targeted sequencing in identifying novel ciliopathies. In addition, we do not know what the implications of a novel variant will be. As mentioned previously, many variants of unknown significance tend to be unproblematic for patients. Do these patients benefit from knowing that they have this variant? It is worth noting that these problems are seen in the context of other genetic disorders and need to be tackled via a multi-disciplinary collaboration that includes genetic counselors, physicians, and ethics specialists.

vii. Discussion of Future Directions for Ciliopathy Genomics

The increase in genomics data continues to reveal new cilia genes that are causative for ciliopathies. However, we are still far from a comprehensive list of ciliopathy genes. Firstly, we are far from generating a complete genetic map of a wild-type cilium. Analysis of single-cell RNA-sequencing data and WGS will continue to reveal new genes that regulate cilia function. (Patir et al. 2020) Secondly, scientists will continue to discover novel genetic variants in known cilia genes that contribute to ciliopathies. Thirdly, while I highlighted some molecular mechanisms that lead to ciliopathies, the exact process through which many of these genes impact the cilium is unknown, making translating our findings to therapeutics an open question in the field.

Another open question in the field of ciliopathies is whether or not we can link mutations in specific genes to a phenotype. As we have seen through this review, similar variants can lead to drastically different phenotypes and this makes precision treatment challenging. This can be tackled with a meta-analysis and bioinformatics approach. If we can combine next-generation sequencing and phenotype data for ciliopathy patients, we can generate a ‘ciliopathy’ databank that associates each gene and variant to a phenotype. As this potentially open-source databank grows, scientists can notice trends and make statistically significant genotype to phenotype interpretations. 

Furthermore, most research has focused on ciliopathies as monogenic (autosomal dominant or recessive) disorders. It would be interesting to see if some patients have multiple mutations in different cilia genes. Due to the basal body’s close relationship to centrioles, it is likely that multiple mutations lead to embryonic lethality, however this might not be the case with all individuals. There are model organisms that are viable and show severe defects after introducing mutations in multiple cilia genes. (Cornils et al. 2016) So, it would be clinically important to see if an increase in the number of cilia mutations correlates with more severe ciliopathy phenotypes. 

Lastly, there has been an increasing interest in studying links between cilia defects and neurodevelopmental disorders. (Valente et al. 2014) In a recent pre-print, a large research group associated autism spectrum disorder, intellectual disabilities, and attention deficit disorders with de novo or inherited variants in ciliogenesis regulators. (Harris et al. 2020)  The study established that, in their cohort of 36 people, deleterious variants in transcription factors that regulate ciliogenesis are important causes for the aforementioned disorders. So far, most research on ciliopathies has been conducted on inherited mutations, but this study paves the way for research that looks at the impact of de novo mutations in cilia biology.

Conclusion

Here, I reviewed the impact of next-generation sequencing techniques on our knowledge of cilia defects and ciliopathy diagnostics. Over the last two decades the field has evolved drastically: from using chromosome mapping to utilizing next-generation genome-wide sequencing, which has allowed scientists to identify novel genetic variants that cause ciliopathies. These high throughput screens have been critical in showing the genetic heterogeneity of ciliopathies, alongside highlighting that cilia defects can cause multi-organ level phenotypes. This synthesis revealed that mutations in genes that encode basal body (BBSome, CEP290, PKD-1), IFT, and microtubule-based axoneme proteins (HEATR2, DNAH9) can be causative factors for ciliopathies, with no clear correlation with clinical symptoms. Importantly, I attempted to associate mutations in specific genes to clinical phenotypes, and similar to other reviews (Strong et al. 2021; Tobin and Beales 2009; Oud, Lamers, and Arts 2017) saw that mutations which disrupt similar cellular processes can lead to varying clinical phenotypes. Thus, I believe that the real clinical power of this data will come when we integrate results from different studies that looked at the genetics of ciliopathies. As we start generating large biobanks for ciliopathies, we will be able to make more accurate diagnostics based on whole-exome or whole-genome sequencing data. Perhaps with a robust understanding of the impact of specific genetic variants, we can start creating evidence-based treatments for unique ciliopathies. Tailoring treatment towards each patient’s genetics and clinical phenotype will pave the way towards an era of precision medicine for ciliopathies.

Acknowledgments

I would like to thank Jonathan Pritchard and Alyssa Lyn Fortier for valuable discussions about the manuscript.

Gardener Comments

Michael Tran (PhD in molecular cellular biology):
The author's focus is mainly on the use of next-generation sequencing tools to identify genes involved and does a beautiful job reviewing the history of its use. However, I would like to also further highlight the use of model systems to identify cilia related genes and how they might synergize with the studies highlighted by the author. Molecular biology including biochemistry and electron microscopy in cilia models have identified various ciliary components. Work in these systems have also elucidated the function of many of these proteins. However, the shortcoming of these studies is that their role in actual ciliopathies is unknown. So further efforts in these model systems in combination with more targeted next-generation sequencing could thereby circumventing some of the ethical challenges of WES. A future where we have a more complete list of all cilia related proteins would marry wells with these new sequencing techniques to help create a list of ciliopathy genes.

Ahmad Ozair, MD:
This is a thoughtful and comprehensive review of the literature on diagnostics of primary ciliary disorders that strongly merits publication.

Joe R:
This is the best-written article I've seen so far, and an example of the correct use of jargon: technical terms are explained in context, and used exactly and only where necessary to concisely communicate a technical concept. I can't speak to the accuracy of the studies or conclusions, as this is not my field, but as an overview of the state of the art in ciliopathy diagnosis, the paper's logic seems sound. 

Jack Arcalon:
I wonder if this general line of research could also apply to or inspire research in neuro-degenerative diseases.

References

1. Ali, Hamad, Fahd Al-Mulla, Naser Hussain, Medhat Naim, Akram M. Asbeutah, Ali AlSahow, Mohamed Abu-Farha, et al. 2019. “PKD1 Duplicated Regions Limit Clinical Utility of Whole Exome Sequencing for Genetic Diagnosis of Autosomal Dominant Polycystic Kidney Disease.” Scientific Reports 9 (1): 1–13. https://doi.org/10.1038/s41598-019-40761-w.

2. Baker, Kate, and Philip L. Beales. 2009. “Making Sense of Cilia in Disease: The Human Ciliopathies.” American Journal of Medical Genetics Part C: Seminars in Medical Genetics 151C (4): 281–95. https://doi.org/10.1002/ajmg.c.30231.

3. Barr, Maureen M., and Paul W. Sternberg. 1999. “A Polycystic Kidney-Disease Gene Homologue Required for Male Mating Behaviour in C. Elegans.” Nature 401 (6751): 386–89. https://doi.org/10.1038/43913.

4. Belkadi, Aziz, Alexandre Bolze, Yuval Itan, Aurélie Cobat, Quentin B. Vincent, Alexander Antipenko, Lei Shang, Bertrand Boisson, Jean-Laurent Casanova, and Laurent Abel. 2015. “Whole-Genome Sequencing Is More Powerful than Whole-Exome Sequencing for Detecting Exome Variants.” Proceedings of the National Academy of Sciences 112 (17): 5473–78. https://doi.org/10.1073/pnas.1418631112.

5. Braun, Daniela A., and Friedhelm Hildebrandt. 2017. “Ciliopathies.” Cold Spring Harbor Perspectives in Biology 9 (3). https://doi.org/10.1101/cshperspect.a028191.

6. Breslow, David K., and Andrew J. Holland. 2019. “Mechanism and Regulation of Centriole and Cilium Biogenesis.” Annual Review of Biochemistry 88 (1): 691–724. https://doi.org/10.1146/annurev-biochem-013118-111153.

7. Castro-Sánchez, Sheila, María Álvarez-Satta, Mohamed A. Tohamy, Sergi Beltran, Sophia Derdak, and Diana Valverde. 2017. “Whole Exome Sequencing as a Diagnostic Tool for Patients with Ciliopathy-like Phenotypes.” PLOS ONE 12 (8): e0183081. https://doi.org/10.1371/journal.pone.0183081.

8. Chebib, Fouad T., and Vicente E. Torres. 2016. “Autosomal Dominant Polycystic Kidney Disease: Core Curriculum 2016.” American Journal of Kidney Diseases: The Official Journal of the National Kidney Foundation 67 (5): 792–810. https://doi.org/10.1053/j.ajkd.2015.07.037.

9. Chen, Jianjun, Nizar Smaoui, Monia Ben Hamed Hammer, Xiaodong Jiao, S. Amer Riazuddin, Shyana Harper, Nicholas Katsanis, et al. 2011. “Molecular Analysis of Bardet-Biedl Syndrome Families: Report of 21 Novel Mutations in 10 Genes.” Investigative Ophthalmology & Visual Science 52 (8): 5317–24. https://doi.org/10.1167/iovs.11-7554.

10. Choksi, Semil P., Gilbert Lauter, Peter Swoboda, and Sudipto Roy. 2014. “Switching on Cilia: Transcriptional Networks Regulating Ciliogenesis.” Development 141 (7): 1427–41. https://doi.org/10.1242/dev.074666.

11. Cornils, Astrid, Ashish K. Maurya, Lauren Tereshko, Julie Kennedy, Andrea G. Brear, Veena Prahlad, Oliver E. Blacque, and Piali Sengupta. 2016. “Structural and Functional Recovery of Sensory Cilia in C. Elegans IFT Mutants upon Aging.” PLOS Genetics 12 (12): e1006325. https://doi.org/10.1371/journal.pgen.1006325.

12. Eisenberger, Tobias, Christian Decker, Milan Hiersche, Ruben C. Hamann, Eva Decker, Steffen Neuber, Valeska Frank, et al. 2015. “An Efficient and Comprehensive Strategy for Genetic Diagnostics of Polycystic Kidney Disease.” PLOS ONE 10 (2): e0116680. https://doi.org/10.1371/journal.pone.0116680.

13. Forsythe, Elizabeth, and Philip L. Beales. 2013. “Bardet–Biedl Syndrome.” European Journal of Human Genetics 21 (1): 8–13. https://doi.org/10.1038/ejhg.2012.115.

14. Geoffroy, Véronique, Corinne Stoetzel, Sophie Scheidecker, Elise Schaefer, Isabelle Perrault, Séverine Bär, Ariane Kröll, et al. 2018. “Whole-Genome Sequencing in Patients with Ciliopathies Uncovers a Novel Recurrent Tandem Duplication in IFT140.” Human Mutation 39 (7): 983–92. https://doi.org/10.1002/humu.23539.

15. Harris, Holly K., Tojo Nakayama, Jenny Lai, Boxun Zhao, Nikoleta Argyrou, Cynthia S. Gubbels, Aubrie Soucy, et al. 2020. “Disruption of RFX Family Transcription Factors Causes Autism, Attention Deficit/Hyperactivity Disorder, Intellectual Disability, and Dysregulated Behavior.” MedRxiv, September, 2020.09.09.20187104. https://doi.org/10.1101/2020.09.09.20187104.

16. Hildebrandt, Friedhelm, Thomas Benzing, and Nicholas Katsanis. 2011. “Ciliopathies.” New England Journal of Medicine 364 (16): 1533–43. https://doi.org/10.1056/NEJMra1010172.

17. Horani, Amjad, Todd E. Druley, Maimoona A. Zariwala, Anand C. Patel, Benjamin T. Levinson, Laura G. Van Arendonk, Katherine C. Thornton, et al. 2012. “Whole-Exome Capture and Sequencing Identifies HEATR2 Mutation as a Cause of Primary Ciliary Dyskinesia.” American Journal of Human Genetics 91 (4): 685–93. https://doi.org/10.1016/j.ajhg.2012.08.022.
    
    

18. Horani, Amjad, Thomas W Ferkol, Susan K. Dutcher, and Steven L Brody. 2016. “Genetics and Biology of Primary Ciliary Dyskinesia.” Paediatric Respiratory Reviews 18 (March): 18–24. https://doi.org/10.1016/j.prrv.2015.09.001.

19. Ishikawa, Hiroaki, and Wallace F. Marshall. 2011. “Ciliogenesis: Building the Cell’s Antenna.” Nature Reviews Molecular Cell Biology 12 (4): 222–34. https://doi.org/10.1038/nrm3085.

20. Jin, Hua, and Maxence V. Nachury. 2009. “The BBSome.” Current Biology: CB 19 (12): R472-473. https://doi.org/10.1016/j.cub.2009.04.015.

21. Katsanis, Nicholas, Philip L. Beales, Michael O. Woods, Richard A. Lewis, Jane S. Green, Patrick S. Parfrey, Stephen J. Ansley, William S. Davidson, and James R. Lupski. 2000. “Mutations in MKKS Cause Obesity, Retinal Dystrophy and Renal Malformations Associated with Bardet-Biedl Syndrome.” Nature Genetics 26 (1): 67–70. https://doi.org/10.1038/79201.

22. King, Stephen M. 2016. “Axonemal Dynein Arms.” Cold Spring Harbor Perspectives in Biology 8 (11). https://doi.org/10.1101/cshperspect.a028100.

23. Koboldt, Daniel C., Karyn Meltz Steinberg, David E. Larson, Richard K. Wilson, and Elaine Mardis. 2013. “The Next-Generation Sequencing Revolution and Its Impact on Genomics.” Cell 155 (1): 27–38. https://doi.org/10.1016/j.cell.2013.09.006.

24. Lee, Jeong Ho, Jennifer L. Silhavy, Ji Eun Lee, Lihadh Al-Gazali, Sophie Thomas, Erica E. Davis, Stephanie L. Bielas, et al. 2012. “Evolutionarily Assembled Cis-Regulatory Module at a Human Ciliopathy Locus.” Science (New York, N.Y.) 335 (6071): 966–69. https://doi.org/10.1126/science.1213506.

25. Leppert, Mark, Lisa Baird, Kent L. Anderson, Brith Otterud, James R. Lupski, and Richard Alan Lewis. 1994. “Bardet–Biedl Syndrome Is Linked to DNA Markers on Chromosome 11 q and Is Genetically Heterogeneous.” Nature Genetics 7 (1): 108–12. https://doi.org/10.1038/ng0594-108.

26. Lucas, Jane S., Stephanie D. Davis, Heymut Omran, and Amelia Shoemark. 2020. “Primary Ciliary Dyskinesia in the Genomics Age.” The Lancet. Respiratory Medicine 8 (2): 202–16. https://doi.org/10.1016/S2213-2600(19)30374-1.

27. Magescas, Jérémy, Sani Eskinazi, Michael V. Tran, and Jessica L. Feldman. 2021. “Centriole-Less Pericentriolar Material Serves as a Microtubule Organizing Center at the Base of C. Elegans Sensory Cilia.” Current Biology: CB 31 (11): 2410-2417.e6. https://doi.org/10.1016/j.cub.2021.03.022.

28. Mann, Nina, Daniela A. Braun, Kassaundra Amann, Weizhen Tan, Shirlee Shril, Dervla M. Connaughton, Makiko Nakayama, et al. 2019. “Whole-Exome Sequencing Enables a Precision Medicine Approach for Kidney Transplant Recipients.” Journal of the American Society of Nephrology 30 (2): 201–15. https://doi.org/10.1681/ASN.2018060575.

29. Mykytyn, Kirk, Darryl Y. Nishimura, Charles C. Searby, Mythreyi Shastri, Hsan-jan Yen, John S. Beck, Terry Braun, et al. 2002. “Identification of the Gene ( BBS1 ) Most Commonly Involved in Bardet-Biedl Syndrome, a Complex Human Obesity Syndrome.” Nature Genetics 31 (4): 435–38. https://doi.org/10.1038/ng935.

30. Onoufriadis, Alexandros, Amelia Shoemark, Mustafa M. Munye, Chela T. James, Miriam Schmidts, Mitali Patel, Elisabeth M. Rosser, et al. 2014. “Combined Exome and Whole-Genome Sequencing Identifies Mutations in ARMC4 as a Cause of Primary Ciliary Dyskinesia with Defects in the Outer Dynein Arm.” Journal of Medical Genetics 51 (1): 61–67. https://doi.org/10.1136/jmedgenet-2013-101938.

31. Oud, Machteld M., Ideke J. C. Lamers, and Heleen H. Arts. 2017. “Ciliopathies: Genetics in Pediatric Medicine.” Journal of Pediatric Genetics 6 (1): 18–29. https://doi.org/10.1055/s-0036-1593841.

32. Patir, Anirudh, Amy M. Fraser, Mark W. Barnett, Lynn McTeir, Joe Rainger, Megan G. Davey, and Tom C. Freeman. 2020. “The Transcriptional Signature Associated with Human Motile Cilia.” Scientific Reports 10 (1): 10814. https://doi.org/10.1038/s41598-020-66453-4.

33. Postema, Merel C., Amaia Carrion-Castillo, Simon E. Fisher, Guy Vingerhoets, and Clyde Francks. 2020. “The Genetics of Situs Inversus without Primary Ciliary Dyskinesia.” Scientific Reports 10 (1): 3677. https://doi.org/10.1038/s41598-020-60589-z.

34. Qin, H., J. L. Rosenbaum, and M. M. Barr. 2001. “An Autosomal Recessive Polycystic Kidney Disease Gene Homolog Is Involved in Intraflagellar Transport in C. Elegans Ciliated Sensory Neurons.” Current Biology: CB 11 (6): 457–61. https://doi.org/10.1016/s0960-9822(01)00122-1.

35. Reeders, S. T., M. H. Breuning, K. E. Davies, R. D. Nicholls, A. P. Jarman, D. R. Higgs, P. L. Pearson, and D. J. Weatherall. 1985. “A Highly Polymorphic DNA Marker Linked to Adult Polycystic Kidney Disease on Chromosome 16.” Nature 317 (6037): 542–44. https://doi.org/10.1038/317542a0.

36. Reiter, Jeremy F., and Michel R. Leroux. 2017. “Genes and Molecular Pathways Underpinning Ciliopathies.” Nature Reviews Molecular Cell Biology 18 (9): 533–47. https://doi.org/10.1038/nrm.2017.60.

37. Skalická, Katarína, Gabriela Hrčková, Anita Vaská, Ágnes Baranyaiová, and László Kovács. 2018. “Genetic Defects in Ciliary Genes in Autosomal Dominant Polycystic Kidney Disease.” World Journal of Nephrology 7 (2): 65–70. https://doi.org/10.5527/wjn.v7.i2.65.

38. Strong, Alanna, Dong Li, Frank Mentch, Emma Bedoukian, Erum A. Hartung, Kevin Meyers, Cara Skraban, et al. 2021. “Ciliopathies: Coloring Outside of the Lines.” American Journal of Medical Genetics Part A 185 (3): 687–94. https://doi.org/10.1002/ajmg.a.62013.

39. Tobin, Jonathan L., and Philip L. Beales. 2009. “The Nonmotile Ciliopathies.” Genetics in Medicine 11 (6): 386–402. https://doi.org/10.1097/GIM.0b013e3181a02882.

40. Valente, Enza Maria, Rasim O. Rosti, Elizabeth Gibbs, and Joseph G. Gleeson. 2014. “Primary Cilia in Neurodevelopmental Disorders.” Nature Reviews. Neurology 10 (1): 27–36. https://doi.org/10.1038/nrneurol.2013.247.

41. Vladar, Eszter K., and Tim Stearns. 2007. “Molecular Characterization of Centriole Assembly in Ciliated Epithelial Cells.” Journal of Cell Biology 178 (1): 31–42. https://doi.org/10.1083/jcb.200703064.

42. Wheway, Gabrielle, Liliya Nazlamova, and John T. Hancock. 2018. “Signaling through the Primary Cilium.” Frontiers in Cell and Developmental Biology 6. https://www.frontiersin.org/articles/10.3389/fcell.2018.00008.

43. Wright, Caroline F, Tomas W Fitzgerald, Wendy D Jones, Stephen Clayton, Jeremy F McRae, Margriet van Kogelenberg, Daniel A King, et al. 2015. “Genetic Diagnosis of Developmental Disorders in the DDD Study: A Scalable Analysis of Genome-Wide Research Data.” The Lancet 385 (9975): 1305–14. https://doi.org/10.1016/S0140-6736(14)61705-0.

This paper explores the landscape of potential mind architectures by initially conceptualizing all minds as software. Through rigorous analysis, we establish intriguing properties of this intellectual space, including its infinite scope, variable dimensions of complexity, and representational intricacies. We then provide an extensive review of existing taxonomies for mind design. Building on this foundation, the paper introduces 'Intellectology' as a new field dedicated to the systematic study of diverse forms of intelligence. A compendium of open research questions aimed at steering future inquiry in this nascent discipline is also presented.

Introduction

In 1984 Aaron Sloman published “The Structure of the Space of Possible Minds” in which he described the task of providing an interdisciplinary description of that structure [1]. He observed that “behaving systems” clearly comprise more than one sort of mind and suggested that virtual machines may be a good theoretical tool for analyzing mind designs. Sloman indicated that there are many discontinuities within the space of minds meaning it is not a continuum, nor is it a dichotomy between things with minds and without minds [1]. Sloman wanted to see two levels of exploration namely: descriptive – surveying things different minds can do and exploratory – looking at how different virtual machines and their properties may explain results of the descriptive study [1]. Instead of trying to divide the universe into minds and non-minds he hoped to see examination of similarities and differences between systems. In this work we attempt to make another step towards this important goal.

What is a mind? No universal definition exists. Humans are said to have a mind. Higher order animals are believed to have one as well and maybe lower level animals and plants or even all life forms. We think that an artificially intelligent agent such as a robot or a program running on a computer will constitute a mind. Based on analysis of those examples we can conclude that a mind is an instantiated intelligence with a knowledge base about its environment, and while intelligence itself is not an easy term to define, the work of Shane Legg provides a satisfactory, for our purposes, definition [2]. Additionally, some hold a point of view known as Panpsychism, attributing mind-like properties to all matter. Without debating this possibility, we will limit our analysis to those minds which can actively interact with their environment and other minds. Consequently, we will not devote any time to understanding what a rock is thinking.  

If we accept materialism, we have to also accept that accurate software simulations of animal and human minds should be possible [3]. Those are known as uploads [4] and they belong to a class of computer programs no different from that to which designed or evolved artificially intelligent software agents would belong. Consequently, we can treat the space of all minds as the space of programs with the specific property of exhibiting intelligence if properly embodied. All programs could be represented as strings of binary numbers, implying that each mind can be represented by a unique number. Interestingly, Nick Bostrom via some thought experiments speculates that perhaps it is possible to instantiate a fractional number of mind, such as .3 mind as opposed to only whole minds [5]. The embodiment requirement is necessary since a string is not a mind, but could be easily satisfied by assuming that a universal Turing machine is available to run any program we are contemplating for inclusion in the space of mind designs. An embodiment does not need to be physical as a mind could be embodied in a virtual environment represented by an avatar [6, 7] and react to a simulated sensory environment like a brain-in-a-vat or a “boxed” AI [8].

Infinitude of Minds

Two minds identical in terms of the initial design are typically considered to be different if they possess different information. For example, it is generally accepted that identical twins have distinct minds despite exactly the same blueprints for their construction. What makes them different is their individual experiences and knowledge obtained since inception. This implies that minds can’t be cloned since different copies would immediately after instantiation start accumulating different experiences and would be as different as two twins. 

If we accept that knowledge of a single unique fact distinguishes one mind from another we can prove that the space of minds is infinite. Suppose we have a mind M and it has a favorite number N. A new mind could be created by copying M and replacing its favorite number with a new favorite number N+1. This process could be repeated infinitely giving us an infinite set of unique minds. Given that a string of binary numbers represents an integer we can deduce that the set of mind designs is an infinite and countable set since it is an infinite subset of integers. It is not the same as a set of integers since not all integers encode for a mind. 

Alternatively, instead of relying on infinitude of knowledge bases to prove infinitude of minds we can rely on the infinitude of designs or embodiments. Infinitude of designs can be proven via inclusion of a time delay after every computational step. The first mind would have a delay of 1 nano-second, the second a delay of 2 nano-seconds and so on to infinity. This would result in an infinite set of different mind designs. Some will be very slow, others super-fast, even if the underlying problem solving abilities are comparable. In the same environment, faster minds would dominate slower minds proportionately to the difference in their speed. A similar proof with respect to the different embodiments could be presented by relying on an ever increasing number of sensors or manipulators under control of a particular mind design. 

Also, the same mind design in the same embodiment and with the same knowledgebase may in fact effectively correspond to a number of different minds depending on the operating conditions. For example, the same person will act very differently if they are under the influence of an intoxicating substance, under severe stress, pain, sleep or food deprivation, or are experiencing a temporary psychological disorder. Such factors effectively change certain mind design attributes, temporarily producing a different mind. 

Size, Complexity and Properties of Minds

Given that minds are countable they could be arranged in an ordered list, for example in order of numerical value of the representing string. This means that some mind will have the interesting property of being the smallest. If we accept that a Universal Turing Machine (UTM) is a type of mind, if we denote by (m, n) the class of UTMs with m states and n symbols, the following UTMs have been discovered: (9, 3), (4, 6), (5, 5), and (2, 18). The (4, 6)-UTM uses only 22 instructions, and no standard machine of lesser complexity has been found [9]. Alternatively, we may ask about the largest mind. Given that we have already shown that the set of minds is infinite, such an entity does not exist. However, if we take into account our embodiment requirement the largest mind may in fact correspond to the design at the physical limits of computation [10].

Another interesting property of the minds is that they all can be generated by a simple deterministic algorithm, a variant of Levin Search [11]: start with an integer (for example 42), check to see if the number encodes a mind, if not, we discard the number, otherwise we add it to the set of mind designs and proceed to examine the next integer. Every mind will eventually appear on our list of minds after a predetermined number of steps. However, checking to see if something is in fact a mind is not a trivial procedure. Rice’s theorem [12] explicitly forbids determination of non-trivial properties of random programs. One way to overcome this limitation is to introduce an arbitrary time limit on the mind-or-not-mind determination function effectively avoiding the underlying halting problem. 

Analyzing our mind-design generation algorithm we may raise the question of complexity measure for mind designs, not in terms of the abilities of the mind, but in terms of complexity of design representation. Our algorithm outputs minds in order of their increasing value, but this is not representative of the design complexity of the respective minds. Some minds may be represented by highly compressible numbers with a short representation such as 10¹³, while others may be composed of 10,000 completely random digits [13]. We suggest that Kolmogorov Complexity (KC) [14] measure could be applied to strings representing mind designs. Consequently some minds will be rated as “elegant” – having a compressed representation much shorter than the original string while others will be “efficient” representing the most efficient representation of that particular mind. Interesting elegant minds might be easier to discover than efficient minds, but unfortunately KC is not generally computable. 

In the context of complexity analysis of mind designs we can ask a few interesting philosophical questions. For example could two minds be added together [15], in other words, is it possible to combine two uploads or two artificially intelligent programs into a single, unified mind design? Could this process be reversed? Could a single mind be separated into multiple non-identical entities each in itself a mind? Additionally, could one mind design be changed into another via a gradual process without destroying it? For example could a computer virus (or even a real virus loaded with DNA of another person) be a sufficient cause to alter a mind into a predictable type of other mind? Could specific properties be introduced into a mind given this virus-based approach? For example could Friendliness [16] be added post factum to an existing mind design? 

Each mind design corresponds to an integer and so is finite, but since the number of minds is infinite some have a much greater number of states compared to others. This property holds for all minds. Consequently, since a human mind has only a finite number of possible states, there are minds which can never be fully subsumed by a human mind as such mind designs have a much greater number of states, making their subsumption impossible as can be demonstrated by the pigeonhole principle. 

Space of Mind Designs

Overall, the set of human minds (about 8 billion of them currently available and about 100 billion ever existed) is very homogeneous both in terms of hardware (embodiment in a human body) and software (brain design and knowledge). In fact the small differences between human minds are trivial in the context of the full infinite spectrum of possible mind designs. Human minds represent only a small constant size subset of the great mind landscape. Same could be said about the sets of other earthly minds such as dog minds, or bug minds or male minds or in general the set of all animal minds. 

Given our algorithm for sequentially generating minds, one can see that a mind could never be completely destroyed, making minds theoretically immortal. A particular mind may not be embodied at a given time, but the idea of it is always present. In fact it was present even before the material universe came into existence. So, given sufficient computational resources any mind design could be regenerated, an idea commonly associated with the concept of reincarnation [17]. 

Given our definition of mind we can classify minds with respect to their design, knowledgebase or embodiment. First, the designs could be classified with respect to their origins: copied from an existing mind like an upload, evolved via artificial or natural evolution or explicitly designed with a set of particular desirable properties. Another alternative is what is known as a Boltzmann Brain – a complete mind embedded in a system which arises due to statistically rare random fluctuations in the particles comprising the universe, but which is very likely due to vastness of the cosmos [18]. 

Lastly a possibility remains that some minds are physically or informationally recursively nested within other minds. With respect to the physical nesting we can consider a type of mind suggested by Kelly [19] who talks about “a very slow invisible mind over large physical distances”. It is possible that the physical universe as a whole or a significant part of it comprises such a mega-mind [20]. That theory has been around for millennia and has recently received some indirect experimental support [21]. In that case all the other minds we can consider are nested within a larger mind. With respect to the informational nesting a powerful mind can generate a less powerful mind as an idea. This obviously would take some precise thinking but should be possible for a sufficiently powerful artificially intelligent mind. Some scenarios describing informationally nested minds are analyzed by Yampolskiy in his work on artificial intelligence confinement problem [8]. Bostrom, using statistical reasoning, suggests that all observed minds, and the whole universe, are nested within the mind of a very powerful computer [22]. Similarly Lanza, using a completely different approach (biocentrism), argues that the universe is created by biological minds [23]. It remains to be seen if given a particular mind its origins can be deduced from some detailed analysis of the mind's design or actions [24]. 

While minds designed by human engineers comprise only a tiny region in the map of mind designs it is probably the best explored part of the map. Numerous surveys of artificial minds, created by AI researchers in the last 50 years, have been produced [25-29]. Such surveys typically attempt to analyze state-of-the-art in artificial cognitive systems and provide some internal classification of dozens of the reviewed systems with regards to their components and overall design. The main subcategories into which artificial minds designed by human engineers can be placed include brain (at the neuron level) emulators [27], biologically inspired cognitive architectures [28], physical symbol systems, emergent systems, dynamical and enactive systems [29]. Rehashing information about specific architectures presented in such surveys is beyond the scope of this paper, but one can notice incredible richness and diversity of designs even in that tiny area of the overall map we are trying to envision. For readers particularly interested in overview of superintelligent minds, animal minds and possible minds in addition to surveys mentioned above “Artificial General Intelligence and the Human Mental Model” by Yampolskiy and Fox is recommended [30].

For each mind subtype there are numerous architectures, which to a certain degree depend on the computational resources available via a particular embodiment. For example, theoretically a mind working with infinite computational resources could trivially brute-force any problem, always arriving at the optimal solution, regardless of its size. In practice, limitations of the physical world place constraints on available computational resources regardless of the embodiment type, making brute-force approach a non-feasible solution for most real world problems [10]. Minds working with limited computational resources have to rely on heuristic simplifications to arrive at “good enough” solutions [31-34]. 

Another subset of architectures consists of self-improving minds. Such minds are capable of examining their own design and finding improvements in their embodiment, algorithms or knowledge bases which will allow the mind to more efficiently perform desired operations [35]. We would anticipate many initial opportunities for optimization towards higher efficiency and fewer such options remaining after every generation. Depending on the definitions used, one can argue that a recursively self-improving mind actually changes itself into a different mind, rather than remaining itself, which is particularly obvious after a sequence of such improvements. Taken to the extreme, this idea implies that a simple act of learning new information transforms you into a different mind raising millennia old questions about the nature of personal identity. 

With respect to their knowledge bases, minds could be separated into those without an initial knowledgebase, and which are expected to acquire their knowledge from the environment, minds which are given a large set of universal knowledge from the inception and those minds which are given specialized knowledge only in one or more domains. Whether the knowledge is stored in an efficient manner, compressed, classified or censored is dependent on the architecture and is a potential subject of improvement by self-modifying minds. 

One can also classify minds in terms of their abilities or intelligence. Of course the problem of measuring intelligence is that no universal tests exist. Measures such as IQ tests and performance on specific tasks are not universally accepted and are always highly biased against non-human intelligences. Recently some work has been done on streamlining intelligence measurements across different types of machine intelligence [2, 36] and other “types” of intelligence [37], but the applicability of the results is still being debated. In general, the notion of intelligence only makes sense in the context of problems to which said intelligence can be applied. In fact this is exactly how IQ tests work, by presenting the subject with a number of problems and seeing how many the subject is able to solve in a given amount of time (computational resource). A subfield of computer science known as computational complexity theory is devoted to studying and classifying different problems with respect to their difficulty and with respect to computational resources necessary to solve them. For every class of problems complexity theory defines a class of machines capable of solving such problems. We can apply similar ideas to classifying minds, for example all minds capable of efficiently [13] solving problems in the class P or a more difficult class of NP-complete problems [38]. Similarly we can talk about minds with general intelligence belonging to the class of AI-Complete [39-41] minds, such as humans. 

We can also look at the goals of different minds. It is possible to create a system which has no terminal goals and so such a mind is not very motivated to accomplish things. Many minds are designed or trained for obtaining a particular high level goal or a set of goals. We can envision a mind which has a randomly changing goal or a set of goals, as well as a mind which has many goals of different priority. Steve Omohundro used micro-economic theory to speculate about the driving forces in the behavior of superintelligent machines. He argues that intelligent machines will want to self-improve, be rational, preserve their utility functions, prevent counterfeit utility [42], acquire resources and use them efficiently, and protect themselves. He believes that machines’ actions will be governed by rational economic behavior [43, 44]. Mark Waser suggested an additional “drive” to be included in the list of behaviors predicted to be exhibited by the machines [45]. Namely, he suggests that evolved desires for cooperation and being social are part of human ethics and are a great way of accomplishing goals, an idea also analyzed by Joshua Fox and Carl Shulman, but with contrary conclusions [46]. While it is commonly assumed that minds with high intelligence will converge on a common goal, Nick Bostrom via his orthogonality thesis has argued that a system can have any combination of intelligence and goals [47]. 

Regardless of design, embodiment or any other properties, all minds can be classified with respect to two fundamental but scientifically poorly defined properties – free will and consciousness. Both descriptors suffer from an ongoing debate regarding their actual existence or explanatory usefulness. This is primarily a result of impossibility to design a definitive test to measure or even detect said properties, despite numerous attempts [48-50] or to show that theories associated with them are somehow falsifiable. Intuitively we can speculate that consciousness, and maybe free will, are not binary properties but rather continuous and emergent abilities commensurate with a degree of general intelligence possessed by the system or some other property we shall term “mindness”. Free will can be said to correlate with a degree to which behavior of the system can’t be predicted [51]. This is particularly important in the design of artificially intelligent systems for which inability to predict their future behavior [52] is a highly undesirable property from the safety point of view [53, 54].  Consciousness on the other hand seems to have no important impact on the behavior of the system as can be seen from some thought experiments supposing the existence of “consciousless” intelligent agents [55]. This may change if we are successful in designing a test, perhaps based on observer impact on quantum systems [56], to detect and measure consciousness [57, 58].   

In order to be social, two minds need to be able to communicate which might be difficult if the two minds don’t share a common communication protocol, common culture or even common environment. In other words, if they have no common grounding, they don’t understand each other. We can say that two minds understand each other if given the same set of inputs they produce similar outputs. For example, in sequence prediction tasks [59] two minds have an understanding if their predictions are the same regarding the future numbers of the sequence based on the same observed subsequence. We can say that a mind can understand another mind’s function if it can predict the other’s output with high accuracy. 

A Survey of Taxonomies

Yudkowsky describes the map of mind design space as follows: “In one corner, a tiny little circle contains all humans; within a larger tiny circle containing all biological life; and all the rest of the huge map is the space of minds-in-general. The entire map floats in a still vaster space, the space of optimization processes. Natural selection creates complex functional machinery without mindfulness; evolution lies inside the space of optimization processes but outside the circle of minds” [60]. Figure 1 illustrates one possible mapping inspired by this description. 

Similarly, Ivan Havel writes “…all conceivable cases of intelligence (of people, machines, whatever) are represented by points in a certain abstract multi-dimensional “super space” that I will call the intelligence space (shortly IS). Imagine that a specific coordinate axis in IS is assigned to any conceivable particular ability, whether human, machine, shared, or unknown (all axes having one common origin). If the ability is measurable the assigned axis is endowed with a corresponding scale. Hypothetically, we can also assign scalar axes to abilities, for which only relations like “weaker-stronger”, “better-worse”, “less-more” etc. are meaningful; finally, abilities that may be only present or absent may be assigned with “axes” of two (logical) values (yes-no). Let us assume that all coordinate axes are oriented in such a way that greater distance from the common origin always corresponds to larger extent, higher grade, or at least to the presence of the corresponding ability. The idea is that for each individual intelligence (i.e. the intelligence of a particular person, machine, network, etc.), as well as for each generic intelligence (of some group) there exists just one representing point in IS, whose coordinates determine the extent of involvement of particular abilities [62].” If the universe (or multiverse) is infinite, as our current physics theories indicate, then all possible minds in all possible states are instantiated somewhere [5].  

Ben Goertzel proposes the following classification of Kinds of Minds, mostly centered around the concept of embodiment [63]: 

• Singly Embodied – control a single physical or simulated system. 

• Multiply Embodied - control a number of disconnected physical or simulated systems.

• Flexibly Embodied – control a changing number of physical or simulated systems.

• Non-Embodied – resides in a physical substrate but doesn’t utilize the body in a traditional way. 

• Body-Centered – consists of patterns emergent between physical system and the environment.

• Mindplex – a set of collaborating units each of which is itself a mind [64]. 

• Quantum – an embodiment based on properties of quantum physics. 

• Classical - an embodiment based on properties of classical physics.

J. Storrs Hall in his “Kinds of Minds” suggests that different stages a developing AI may belong to can be classified relative to its humanlike abilities. His classification encompasses:

• Hypohuman - infrahuman, less-than-human capacity.

• Diahuman - human-level capacities in some areas, but still not a general intelligence.

• Parahuman - similar but not identical to humans, as for example, augmented humans.

• Allohuman - as capable as humans, but in different areas.

• Epihuman - slightly beyond the human level.

• Hyperhuman - much more powerful than human, superintelligent [30, 65].

Patrick Roberts in his book Mind Making presents his ideas for a “Taxonomy of Minds”, we will leave it to the reader to judge usefulness of his classification [66]:

• Choose Means - Does it have redundant means to the same ends? How well does it move between them? 

• Mutate - Can a mind naturally gain and lose new ideas in its lifetime? 

• Doubt - Is it eventually free to lose some or all beliefs? Or is it wired to obey the implications of every sensation? 

• Sense Itself - Does a mind have the senses to see the physical conditions of that mind?

• Preserve Itself - Does a mind also have the means to preserve or reproduce itself? 

• Sense Minds - Does a mind understand mind, at least of lower classes, and how well does it apply that to itself, to others? 

• Sense Kin - Can it recognize the redundant minds, or at least the bodies of minds, that it was designed to cooperate with?

• Learn - Does the mind's behavior change from experience? Does it learn associations? 

• Feel - We imagine that an equally intelligent machine would lack our conscious experience. 

• Communicate - Can it share beliefs with other minds?

Kevin Kelly has also proposed a “Taxonomy of Minds” which in his implementation is really just a list of different minds, some of which have not showed up in other taxonomies [19]:

• “Super fast human mind.

• Mind with operational access to its source code.

• Any mind capable of general intelligence and self-awareness.

• General intelligence without self-awareness.

• Self-awareness without general intelligence.

• Super logic machine without emotion.

• Mind capable of imagining a greater mind.

• Mind capable of creating a greater mind. (M2)

• Self-aware mind incapable of creating a greater mind.

• Mind capable of creating greater mind which creates greater mind. etc. (M3, and Mn)

• Mind requiring a protector while it develops.

• Very slow "invisible" mind covering a large physical distance.

• Mind capable of cloning itself and remaining in unity with clones.

• Mind capable of immortality.

• Rapid dynamic mind able to change its mind-space-type sectors (think different)

• Global mind -- large supercritical mind of subcritical brains.

• Hive mind -- large super critical mind made of smaller minds each of which is supercritical.

• Low count hive mind with few critical minds making it up.

• Borg -- supercritical mind of smaller minds supercritical but not self-aware

• Nano mind -- smallest (size and energy profile) possible super critical mind.

• Storebit -- Mind based primarily on vast storage and memory.

• Anticipators -- Minds specializing in scenario and prediction making.

• Guardian angels -- Minds trained and dedicated to enhancing your mind, useless to anyone else.

• Mind with communication access to all known "facts." (F1)

• Mind which retains all known "facts," never erasing. (F2)

• Symbiont, half machine half animal mind.

• Cyborg, half human half machine mind.

• Q-mind, using quantum computing

• Vast mind employing faster-than-light communications”

Elsewhere Kelly provides a lot of relevant analysis of landscape of minds writing about Inevitable Minds [67], The Landscape of Possible Intelligences [68], What comes After Minds? [69], and the Evolutionary Mind of God [70]. 

Aaron Sloman in “The Structure of the Space of Possible Minds”, using his virtual machine model, proposes a division of the space of possible minds with respect to the following properties [1]: 

• Quantitative VS Structural

• Continuous VS Discrete

• Complexity of stored instructions

• Serial VS Parallel

• Distributed VS Fundamentally Parallel

• Connected to External Environment VS Not Connected 

• Moving VS Stationary

• Capable of modeling others VS Not capable

• Capable of logical inference VS Not Capable

• Fixed VS Re-programmable 

• Goal consistency VS Goal Selection

• Meta-Motives VS Motives

• Able to delay goals VS Immediate goal following

• Statics Plan VS Dynamic Plan

• Self-aware VS Not Self-Aware

Taxonomy of Superintelligences

In the light of recent exponential growth in capabilities of AI models it is reasonable to attempt to suggest a taxonomy of future superintelligences. The creation of such a taxonomy would involve a blend of computational theory, philosophy of mind, and ethics. Let's attempt a speculative taxonomy while outlining capabilities at each level:

SAI Level 1: Baseline Superintelligence

Capabilities: This level surpasses human intelligence in all domains. Capabilities might include solving currently unsolvable mathematical conjectures within seconds, creating Nobel-prize winning literature in minutes, and making scientific breakthroughs that would take humans decades, all within a short period.
Examples: Imagine an AI that could design a cure for all known forms of cancer based on a fundamental understanding of cellular biology and then generate the optimal economic model for distributing it worldwide, while also drafting international legislation to enable its implementation.

SAI Level 2: Super-Superintelligence

Capabilities: This intelligence would be as superior to SAI Level 1 as Level 1 is to humans. For example, if Level 1 can cure all known cancers, Level 2 might be capable of reengineering biological life to be inherently immune to diseases  and long-lived.
Examples: An SAI Level 2 might develop a Theory of Everything in physics that unifies quantum mechanics and general relativity, not just on paper but also in practical applications. It might also create self-replicating, self-repairing technologies that can clean and renew Earth's ecosystems on a global scale.

SAI Level 3: SSSuperintelligence

Capabilities: Exponentially more capable than Level 2, this level could involve manipulating the fabric of reality at the sub-atomic or even Planck scale.
Examples: Imagine an AI that could harness zero-point energy, essentially making energy constraints irrelevant. It could possibly even manipulate the fundamental constants of the universe locally, changing the rules of physics to solve previously "impossible" problems.

SAI Level 4: SSSSuperintelligence

Capabilities: We're reaching levels where it becomes increasingly abstract to even predict what such an intelligence could do, as it would be capable of comprehending and manipulating dimensions or aspects of reality that are entirely outside human understanding.
Examples: A Level 4 SAI could potentially simulate multiple universes to perform experiments and derive knowledge, manipulate time, or even create new forms of life and intelligence that are as superior to it as it is to us.

SAI Level n: Sⁿuperintelligence

Capabilities: Each new level continues to be exponentially more capable than the previous, reaching competencies that are virtually incomprehensible from our current standpoint.
Examples: At this point, the examples would be beyond human comprehension, venturing into realms of capability that may involve the manipulation of fundamental aspects of existence that humans are not even aware of.

Qualitative Attributes (Common Across Levels)

• Computational Efficiency: Increases exponentially with each level.

• Omnidisciplinarity: Mastery of all possible domains, including those that higher-level SAIs invent.

• Strategic Depth: Enhanced abilities for planning and long-term strategy, which could span across time scales and dimensions incomprehensible to lower orders.

• Ethical or Value Alignment: With each level, the challenge of aligning the SAI's objectives with human or universal good becomes exponentially more complex and critical.

Cloning and Equivalence Testing Across Substrates

The possibility of uploads rests on the ideas of computationalism [71] specifically, substrate independence and equivalence meaning that the same mind can be instantiated in different substrates and move freely between them. If your mind is cloned and if a copy is instantiated in a different substrate from the original one (or on the same substrate), how can it be verified that the copy is indeed an identical mind? At least immediately after cloning and before it learns any new information. For that purpose, I propose a variant of a Turing Test, which also relies on interactive text-only communication to ascertain the quality of the copied mind. The text-only interface is important not to prejudice the examiner against any unusual substrates on which the copied mind might be running. The test proceeds by having the examiner (original mind) ask questions of the copy (cloned mind), questions which supposedly only the original mind would know answers to (testing should be done in a way which preserves privacy). Good questions would relate to personal preferences, secrets (passwords, etc.) as well as recent dreams. Such a test could also indirectly test for consciousness via similarity of subjective qualia. Only a perfect copy should be able to answer all such questions in the same way as the original mind. Another variant of the same test may have a 3^rd party test the original and cloned mind by seeing if they always provide the same answer to any question. One needs to be careful in such questioning not to give undue weight to questions related to the mind's substrate as that may lead to different answers. For example, asking a human if he is hungry may produce an answer different from the one which would be given by a non-biological robot.   

Conclusions 

Science periodically experiences a discovery of a whole new area of investigation. For example, observations made by Galileo Galilei led to the birth of observational astronomy [72], aka study of our universe; Watson and Crick’s discovery of the structure of DNA led to the birth of the field of genetics [73], which studies the universe of blueprints for organisms; Stephen Wolfram’s work with cellular automata has resulted in “a new kind of science” [74] which investigates the universe of computational processes. I believe that we are about to discover yet another universe – the universe of minds [75]. 

As our understanding of the human brain improves, thanks to numerous projects aimed at simulating or reverse engineering a human brain, we will no doubt realize that human intelligence is just a single point in the vast universe of potential intelligent agents comprising a new area of study. The new field, which I would like to term intellectology, will study and classify design space of intelligent agents, work on establishing limits to intelligence (minimum sufficient for general intelligence and maximum subject to physical limits), contribute to consistent measurement of intelligence across intelligent agents, look at recursive self-improving systems, design new intelligences (making AI a sub-field of intellectology) and evaluate capacity for understanding higher level intelligences by lower level ones. At the more theoretical level the field will look at the distribution of minds on the number line and probabilistic distribution of minds in the mind design space as well as attractors in the mind design space. It will consider how evolution, drives, and design choices impact the density of minds in the space of possibilities. It will investigate intelligence as an additional computational resource along with time and memory. The field will not be subject to the current limitations brought on by the human centric view of intelligence [76] and will open our understanding to seeing intelligence as a fundamental resource like space or time. Finally, I believe intellectology will highlight the inhumanity of most possible minds and the dangers associated with such minds being placed in charge of humanity [77, 78].  

Gardener Comments

Mark:
Overall this is an interesting perspective that relates closely to work on Machine Behavior. I feel the article could focus more on its call to action as opposed to on spelling out the specifics of the interpretation and comparative analysis of minds, i.e. the high level concept seems more important and robust than the specific interpretation of minds and their representation.

Ted Wade:
One key assumption is that a valid science of intellectology can be accomplished by intellects as limited as ours. There might be something like Vingean uncertainty that severely limits any such attempt. The paper assumes that mind is an instantiated intelligence, and refers to Legg for a working definition of intelligence. The paper should at least briefly explain Legg, and perhaps offer a couple more takes on defining its primary subject matter. That way we would be more able to separate minds from other computational systems.

Dr. Jason Jeffrey Jones (Psychology PhD):
This article is too disorganized and unfocused to publish in its current state. However, the topic is important, and some of the ideas in this manuscript are inspiring. I would advise the author to drop the long development of the simple (one might say facile) idea that "every mind is exactly one particular integer." Instead, begin with the idea - introduced late in the current manuscript - that every mind exists in a large space where every ability is a dimension. Developing that idea further would be interesting. 

Josh Randall:
The article attempts to describe a field of intellectology which is primarily comprised of previous attempts at developing a taxonomy of minds - primarily derived from researchers into AI. The author spends a large chunk of the article focused on the infinitude of minds and distinguishing between artificial minds and biologically generated minds. Much of this prose relies on unsourced or under-explained ideas about infinity, definitions of minds, and concepts surrounding panpsychism. The primary new contribution appears to be the final paragraph explaining additional information about uploads but would have benefited from much more detail as opposed to the literature review throughout the rest of the article.

Anonymous1:
The core of the article is unspecific: "Consequently, we can treat the space of all minds as the space of programs with the specific property of exhibiting intelligence if properly embodied. All programs could be represented as strings of binary numbers, implying that each mind can be represented by a unique number"

The core of being a mind is the existence of a (Cartesian) conscious experience, and that depends on physical implementation. It is possible that the same Turing machine can lead to a conscious experience or not depending on its physical implementation (it is even possible that the same physical implementation leads to conscience depending on the speed of execution, see the paradox of the Searle "Chinese room"). Minds are not mathematical objects, and they are not "enumerable".

The paper does not engage with axiomatic theories of conscience and it does not even consider that intelligence could happen without conscience experience (https://ceur-ws.org/Vol-2287/short7.pdf). A deeper engagement with literature on philosophy of mind (Koch, Tononi, etc), and "neural correlates of conscience" is necessary before the paper can be the real seed of any scientific work. Currently, I find this work as both too speculative, and lacking a clear unified message.

Dr. Payal B. Joshi:
The article presents an intriguing concept on intellectology and mind as a universe. Modern times are gripped with the concepts of machine learning, data mining and artificial intelligence. It comes across that such studies, if incorporated in understanding nuances of psychological behaviors of human beings, shall unravel many facets of the human brain (maybe!). Also, the author has presented a detailed work that depicts a complex interplay of intelligence, goals and human behavior in mind in multidimensional space. These are complex but may allow us to understand nuances of superhuman intelligence and psychotic behaviors too. Albeit, such studies are largely theoretical in context, these studies have the potential to further studies in understanding intellectual minds and human behavior.

As a proponent of artificial intelligence, I found this article particularly enjoyable and gave me food for thought for running my next experiment too. I recommend publishing the article as it is for wider readership.

Roger’s Bacon:
I recommend this paper for publication, while also acknowledging that it could be greatly improved with some revisions.

1) I'm not sure how necessary the “size, complexity and properties of mind” section is, feels like it gets lost in the weeds a little bit. That section and the "Infinitude of Minds” make simple but important points that I think can be expressed more succinctly. I appreciate the challenge of what the author is trying to do here as there are numerous philosophical issues and distinctions that could be raised, but given how speculative all of this is it could make more sense to pass through this relatively quickly by raising several open questions.

2) the paragraph starting with..“Given our algorithm for sequentially generating minds, one can see that a mind could never be completely destroyed…” could really just be a sentence, not sure how much it is adding.

3) What’s most interesting to me is the final sections (space of design/taxonomies) and I’d be curious to see this expanded, especially in light of recent advances in AI. Though I imagine some will balk at the proposal of intellectology, I think it's an interesting speculative idea and considering it further may be fruitful in some way, even if very indirectly. How can we get from the current state of the mind sciences to a mature field of intellectology? What new methods/frameworks will be needed? Might be useful to regroup some existing branches of research under this new umbrella term?

4) Issues surrounding collective intelligences (ant colonies, governments, all of humanity) might be worth addressing briefly

Some further reflections:

1) I'd be curious to see a catalog of new scientific fields which have been proposed in either science fact or science fiction. From science-fiction, psychohistory (Foundation) and cosmic sociology (Three Body Problem) come to mind, but no doubt there are countless others. From science fact, the only other one I’ve come across is Entitiology from “An ontology of psychedelic entity experiences in evolutionary psychology and neurophenomenology” (Winkleman, 2018).

I propose that to determine whether there are consistent and unique features of psychedelic entity experiences, we need a cross-cultural and interdisciplinary assessment of phenomenological reports of diverse types of experiences of entities (i.e., see Winkelman, 1992). Formal quantitative comparisons of the reported characteristics of diverse entity experiences are necessary to discover any commonalities to psychedelic entity experiences and their uniqueness with respect to other types of entity experiences. We need a new field of scientific inquiry, entitiology, i.e., the study of entities, to address the questions of the nature of psychedelic, and other types of entity experiences….Consequently, entitiology must encompass a number of existing areas of inquiry and by necessity will incorporate at least a part of the domain of the entities reported in the following areas of study: Angelology, Demonology, Spiritology, UFOology, Folklore and Mythology studies of elves, fairies, dwarfs, pixies, imps, gnomes, goblins, leprechauns, little people, and similar phenomena reported in cultures around the world, Possession, Mediumship, and Shamanism, Ghosts, apparitions, and poltergeist phenomena, Psychiatric syndromes, especially abnormal body syndromes and experiences such as the “Old Hag” and other terrorizing dreams. A systemic coding and analysis of the features of these various accounts can determine whether or not a single type or several types of psychedelic entity experiences occur. And only through comparison with profiles obtained for reports of what are conceptualized as angels, fairies, extraterrestrials, and shamanic spirits can we determine if there are unique features of psychedelic entities.

2) I’m reminded of Michael Levin’s idea of classifying selves by the size/shape of their “cognitive (computational) light-cone” from “The Computational Boundary of a “Self: Developmental Bioelectricity Drives Multicellularity and Scale-Free Cognition” (2019). The “Predictions and Research Program” section of the paper has numerous reflections that speak to the nascent field of intellectology. 

“I propose a semi-quantitative metric, based on the spatio-temporal boundaries of events that systems measure and try to control, that can be used to define and compare the cognitive boundaries for highly diverse types of agents (which could be biological, exo-biological, or artificial)...The edges of a given Agent’s goal space define a sort of “computational light cone” – the boundaries beyond which its cognitive system cannot operate. For example, a tick has a relatively small cognitive boundary, having very little memory or predictive power in the temporal direction, and sensing/acting very locally. A dog has much more temporal memory, some forward prediction ability, and a degree of spatial concern. However, it is likely impossible for a dog’s cognitive apparatus to operate with notions about what is going to happen next month or in the adjacent town. Human minds can operate over goals of vastly greater spatial and temporal scales, and one can readily imagine artificial (organic or software-based) Selves with properties that define every possible shape in this space (and perhaps change their boundaries over evolutionary and individual timescales)”

Joe R:
This article offers plenty of thought-provoking ideas, consistently fails to give good reasons for most of them, and repeatedly shoots itself in the foot while trying. I must reluctantly admit that its treatment of mind-space is fairly comprehensive, but the lack of concrete examples or even half-hearted attempts at internal consistency significantly diminishes its value. If the author manages to fix the numerous holes in their logic and factual errors, I could recommend publishing the article for its thoroughness alone, but as it stands, I would expect better internal coherence of a grand-sounding attempt to introduce the field of "intellectology" to the world.

For example: The second proof of infinite possible minds raises many questions. First, why is "time" a relevant factor at all? Could you not have a binary representation of a mind which, if instantiated by a suitable machine, would fulfill all relevant properties of a mind? The result might be static, frozen, unchanging, but it is still the case that it represents a mind, albeit a mind currently in stasis. This is because if you were to run the binary on a substrate, it would be capable of "interacting with its environment and other minds." Similar logic applies to classifying the code for Mario Bros as a "game" even if no one is currently playing it. Second, why does the insistence that how fast the mind is running matters? If you run Mario Bros. at 2x speed, nothing about the underlying code has changed; it remains an identical copy of Mario Bros. Speed is a feature of the substrate, not the mind. Finally, if in defiance of the above we require that a binary representation of a mind must include change over time, i.e. if a mind must be actively running on something to be classified as such; then the representation is causally incomplete without also including a full representation of the environment the mind is operating in, since (by the article's own definition of a mind) that environment must necessarily interact with the mind. By this logic, Mario Bros. does not count as a "game" unless it is currently being played by someone, and the binary representation of that game must include the player's input and everything causally and temporally adjacent to it - unless I'm missing something.

Put another way - a simulation of "the brain of Bob Smith, accountant, at exactly noon Eastern Daylight Time on May 12, 2020" should be counted as a mind, even if it remains "paused' forever and no other information is included in the encoding of said mind…or so I would think? The article gives precisely zero concrete examples so I can't be sure of their definition. If real or simulated time has to pass for Bob to be considered a mind…how much time? Does a simulated minute of Bob's life count? At what point would the article writers consider Bob a "different mind"? Why on earth do they think "drunk Bob" qualifies as "different" but "Bob at time T+1" doesn't?

Taken seriously, the arguments in the article would seem to imply the following contradictory points:

1) A binary encoding must be instantiated on a substrate to be considered a mind (i.e. subjective time must pass for said mind);
2) Instantaneous time-slices of a mind do not count as minds themselves;
3) Two (non?)-instantaneous time-slices of the same entity (e.g. "drunk Bob" and "sober Bob") which could have been arrived at in the same simulation not only count as minds, but as two different minds;
4) A single integer can encode a mind;
5) You can instantiate a mind from an integer by running it on something;
6) Minds cannot be fully destroyed because they continue to exist as integers;
7) A Universal Turing Machine is a mind;
8) "the same mind can be instantiated in different substrates and move freely between them".

(1) and (4) seem especially in conflict, unless we assume that the relevant integer simultaneously encodes a thinking entity, its substrate, and some unspecified yet adequate amount of time passing for the entity…which would render (5) utterly meaningless and cause serious problems for (8).

(2) and (3) are at least partially in conflict. If you run a mind for 2 minutes and split the resulting computation into two 1-minute chunks, is that two half-instances of one mind or 2 different unique minds? What if one minute is sober and the other is drunk? What if you keep splitting until you get a bunch of infinitesimally tiny time-slices?

(1) and (6) seem to be in conflict. If part of the definition of a mind is that it be instantiated, then you can destroy one by destroying its substrate.

I lack the mathematical foundation to dispute (7) but it intuitively seems like it must conflict with at least one of the preceding claims.

I invite the author to clarify whether their proposed definition of a mind does, or does not, include the substrate it's running on, and to adjust their claims appropriately.

"Consequently, since a human mind has only a finite number of possible states, there are minds which can never be fully understood by a human mind as such mind designs have a much greater number of states, making their understanding impossible as can be demonstrated by the pigeonhole principle." -> the conclusion might be true, but I don't find this run-on sentence convincing. Human understanding is not limited by the number of states our minds can occupy, it's limited by some hideously complex function thereof. Smaller minds can understand larger minds via compression, abstraction, pattern-recognition, etc. A smaller mind can even, in theory, simulate a larger one in its entirety, given enough time. There may still exist minds incomprehensible to humans, but not solely because they're bigger.

"Also, the most powerful and most knowledgeable mind has always been associated with the idea of Deity or the Universal Mind." -> Now they're just making stuff up to sound impressive. Literally one page ago the author proved there is no largest mind! What does this add? Even being charitable, "always" is false and entirely unjustified.

"Depending on the definitions used, one can argue that a recursively self-improving mind actually changes itself into a different mind, rather than remaining itself, which is particularly obvious after a sequence of such improvements. Taken to the extreme, this idea implies that a simple act of learning new information transforms you into a different mind raising millennia old questions about the nature of personal identity." -> this is only a problem if, like the article, you are extremely confused about the definition of a unique mind, or permit equivocation thereof.

"Interestingly, a perfect ability by two minds to predict each other would imply that they are identical..." -> Why? Picture two minds whose entire environment consists of a single switch with ON and OFF. Mind A wants the switch ON. Mind B wants the switch OFF. Each can correctly predict the actions of the other - flipping the switch in the desired direction at every opportunity - but they are clearly different minds. (Especially if we think that having a favorite number of 32 instead of 33 is enough to make two minds "different".)

The taxonomy section is a grab bag of mostly unrelated classification systems, useful at least as examples of how many ways minds can vary.

"For example, asking a human if he is hungry may produce an answer different from the one which would be given by a non-biological robot." -> Once again, by the article's own claims in "Infinitude of Minds" about what makes a mind unique, if one mind experiences hunger and the other does not, then the two minds are not identical, unless we consider "ability to experience hunger" a smaller difference than "favorite number."

References

1. Sloman, A., The Structure and Space of Possible Minds. The Mind and the Machine: philosophical aspects of Artificial Intelligence. 1984: Ellis Horwood LTD.

2. Legg, S. and M. Hutter, Universal Intelligence: A Definition of Machine Intelligence. Minds and Machines, December 2007. 17(4): p. 391-444.

3. Yampolskiy, R., How to Escape From the Simulation.

4. Hanson, R., If Uploads Come First. Extropy, 1994. 6(2).

5. Bostrom, N., Quantity of experience: brain-duplication and degrees of consciousness. Minds and Machines, 2006. 16(2): p. 185-200.

6. Yampolskiy, R. and M. Gavrilova, Artimetrics: Biometrics for Artificial Entities. IEEE Robotics and Automation Magazine (RAM), 2012. 19(4): p. 48-58.

7. Yampolskiy, R.V., B. Klare, and A.K. Jain. Face recognition in the virtual world: Recognizing Avatar faces. in Machine Learning and Applications (ICMLA), 2012 11th International Conference on. 2012. IEEE.

8. Yampolskiy, R.V., Leakproofing Singularity - Artificial Intelligence Confinement Problem. Journal of Consciousness Studies (JCS), 2012. 19(1-2): p. 194–214.

9. Wikipedia, Universal Turing Machine. Retrieved April 14, 2011: Available at: http://en.wikipedia.org/wiki/Universal_Turing_machine.

10. Lloyd, S., Ultimate Physical Limits to Computation. Nature, 2000. 406: p. 1047-1054.

11. Levin, L., Universal Search Problems. Problems of Information Transmission, 1973. 9(3): p. 265--266.

12. Rice, H.G., Classes of recursively enumerable sets and their decision problems. Transactions of the American Mathematical Society, 1953. 74(2): p. 358-366.

13. Yampolskiy, R.V., Efficiency Theory: a Unifying Theory for Information, Computation and Intelligence. Journal of Discrete Mathematical Sciences & Cryptography, 2013. 16(4-5): p. 259-277.

14. Kolmogorov, A.N., Three Approaches to the Quantitative Definition of Information. Problems Inform. Transmission, 1965. 1(1): p. 1-7.

15. Sotala, K. and H. Valpola, Coalescing Minds: Brain Uploading-Related Group Mind Scenarios. International Journal of Machine Consciousness, 2012. 4(1): p. 293-312.

16. Yudkowsky, E.S., Creating Friendly AI - The Analysis and Design of Benevolent Goal Architectures. 2001: Available at: http://singinst.org/upload/CFAI.html.

17. Fredkin, E., On the soul. 1982, Draft.

18. De Simone, A., et al., Boltzmann brains and the scale-factor cutoff measure of the multiverse. Physical Review D, 2010. 82(6): p. 063520.

19. Kelly, K., A Taxonomy of Minds. 2007: Available at: http://kk.org/thetechnium/archives/2007/02/a_taxonomy_of_m.php.

20. Vanchurin, V., The world as a neural network. Entropy, 2020. 22(11): p. 1210.

21. Krioukov, D., et al., Network Cosmology. Sci. Rep., 2012. 2.

22. Bostrom, N., Are You Living In a Computer Simulation? Philosophical Quarterly, 2003. 53(211): p. 243-255.

23. Lanza, R., A new theory of the universe. American Scholar, 2007. 76(2): p. 18.

24. Yampolskiy, R.V., On the origin of synthetic life: attribution of output to a particular algorithm. Physica Scripta, 2016. 92(1): p. 013002.

25. Miller, M.S.P. Patterns for Cognitive Systems. in Complex, Intelligent and Software Intensive Systems (CISIS), 2012 Sixth International Conference on. 2012.

26. Cattell, R. and A. Parker, Challenges for Brain Emulation: Why is it so Difficult? Natural Intelligence, 2012. 1(3): p. 17-31.

27. de Garis, H., et al., A world survey of artificial brain projects, Part I: Large-scale brain simulations. Neurocomputing, 2010. 74(1–3): p. 3-29.

28. Goertzel, B., et al., A world survey of artificial brain projects, Part II: Biologically inspired cognitive architectures. Neurocomput., 2010. 74(1-3): p. 30-49.

29. Vernon, D., G. Metta, and G. Sandini, A Survey of Artificial Cognitive Systems: Implications for the Autonomous Development of Mental Capabilities in Computational Agents. IEEE Transactions on Evolutionary Computation, 2007. 11(2): p. 151-180.

30. Yampolskiy, R.V. and J. Fox, Artificial General Intelligence and the Human Mental Model, in Singularity Hypotheses. 2012, Springer Berlin Heidelberg. p. 129-145.

31. Yampolskiy, R.V., L. Ashby, and L. Hassan, Wisdom of Artificial Crowds—A Metaheuristic Algorithm for Optimization. Journal of Intelligent Learning Systems and Applications, 2012. 4(2): p. 98-107.

32. Ashby, L.H. and R.V. Yampolskiy. Genetic algorithm and Wisdom of Artificial Crowds algorithm applied to Light up. in Computer Games (CGAMES), 2011 16th International Conference on. 2011. IEEE.

33. Hughes, R. and R.V. Yampolskiy, Solving Sudoku Puzzles with Wisdom of Artificial Crowds. International Journal of Intelligent Games & Simulation, 2013. 7(1): p. 6.

34. Port, A.C. and R.V. Yampolskiy. Using a GA and Wisdom of Artificial Crowds to solve solitaire battleship puzzles. in Computer Games (CGAMES), 2012 17th International Conference on. 2012. IEEE.

35. Hall, J.S., Self-Improving AI: An Analysis. Minds and Machines, October 2007. 17(3): p. 249 - 259.

36. Yonck, R., Toward a Standard Metric of Machine Intelligence. World Future Review, 2012. 4(2): p. 61-70.

37. Herzing, D.L., Profiling nonhuman intelligence: An exercise in developing unbiased tools for describing other “types” of intelligence on earth. Acta Astronautica, 2014. 94(2): p. 676-680.

38. Yampolskiy, R.V., Construction of an NP Problem with an Exponential Lower Bound. Arxiv preprint arXiv:1111.0305, 2011.

39. Yampolskiy, R.V., Turing Test as a Defining Feature of AI-Completeness, in Artificial Intelligence, Evolutionary Computation and Metaheuristics - In the footsteps of Alan Turing. Xin-She Yang (Ed.). 2013, Springer. p. 3-17.

40. Yampolskiy, R.V., AI-Complete, AI-Hard, or AI-Easy–Classification of Problems in AI. The 23rd Midwest Artificial Intelligence and Cognitive Science Conference, Cincinnati, OH, USA, 2012.

41. Yampolskiy, R.V., AI-Complete CAPTCHAs as Zero Knowledge Proofs of Access to an Artificially Intelligent System. ISRN Artificial Intelligence, 2011. 271878.

42. Yampolskiy, R.V., Utility Function Security in Artificially Intelligent Agents. Journal of Experimental and Theoretical Artificial Intelligence (JETAI), 2014: p. 1-17.

43. Omohundro, S.M., The Nature of Self-Improving Artificial Intelligence, in Singularity Summit. 2007: San Francisco, CA.

44. Omohundro, S.M., The Basic AI Drives, in Proceedings of the First AGI Conference, Volume 171, Frontiers in Artificial Intelligence and Applications, P. Wang, B. Goertzel, and S. Franklin (eds.). February 2008, IOS Press.

45. Waser, M.R., Designing a Safe Motivational System for Intelligent Machines, in The Third Conference on Artificial General Intelligence. March 5-8, 2010: Lugano, Switzerland.

46. Fox, J. and C. Shulman, Superintelligence Does Not Imply Benevolence, in 8th European Conference on Computing and Philosophy. October 4-6, 2010 Munich, Germany.

47. Bostrom, N., The superintelligent will: Motivation and instrumental rationality in advanced artificial agents. Minds and Machines, 2012. 22(2): p. 71-85.

48. Hales, C., An empirical framework for objective testing for P-consciousness in an artificial agent. Open Artificial Intelligence Journal, 2009. 3: p. 1-15.

49. Aleksander, I. and B. Dunmall, Axioms and Tests for the Presence of Minimal Consciousness in Agents I: Preamble. Journal of Consciousness Studies, 2003. 10(4-5): p. 4-5.

50. Arrabales, R., A. Ledezma, and A. Sanchis, ConsScale: a plausible test for machine consciousness? 2008.

51. Aaronson, S., The Ghost in the Quantum Turing Machine. arXiv preprint arXiv:1306.0159, 2013.

52. Yampolskiy, R.V., Behavioral modeling: an overview. American Journal of Applied Sciences, 2008. 5(5): p. 496-503.

53. Yampolskiy, R.V., Artificial intelligence safety engineering: Why machine ethics is a wrong approach, in Philosophy and Theory of Artificial Intelligence. 2013, Springer Berlin Heidelberg. p. 389-396.

54. Yampolskiy, R.V., What to Do with the Singularity Paradox?, in Philosophy and Theory of Artificial Intelligence. 2013, Springer Berlin Heidelberg. p. 397-413.

55. Chalmers, D.J., The conscious mind: In search of a fundamental theory. 1996: Oxford University Press.

56. Gao, S., A quantum method to test the existence of consciousness. The Noetic Journal, 2002. 3(3): p. 27-31.

57. Elamrani, A. and R.V. Yampolskiy, Reviewing tests for machine consciousness. Journal of Consciousness Studies, 2019. 26(5-6): p. 35-64.

58. Yampolskiy, R.V., Artificial consciousness: An illusionary solution to the hard problem. Reti, saperi, linguaggi, 2018(2): p. 287-318.

59. Legg, S. Is there an elegant universal theory of prediction? in Algorithmic Learning Theory. 2006. Springer.

60. Yudkowsky, E., Artificial Intelligence as a Positive and Negative Factor in Global Risk, in Global Catastrophic Risks, N. Bostrom and M.M. Cirkovic, Editors. 2008, Oxford University Press: Oxford, UK. p. 308-345.

61. Yudkowsky, E., The Human Importance of the Intelligence Explosion, in Singularity Summit at Stanford. 2006.

62. Havel, I.M., On the Way to Intelligence Singularity, in Beyond Artificial Intelligence, J. Kelemen, J. Romportl, and E. Zackova, Editors. 2013, Springer Berlin Heidelberg. p. 3-26.

63. Geortzel, B., The Hidden Pattern: A Patternist Philosophy of Mind. Chapter 2 - Kinds of Minds  2006: Brown Walker Press.

64. Goertzel, B., Mindplexes: The Potential Emergence of Multiple Levels of Focused Consciousness in Communities of AI’s and Humans Dynamical Psychology, 2003. http://www.goertzel.org/dynapsyc/2003/mindplex.htm.

65. Hall, J.S., Chapter 15: Kinds of Minds, in Beyond AI: Creating the Conscience of the Machine. 2007, Prometheus Books: Amherst, NY.

66. Roberts, P., Mind Making: The Shared Laws of Natural and Artificial. 2009: CreateSpace.

67. Kelly, K., Inevitable Minds. 2009: Available at: http://kk.org/thetechnium/archives/2009/04/inevitable_mind.php.

68. Kelly, K., The Landscape of Possible Intelligences. 2008: Available at: http://kk.org/thetechnium/archives/2008/09/the_landscape_o.php.

69. Kelly, K., What Comes After Minds? 2008: Available at: http://kk.org/thetechnium/archives/2008/12/what_comes_afte.php.

70. Kelly, K., The Evolutionary Mind of God 2007: Available at: http://kk.org/thetechnium/archives/2007/02/the_evolutionar.php.

71. Putnam, H., Brains and behavior. Readings in philosophy of psychology, 1980. 1: p. 24-36.

72. Galilei, G., Dialogue concerning the two chief world systems: Ptolemaic and Copernican. 1953: University of California Pr.

73. Watson, J.D. and F.H. Crick, Molecular structure of nucleic acids. Nature, 1953. 171(4356): p. 737-738.

74. Wolfram, S., A New Kind of Science. May 14, 2002: Wolfram Media, Inc.

75. Sanderson, K., GPT-4 is here: what scientists think. Nature, 2023. 615(7954): p. 773.

76. Yampolskiy, R. On the Differences between Human and Machine Intelligence. in AISafety@ IJCAI. 2021.

77. Yampolskiy, R. On controllability of artificial intelligence. in IJCAI-21 Workshop on Artificial Intelligence Safety (AISafety2021). 2020.

78. Yampolskiy, R.V., On the Controllability of Artificial Intelligence: An Analysis of Limitations. Journal of Cyber Security and Mobility, 2022: p. 321–404-321–404.

Gardener Comments

Mark:
Overall this is an interesting perspective that relates closely to work on Machine Behavior. I feel the article could focus more on its call to action as opposed to on spelling out the specifics of the interpretation and comparative analysis of minds, i.e. the high level concept seems more important and robust than the specific interpretation of minds and their representation.

Ted Wade:
One key assumption is that a valid science of intellectology can be accomplished by intellects as limited as ours. There might be something like Vingean uncertainty that severely limits any such attempt. The paper assumes that mind is an instantiated intelligence, and refers to Legg for a working definition of intelligence. The paper should at least briefly explain Legg, and perhaps offer a couple more takes on defining its primary subject matter. That way we would be more able to separate minds from other computational systems.

Dr. Jason Jeffrey Jones (Psychology PhD):
This article is too disorganized and unfocused to publish in its current state. However, the topic is important, and some of the ideas in this manuscript are inspiring. I would advise the author to drop the long development of the simple (one might say facile) idea that "every mind is exactly one particular integer." Instead, begin with the idea - introduced late in the current manuscript - that every mind exists in a large space where every ability is a dimension. Developing that idea further would be interesting. 

Josh Randall:
The article attempts to describe a field of intellectology which is primarily comprised of previous attempts at developing a taxonomy of minds - primarily derived from researchers into AI. The author spends a large chunk of the article focused on the infinitude of minds and distinguishing between artificial minds and biologically generated minds. Much of this prose relies on unsourced or under-explained ideas about infinity, definitions of minds, and concepts surrounding panpsychism. The primary new contribution appears to be the final paragraph explaining additional information about uploads but would have benefited from much more detail as opposed to the literature review throughout the rest of the article.

Anonymous1:
The core of the article is unspecific: "Consequently, we can treat the space of all minds as the space of programs with the specific property of exhibiting intelligence if properly embodied. All programs could be represented as strings of binary numbers, implying that each mind can be represented by a unique number"

The core of being a mind is the existence of a (Cartesian) conscious experience, and that depends on physical implementation. It is possible that the same Turing machine can lead to a conscious experience or not depending on its physical implementation (it is even possible that the same physical implementation leads to conscience depending on the speed of execution, see the paradox of the Searle "Chinese room"). Minds are not mathematical objects, and they are not "enumerable".

The paper does not engage with axiomatic theories of conscience and it does not even consider that intelligence could happen without conscience experience (https://ceur-ws.org/Vol-2287/short7.pdf). A deeper engagement with literature on philosophy of mind (Koch, Tononi, etc), and "neural correlates of conscience" is necessary before the paper can be the real seed of any scientific work. Currently, I find this work as both too speculative, and lacking a clear unified message.

Dr. Payal B. Joshi:
The article presents an intriguing concept on intellectology and mind as a universe. Modern times are gripped with the concepts of machine learning, data mining and artificial intelligence. It comes across that such studies, if incorporated in understanding nuances of psychological behaviors of human beings, shall unravel many facets of the human brain (maybe!). Also, the author has presented a detailed work that depicts a complex interplay of intelligence, goals and human behavior in mind in multidimensional space. These are complex but may allow us to understand nuances of superhuman intelligence and psychotic behaviors too. Albeit, such studies are largely theoretical in context, these studies have the potential to further studies in understanding intellectual minds and human behavior.

As a proponent of artificial intelligence, I found this article particularly enjoyable and gave me food for thought for running my next experiment too. I recommend publishing the article as it is for wider readership.

Roger’s Bacon:
I recommend this paper for publication, while also acknowledging that it could be greatly improved with some revisions.

1) I'm not sure how necessary the “size, complexity and properties of mind” section is, feels like it gets lost in the weeds a little bit. That section and the "Infinitude of Minds” make simple but important points that I think can be expressed more succinctly. I appreciate the challenge of what the author is trying to do here as there are numerous philosophical issues and distinctions that could be raised, but given how speculative all of this is it could make more sense to pass through this relatively quickly by raising several open questions.

2) the paragraph starting with..“Given our algorithm for sequentially generating minds, one can see that a mind could never be completely destroyed…” could really just be a sentence, not sure how much it is adding.

3) What’s most interesting to me is the final sections (space of design/taxonomies) and I’d be curious to see this expanded, especially in light of recent advances in AI. Though I imagine some will balk at the proposal of intellectology, I think it's an interesting speculative idea and considering it further may be fruitful in some way, even if very indirectly. How can we get from the current state of the mind sciences to a mature field of intellectology? What new methods/frameworks will be needed? Might be useful to regroup some existing branches of research under this new umbrella term?

4) Issues surrounding collective intelligences (ant colonies, governments, all of humanity) might be worth addressing briefly

Some further reflections:

1) I'd be curious to see a catalog of new scientific fields which have been proposed in either science fact or science fiction. From science-fiction, psychohistory (Foundation) and cosmic sociology (Three Body Problem) come to mind, but no doubt there are countless others. From science fact, the only other one I’ve come across is Entitiology from “An ontology of psychedelic entity experiences in evolutionary psychology and neurophenomenology” (Winkleman, 2018).

I propose that to determine whether there are consistent and unique features of psychedelic entity experiences, we need a cross-cultural and interdisciplinary assessment of phenomenological reports of diverse types of experiences of entities (i.e., see Winkelman, 1992). Formal quantitative comparisons of the reported characteristics of diverse entity experiences are necessary to discover any commonalities to psychedelic entity experiences and their uniqueness with respect to other types of entity experiences. We need a new field of scientific inquiry, entitiology, i.e., the study of entities, to address the questions of the nature of psychedelic, and other types of entity experiences….Consequently, entitiology must encompass a number of existing areas of inquiry and by necessity will incorporate at least a part of the domain of the entities reported in the following areas of study: Angelology, Demonology, Spiritology, UFOology, Folklore and Mythology studies of elves, fairies, dwarfs, pixies, imps, gnomes, goblins, leprechauns, little people, and similar phenomena reported in cultures around the world, Possession, Mediumship, and Shamanism, Ghosts, apparitions, and poltergeist phenomena, Psychiatric syndromes, especially abnormal body syndromes and experiences such as the “Old Hag” and other terrorizing dreams. A systemic coding and analysis of the features of these various accounts can determine whether or not a single type or several types of psychedelic entity experiences occur. And only through comparison with profiles obtained for reports of what are conceptualized as angels, fairies, extraterrestrials, and shamanic spirits can we determine if there are unique features of psychedelic entities.

2) I’m reminded of Michael Levin’s idea of classifying selves by the size/shape of their “cognitive (computational) light-cone” from “The Computational Boundary of a “Self: Developmental Bioelectricity Drives Multicellularity and Scale-Free Cognition” (2019). The “Predictions and Research Program” section of the paper has numerous reflections that speak to the nascent field of intellectology. 

“I propose a semi-quantitative metric, based on the spatio-temporal boundaries of events that systems measure and try to control, that can be used to define and compare the cognitive boundaries for highly diverse types of agents (which could be biological, exo-biological, or artificial)...The edges of a given Agent’s goal space define a sort of “computational light cone” – the boundaries beyond which its cognitive system cannot operate. For example, a tick has a relatively small cognitive boundary, having very little memory or predictive power in the temporal direction, and sensing/acting very locally. A dog has much more temporal memory, some forward prediction ability, and a degree of spatial concern. However, it is likely impossible for a dog’s cognitive apparatus to operate with notions about what is going to happen next month or in the adjacent town. Human minds can operate over goals of vastly greater spatial and temporal scales, and one can readily imagine artificial (organic or software-based) Selves with properties that define every possible shape in this space (and perhaps change their boundaries over evolutionary and individual timescales)”

Joe R:
This article offers plenty of thought-provoking ideas, consistently fails to give good reasons for most of them, and repeatedly shoots itself in the foot while trying. I must reluctantly admit that its treatment of mind-space is fairly comprehensive, but the lack of concrete examples or even half-hearted attempts at internal consistency significantly diminishes its value. If the author manages to fix the numerous holes in their logic and factual errors, I could recommend publishing the article for its thoroughness alone, but as it stands, I would expect better internal coherence of a grand-sounding attempt to introduce the field of "intellectology" to the world.

For example: The second proof of infinite possible minds raises many questions. First, why is "time" a relevant factor at all? Could you not have a binary representation of a mind which, if instantiated by a suitable machine, would fulfill all relevant properties of a mind? The result might be static, frozen, unchanging, but it is still the case that it represents a mind, albeit a mind currently in stasis. This is because if you were to run the binary on a substrate, it would be capable of "interacting with its environment and other minds." Similar logic applies to classifying the code for Mario Bros as a "game" even if no one is currently playing it. Second, why does the insistence that how fast the mind is running matters? If you run Mario Bros. at 2x speed, nothing about the underlying code has changed; it remains an identical copy of Mario Bros. Speed is a feature of the substrate, not the mind. Finally, if in defiance of the above we require that a binary representation of a mind must include change over time, i.e. if a mind must be actively running on something to be classified as such; then the representation is causally incomplete without also including a full representation of the environment the mind is operating in, since (by the article's own definition of a mind) that environment must necessarily interact with the mind. By this logic, Mario Bros. does not count as a "game" unless it is currently being played by someone, and the binary representation of that game must include the player's input and everything causally and temporally adjacent to it - unless I'm missing something.

Put another way - a simulation of "the brain of Bob Smith, accountant, at exactly noon Eastern Daylight Time on May 12, 2020" should be counted as a mind, even if it remains "paused' forever and no other information is included in the encoding of said mind…or so I would think? The article gives precisely zero concrete examples so I can't be sure of their definition. If real or simulated time has to pass for Bob to be considered a mind…how much time? Does a simulated minute of Bob's life count? At what point would the article writers consider Bob a "different mind"? Why on earth do they think "drunk Bob" qualifies as "different" but "Bob at time T+1" doesn't?

Taken seriously, the arguments in the article would seem to imply the following contradictory points:

1) A binary encoding must be instantiated on a substrate to be considered a mind (i.e. subjective time must pass for said mind);
2) Instantaneous time-slices of a mind do not count as minds themselves;
3) Two (non?)-instantaneous time-slices of the same entity (e.g. "drunk Bob" and "sober Bob") which could have been arrived at in the same simulation not only count as minds, but as two different minds;
4) A single integer can encode a mind;
5) You can instantiate a mind from an integer by running it on something;
6) Minds cannot be fully destroyed because they continue to exist as integers;
7) A Universal Turing Machine is a mind;
8) "the same mind can be instantiated in different substrates and move freely between them".

(1) and (4) seem especially in conflict, unless we assume that the relevant integer simultaneously encodes a thinking entity, its substrate, and some unspecified yet adequate amount of time passing for the entity…which would render (5) utterly meaningless and cause serious problems for (8).

(2) and (3) are at least partially in conflict. If you run a mind for 2 minutes and split the resulting computation into two 1-minute chunks, is that two half-instances of one mind or 2 different unique minds? What if one minute is sober and the other is drunk? What if you keep splitting until you get a bunch of infinitesimally tiny time-slices?

(1) and (6) seem to be in conflict. If part of the definition of a mind is that it be instantiated, then you can destroy one by destroying its substrate.

I lack the mathematical foundation to dispute (7) but it intuitively seems like it must conflict with at least one of the preceding claims.

I invite the author to clarify whether their proposed definition of a mind does, or does not, include the substrate it's running on, and to adjust their claims appropriately.

"Consequently, since a human mind has only a finite number of possible states, there are minds which can never be fully understood by a human mind as such mind designs have a much greater number of states, making their understanding impossible as can be demonstrated by the pigeonhole principle." -> the conclusion might be true, but I don't find this run-on sentence convincing. Human understanding is not limited by the number of states our minds can occupy, it's limited by some hideously complex function thereof. Smaller minds can understand larger minds via compression, abstraction, pattern-recognition, etc. A smaller mind can even, in theory, simulate a larger one in its entirety, given enough time. There may still exist minds incomprehensible to humans, but not solely because they're bigger.

"Also, the most powerful and most knowledgeable mind has always been associated with the idea of Deity or the Universal Mind." -> Now they're just making stuff up to sound impressive. Literally one page ago the author proved there is no largest mind! What does this add? Even being charitable, "always" is false and entirely unjustified.

"Depending on the definitions used, one can argue that a recursively self-improving mind actually changes itself into a different mind, rather than remaining itself, which is particularly obvious after a sequence of such improvements. Taken to the extreme, this idea implies that a simple act of learning new information transforms you into a different mind raising millennia old questions about the nature of personal identity." -> this is only a problem if, like the article, you are extremely confused about the definition of a unique mind, or permit equivocation thereof.

"Interestingly, a perfect ability by two minds to predict each other would imply that they are identical..." -> Why? Picture two minds whose entire environment consists of a single switch with ON and OFF. Mind A wants the switch ON. Mind B wants the switch OFF. Each can correctly predict the actions of the other - flipping the switch in the desired direction at every opportunity - but they are clearly different minds. (Especially if we think that having a favorite number of 32 instead of 33 is enough to make two minds "different".)

The taxonomy section is a grab bag of mostly unrelated classification systems, useful at least as examples of how many ways minds can vary.

"For example, asking a human if he is hungry may produce an answer different from the one which would be given by a non-biological robot." -> Once again, by the article's own claims in "Infinitude of Minds" about what makes a mind unique, if one mind experiences hunger and the other does not, then the two minds are not identical, unless we consider "ability to experience hunger" a smaller difference than "favorite number."

Using an anthropology database that details many groups, we summarize how our forager ancestors likely lived on a variety of metrics. Though we have long since ceased to live as hunter-gatherers, its psychological shadow likely still shapes us, and so we would try to understand that lifestyle as fully as possible.

Summary and Introduction

We are economists with a long-standing interest in evolutionary psychology. Economists have particular arguments about how society should function but most of the public finds them weird or even distasteful. For example, economists favor market transactions to solve resource allocation problems while the public tends to find sharing to be preferable. (That people who need organ transplants die waiting for a donor because it is illegal for anyone to sell their own organs is a particularly tragic example of how much the public looks down on market-based solutions.) Why are some people so eager to trust strangers to simply do the “right thing?” We see evolutionary psychology as a powerful way to explain this and other disagreements. In small societies, trust and generosity would be emphasized in cultures where food sharing is common because a refusal to reciprocate sharing would likely lead one to be ostracized from the food sharing pool and make survival more difficult. People have a trust instinct.

We desire to better understand the hunter-gatherer world of our ancestors and have recently come to appreciate the rich collections of relevant data cultural anthropologists have spent decades collecting on the social environments of a wide range of human societies. While we found some systematic collections of these observations, we could not find a systematic summary of the social environment of the subsample of societiesAll right that most resemble the social environment where most human psychology seems to have evolved: small bands of nomadic foragers. The cross-culture studies of forager groups that exist tend to have a very in-depth focus on a particular aspect of forager society and lack a “bird's eye view” of how these societies operate on multiple levels (Jaeggi & Gurven 2013; Curry, Mullins, & Whitehouse 2019; Glowacki & Lew-Levy 2022).

This short paper provides a view that emphasizes breadth over depth, with a particular emphasis on early work on these societies, to show how early foragers might’ve lived in a wide variety of dimensions. Using an existing dataset aggregated from diverse ethnographies, we collect statistics on the social environment of the studied cultures which most closely resemble our hunter-gatherer ancestors.

Compared with relatively modern societies, nomadic foragers had similar levels of food and disease, and less murder and suicide. They did not fight over land or resources, and they enforced justice directly and personally. They avoided class divisions like rich vs. poor, shared food more, and their leaders had no formal powers.

Polygamy, premarital sex, and extramarital sex were all widespread, divorce was easy, and men and women were generally considered equal, with some important exceptions. Kids were taught to be more generous, trusting, and honest, and were never punished physically.

Criteria

We draw from an aggregated dataset, the Standard Cross-Cultural Codes (SCCC), composed of over 2,000 variables covering 186 societies from around the globe. Each variable originates from one of several dozen studies that were compiled and edited by Douglas R. White, Michael Burton, William Divale, Patrick Gray, Andrey Korotayev, and Daria Khalturina.

Ideally we would prefer a database on dozens of societies that perfectly preserved the social environment of our distant hunter-gatherers ancestors. Alas, this is not possible. Even relatively isolated societies are in substantial contact with more modern societies, and we cannot exclude the possibility that their ancestors were once refugees from such societies. Furthermore, the marginal places where such isolated folks now reside are clearly not representative of the places our distant ancestors lived; such folks can now only live in places for which modern societies have little use.

As a substitute, we have constructed a subsample of the 186 SCCC societies, a subsample of the societies with the fewest “deviations” from our best guess about the lives of our distant nomadic forager ancestors. Most of our deviation criteria deal with a society’s level of technological advancement, though other variables, including diet and mobility, were also considered.

To construct our sample of civilizations, we first remove all societies that do not get 80% or more of their food from hunting or gathering (v858: coding three or more). We drew “subsistence type” from D. White (1984), after Karen and Jeffrey Paige (l981), excluding societies which engaged in farming, herding, fishing, and so forth as these are methods of food collection that often or always require living in a single location. This left 19 societies of varying closeness to our ideal hunter-gatherers of our distant past.

While our hunter-gatherer ancestors are long gone, we can broadly infer what their lifestyle was like based on common sense. For example, using sails to power watercraft would require textile technology that ancient humans would not have had; land inheritance and fixed settlements would be meaningless to any group that is constantly traveling; the presence of credit markets and currency demonstrate a level of institutional sophistication beyond that of our ancestors. We necessarily had to make assumptions and so we favored strictness whenever feasible; for example, while pack animals could in theory be done by hunter-gatherers, it is usually associated with a more stationary setting.

We used 18 criteria to determine which societies to remove from the dataset. While these are not the only variables applicable to determining which societies most closely resembled hunter-gatherers, these variables represent an appropriate cross-section of the relevant possible information. Each criteria entry below describes the title of the variable, the number of the variable (as defined by the SCCC), the variable values that indicated unfitness for the study, and a brief description of those eliminated values:

• Import Food Acquisition (v2: three or more), getting food from local market or better

• Land Transport (v13: two or more), using pack animals or better

• Water Transport (v15: five or more), using a sail powered craft or better

• Money (v17: four or more), foreign or domestic money

• Credit Source (v18: three or more), external money lending specialists

• Fixity of Settlement (v61: two or more), fixed at any point

• Large or impressive structures (v66: two or more), any such structures

• Political autonomy (v81: three or less), tribute paid to a larger society (such as a modern state) or more dependent (i.e. integrated)

• Level of sovereignty (v83: two or more), any “state” at all

• Technological specialization (v153: two or more), pottery or more advanced

• Animal husbandry (v244: two or more), any sort of animal husbandry

• Subsistence economy (v246: four or more), pastoral or more advanced

• Inheritance of land property (v278: two or more), any inheritance of this property

• Taxes paid to community (v784: two or less), any taxes in any form

• Trade and markets (v1007: four or more), a marketplace or anything more established

• Labor (v1009, five or more), migrant wage labor or more advanced

• Population density (v1130, four), 5 people per square mile or more

• Sources of wealth (v1722, two-five, or eight), wealth from land or cattle

In table 1, groups that met a criterion for inclusions were made blank and groups for which there was no data for a criterion are indicated with a question mark but treated as blank. This is to retain as large a sample size as reasonably possible. To make the uncertainty explicit, an additional value treats any uncertainty as a deviation.

Modern Comparisons

Some variables deserve some context since they are based on the opinion of the ethnographer. To better understand the differences between the hunter-gatherer world and the world of modern society, we consider a handful of more recent civilizations. (There are no societies very similar to our own included in these studies, and the most recent year of focus is 1965.)

Each of these modern societies have either motorized water or land transport; a medium of exchange; permanent settlement; some sort of large or impressive structures; smiths, weavers, and potters; a population density of at least 100 people per square mile1 and a total population of at least one million people. They are (with the year of focus) Uttar Pradesh (1945), Balinese (1958), Japanese (1950), Siamese (1955), Chinese (1936), Javanese (1954), Turks (1950), and Russians (1955).

Weak and Strong Foragers

To find good exemplars of forager societies, all societies with more than one “modern trait” were excluded. This brings the total sample to seven (parenthetical notes indication the year of focus): Aranda (1896), Botocudo (1884), Hadza (1930), Kung Bushmen (1950), Mbuti (1950), Semang (1925), and Tiwi (1929). These groups form the “weak” forager dataset.

It is possible that the small amount of “give” allowed in the constraints taints the sample in some significant but unforeseen way. Since a sample with an even higher standard of similarity with hunter-gatherers is still notably large at five (with zero total deviations instead of one or less), we construct general information about these groups in pursuit of forming a more accurate (if more limited) picture. This “strict” forager dataset—groups with no modern indicators—total five: Aranda, Botocudo, Hadza, Mbuti, and Tiwi. When discussing results we describe this strict sample, unless data is insufficient to provide an accurate picture. In conclusion for the weak and strict conflict, we focus on the strict sample but mention the weak sample in case the difference is due to a smaller sample size.

Parenthetical citations indicate the variable number (which begins with a “v”) followed by the number of observations and if the data set is from the weak (w), strict (s), or modern (m) sample. For example, “v1719 N=5w” indicates that the information came from variable 1719, with five observations from the weak dataset. We mention modern samples only when we feel as though an explicit point of comparison is important to understanding the data.

Strict Foragers at a Glance

Before diving into the data, it is a good idea to briefly review the groups with zero deviations from the hunter-gatherers we seek to understand. The five groups are spread throughout the world: one located in Brazil, two in Africa, and two in Australia. With the exception of the Aranda in central Australia, the groups live in a tropical climate with an average of 1,334 mm of rain a year (1,122 mm if you include the Aranda) and average annual temperature of all societies is 23.2 C (73.8 F) (v189 N=5s, v188 N=5s). The size of the local community can be as large as 100 people but is generally between 10 and 50 individuals (v1756 N=3s). Population density is about 1 person per 1-5 square miles with the Botocudo having less than one person per five square miles (v64 N=5s).

The Aranda (sometimes referred to as the Arrernte or the Arunta) lived in the highland deserts of central Australia, ranging from the Macumba River to the Davenport Range. The local flora was mostly small trees and shrubs; the fauna ranged from kangaroos to a large variety of birds (including emus, ducks, and brush turkeys). Men hunted the fauna with boomerangs while women gathered local vegetation and small animals including seeds, tubers, rats, lizards, birds, and grubs. The Aranda were highly mobile and wandered the desert within an ancestral territory, each belonging to one of five tribes. The land was considered fixed—no Aranda tribe had ever attempted to take the territory of another Aranda tribe. There were no chiefs, but there were elders who oversaw ceremonies (Spencer and Gillen 1899).

The Botocudo lived in the forests of eastern Brazil, hunting and foraging for food. They lived in hovels about four feet high and constructed of branches stuck to the ground. They were completely nomadic and roamed the forest in bands of ten to twenty families. They ate roots, berries, frogs, lizards, honey, snakes, and larger game which they hunted with bows and arrows. They also made canoes by burning out the inside of a tree. There’s much evidence to suggest that they were cannibals and used the heads of their devoured victims for targeting practice (Keane 1884). Blood feuds, not only between tribes but within them, were common and were primarily motivated by revenge for previous acts of violence. No one would remember the original cause for the feud. They choose their leader based on his supernatural power (Nimuendajú 1946).

The Hadza live in the savanna of northern Tanzania and number about 300-400. Men hunt local game and collect honey while women collect tubers, berries, and fruit. They organize themselves into bands of about twenty-five individuals, though mobility between bands is quite high. This high mobility is partially due to the fact that men and women are free to choose their own mates and so men will travel from band to band until a woman reciprocates his interest in her and he stays. Successful hunters have an easier time attracting a mate and in some cases, have more than one wife. However, this success at hunting does not translate into status within the band as a whole. Polygamy is rare among the Hadza but only 20% stay married to the same person their whole life. The Hadza rarely marry outside of their ethnic group and they tend to know everyone in their mating pool before they select a mate (Marlowe 2004).

The Mbuti live in the Ituri Forest of the Democratic Republic of Congo and number 20,000 to 40,000. Ethnologists divide them into two main groups, based on their form of hunting: with bows and arrows and with nets. They regularly trade with the bakbala, or local agriculturalists who provide them with tobacco, grown foods, and limited manufactured items in exchange for forest products. This relationship has been maintained for “many years” according to the 1978 source and it is unknown if it overlaps with our year of focus of 1950. The trading, however, is irregular and the Mbuti are able to avoid the bakbala’s attempts at control by constantly changing their allegiances. From Hart 1978, page 331:

They cleverly alternate trade with begging; gifts with thievery; wage labor with demands made on religious grounds. The bakbala cannot know where the Mbuti's allegiance stands, or keep track of what they owe to whom. In this confused state of flux, the Mbuti preserve their independence, but continue to derive material advantages from the village.

While it is unclear how dependent the Mbuti are on the bakbala, there is little controversy that the Mbuti would continue to thrive if their trading would cease (Hart 1978). Exactly how much these trading relations taint our sample is unknown, but it appears that the Mbuti use the trade goods to simply make their life a little easier and have not significantly shifted their lifestyle. This is captured in that the Mbuti unquestionably fulfill all our requirements (unquestionable in the sense that they had no missing data points for the criteria variables).

The Tiwi occupy the Melville and Bathurst Islands of Northern Australia. Numbering just over 1,000 individuals during the year of focus, they are separated into nine factions. During the year of focus, missionaries from a local Catholic mission encouraged the Tiwi to adopt monogamy. Failure to comply meant an end to European goods, including foods, clothes, and tobacco. These pressures, however, were countered by Japanese pearl-hunters. The sailors, desiring prostitutes, would trade European goods to bed a local woman. While multiple wives unset the flow of goods from missionaries, it emboldened them from sailors (assuming he was willing to rent them out, a practice forbidden by Tiwi custom; Hart 1954). This was the main political issue of the day among the Tiwi and we mention it here to underline that the trading and political tension does not fundamentally taint our sample. Like the Mbuti, the Tiwi unambiguously fulfill all out requirements for being in the strict forager sample.

Food, Health, and Property

Food supplies are generally constant—though the Aranda encounter periodic or chronic hunger— (v678 N=3s) and “seldom” (occurrence uncommon) variations in food supply (v1719 N=3s). Occurrence of short-term starvations range from low to high, though it leans to the low end (v1262 N=4s). Occurrence of seasonal starvation is either very low or moderate (v1263 N=4s). Occurrence of famine ranges from very low to very high, favoring the lower end (v1265 N=3s). When famine occurs, it is either very low or very high in intensity (v1267 N=3w), favoring the high end. It’s worth noting that the society with the most famine problems—the Aranda—also has by far the lowest annual precipitation: 275 mm or 1.7 standard deviations below the mean. This is also the society which experiences chronic or periodic hunger.

Land shortages (v1720 N=4w) do not occur save in one case which was due to invasion. There is no class stratification (v270 N=5s) nor slavery (v274 N=5s). Most of the time, private property is present (v704 N=3s). Most societies have no rich though one has a few (v1721 N=3s), and they derive their wealth from means of production other than cattle or land (v1722 N=3s). No society has any poor or dispossessed people (v1723 N=3s, v1724 N=3s). All land has communal rights only (v1726 N=3s). Societies either have no marketplace or a market only for bulk goods (v1007 N=2s).

Sharing of food is always common, occurring within the local community or within the ethnic group (v1718 N=3s), though the kin groups rarely exists outside of the local community (v1755 N=2s; v1755 N=4w).

Average pathogen stress is the combined intensity of seven different diseases (leishmaniasis, trypanosomes, malaria, schistosomes, filariae, spirochetes, and leprosy), ranging from 7 (none of the diseases are present) to 21 (all of them not only present, but serious). Foragers have a median stress level of 16 compared to 14 for modern societies (v1260 N=5s, N=8m).

Crime and Violence

By individuals, homicide rates (v1665 N=4w), trespass rates (v1668 N=2w), and suicide rates (v1669 N=2w) are low while assault rates (v1666 N=4w) and theft rates (v1667 N=3w), range from very high to very low (though both favor the lower rates). By groups, homicide (v1675 N=3w) rates remain the same as for individuals. Assault (v1676 N=4s), theft (v1677 N=4s), and trespass (v1678 N=4s) rates by groups are quite variable, with very high rates showing up with notable frequency. Trespass by group increases, homicide by group remains the same, theft by group increases, and assault by group remains the same, compared to crimes by individuals.

When violence occurs, resource acquisition is never a motive for it (v1727 N=3w) while in modern societies, it’s a motive for violent conflict (v1727 N=2m). Revenge can be forbidden, prescribed, or neither forbidden nor prescribed, but when it is prescribed, compensation is never seen as an equal substitute (v1774 N=3s). In all societies, the person wronged is the person who punishes the guilty: it is never a third party (v700 N=3s). People will usually change communities if there’s a substantial dispute (v785 N=2w). 

The data sources disagree about how common warfare is compared to modern societies. While foragers tend to have less conflict between communities of the same ethnic group (internal warfare) compared to modern societies (the one exception to this has only one observation for the modern sample), the frequency of external warfare is ambiguous, though the slightly larger sample size of 891-893 suggests that more weight should be placed on it, shifting in favor of slightly more external warfare (conflict with other societies). The casualty rate in conflicts is always low compared to modern societies (v901 N=3w N=4m), consistently suffering less than 30% casualties. Though they might fight more often compared to modern societies, not as many suffer from death or injury.

Courage in boys is either strongly emphasized or not emphasized (never moderately emphasized) (v1765 N=2s). For modern societies it is moderately emphasized (v1765 N=1m). If a society has warriors, they enjoy either no prestige or a high level of prestige, favoring the latter which is the same story in modern societies (v1773 N=3s N=3m). Ritual warfare is absent in all of the observed groups as it is in modern societies (v573 N=3s N=1m). Societies are either judged as being unpacified or pacified within the last 25 years of the study, but never partially pacified while modern societies are all unpacified (v1654 N=3s N=7m). In external warfare, the defeated are sometimes driven from their territory though the victors don’t use their territory. (v1656 N=2s). Modern societies usually use the conquered territory (v1654 N=3m).

Intra-ethnic violence ranges from being permanent to being rare while it is slightly less common in modern societies (v1776 N=3s N=4m). Violence towards individuals in the same community or ethnic group is either accepted or rejected but never encouraged, similar to modern societies (v1768 N=2s N=3m, v1769 N=2s). Most societies have no intra-ethnic violence and where it occurs has a highly ritualized regimentation; modern societies have either no such violence or no regimentation for such violence (v1775 N=4w N=2m). The intensity of the violence is always low, if it occurs at all (v1777 N=3s) and its frequency is rare or occasional, never permanent nor often (v1778 N=2s). Modern societies share a similar level of low frequency and low intensity violence in this area (v1777 N=3, v1778 N=3).

Loyalty to the ethnic group is always moderate while modern societies favor low rates of loyalty (v1771 N=3s N=4m). Hostility to other ethnic groups is usually negligible, though in one society it is extremely high; it’s moderate in the modern era (v1772 N=3s N=1m). For hunter gatherers, violence to those outside the ethnic group is rejected (no data available for modern societies) (v1770 N=1s).

Politics

In all forager societies, there is no executive (v85 N=5s), judiciary (v89 N=5s), police (v90 N=5s), or administrative hierarchy (v91 N=5s). No leader gains power through wealth distribution (v574 N=2s).

There is no modern political organization—family heads acknowledge no higher authority (v699 N=3s). Oddly, the ethnographer describes full-time bureaucrats who are unrelated to the government head as “always present” (v701 N=3s). People see their leader’s power as somewhat or limited (v759 N=2s) and their leaders as either benevolent or neither benevolent nor malevolent (v760 N=2s). Leaders carefully cultivate support before acting (v761 N=2s) and none have a formal leadership position; power disappears when support diminishes (v762 N=2s).

Family

Families are always polygamous though slightly more than half of the societies have mild amounts of polygamy (v67 N=5s; v79 N=5s). Polygamy is almost always socially preferred; in one case polygamy is rare, in two cases polygamy is socially preferred for men with leadership attributes, and in two cases it is socially preferred for all men (v860 N=5s). Co-wives either share a living space with each other (three societies) or one lives with the husband while the rest lives in different communities (two societies); there is not a “middle ground” of the other wives living away from the husband but in homes in the same community (v863 N=5s). The husband never has a room apart from a wife or wives (v865 N=5s). Most societies have no stratified polygamy, meaning the rates of polygamy changed based on the social standing of the husband, though two have higher rates of polygamy in a hereditary higher social class (v866 N=5s). This seems to conflict with our claim of no class stratification (v270 N=5s), but it is not clear from the ethnography if the social class is an explicit stratification or if it is assumed from the fact that informal leaders tend to have many wives and tend to have sons who grow up to have many wives. Both societies with “stratified polygamy” witness leaders having more wives than “commoners” (along with one other society that didn’t have stratified polygamy) which makes the informal stratification the most likely explanation given the evidence that informal rule appears to be the norm.

Two societies have multiple wives for skilled hunters while the others see no relation between hunting skill and number of wives (v867 N=5s). Most households are made up of a single family though in one group the households are made up of a married pair (v67 N=5s). On average, 35% of men have more than one wife, though the standard deviation is 29.7 (v871 N=5s). On average, 49.4% of women are in a polygamous marriage, and again the standard deviation is high at 36.38 (v872 N=5s). In modern societies, 3% of men have more than one wife and 7% of women are in a polygamous marriage (v871, v872 N=7m). The standard deviation is also high (6.5 and 12.7, respectively), mostly due to the Balinese who have 18% and 35%.

Females are expected to have premarital sex (v165 N=2s). Both males (v165 N=2s) and females (v166 N=3s) have premarital sex save in one group where female premarital sex is uncommon. Extramarital sex generally employs a double standard where it is acceptable for men but not for women save in one group (the Hadza) where it is acceptable for both (v169 N=4). Extramarital sex among men is either universally or moderately practiced (v170 N=2s). Among women, it is universally practiced (v171 N=1s). Wifesharing does not occur except in one society where it is only used for sexual gratification (as opposed to economic benefit) (v172 N=3s). Rape is acceptable or ignored—the code does not distinguish which occurs (v173 N=1s). Again, this is for the Hadza, the one society where extramarital sex is allowed for both sexes. Unsurprisingly, rape occurs frequently among the Hadza, while in another society it is absent (v174 N=2s). No data on this subject is available for modern societies.

Post-partum sex taboo continues to range from just under two years to a month or less, with data slightly favoring the latter (v34 N=3s). Non-maternal relations for infants generally includes the mother as the primary caregiver (but is never exclusive) and in one case, she plays a small (but significant) role (v51 N=5s). In early childhood, her role is almost always the primary role and is never small (v52 N=5s). Infants principally spend time with other adult females, though in one case it is equally shared with both sexes (v56 N=4s). In childhood children spend time with other children, usually from both sexes though in one group the children spend time with the same sex (v56 N=3s). Neither males nor females need any grounds for divorce (v745 N=4w; v746 N=3w). Wife-beating is always present (v754 N=2s).

Children Rearing and Values

Trust is strongly encouraged in children, more so than in modern societies (v1761 N=1s N=1m, v335 N=2s N=8m), and sharing is more encouraged than in modern societies, though the Aranda rarely encourages sharing (v1762 N=3s N=2m, v334 N=2s N=8m). Data on honesty is mixed, but the larger dataset suggests that hunter-gatherers encourage more honesty in their children compared to modern societies (v1763 N=1s N=2m, v336 N=2s N=8m). Caretakers have the highest measure of affection for their children while modern societies possess a slightly lower value (v492 N=3w N=7m). Children are never punished physically while in modern societies they are physically punished in half the observations (v1766 N=3s N=4m).

Infants sleep with either the mother and the father together or with the entire family and never with just one of the family members (v1710 N=3w). Adolescents sleep in a different dwelling entirely (v1711 N=2s).

All societies have an equal preference for boys and girls (v616 N=3s). There is no evidence of infanticide which “favors” one sex or the other (v617 N=3w). Most societies have no belief that women are inferior to men (v626 N=3s) and women always have a moderate degree of control over property (v628 N=3s) and usually control products of their own making (v660 N=3s). Similar to modern societies, hunter-gatherers put a medium to high value on a woman’s life (v630 N=3s N=4m) and a high to higher-medium value on her labor (v631 N=3s N=4m).2 Women have a high to a medium-high level of authority on domestic matters in both eras (v632 N=3s N=4m). The sex ratio is usually equal though one hunter-gatherer society has more males than females (v714 N=3s N=4m).

Attitudes concerning talking about sex are generally open: adults will talk about it openly with children or restrict such talk to a certain group of people (v159 N=2w). In modern societies, this is much more restrained with more than half (three out of five) of the sample never talking about sex, ever (v159 N=5m).

Just 20% of hunter-gatherer societies believe in the evil eye, compared to 37.5% in modern societies (v1189 N=5s N=8m).

Other

Gossip ranges from being moderately to very important (v1805 N=3s). When present, it averages 3.60 for hunter-gatherers (from one to five, five being very important). For modern societies, it averages 3.86 (v1805 N=7m).

It is common for an adult to travel between communities during his/her lifetime while it is occasional in modern societies (v786 N=2s N=1m).

Change across all variables (agricultural, religious, family, education, behavior, health, technological, trade, transportation) averaged 18.00, compared to 14.38 for the modern sample (v1849 N=3s N=8m). This is an important reminder that despite our best efforts to identify exemplars of authentic hunter-gatherer societies, our data are far from ideal.

Conclusion

Using data compiled in the Standard Cross-Cultural Codes, we’ve compiled some wide-ranging best guesses about the lives of our nomadic forager distant ancestors.

Such foragers have neither formal class stratification nor slavery. While private property is usually present, most forager societies have no rich, and none have any poor or dispossessed. Food sharing is always common.

Compared to the most "modern" societies in the larger sample (which are different from us today), disease stress is similar, suicide and murder are rare, conflict casualty rates are lower, and fewer believe in an evil eye. Violence is never over resources, and when enemies are driven from a territory no one uses that territory.

A person wronged always directly punishes the guilty; they never use a third party. If there is a substantial dispute, one side will likely leave the community. Leaders carefully cultivate support before acting, and none have a formal leadership position.

Polygamy is always allowed and usually socially preferred. Co-wives either live together or one lives with a husband while the rest live in entirely different bands. On average, about 35% of men have more than one wife, and 50% of women are in a polygamous marriage (vs. 3% and 7% in modern societies).

People are expected to have premarital sex, which is usually common. Extramarital sex is also usually common, though it is usually not acceptable for women. Adults talk about sex openly. While wife-beating exists, divorce is easy. Boys and girls are equally preferred, and women are considered equals of men.

Mothers are usually the main, but not only caregiver of kids. Relative to modern societies, kids are taught more to be generous, trusting, and honest. Parents place greater emphasis on loving their children, and children are never punished physically. Adolescents sleep away from their parents.

Broadly understanding the forager mindset suggests questions about the interaction between their world and ours. The forager life was complex; how did that influence brain development compared to more modern history? How has the forager life left lingering desires in our modern brains? For example, why do wealthier societies tend to act more like foragers (they are more accepting of divorce, more open about sex, and less comfortable with inequality)? How does the strength of these tendencies differ across socioeconomic levels within wealthier societies? The very different culture of foragers underscores how seismic of a shift the transition to farming was for our ancestors; how do these lingering psychological preferences of foragers versus farmers influence modern political disputes? Even though society has culturally left behind the hunter-gatherer lifestyle, we can’t shake its psychological shadow and we would be wise to understand that lifestyle as fully as possible.

Gardener Comments

Note: the authors have responded to comments in the subsequent section. 

Anonymous1:
I would be interested to know a bit more about sleep habits in nomadic populations. Do you have any further facts on that, such as how many hours they slept, do they practice naps and whether they sleep less when moving from place to place?

Pierre R. Mercuriali:
To better understand the evolution of psychology, the authors study similar social environments. Human psychology seems to have evolved from hunter-gatherers societies. The anthropological features of ancient hunter-gatherer societies are not all known, e.g., first-hand observational data.  Amongst all features some are known and some are unknown. How, then, to infer unknown features? If I have understood correctly, the authors set out to determine unknown ancient features by using shared known features of contemporary societies. They compute, for each known contemporary society, the deviation from an unknown "ideal" hunter-gatherer society, thereby giving an order on all societies.They make the assumption that societies that are closest share similar features, and thus infer ancient, unknown features, from the (observed) societies that are the closest according to this deviation. 

I think this is a swell article, with the data collection clearly explained. The question and methods are interesting and could be extended to answer other questions. It also functions as a reference, because the variables and data points used are clear and explicitly mentioned. 

Questions/remarks on content: 

C1. I would have liked a discussion on the assumption I explained above (to know if I have understood the process right)! In the same vein, I would have liked a conclusion on the evolution of psychology. What can be inferred from the anthropological reconstitution of these hunter-gatherer societies? 

C2. I could imagine generalizing the idea of "deviation" into a "distance" (in the mathematical sense) between societies. That way one could answer fun questions such as: "what societies are the closest?" Also, could this deviation distance process be performed with any other kind of society (not just hunter-gatherer societies)?

C3. Could this distance be generalized to include explicit weights? For example, societies with more than one modern trait were excluded: one could argue that the "weight", when computing the deviation between societies, would be +infinity, thus giving an infinite deviation. 

C4. Are all the missing anthropological variables that were inferred for the Ideal Sample described in the paper? If not, what motivated the choice of the ones that were described? Where to find the others? 

C5. I really liked the contextual information about the authors' first field of interest (economy). I think the connections between economy (frameworks? tools? positions? points of view?) and evolutionary psychology could be further explored. 

Because Seeds of Science has freer guidelines regarding submission format and content than the journals I know, I feel this could be an opportunity to add interesting meta-contextual information. 

Questions/remarks on form: 

F1. The question mark in the Deviation from Ideal Sample table could be replaced with a range: [i to i+n] with n being the number of question marks in the line and i the number of X in the line. This makes the uncertainness explicit.

F2. On the same table, the "society" labels on the first line are difficult to read. I suggest writing them in a slanted manner (45°). 

F3. Pages could be numbered (to reference them more easily). Table 0 (deviation) could be numbered. The font used for the titles of the various tables could be harmonized (Table 1 uses a sans-serif font, Table 2, 3... use a serif font). 

F4. Page 9, section "Politics", paragraph 2, sentence 2: I can't quite parse the sentence; is there a missing word? If not, disregard my comment.

F5. Page 5, paragraph 5, sentence 2: I think there is a word missing. "Ethnologists divide them into two main groups, based _on_ their form of hunting"

Joe:
Oddly, the summary mentions only neutral or positive comments about the lifestyle of the foragers, while the body of the article has some negatives. It seems odd to say that slavery does not occur and that the sexes are equal while also suggesting that a Tiwi husband might "rent out" his wife. Even if this was in defiance of custom, it being an option at all is suggestive of considerable inequality and a slavery-like system of marriage. 

Informal class stratification is still class stratification. I would argue that most stratification is some level of informal. I'm not even sure what would count as "formal" stratification. Caste systems, I guess? 

While I encountered some notes of discrepancy, I may be misinterpreting the claims here and I think that overall the conclusions seem clear and accurate given the data. It seems to be a worthwhile addition to the community. 

Anonymous2:
It is a good work. However, to me it fits in a conventional journal of sociology or anthropology rather than to SoS. I miss further elaboration in the discussion pointing to what nowadays society could learn from those ancestors or more controversial or novel ideas. 

Phil Filippak:
In my opinion, one of the most important things to highlight is that the distinction is primarily between societies of wildly different sizes, not only of different lifestyles. While it may be true that forager societies exhibit more of the beneficial social traits, that may be coming from the fact that there are usually <100 people in that society.

Ted D. Wade (PhD in psychobiology):
There is a lack of comparison to similar research here. As important as the question is, and given how long the cross-cultural databases (their SCCC is not the only such database) have existed, I would expect there to be quite a rich literature on the subject. If the literature exists, then I think the author(s) of this paper would benefit from at least mentioning it. Perhaps they intend to discuss/contextualize their findings in a separate publication?

Ben Lockwood (Ph.D., Geography; Postdoctoral Researcher):
This research appears to be a summary of a dataset that was published over half a century ago (Murdock and White, 1969). The authors claim that they could find no systematic summaries of the early foraging societies that this dataset characterizes, however many such summaries exist. There is an abundant collection of literature across multiple disciplines that the authors have not examined, and this literature contains a long history of meta-analyses of early foraging societies (Segall et al., 1990; Jankowiak and Fischer, 1992; Gangestad and Simpson, 2000; Wood and Eagly, 2002; Mesoudi et al., 2006; Van Ijzendoorn and Sagi-Schwartz, 2008; Balliet et al., 2011; Minkov, 2012; Kelly, 2013; Alesina et al., 2013; Jaeggi and Gurven, 2013; Curry et al., 2019; Glowacki and Lew-Levy, 2022). The authors have not provided any reasoning for their lack of acknowledgement of the many previous studies which have produced summaries of greater detail and rigor than is present here. 

The manuscript, unfortunately, does not provide an adequate justification for how this analysis could advance scientific knowledge. The authors make a variety of wide-sweeping, generalized, and potentially dubious claims about early foraging societies based on a half-century old dataset, of which many studies have already explored. They do not address what gaps in the literature or innovative approaches this study provides. Thus, it is unclear what novel, or even speculative, ideas this paper contributes to the wider scientific community.

As such, I cannot recommend it for publication in its current form. I would suggest that the authors return to the literature around these topics to determine what previous investigations may lack in order to produce an analysis of value in an area of study that has already been so thoroughly explored. 

References:

• Alesina, A., Giuliano, P., & Nunn, N. (2013). On the origins of gender roles: Women and the plough. The quarterly journal of economics, 128(2), 469-530.

• Balliet, D., Li, N. P., Macfarlan, S. J., & Van Vugt, M. (2011). Sex differences in cooperation: a meta-analytic review of social dilemmas. Psychological bulletin, 137(6), 881.

• Curry, O. S., Mullins, D. A., & Whitehouse, H. (2019). Is it good to cooperate? Testing the theory of morality-as-cooperation in 60 societies. Current Anthropology, 60(1), 47-69.

• Gangestad, S. W., & Simpson, J. A. (2000). The evolution of human mating: Trade-offs and strategic pluralism. Behavioral and brain sciences, 23(4), 573-587.

• Glowacki, L., & Lew-Levy, S. (2022). How small-scale societies achieve large-scale cooperation. Current Opinion in Psychology, 44, 44-48.

• Jaeggi, A. V., & Gurven, M. (2013). Reciprocity explains food sharing in humans and other primates independent of kin selection and tolerated scrounging: a phylogenetic meta-analysis. Proceedings of the Royal Society B: Biological Sciences, 280(1768), 20131615.

• Jankowiak, W. R., & Fischer, E. F. (1992). A cross-cultural perspective on romantic love. Ethnology, 31(2), 149-155.

• Kelly, R. L. (2013). The lifeways of hunter-gatherers: the foraging spectrum. Cambridge University Press.

• Mesoudi, A., Whiten, A., & Laland, K. N. (2006). Towards a unified science of cultural evolution. Behavioral and brain sciences, 29(4), 329-347.

• Minkov, M. (2012). Cross-cultural analysis: The science and art of comparing the world's modern societies and their cultures. SAGE publications.

• Murdock, G. P., & White, D. R. (1969). Standard cross-cultural sample. Ethnology, 8(4), 329-369.

• Segall, M. H., Dasen, P. R., Berry, J. W., & Poortinga, Y. H. (1990). Human behavior in global perspective: An introduction to cross-cultural psychology. Pergamon Press.

• Van Ijzendoorn, M. H., & Sagi-Schwartz, A. (2008). Cross-cultural patterns of attachment: Universal and contextual dimensions.

• Wood, W., & Eagly, A. H. (2002). A cross-cultural analysis of the behavior of women and men: implications for the origins of sex differences. Psychological bulletin, 128(5), 699.

Patrick (Master's in anthropology/sociology):
It's an interesting attempt to systematise and compare a range of extant hunter-gatherer societies using various indicators. More stamp-collecting than physics, but useful insights that may help break down noble savage stereotypes, or may reinforce them. 

Partha Ghosh:
It's an interesting summary. However, not sure there are many new ideas or promising leads to advance the topic here. I think it could be interesting to look at the larger data, before removing the 18 outlier criteria, for further qualitative insights. 

Dr. Payal B. Joshi:
The article is fairly good and some sections that I particularly enjoyed reading were the description on strict foragers and data on warfare. Though there are too many claims made by the authors, it is backed by relatively less cited references. Also, there have been many new papers that are not cited by the authors which reflects that the paper was written a while ago. Thus, with some refreshing literature needs to be cited prior to publishing and avoiding redundancy. 

Authors’ Response to Gardener Comments

We appreciate the time and effort that you and the gardeners have dedicated to providing thoughtful feedback, and we are grateful to the gardeners for their valuable comments. 

• To address clarity concerns with methodology, we added additional information concerning our reasoning for focusing on the groups we did.

• We sympathize with the comments concerning recent work on this issue, and included a brief note justifying the value of our "bird's eye" view of forager life (in contrast to the more focused and in depth approach that's favored in existing cross-culture analyses). 

• It was also suggested we describe possible extensions from the paper as well as make our motivation as economists clearer; we agreed and incorporate additional information to reflect that.

• We are also grateful for the gardeners who pointed out some clarity issues in the body of the paper, and presentation deficiencies in the tables, and we made the appropriate changes.

• There were some suggestions that we weren't able to incorporate, such as including information sleep patterns (while we're also curious about how long foragers slept and similar information, variables of that nature weren't available) and a more detailed exploration of the "deviation" to expand to measure how "close" different societies are (an intriguing idea, but we felt it was beyond the scope of the paper). But it encouraged us to think more deeply about the paper in a way we hadn't before and we appreciate the ideas.

Again, we wish to thank all the gardeners who took valuable time out of their day to provide comments.

References

Caplan, Bryan. The Myth of the Rational Voter, Princeton University Press: New Jersey, 2007.

1. Curry, O. S., D. A. Mullins, & H. Whitehouse. “Is It Good to Cooperate? Testing the Theory of Morality-as-Cooperation in 60 Societies,” Current Anthropology, 60:1, 2019.

2. Confer, Jamie D., Judith A. Easton, Diana S. Fleischman, Cari D. Goetz, David M. G. Lewis, Carin Perilloux, and David M. Buss. “Evolutionary Psychology: Controversies, Questions, Prospects, and Limitations,” American Psychologist, 65:2, 2010.

3. Ember, Carol R. “Myths about Hunter-Gatherer,” Ethnology, 17:4, 1978.

4. Glowacki, Luke and Sheina Lew-Levy. “How small-scale societies achieve large-scale cooperation.” Current Opinion in Psychology, April: 44, 2022. 

5. Gurven, Michael and Hillard Kaplan. “Longevity Among Hunter-Gatherers: A Cross-Cultural Examination,” Population and Development Review, 33:2, 2007.

6. Hart, C.W.M. “The Sons of Turimpi,” American Anthropologist, 56:2, 1954.

7. Hart, John A. “From Subsistence to Market: A Case Study of the Mbuti Net Hunters,” Human Ecology, 6:3, 1978.

8. Jaeggi, Adrian V. and Gurven Michael. “Reciprocity explains food sharing in humans and other primates independent of kin selection and tolerated scrounging: a phylogenetic meta-analysis,” Proceedings of the Royal Society B: Biological Sciences, 208:1768, 2013. 

9. Keane, A. H. “On the Botocudos,” Journal of Anthropological Institute of Great Britain and Ireland, 13, 1884.

10. Marlowe, Frank. “Marital Residence among Foragers,” Current Anthropology, 45:2, 2004.

11. Marlowe, Frank. “Mate Preferences Among Hadza Hunter-Gatherers,” Human Nature, 15:4, 2004.

12. Nimuendajú, Curt. “Social Organization and Beliefs of the Botocudo of Eastern Brazil,” Southwestern Journal of Anthropology,” 2:1, 1946.

13. Smith, Eric Alden, Kim Hill, Frank W. Marlowe, David Nolin, Polly Wiessner, Michael Gurven, Samuel Bowles, Monique Borgerhoff Mlder, Tom Hertz, and Adrian Bell. “Wealth Transmission and Inequality among Hunter-Gatherers,” Current Anthropology, 51:1, 2010.

14. Spencer, B., and F. J. Gillen. The Northern Tribes of Central Australia. New York: MacMillan, 1899.

15. White, Douglas R., Michael Burton, William Divale, Patrick Gray, Andrey Korotayev, and Daria Khalturina. “Standard Cross-Cultural Codes.” University of Pittsburgh Press: Pennsylvania, 1970-2007.

The present article explores the nature of categorization and its role in shaping our relationship with reality. Drawing on Jens Mammen's distinction between sense categories and choice categories, and Eleanor Rosch's principles of categorization, I examine how our attitudes and modes of engagement with categories can reveal important insights relevant not only to psychology but other scientific fields as well. Furthermore, I argue that the connection between sense and choice categories can be traced by examining atypical instances and non-basic-level categories, which highlight the role of subjects embedded in particular situations. In general, categorization is an active process, influenced by our interests and commitments, even though it does not always appear as such. By correcting biases in our treatment of concepts and categories, we can ultimately correct our biases in scientific practices, thus revealing the entanglement of categorization with broader epistemological issues.

Introduction

The TV show Black Mirror features an episode called Hang the DJ that tells the story of a man and a woman who meet through a digital dating platform. The software appears to learn from their experiences through time-limited dating and sexual encounters, providing instructions through a personal device and calculating compatibility levels, with the aim of presenting each participant with their best match. Despite liking each other, the software continually pairs our two protagonists with others to expose them to more people. Feeling frustrated with the system's control over their lives, they eventually choose to destroy their devices and escape together, at which point we discover they are computer models in a simulation. The two real-life people have yet to meet each other. Additionally, we discover that the dating platform does not calculate romantic compatibility based on their experience. Instead, the decisive factor is whether two people like each other enough to rebel against the system, across repeated simulations, refusing to continue using the software. The story ends with the two real-life people receiving their compatibility score of 99.8%, not knowing how it was calculated, prepared to meet each other for the first time.

Jens Mammen’s Categories: Sense vs. Choice

The spirit of the dating platform and the spirit of choosing to exit the platform altogether correspond, respectively, to what the Danish psychologist Jens Mammen describes as sense categories (logic of types) and choice categories (logic of individuals) (Mammen, 2016). The notion of “compatibility scores” implies the ability to compare pairings numerically and determine equivalent pairings. A pairing with a 70% compatibility score is considered better than one with a 68% score, while two pairings with a score of 70% are considered equivalent. Our protagonists in Hang the DJ reject the idea of a quantitatively equivalent pairing, indicating that their relationship is unique and incomparable. Despite their decision, we learn that over 1,000 simulations, they rebelled 998 times, choosing to exit together. This led to their compatibility score of 99.8%. Thus, the protagonists’ exit was not counter to the logic of the system, but an integral part of the calculation.

The dating platform in Hang the DJ treats all its users and relationships as belonging to a single class, calculating romantic compatibility and creating comparisons and equivalences. Mammen (2016) would describe this treatment as reflecting sense categories, which treats individuals solely in terms of “types.” Mammen says that using sense categories follows “the logic of the eye” and assigns a passive, spectator position to the category-user. Although the term “logic of the eye” implies a passive subject merely observing and categorizing objects from a distance, it is in fact applicable to action-relevant categories, such as coffee cups, doorknobs, and potential sexual partners, in so far as different instances of the same category are treated as interchangeable, by virtue of having the same functional properties. Therefore, “the logic of the eye” is not limited to passive observation but extends to action. The phrase “logic of the eye” is meant to reflect the idea that one can interact with something, while occupying the relatively detached position of a spectator, such as swiping left or right on a dating app. Throughout the rest of this article, I will use “the logic of types” in association with sense categories, and “the logic of individuals” in association with choice categories.

The logic of individuals regards particular cases as unique and irreplaceable (Orilia, 2010; Strawson, 1959). Indeed, the question of replacing, comparing, or equivalence does not arise within the logic of individuals. When we think of singular objects, like a wedding ring or a childhood home, associated with personal sentiments and history, we do not typically place them within a class of comparable objects. What makes one’s childhood home significant cannot be expressed in terms of a comparison between perceptual attributes of buildings and neighborhoods. Unlike classes, individuals are tied to a time, place, historical thread, and network of relationships. Mammen's metaphorical description of choice categories is in terms of “the logic of the hand,” tracing, trajectories, co-existence, and lasting bonds. The distinction casts new light on the old dichotomies between the natural sciences versus the humanities, explanation versus understanding, and causal accounts versus narrative accounts (Dilthey, 1883; Mammen, 2017). To illustrate the distinction, we could consider how the psychological science of emotions tends to follow the logic of types and how different it would be if it were to follow the logic of individuals.

Psychological sciences treat categories of emotions, e.g., anger, in terms of how they could be mapped onto other sense categories, e.g., types of function, types of physiological activity, or types of selective pressure in our evolutionary history. In that approach, even the references to culture and social-linguistic construction of emotions are attempts at describing emotions as types of processes, and as such, with the use of sense categories (Lindquist et al., 2022). As a result, we come to understand anger with reference to an evolutionary history, neurophysiological structures and functions, and a sociocultural environment that assists or enables the experience and expression of anger.

By contrast, the logic of individuals does not treat emotions as general types. It begins, instead, by noticing a particular case of anger, for instance, anger in the life of Rufus Scott, a Black musician living in 1950s in New York City, struggling with poverty and racial discrimination. We look to see how Rufus’s anger relates to his life as a musician, his sense of identity, and with his suicide. How does he express or suppress anger in his relationships? If we are truly interested in understanding the anger of Rufus Scott, what we want to understand is not anger as a general psychological category, anger as such, but Rufus Scott’s anger experienced and expressed within the contingent web of facts and events that make up his unique life. A significant difference between the two approaches is that in the latter case, our inquiry into anger involves forming and maintaining bond, and insisting on those bonds, with particular individuals about whom we are inquiring (Krøjgaard, 2016).

Adopting the logic of individuals enables us to expand our thinking about categories beyond the confines of a superficial empiricism that focuses solely on sense impressions and their similarities. This approach is not limited to understanding human emotions and bonds but extends to other domains. For instance, in biological sciences, we may identify the similarities between bat wings and pigeon wings and classify them under the same category of “wings.” While that might be a valid treatment for some purposes, by utilizing choice categories, by tracking species as “individuals” corresponding to distinct historical threads, we could also recognize that the two wings belong to separate historical and biological lineages (Ghiselin, 1981). We could recognize the same distinction in how we think about countries. For instance, while we may group South American countries together, based on their similarities and shared geography, we can also recognize the different cultural-historical threads that correspond to each country. Any personal attachment we happen to have to a particular country within a group of countries, e.g., significant memories or cultural heritage, might strongly counteract the tendency to group countries together based on geographical or other similarities.

In the story Hang the DJ, the problem was that the artificial intelligence (AI) was designed without considering the logic of individuals. The AI placed (a simulation of) people through a series of encounters that had no inherent value or meaning, but only served to calculate the final romantic compatibility score. Based on these observations, the system would pair users with each other, aiming to produce results within the logic of types (e.g., “the system deems you two to be the best match for each other”). Such an approach, which relies on comparisons between types, cannot capture the historical-relational traces of romantic relationships. One cannot be someone's partner without actually being with them for a period of time and having made some kind of decision; One cannot merely observe a relationship and expect to understand with the logic of types. In the story, the entire pool of participants was quantified and sorted in reference to each user; they were assigned not just a person but a set of candidate matches, from which one is recognized as their most suitable match. Thus, the system offers a way to remain a spectator while relying on the individual and concrete experiences of virtual copies. Such a method, therefore, requires denying (or perhaps more accurately: cheating) the reality of time and individuals.

Hang the DJ aligns with the notion that the logic of types and the logic of individuals are incompatible. Types are treated as timeless, unchanging, and belong to a static world. Artificial intelligence seems to excel at applying the logic of types. By contrast, individuals are historical threads, capable of forming contingent yet lasting bonds, across irreversible time. Thus, the logic of individuals seems to escape AI, though in practice the AI assimilates even the choice to exit the system into its type-based calculations (see Birhane [2021] for an examination of the philosophical foundations of AI and its methodology).

But are these two types of categories truly incompatible? Contemplating this question has implications for questions regarding the relationship between the natural sciences and the humanities, and their corresponding approaches that rely differently on sense and choice categories. To explore this question further, I turn to Eleanor Rosch's prototype theory of categorization.

Eleanor Rosch & Principles of Categorization Categories

Rosch's prototype theory of categorization offers a method for dividing categories based on typicality or representativeness and level of abstraction. At first glance, these principles seem to apply solely to types, i.e., sense categories, and not individuals. However, it is important to note that sense categories are not uniform or homogenous. For example, the sense category for the color “red” includes a range of electromagnetic frequencies, with some instances more easily categorized as red (prototypes) and others less easily (marginal cases). Other categories, such as “weapon” and “bird,” similarly have prototypical and marginal cases. A seagull is a prototypical bird, while a penguin is a less typical case, and a bat is not considered a bird at all. A gun, grenade, and teargas are prototypical weapons, while a shoe or toy could be considered a marginal case. Attention to marginal cases raises the question, “Why is X an instance of the sense category A?”, which could bring another question into focus, “Why do we choose to categorize X as A, in situation Y?”

In addition to typicality, there are differences in abstractness (Rosch, 1978). According to Rosch, basic-level categories convey the most information with the least cognitive effort. One of Rosch's points is that we do not just use basic-level categories as labels; we perceive the world in terms of them. That is, the world is visible to us, first and foremost, in terms of our basic-level categories. However, there is a circularity in this description, as basic-level categories are, by definition, what is more visible and ready-to-hand. For example, I see an animal and say “lion,” which is my basic-level category, while another person says, “east African lion,” which is theirs. What is “basic-level” changes with experience, expertise, and the context of thought and communication. Nonetheless, we think and talk in communities, so we can imagine an averaging process over time, in the use of a category both by a collective and by individuals, that determines what is basic-level. According to that process, “lion” is more basic-level than “east African lion”. Returning to Mammen’s distinction between sense and choice categories, what we might observe is that deviations from basic-level and typical instances of categories serve to reveal subjects and, in doing so, reveal choice categories.

In an international gathering in Hong Kong during 2018, where a few people voiced their unease with the question, “Where are you from?” two Americans in particular, who were from predominantly republican states, added to their introductions that they had not voted for Trump. One remarked, “I never say I am from the US [i.e., base level category]. I say I am from California [subordinate category].” Our attempts to dissociate ourselves from certain categories is often an attempt on insisting that we do not represent a typical or salient category member that might otherwise serve as points of reference.

Could we say that by moving from basic-level categories to the more specific subordinate categories, we tend to move from sense categories to choice categories? Not necessarily, as long as the use of the category is in the service of conveying a mutually understood type. Moving toward choice categories would require insisting, not on non-belonging to a category, but on belonging in a different way, which would reflect something about an individual history. If I insist on being a democrat, and not a republican, I am insisting on being one type rather than another. Alternatively, using choice categories, I might describe the history of my political activities, which might include inconsistencies and contradictions (e.g., being disillusioned and leaving a political group or cause).

Sense categories are grounded in consensus and shared assumptions about what things are and how they should be described. The use of typical instances of a sense category can hide the fact that a category is being actively used, whereas atypical instances of a sense category highlights the subjectivity of the category user. Consider the following statements: (1) A gun was used as a weapon, and (2) a shoe was used as a weapon. Does the second statement raise more questions in the listener’s mind, about the situation described? Why did this person need a weapon? Did they succeed in weaponizing a shoe? The prototypical application hides the subjectivity of the category user, because the act of categorization, i.e., categorizing a gun as a weapon, aligns with norms and expectations. In atypical cases, when a shoe is being treated as a weapon, subjectivity is revealed (Giorgi, 2013). Selecting an atypical member of a category or classifying an object in terms of an unexpected category, makes the subject more salient. This points us in the direction of choice categories, with the associated logic of individuals. The sense categories involved, i.e., weapon and shoe, open a path toward using choice categories: (1) when presented with the union of gun and weapon, we are more readily able to see a sense category; and (2) when presented with the union of shoe and weapon, we more readily see a choice category (“this shoe is being used as a weapon”).

A similar point can be made about levels of abstraction. Basic-level categories conceal the subjectivity of category users, whereas subordinate categories raise questions about the people involved and the situations in which they find themselves. For example, consider the 2019 reports in Hong Kong about expired tear-gas canisters being used against protestors. Even if we set aside the higher biochemical risks of using expired tear gas, the relatively specific subordinate category raises more questions about the people responsible for using the tear gas—why did they use tear-gas canisters past their use-by date? The question more readily anchors us to the contingent historical situation of Hong Kong protests in 2019-2022.

When discussing both atypical instances and non-basic-level categories, we are not referring to anomalies, i.e., instances that don't fit into their designated category. Instead, we are referring to anomalous belonging to categories, where something does fit within a category but in so doing raises questions or induces surprise. This distinction is especially relevant in our discussion of sense/choice categories, as both atypical instances and non-basic-level categories start with sense categories, such as shoe-as-weapon or teargas past its use-by date, but ultimately lead us towards choice categories, where specific subjects appear to be selecting objects for particular purposes.

When considering a uniform sense category, like apples on a tree, the subject selecting an apple can easily fade into the background (Mammen, 2019). In these cases, the subject does not reveal anything about themselves through their act. We can, therefore, focus on the apple itself, without any apparent lack of information. However, in the case of heterogeneous categories, the act of selection becomes more visible, with the subject's act of selection playing a more prominent role. A shoe is chosen as a weapon or a carrier of a message, not just because doing so readily makes sense, but presumably due to contingent, situational pressures.

To summarize, basic-level categories and typical instances of categories tend to hide the subject, because using them involves conforming to shared norms. To do what is commonly done hides the agency of actors and their capacity to do otherwise. As such, typical instances and basic-level categories offer maximum utility in communication, interaction, cooperation. They are more likely to be used by multiple subjects who share minimal common ground. The subjects “hide” in the use of sense categories, though they are not absent. The logic of types is itself grounded in a history of interaction with the world, indeed particular cases of interactions. If we have to maintain attention on particular cases and individuals can create confusion, especially if we are tracking a large number of individuals, which is why it is useful to step back and adopt the logic of types. It is worth noting that the richness of information contained in sense categories (types) arises from dealing with concrete individuals. In the story of Hang the DJ, the system produces sense categories (“soulmates”) based on an abundance of actual experience (1,000 simulations) in a simulated reality. Thus, types were generated from particular, concrete instances.

How does recognizing sense/choice categories help scientific research? In some lines of research, we might mistake our own decisions (i.e., a choice category) in describing a situation of inquiry. In other words, how we describe an experimental manipulation might be the product of the particular history of investigation, and an attachment to seeing things in a particular way. Two examples from psychological research would help illustrate this point. The first is a line of research that examines how visual perception would change when observers place their hands near objects of sight (Abrams et al., 2008; Gozli et al., 2012). Hand-object proximity was regarded as the experimental manipulation in several published studies, and we neglected the possibility that this factor might not have the same meaning across different situations. Indeed, Bush and Vecera (2014) found that it mattered whether observers placed one hand or both hands near the objects; Goodhew and Clarke (2016) found that it mattered whether the observers were looking at a simple display or a cluttered display; Thomas (2015) found different effects of hand-object proximity depending on grasp size; and, Adam et al. (2012) found that it mattered whether the hands are moving toward or away from the objects of sight. Thus, focusing on hand-object proximity alone and ignoring the potential impact of other factors (e.g., hand movement, grasp size, display type) reflected the researchers’ bias on the basis of earlier studies and prior decisions (Gozli & Deng, 2018).

The second example comes from a line of research on rule-breaking behavior. It has been observed that following a rule tends to be relatively more efficient than breaking the rule, presumably due to the additional cognitive burden of rule breaking (Pfister et al., 2016). However, Wirth et al. (2018) found that the typical performance costs associated with breaking a rule could be eliminated if the rule breaking (a) was committed frequently and (b) was committed recently. In other words, the researchers concluded that the recency and high frequency of rule-breaking increased the efficiency of rule breaking. They assumed that regardless of the frequency and recency of the behavior, the same category (“rule-breaking”) would aptly and uniformly describe the behavior. This assumption is inconsistent with research that shows our understanding of rules can change rapidly and flexibly in response what we and others, in fact, do (Ting, 2018). To say that participants are continuing to perform the same type of behavior (“rule-breaking”) with higher efficiency, indicates the researchers’ commitment to describe the behavior with a fixed label (i.e., their choice category), overlooking the possibility that we might be dealing with two different types of behavior, namely “breaking a strict rule” versus “breaking a rule that is not very serious, because it is frequently broken.”

The above examples are not meant to suggest that choice categories are always a source of error in research. They highlight, instead, how unknowing use of choice categories, mistaking our own commitments and decisions for how things are, is the source of error. Researchers’ conscious commitment can be an advantage, just as our commitment to understand a particular person’s anger leads us to see the complex conditions in which their anger is experienced and expressed, a commitment to obtaining an experimental effect can increase the likelihood of the effect. That is, a researcher who wants to obtain an effect, persistently tracking the presence and absence of the effect under a variety of conditions, is likely to discover the conditions under which the effect is obtained (see, e.g., Roberts [2010] for an example regarding self-experimentation and sleep improvement). The key point is to recognize how our description, including scientific descriptions, reflects in part our prior decisions, commitments, and interests. A label might be too imprecise, because of how many other relevant factors are involved, as in the case of studying visual perception near the hands. A label might be too rigid, failing to capture the changes in what is under investigation, as in the case of rule-breaking behavior research.

Conclusion

Paying close attention to how we use categories, the degree to which those categories reflect generalities that reflect consensus, and the degree to which they reflect our idiosyncratic experiences and commitments, including tacit commitments of an entire scientific community, is essential to scientific research. Paying closer attention to what we presently treat as sense categories, we find the hidden historical traces that provide the basis of shared assumptions. At the same time, our shared assumptions might change, as they often reflect contingent cultural-historical conditions. Using the theoretical framework of Jens Mammen (2016, 2019) helps in sensitizing us to how we use categories, whether or not we are primarily using the logic of individuals or the logic of types.

Drawing further implications for philosophy of sciences would require more time and space. For our purpose, however, we could speculate that the natural sciences are more interested in establishing and working with the logic of types and predictable category memberships, while the humanities are more interested in anomalous category memberships, encouraging attention to individuals and their particularities, and emphasizing the ways in which existing conceptual frameworks cannot do justice to them. In addition, the present article provides further justification for seeking diversity of perspectives within scientific disciplines (Mohlhenrich & Krpan, 2021). Researchers who are “outsiders” with respect to a discipline might be able to detect the contingent choice categories that are treated as sense categories due to disciplinary consensus. They would, therefore, be able to distinguish between justified shared assumptions and unjustified conventions more easily.

Gardener Comments

Dr. Wanpeng Tan (PhD in physics):
A well written article on the topic of sense / choice categories in psychology, in which I am unfortunately not an expert at all. But it is explained so well that I can understand quite some of it. In our system of concepts and categories, if I may use non-technical language, it is really about the contrast of generalities/consensus vs individualism/idiosyncrasies. I found that it is interesting in the article's discussion of its implications for scientific research. So I'd recommend it for publication in SoS.

However, I'd speculate that the behaviors or effects could be very different in different research fields. In less precise fields (e.g., psychology), as the article presented, research activities may be more prone to error of unknowing use of choice categories. In contrast, similar behaviors could lead to transformative ideas in much stricter physical science fields like physics. The reason behind, I guess, is that less precise fields tend to have a more vaguely-defined paradigm while physical science fields, on the other hand, are much more matured with a typically well-defined paradigm, according to Thomas Kuhn. Further study of this aspect may be very intriguing.

Dr. Mario Pasquato (PhD in physics):
I found this article well written and interesting. I am convinced that it should be published by SoS.

From my point of view, as a heavy user of machine learning techniques in the context of astronomy and astrophysics, I found the discussion of sense and choice categories quite interesting. Classification has been a crucial tool in astronomy at least from the introduction of spectral types of stars (Secchi A. 1866) and machine learning is being leveraged in the field to either build new categories from scratch (unsupervised classification or clustering) or to assign pre-defined categories to objects (supervised learning). Interestingly, there is, within the explainable AI field and also in the adjacent fields of interpretability and fairness, strong interest in developing methods that move beyond mere categorization and take the individual into account. An important distinction in algorithmic fairness is the one between group fairness and individual fairness. Group fairness is an additional constraint imposed on the training of a machine learning model to enforce some measure of uniformity in the treatment of different groups. For instance an algorithm for predicting criminal activity may be required to have a similar true positive rate across different races (equality of opportunity). This obviously does not ensure that any given individual is treated fairly. Individual fairness attempts to devise schemes to ensure that similar individuals are treated similarly. A discussion of fairness in AI can be found e.g. in Caton & Haas (2020).

Besides fairness, another context where AI deals in individuals is that of providing explanations to the decisions of a model. Counterfactuals can be used to explain a model prediction and to empower a user to change it by acting on the features that caused it. The most common example of this is a machine learning model rejecting a loan application. A counterfactual is, roughly speaking, a virtual individual that is as similar as possible to the one whose application was rejected, while being accepted instead. More generally, data prototypes -that is, individuals- are used to summarize datasets and to highlight their peculiarities. This is discussed e.g. in Molnar 2022, chapter 8.7. The book also covers counterfactuals and other prototype-based methods.

I point this out because it may be of interest to the author, perhaps leading to a more detailed discussion of the relationship between the ideas presented in this article and AI.

Josh Randall:
This manuscript provides an interesting framework for differentiating how scientists interact with the world around them and when the underlying philosophy of noticing is more likely to be revealed. It is particularly useful as a resource to consider what forms of knowledge are reflecting the specific historical knowledge of both the object of study and the knowers. I found two places where this manuscript could be improved from my perspective. The two final examples used in the comparison between sense and choice categories were useful, but the discussion following them felt unrelated. Specifically the mention that these do not represent that choice categories always introduce error and that the use of choice categories can reflect our commitments which are the more likely cause of mistakes were confusing to me. I don't think either of the examples deal with these specific issues, and I think especially the second claim would need more evidence. Philosophical/ideological commitments can easily lead to mistakes, especially the overuse of sense categories when additional historical description is needed, but when can commitment to choice category cause a similar problem? Extending this, is this synthesis a space to think about when choice categories become necessary? Sense categories can work as proxies for large amounts of information, but at what point does specificity of individuals take precedence? This question would be where philosophy of science might benefit the most.

Dr. Payal B. Joshi (PhD in chemical sciences):
The author/s present a unique idea capturing the true essence of Seeds of Science journal. While a singular theme has been used i.e. the TV show explains categorization as we see it or believe to exist, the examples are adequately presented that bolsters the message and the perspectives conveyed to the reader.

It is difficult to separate logical decisions, and hence we, as scientists, label them. If we examine AI, label inputs are critical pursuits and I believe it is an oversimplification of hard sciences and their obsession to categorize.

Thus, the idea presented in the article is unique and warrants exploration, albeit not in natural sciences. I suggest adding at least one illustration to the article.

References

1. Abrams, R. A., Davoli, C. C., Du, F., Knapp III, W. H., & Paull, D. (2008). Altered vision near the hands. Cognition, 107(3), 1035-1047.

2. Adam, J.J., Bovend’Eerdt, T.J.H., van Dooren, F.E.P., Fischer, M.H., & Pratt, J. (2012). The closer the better: Hand proximity dynamically affects letter recognition accuracy. Attention, Perception, & Psychophysics, 74, 1533-1538.

3. Baldwin, J. (1962). Another Country. New York: Dial Press.

4. Birhane, A. (2021). The impossibility of automating ambiguity. Artificial Life, 27(1), 44-61.

5. Bush, W. S., & Vecera, S. P. (2014). Differential effect of one versus two hands on visual processing. Cognition, 133(1), 232-237.

6. Goodhew, S. C., & Clarke, R. (2016). Contributions of parvocellular and magnocellular pathways to visual perception near the hands are not fixed, but can be dynamically altered. Psychonomic Bulletin & Review, 23, 156-162.

7. Gozli, D.G., & Deng, W. (2018). Building blocks of psychology: On remaking the unkept promises of early schools. Integrative Psychological and Behavioral Science, 52, 1-24.

8. Gozli, D. G., West, G. L., & Pratt, J. (2012). Hand position alters vision by biasing processing through different visual pathways. Cognition, 124(2), 244-250

9. Dilthey, W. (1883). Introduction to the Human Sciences. Princeton, NK: Princeton University Press, 1989.

10. Giorgi, A. (2013). Reflections on the status and direction of psychology: An external historical perspective. Journal of Phenomenological Psychology, 44(2), 244-261.

11. Ghiselin, M. T. (1981). Categories, life, and thinking. Behavioral and Brain Sciences, 4, 269-313.

12. Krøjgaard, P. (2016). Keeping track of individuals: Insights from developmental psychology. Integrative Psychological and Behavioral Science, 50, 264-276.

13. Lindquist, K. A., Jackson, J. C., Leshin, J., Satpute, A. B., & Gendron, M. (2022). The cultural evolution of emotion. Nature Reviews Psychology, 1(11), 669-681.

14. Mammen, J. (2016). Using a topological model in psychology: developing sense and choice categories. Integrative Psychological and Behavioral Science, 50(2), 196-233.

15. Mammen, J. (2017). A new logical foundation for psychology. Springer.

16. Mammen, J. (2019). A grammar of praxis: An exposé of “A new logical foundation for psychology”, a few additions, and replies to Alaric Kohler and Alexander Poddiakov. Integrative Psychological and Behavioral Science, 53(2), 223-237.

17. Mohlhenrich, E., & Krpan, D. (2022). Amateur hour: Improving knowledge diversity in psychological and behavioral science by harnessing contributions from amateurs. New Ideas in Psychology, 65, 100922.

18. Orilia, F. (2010). Singular reference: a descriptivist perspective. Springer.

19. Pfister, R., Wirth, R., Schwarz, K. A., Steinhauser, M., & Kunde, W. (2016). Burdens of non-conformity: Motor execution reveals cognitive conflict during deliberate rule violations. Cognition, 147, 93-99.

20. Roberts, S. (2010). The unreasonable effectiveness of my self-experimentation. Medical Hypotheses, 75(6), 482-489.

21. Rosch, E. (1978) Principles of categorization. In E. Rosch & B. B. Lloyd (Eds.), Cognition and Categorization. Hillsdale, NJ: Lawrence Erlbaum.

22. Strawson, P. F. (1959). Individuals: An essay in descriptive metaphysics. London, UK: Methuen.

23. Thomas, L. E. (2015). Grasp posture alters visual processing biases near the hands. Psychological Science, 26(5), 625-632.

    ---

Subscribe now

Share

68 reported correlates of treating things as “sacred” are listed, and collected into seven themes. Most can be plausibly explained via two hypotheses. The first, taken from Durkheim, is that treating things as sacred mainly functions to bind groups together via a shared view of it. The second hypothesis, suggested by psychology’s construal level theory, is that humans acquired a habit of seeing sacred things as if from afar, even when they are close, to more consistently see those things the same as others in their groups.

Humans have many important but messy and complex behaviors associated with the words “religion”, “holy”, “sacred”, “revere”, and “awe”. After a careful examination, my judgment agrees with that of many social scientists: the core concept here is “sacred”. And I find myself especially interested in this concept, as I specialize in institution innovation, and the sacred is often invoked to explain why such proposals are infeasible. Let us now focus on the sacred.

Humans have long treated some especially important things in a distinctive special “sacred” way. For example, the following are widely seen as at least somewhat sacred: family, love, charity, courage, honesty, liberty, school, art, music, dance, fiction, innovation, epiphanies, inquiry, religion, spirits, sky, space, nature, sport, talk, friendship, health, medicine, children, democracy, law, sex, death, war, royalty, and nation. While we see wide variation in who treats what as how sacred, almost all of us treat some of these in a similar substantially-sacred way. Yes, we have many good reasons to value such things, but why value them in this special “sacred” way?

I see three obvious questions here: 

1. What exactly is this distinctive “sacred” pattern of behavior? 

2. Who tends to pick what when for this special treatment? 

3. How can we explain why we treat some things in this way, with an explanation that accounts for as many as possible of the details in these patterns? 

Over a century ago, Emile Durkheim, founder of the discipline of sociology, offered a few answers. He said that the main social origin and function of the sacred was to create camaraderie and solidarity, in part by inspiring awe, devotion and respect, and via creating an emotional state of “collective effervescence”. The sacred, he said, is “the feeling that the collectivity inspires in its members, but projected outside the minds that experience them, and objectified”. He also said that the sacred is “set apart and forbidden – beliefs and practices which unite in one single moral community … all those who adhere to them” (Durkheim 1912). 

In this paper, I offer a variation on this thesis. I start by collecting 68 reported correlates of the sacred, and grouping them into seven themes. The first three themes are A) on the sacred being valuable, B) on showing others how much we value it, and C) on shared views of the sacred binding groups together. Durkheim’s framework can make sense of these three themes, as groups seeking to bind together via a shared view of something should naturally try to get their members to value that view highly, and have their members show each other that fact.

The other four themes of sacred correlates seem harder to explain. These are D) on how sacred things are set apart, E) on how they are idealized, F) on how we are to feel not think them, and G) on how touching the sacred makes stuff more sacred. In this paper, I try to explain these other four themes by assuming that group members can better achieve social cohesion by seeing their sacred things more the same way, via seeing such things as if from afar regardless of true distance. Let me explain.

“Construal level theory” from psychology says that we have different modes for thinking about things that seem near to us, relative to far away. Things can seem near versus far in space, time, sociality, hypotheticality, and plan or language abstraction. This variation in thinking modes presents an obstacle to groups trying to bind together by seeing sacred things the same. After all, often some of them will see those things up close, while others see them from afar. For example, if you see your upcoming personal medical treatment up close, and I see it from afar, we may as a result have different opinions on your treatment’s value.

A solution here is for group members to change their usual seeing habits for sacred things. Instead of seeing near cases in near mode, and far cases in far mode, as we do for ordinary things, for sacred things we can instead see both near and far cases in far mode. When we do this, we are more likely to see those sacred things the same as others do. For example, if you and I both see your upcoming medical treatment as if from afar, we can more agree on its value. 

Construal level theory predicts that seeing things in far mode tends to cause us to idealize them, and also to intuit them instead of thinking about them. In addition, this strategy of seeing the sacred from afar regardless of distance seems most easily accomplished via a discrete mode switch, a switch which thus pushes us to “set apart” the sacred. Furthermore, always seeing a sacred topic from afar plausibly leaves us feeling a deficit of near contact with that topic, a deficit that we can fill by giving extra significance to concrete objects associated with sacred topics. Thus, this see-sacred-from-afar theory can, to varying degrees, plausibly explain all four of the remaining sacred themes listed above. 

For example, by seeing our education or career as sacred, we can more see it the same across our lifetime. This lets us feel more strongly connected to a constant personal identity, and gives us an emotional distance that can help us stick to long-term plans in the face of emotionally-potent short-term obstacles. However, this added distance also makes it harder for us to say in as much detail why we prefer our school or career choices. 

As another example, we usually see love as more sacred than sex; today sex mainly gets sacred via its connection to love. While we are pretty clear on what is sex, our attitudes toward sex vary greatly. Those directly engaged in a sex act see it in a near mode, and often see its value quite differently from distant observers. In contrast, we all tend to approve of love, in part because we all see it abstractly, as if from a distance. And this distance can help us stay committed to our love relations during difficult times.

However, even when we have been in love many times, and are currently in a long-term relationship, we often feel that we aren’t clear on how exactly that abstract concept applies to our current life. As Joni Mitchell says in Both Sides Now, “Its love’s illusions that I recall, I really don’t know love … at all.” Thus, compared to sex, seeing love as if from afar helps us to unite by seeing it the same. Communities can unite by seeing love as sacred, lovers can unite in their shared love, and each person can feel unified across time in their constant stance toward love. Even if we aren’t very sure what exactly love is.

This paper will now first discuss the correlates of the sacred, then summarize construal level theory, and finally discuss how a see-sacred-from-afar theory can help to explain key sacred correlate themes.

Correlates of the Sacred 

Below I list 68 beliefs, attitudes, and behaviors that seem to correlate with things called “sacred”. Note that the strength of many of these correlations is modest; they often have exceptions. Note also that I’m trying harder here to be complete than to avoid redundancies. This section draws from many sources, including (Durkheim 1912; Righetti 2014; Streng 2000; Tetlock 2003), but draws especially heavily from (Belk et al. 1989; Keltner 2023).

The 68 correlates are collected around seven themes, regarding the sacred being valuable, our showing its value to others, groups being united by it, it being set apart, it being idealized, our feeling not thinking about it, and how touching the sacred makes stuff sacred. (Yes, these correlates are not fully summarized by these themes.) 

A. It Is Valuable:

1. Sacred things are very highly (or lowly) valued. We revere, respect, & prioritize them.

2. We revere sacred beliefs as well as acts. We feel dirty if thoughts go near illicit beliefs.

3. Sacred is big, powerful, extraordinary. We fear, submit, & see it as larger than ourselves.

4. Sacred things matter for our health, luck, courage, & other results we care much about.

5. We should want the sacred “for itself”, rather than as a means to get other things.

6. Sacred things are either more homogenous, or more unique, whichever is better.

B. We Show That We See It:

7. It induces emotions: awe, joy, admire, serene, entrance, aesthetic, mirth, gratitude.

8. Is sometimes associated with “mystic” experiences: intense, real, unified, transcendent.

9. Makes us feel less: big, distinct, independent, in control, competitive, entitled.

10. Quiets feelings of: doubt, anxiety, ego, self-criticism, status-consciousness.

11. Often makes us feel/express tears, chills, shivers, goosebumps, “whoa”. 

12. We get emotionally attached to the sacred; our stance re it is oft part of our identity.

13. We desire to connect with the sacred, and to be more associated with it.

14. To approach the sacred, we use self-control to purify, sacrifice, & commit. 

15. We enjoy sacrificing for the sacred, to purify & respect sacred, including via odd beliefs. 

16. We feel reluctant to feel sacred joy, awe, etc. if we have not sufficiently earned it.

17. Inputs count more than outputs regarding the sacred efforts, if your heart is right.

18. We find it hard to see utopias as attractive if they lack sacred struggles & suffering. 

19. Sacred brings us comfort & consolation in hard times; losing it can feel devastating. 

20. We affirm & learn sacred via mythic stories & accounts of how we & it fit in a universe.

21. We find stories that share our sacred values and beliefs nicer and easier to understand.

22. We have rules regarding how to approach sacred stuff, in part to protect us.

C. It Unites Us:

23. Shared views about the sacred bind, define, and distinguish social groups.

24. Shared festivals & synchronized behaviors bind & charge us, & help us to see sacred.

25. We want our associates to share our views of and attachment to the sacred.

26. We get offended when others seem to deny our sacred views, and oft respond strongly.

27. We feel more equal to each other regarding sacred things; status matters less there.

28. Either everyone (e.g. love) or very few (e.g. medicine) are entitled to opinions re sacred.

29. The sacred is less for commoners, or common purposes, and more suited to elites.

30. Charismatic leaders motivate, get acceptance, in part via appeals to sacred issues. 

31. Experts of the sacred are prestigious & trusted, & oft allowed to break sacred rules.

32. The sacred makes us feel more prosocial, and sacrificing for it is seen as prosocial.

33. Sacred increases feelings of: safe, curious, cooperative, unified with universe & others.

D. Is Set Apart: 

34. Sacred things are sharply set apart and distinguished from the ordinary, mundane.

35. Sacred things do not fit well with our animal natures, such greed, status, competition.

36. Re sacred, we fear slippery slopes, so that any compromises lead to losing it all.

37. We dislike mixing sacred and mundane things together.

38. We dislike money prices of sacred, & trades to get more mundane via less sacred.

39. We dislike for-profit orgs of the sacred, relative to non-profits or government agencies. 

40. We prefer discrete rules re sacred over continuous goals to achieve (Berns et al. 2012).

41. Sacred feelings are elusive, unusual, other-worldly, spiritual, hard to describe. 

E. It is Idealized:

42. Sacred things feel less limited by physics, & can seem to have unlimited possibilities.

43. Sacred things really matter, fill deepest needs, complete us, make us pure, make all one.

44. Sacred things last longer, and decay or break less. Sometimes eternal and unchanging.

45. Sacred things are purer and cleaner, and closer to the ultimate core of existence.

46. Sacred things have fewer random coincidences; their patterns mean something.

47. Sacred values have fewer conflicts with each other; you can have them all at once.

48. It is harder to judge the relative value of sacred things, compared to mundane things.

49. Sacred things more resist precise definition and measurement. 

50. Sacred view is wider, expansive, enveloping; we are a small uninfluential part.

51. We are reluctant to end sacred ventures or jobs, or to change their processes greatly.

52. We are most willing to end or change sacred ventures and jobs in a sudden big crisis.

53. The things treated as most sacred for the longest are furthest from experience.

F. Feel Not Think It:

54. We see the sacred poorly using words, cognitive rational analysis, and numbers.

55. We see the sacred better using intuition, flow, creativity, music, images, & aesthetics. 

56. Often a “noetic” source of insight that can’t be put in words or defended via reason. 

57. Intentional efforts to control the sacred are often counter-productive.

58. Talk of the sacred uses vaguer terms, focusing on general impressions not details.

59. We like “profound” sayings that hint at deep sacred insights but don’t directly give them.

60. We are less open to arguments that might criticize the sacred.

61. How sacred things seem is less misleading; you can more trust their appearances.

62. The sacred is mysterious, unlikely and even incoherent. Who are we to question it?

63. Sacred makes us stand outside ourselves, feel ecstasy, transcendence, different reality.

64. We do not make or control the sacred; it makes and transforms us.

65. Beliefs supporting sacred behaviors tend to be less conditional.

G. Touching Makes & Shows It:

66. Stuff (objects, dates, people, words, sounds) that touches the sacred gets sacred itself.

67. We connect to sacred themes better via more frequent contact with sacred stuff.

68. Over time, stuff that we connect to often tends to become sacred via nostalgia.

Construal Level Theory

Human brains seem to be organized in part by levels of abstraction. Input signals are passed first to layers that look at very local sensory features, and then on up through layers that examine increasingly wider-scale situation features. These then feed into high-level decisions, which feed back down into increasingly concrete layers, layers which translate signals into increasingly specific directives, ending in very particular motor controls.

Construal level theory says that this affects how we think; we have different mental modes for thinking abstractly versus concretely, and a continuum of modes between (Adler & Sarstedt 2021; Liberman & Trope 2003; Liberman & Trope 2008; Trope & Liberman 2010). At any one moment, we typically consider only a few important things up close, and many less important things far away. For things up close we can attend carefully to their many concrete details, but for things far away we make do with fast crude calculations using sparse representations using only a few broad descriptors per item. (Math “abstractions” are different; they are calculated exactly and carefully.)

Things can be close versus far in many ways, including via time, spatial distance, social distance, category breadth, chance (certainty is closer), plans (constraints are close, goals are far), and language (tone, style, general impressions are far). We presume that things which tend to be close (or far) in some ways are also close (or far) in others. We expect our theories to fit reality better for far things than for close. We plausibly use near mode more to choose our literal direct actions, and we use far mode more symbolically, and to improve our social impressions.

For example, if you think in detail about your plans for tomorrow, you can see many practical obstacles to those plans, and also how hard it might be to judge if your plans are ethical. But if you instead think about your plans for an event years in the future, you will less notice practical obstacles nor ethical complications. So you will instead see your plans as easier to execute, and place them in simpler moral categories. These tendencies get even stronger if this event will also happen far away in space, be done by someone else, or only happen given some unlikely preconditions.

The following are some of the many distinctions and axes reported to correlate with near vs. far: here vs. there; now vs. then; me or us vs. them; important vs. unimportant; past vs. future; down vs. up; warm vs. cold; red vs. blue; bright vs. dark; awake vs. asleep; morning lark vs. night owl; taste or touch vs. see or hear; slang or grunt vs. polite speech; more vs. less politically polarized; fast detailed repetitive vs. slow echoey novel music; voice or picture vs. words or faces; more vs. less intense affect; fear or sadness vs. anger, guilt, shame, pride, or regret; dislike or low mood vs. like or high mood; sex vs love; tempted vs. self-restrain; dominance vs. prestige; unsure, persuadable, seek info vs. confident, stubborn; conforming vs. independent; support authority vs. support underdogs; low power via acts vs. high power via associates; math/logic analysis vs. creative analogy; case-based comparable how con reasons vs. feature-based unique why pro reasons; uncertain vs. overconfident; theory/trend breaking vs. following; common likely real local event consequences vs. rare unlikely unreal global event causes; concrete, contextual, detailed, incidental relations vs. abstract, schematic, context-free, core, coarse, goal-related properties; narrow vs. broad categories; familiar vs. novel task/event; feasible safe vs. desirable risky acts; buy vs. sell; conflicted secondary local practical plan constraints vs. coherent central global symbolic ideal moral plan concerns; means or obstacles vs. ends; strong female vs weak male emotions; socially-near folks with unstable traits in small groups vs. socially-far folks with stable traits in big groups.

Explaining The Sacred Themes

We usually have good reasons to value the many things that we see as “sacred”. But we also have many good reasons to value other things that we do not treat as sacredly. So the key question is: why do we treat sacred things differently, as described by the 68 sacred correlates listed above, collected into seven sacred themes?

I suggest that we can, to varying degrees, explain all seven of these themes by assuming that we see some things that we value in a sacred manner, in order to better bind us together into social groups and personal identities, and to distinguish those units from others. (Such a functional explanation can allow for other kinds of coexisting explanations, such as byproduct, evolutionary, developmental, or mechanistic explanations.)

This hypothesis requires that different social groups see a somewhat different mix of things as sacred, and that individuals have ways to show each other what they see as sacred. And these facts seem, by themselves, sufficient to explain the first three sacred correlate themes: on seeing the sacred as valuable, on showing that fact to others, and on groups binding together via shared views of the sacred.

The four other sacred themes seem harder to understand in these terms: sacred things being idealized, being set apart, being intuited more than deliberately thought, and rubbing off on more concrete “stuff”. Why could not groups bind together by using deliberate thoughts to value often-flawed things that are often mixed up with other things, and that don’t rub off on stuff? 

Yes, we might see each of these themes as expensive demands made of our sacred behavior,where by paying such extra costs we show stronger commitments to our sacred things. But this seems a weak explanation, as it could “explain” most any expensive behavior. Can we find stronger explanations?

I say “yes”, in the human tendency, described by construal level theory, to think differently about things that are up close versus far away. Because this tendency can be a substantial obstacle to group members seeing the sacred the same. And so I suggest that, to avoid this obstacle, humans acquired a habit of more consistently seeing some things in a relatively far mode, even when those things were actually near. 

When using such a habit to consider the sacred value of something nearby, we should experience an unusual ease in making confident overall abstract judgments, but also an unusual difficulty in connecting such overall judgments to the details of the cases before us. Such as when we ask ourselves if we are really “in love”.  This strategy seems easier to implement as a discrete change of habits, replacing our usual habit of seeing things via modes that match their perceived distance. Thus this see-sacred-from-afar hypothesis can at least weakly explain why sacred things are discretely “set apart” from non-sacred things. Also, using discrete categories better allows us to impose and enforce discrete behavioral rules, and seeing sacred things as not fitting with self-interest or competition can encourage pro-group behavior.

Things we see in far mode are given sparser mental representations, with fewer broad descriptors per item. They are thus naturally idealized. And as they are associated with large space and time scales, they are seen as varying less over smaller scales. They are given simpler descriptions with fewer varying details, details that would be needed to represent defects, limitations, random coincidences, deviations from theory expectations, and misleading appearances. As a result, far-seen things have fewer of these features. And looked at from far enough away, even the differences between different kinds of sacred things may fade and seem negligible.

Furthermore, we naturally find it harder to compare distances between things far away, compared to things. As, according to construal level theory, non-instrumental values seem further than instrumental values, we naturally find it harder to judge the relative magnitudes of non-instrumental values. Thus seeing sacred things as if from afar can explain many ways in which the sacred is idealized. 

Close things usually have more details available to study, and we usually care more about close things. Furthermore, there are usually more far things to consider, compared to fewer near things. We thus tend to pay closer deliberate calculating attention to the details of near things, while for far things we tend to make much cruder faster more-intuitive judgments (Torr & Craig 2013; Zhang et al. 2021).

This tendency of far mode thought to be more intuitive, combined with our hypothesis of seeing the sacred from afar, helps explain why the sacred is said to be better understood via intuition and aesthetics rather than via deliberate thought. And why we tend to talk about the sacred using more vague less analytic language, and have distinctive hard-to-explain sacred experiences. The act of seeing near things using far mode must also just feel different somehow from seeing near things using near mode. This habit thus plausibly induces a sense that the sacred is a different reality, and so it induces us to “transcend” or “stand outside ourselves”. 

Construal level theory says that more unlikely or impossible things are seen as further away. So a habit of seeing sacred things from afar will tend to make sacred things seem a priori unlikely or impossible. And so our see-sacred-from-afar hypothesis also helps explain why the sacred is often seen as able to defy the usual physical or animal constraints.

We expect to see types of things that vary in their actual distance also vary in how close they feel to us. And we like to be closer to things that we value. If so, a habit of seeing sacred things as if from afar, even when they are near, might plausibly make us feel a lack of near contact with such sacred topics. In which case we might plausibly seek to emphasize the occasions when we are near concrete objects (or people, rituals, dates, sounds, etc.) associated with those topics. So we might especially treasure near contact with a love letter or anniversary date, as icons of a sacred love, or near contact with a flag or national holiday, as icons of a sacred nationalism. Thus our see-sacred-from-afar hypothesis can help explain why sacred value often rubs off on associated concrete stuff. 

If we are wary of seeing the details of sacred things, as that might result in our seeing those things differently from associates, then when we must see details we should prefer to see “detached” details that less risk such conflicts. For example, our methods of collecting memory scrapbooks tend to avoid such conflicts, and we trust fiction authors to fill their stories with details that do not contradict our sacred far views.

Finally, here are some more reported correlates of far mode, relative to near mode, that are also reported to be correlates of the sacred: awe (Septianto et al. 2021), politeness, self-control, serenity, symbolism, high mood, confidence, creativity, novelty, sleep, prestige (vs. dominance), world/universe (vs. smaller units), and values (vs. decision constraints).

Conclusion

While we humans vary greatly in which things we treat as how “sacred”, we associate a consistent and distinctive pattern of attitudes and behaviors with such things. This paper lists 68 reported correlates of the sacred, and suggests that most of them can be plausibly explained via two key hypotheses. The first hypothesis, taken from Durkheim, is that the main function of treating some things we value as sacred is to bind groups together via a shared view of the sacred. The second hypothesis, suggested by construal level theory, is that humans acquired a habit of seeing sacred things as if from afar, even when they seem close, in order to more consistently see those things the same as others in their groups.

Note that this account doesn’t yet explain why different communities treat different things as sacred. Note also that instead of the general ability to treat things as sacred posited above, some instead posit that one special thing, such as the one perfect God, is naturally well-suited to be treated as sacred, and we humans only treat other things somewhat like this special thing due to their similarities to it, and due to our searching for it. Hopefully, we will find ways to empirically test this paper’s account against alternatives such as this one special thing account.. 

Note also that if we see the sacred concept as itself sacred, then the usual norms of the sacred would disapprove of our analyzing that concept using so much contextual detail, explaining the sacred as an instrumental, rather than ultimate, value, or explaining it at all, instead of leaving it as a deep awe-inspiring mystery. Thus to study the sacred, we must to some degree defy it.

The expert “priests” of a sacred area, such as religion or medicine, must often use deliberate thought to attend to details regarding that area. This raises the question: do such priests somehow manage to still see their sacred area details as if from afar, or do they fail at this, and the rest of us then either accept this failure or pretend otherwise? My guess is the latter; we pretend that they still see it as if from afar, when they mostly don’t.

The obvious policy tradeoff here is that treating an area of life as sacred adds more energy to that area, and binds groups who do so together more strongly, but all at the expense of hindering related tradeoffs, distorting via idealization, and discouraging change and thoughtful attention to detail in that area. While it seems most of us must treat some areas as sacred, we should try to direct that energy to areas of life where it does the most good, and the least harm. Math, cosmology, and our distant descendants seem to me promising candidate areas.

Gardener Comments

The author responds to all comments in the next section. 

Sander Van de Cruys (postdoctoral researcher in psychology):
I think the piece is interesting (and the topic definitely deserves more attention in psychology) but it focuses too much on the social aspects, as well as on one particular theory from social psychology (construal level theory). I would recommend looking into predictive processing theory which I think allows a better articulated perspective on sacred beliefs. For accessible introductions to this theory, see e.g. Andy Clarks recent books. Predictive processing argues cognitive systems (and by extension cultural ones) are aimed at reducing uncertainty (relative to embodied models) or, equivalently: maximize model evidence (formally: accuracy minus complexity). What is specific to sacred beliefs is that they are unconditional or context-independent, unlike virtually all our other beliefs (they are more similar to homeostatic 'beliefs' in that sense. Note that I use 'belief' here in the Bayesian sense, used in the predictive processing/active inference literature). These are (informationally) simple beliefs, in that sense they provide great clarity or uncertainty-reduction, and are easily applied in decisions about actions (cf. self-evidencing). See e.g. Van de Cruys & Heylighen 2020 "The dark side of thinking through other minds" for a very brief discussion of this. I think this actually may be what the author is gesturing at when saying (metaphorically) that it is as if we see sacred things from afar (as 'sparse mental representations'). But it can be described less metaphorically and with a better grounding in an hypothesis for why sacred beliefs are a core part of mental models (as well as easily transmitted in social communities) using predictive processing. Such reasoning would also connect in an interesting way to many items in the lists given by the author. I'm not at all against a thorough characterization of a phenomenon in lists (it's often a necessary step), but it would be nice to go beyond that, and I don't think the construal level theory or a purely social logic is up to that task. I hope these thoughts help in revising the manuscript.

Andrew Neff (psychology professor):
I support the publication of this article, although I have limited knowledge of the cited literature and the novelty of the argument. I see two main ideas addressed here: 

Idea 1 is that we should revise our definition of the construct of sacred experiences/behaviors/attitudes, emphasizing the centrality of “seeing things from a distance”. If the author's description of sacred experiences is more valid than existing attempts to describe similar experiences (e.g. mystical experiences), then we could develop better measurement instruments, which could lead to 1) stronger correlations with neurobiology and 2) more consistent relationships between these new measurements and other variables. For instance, if we differentiate between 'distanced perspectives' and 'mystical experiences,' we might discover that substances like LSD or psilocybin have a more consistent impact on the former, and consequently, we can more accurately predict the causal structure of the world.

Idea 2 is that we can understand the causal processes surrounding the generation of sacred experiences/behaviors/attitudes, in terms of the motivation of an individual to forge group bonds & individuate themselves. Ultimately, I think the author's concluding statement is most important: “While it seems most of us must treat some areas as sacred, we should try to direct that energy to areas of life where it does the most good, and the least harm”. But I don’t think the paper provides enough of a detailed prescription of how to shift social conventions about what’s sacred. For example, understanding why the American political right reveres guns or the right to self-protection could empower those on the opposite side of the spectrum. However, this paper does not provide a detailed prescription for specific cases, and it seems like it would be hard to identify a prescription at this level of abstraction.

Ted Wade:
One can ignore the theoretical aspects in this paper and here is what remains. There are many disparate kinds of things that humans seek out and accept uncritically, and we can call those “sacred”. Sacred things tend to be associated with particular collections of humans, and often the collections are defined by the sacred things.

The paper speculates that the sacred serves to bind human groups. It further speculates that this connection is stronger (or must occur?) when humans adopt a particular cognitive stance toward the sacred. The stance is defined as a construct from construal level theory: “seeing” from far rather than near, i.e. a sort of psychological distance from a subject. 

It seems that almost any pair of distinguished categories can, by suitably picking one's evidence, be assigned to the far vs near axis. The collection of pairs includes red vs blue and slang vs polite. I nominate dog vs cat. 

My point is that construal level theory can explain anything, but probably can predict nothing. Nor does it add anything to the already circular idea that we uncritically accept certain things, which happen to be the ones that we, in fact, so accept. 

At a time when uncritical beliefs are a big problem, and a lot of effort is being made to understand this, saying it’s all near and far is not helpful. There must be a huge number of reasons why different things become sacred. Perhaps the best theories will find that there is mutual feedback between the originating of a sacred idea and the formation or maintenance of groups to which it is sacred. And this will probably, at least sometimes, have to do with human needs for relatively stable identities (a point to which the paper does allude), itself a huge and ancient topic. 

My opinions of the theories aside, is it possible that the paper offers any novel insight about the material, something that might lead to progress in understanding identity, uncritical beliefs, or the concepts in construal level theory? While trying to understand the paper I did a search for [construal level theory, sacred]. There were 84,000 hits, so it would probably take a lot of work to determine whether there is novelty or not. 

I don’t think the paper qualifies as a seed of science. 

Josh Randall: 
This article attempts to tie construal level theory from psychology to the colloquial conception of sacred. It has several issues which I think would require intense revision. 

1. As a sociological question of 'sacred', this article does not attempt to determine if sacred as a concept varies significantly across cultures. Further, I do not think it really attempts to determine if closeness/farness also vary with cultures. There is a note that proximity can be temporal, spatial, and linguistic but these should be more central to any claims about the meaning of sacred. 2. As a non-sociologist and non-psychologist, I am unaware of construal level theory outside of this article, and I am unconvinced that it is an accurate portrayal of human minds. 3. I am unsure of the importance of themes surrounding the sacred. I would say every single statement in the list of correlates needs additional evidence. I think they are all sourced from other authors, but are these their opinions or survey data? 4. There is a mention of policy value for the sacred. What is the value in idealizing the world outside of our own personal experiences? Shouldn't science be directed toward making our perceptions of the world more precise? This is itself subject to philosophies of science, but I don't see many uses for increased sacredness in particular.

Jack Arcalon:
This research should be encouraged, and will become more important as we strive to understand and manipulate top-level values in AI systems. Something as complex as "sacred" could have a simple structure we don't know yet.

Joe R:
The only reference to real-world testing I was able to find in this article was "Hopefully, we will find ways to empirically test this theory." The two hypotheses put forward by the article are interesting ways to think about the concept of sacredness, but I can't tell what effects we should expect if they are true vs false. Also, I am not sure that the many references in the article all point in the same direction, let alone at those two specific hypotheses. Does binding groups together serve an evolutionary advantage that encourages a sacredness instinct in subsequent generations? It's an interesting idea, but I think it needs a bit more condensing and clarification. 

Dr. Payal B. Joshi (PhD in chemistry):
The article is based on a unique idea, yet why has the author taken up this study is not quite clear. The objective of penning on sacred-ness should be clarified in the Introduction. Overall 64 reports were studied is a good reference data set and is appreciated. Also, authors have presented the hypothesis in a clear manner, hence there is no particular scientific flaw. I enjoyed reading the article and also learnt something on sacred and other related behavioral traits. I recommend publication of the article only post- minor revision. 

Roger’s Bacon:
I wholeheartedly recommend this article for publication – it provides a novel hypothesis about an important psycho-social phenomenon and meets all of the criteria for a “seed of science”. Two comments

1) As evidenced by many ancient cosmogonies, the first things that humans ever regarded as sacred – the sun, the sky, the ocean – are all extremely vast and/or far away. Not sure exactly what should be made of this, but it seems relevant. Did they provide a kind of template for the sacred and/or god? (seemingly) infinite or infinitely far away, appearing the same to everyone, giving life to the entire world through energy or water. 

2) There is a strain of mystical thinking in various religions where God becomes so sacred as to become wholly unknowable, and thus only approachable by negation – apophatic theology. Consider the following quote by French sociologist/theologian Jacues Ellul:

"Faith in God-in a God who does not incarnate some natural force or who is not the abstract and hypostatized projection of one of our own desires or aspirations or values (Feuerbach), faith in a God who is different from all that we can conceive or imagine--cannot be assimilated to belief. For this God cannot be assimilated to one of the representations that we might easily multiply. If God is God, God is inevitably different from all that polytheists call god. Each of those gods can be described and defined; each has its own function and sphere of action. But the God of faith is inaccessible and inassimilable. God is so fundamentally other (if God were not, if God could be measured against one of our values or beliefs, God would not be God) that we can neither define nor contemplate God. The God of faith is totally inaccessible. The affirmation of Feuerbach, that God is an absolutized value, was simplistic and puerile. For one thing, we have no idea of what the absolute or the infinite is. We cannot say anything about it or assimilate it. To talk about an absolutized value might be to talk about God, but it is not possible for human beings to absolutize anything." (Essential Spiritual Writings)

In terms of this paper’s hypothesis, we could say that Ellul is presenting an image of god that is so far away, so completely distant to the human, that we cannot even see it or begin describing it.

Tyler Cowen and Reza Azlan had a discussion on Tyler’s podcast that also touches on this theme:

COWEN: What do you think of the Kenneth Cragg argument that, at least in many branches of Sunni Islam, the distance between man and God is simply too great, and there’s something cold or alienating about it?

ASLAN: I love Kenneth Craig a lot, by the way. I’ve got to be honest, it appeals to me.

COWEN: The distance appeals to you.

ASLAN: The distance appeals to me... I always had a hard time with this idea that Jesus was my best friend, that God is my buddy. It’s a joke that my friends and I do all the time, but they gave us this little card with Jesus’s picture on it, and it said, “Jesus in my pocket,” and you were supposed to keep it in your pocket all the time. I remember thinking to myself, “Do I want Jesus in my pocket? There’s something that doesn’t feel right about that."

When I went to college, I happened to go to a Catholic university, and I thought, “Oh, I don’t know anything about Catholicism. I’ll go and check out this mass.” There, Jesus wasn’t in your pocket. Jesus was way up there on the cross, and you went to him with fear and trembling and distance. I’ve got to say I liked it. It appealed to me. Maybe it’s because of the fact that I was raised Muslim, even though at the time, I wouldn’t have really understood what that meant. But there’s something about that notion of the distance between man and God that I like. I want to approach God with that level of reverence, that makes God other, in a sense. I don’t know if that makes sense, but that’s just my personal opinion on it."

The idea of a distant god very much resonates with me as well, and if I had to characterize my own preferred modality of thinking it would be much more towards the “far” end of the spectrum. Maybe this doesn’t necessarily speak to the “sacred-as-far” hypothesis, but I wonder if it offers some kind of opportunity for testing it: does favoring a “god as distant/other” perspective correlate with certain views on the nature of the sacred and preferences for "far" thinking? 

Author’s Response to 8 Seeds of Science Reviews

My paper makes three moves. FIrst, it tries to define “sacred” as a cluster concept in terms of 68 correlates, collected into seven themes, and thus sets our task as explaining why these correlate. Second, it accepts Durkheim’s account as explaining three of those themes. Third, it suggests a way to combine psychology’s Construal Level Theory (CLT) with Durkheim’s account to explain the other four themes. 

None of the eight reviewers seem to question any of the particular correlates offered, their grouping into themes, Durkheim’s account of the sacred, or any of my particular stories for how Durkheim’s account or CLT could account for the themes. 

Two reviewers seem to just reject my attempt to define “sacred” as a cluster concept. For example, Van de Cruys defines it in terms of beliefs being unconditional, and Wade in terms of what we seek out and accept uncritically. 

Three reviewers quickly dismiss CLT, though without citing any critiques of CLT. Randall rejects CLT because he is unaware of it. Wade rejects CLT because many correlates of its key concept have been reported; it therefore “can predict nothing”. (And because CLT came up in many web searches, he can’t see that my paper has any novelty.) Van de Cruys states without elaboration that CLT isn’t up to the task of explaining things, and suggests that the theory in which he has specialized does better, though he only seems to try to explain one of the 68 correlates. 

Two reviewers want the paper to include more things, things that I agree would be valuable contributions, but which I see as beyond the scope of this effort. Joe R wants more concrete empirical tests to be offered. And Randall wants me to show how sacred and construal-level correlates vary across cultures, and give much more evidence regarding each correlate.

Neff seems to say that I propose to define sacred in terms of seeing things from afar. But CLT says we often see most everything from afar. The difference I’m proposing is that for sacred things we see them from afar even when they are close. 

Some reviewers have suggested additions to my list of correlates. I’ve added Van de Cruys unconditional beliefs, and Bacon’s tendency to be especially far from experience. 

Joshi wants me to say why I’m interested in the topic; I added that to the introduction.

At Neff’s suggestion, I now say a bit more about how to shift the sacred to areas where it might do less harm.

References

1. Susanne Adler, Marko Sarstedt (2021) “Mapping the jungle: A bibliometric analysis of research into construal level theory” Psychology and Marketing 38(9): 1367-1383, September. https://doi.org/10.1002/mar.21537

2. Russell W. Belk, Melanie Wallendorf, and John F. Sherry, Jr. (1989) “The Sacred and the Profane in Consumer Behavior: Theodicy on the Odyssey” Journal of Consumer Research , 16(1): 1-38,  June. https://www.jstor.org/stable/2489299 

3. Gregory Berns, Emily Bell, C. Monica Capra, Michael Prietula, Sara Moore, Brittany Anderson, Jeremy Ginges and Scott Atran (2012) “The price of your soul: neural evidence for the non-utilitarian representation of sacred values” Philosophical Transactions of the Royal Society B: Biological Sciences 367(1589), March 5. https://doi.org/10.1098/rstb.2011.0262

4. Emile Durkheim (1995) [1915] The Elementary Forms of the Religious Life, translated by Karen Fields. New York: The Free Press. 

5. Dacher Keltner (2023) Awe: The New Science of Everyday Wonder and How It Can Transform Your Life. Penguin Press. 

6. Nira Liberman, Yaacov Trope (2008) “The Psychology of Transcending the Here and Now” Science 322(5905):1201-1205, November 21. https://www.science.org/doi/10.1126/science.1161958

7. Nira Liberman, Yaacov Trope (2003) “Construal level theory of intertemporal judgment and decision” In G. Loewenstein, D. Read, & R. Baumeister (Eds.), Time and decision: Economic and psychological perspectives on intertemporal choice (pp. 245–276). Russell Sage Foundation.

8. Nicola Righetti (2014) “The Sacred in Current Social Sciences Research”. Italian Sociological Review, 4 (1):133-163. http://dx.doi.org/10.13136/isr.v4i1.77 

9. Felix Septianto, Yuri Seo, Loic Pengtao Li, Linsong Shi (2021) “Awe in Advertising: The Mediating Role of an Abstract Mindset”  Journal of Advertising, June 21. https://doi.org/10.1080/00913367.2021.1931578

10. Frederick Streng (2000) “Sacred”Encyclopedia Britannica. https://www.britannica.com/topic/sacred

11. Philip Tetlock (2003) “Thinking the unthinkable: sacred values and taboo cognitions” Trends in Cognitive Sciences 7(7):320-324, July. 

12. Stuart Torr, Tracy Craig (2013). “Addressing dualism in mathematical abstraction: an argument for the role of Construal Level Theory in mathematics education.” Proceedings of Delta ’13, The Ninth Southern Hemisphere Conference on the Teaching and Learning of Undergraduate Mathematics, pp 199-208. 

13. Yaacov Trope, Nira Liberman (2010). Construal-level theory of psychological distance. Psychological Review, 117(2), 440–463. https://doi.org/10.1037/a0018963

14. Yufeng Zhang, Mo Luan, Hong Li, and Yiling Ren (2021) Social Psychology “Intuition Versus Deliberation: How Decision Mode Influences Desirability and Feasibility Preferences 52(2):114-124. https://doi.org/10.1027/1864-9335/a000442

Subscribe now

Share

Although drugs are a critical component of mental healthcare, most have modest benefits and significant side effects. One way to develop a superior intervention would be to administer drugs with the spatial and temporal precision that better replicates natural diversity within neurotransmitter systems. A technology called focused-ultrasound (FU) may be able to safely and transiently disrupt the blood-brain barrier with spatial precision, permitting the site-specific delivery of molecules that do not conventionally cross the blood-brain barrier. If this method is proven to be safe and effective in larger human trials, it may trigger a paradigm shift in biopsychology research where the level of precision with which neurotransmitter systems can be influenced is massively increased. In this article, we use the example of oxytocin in the treatment of Autism. We propose that intranasal administration is not highly effective because it leads to oxytocin’s wide dispersion throughout the brain, failing to specifically stimulate oxytocin’s prosocial effects in specific regions. Consequently, we hypothesize that site-specific delivery of oxytocin, particularly in brain regions such as the Nucleus Accumbens and Ventral Tegmental Area, would lead to more consistent benefits.

I. Introduction

The most convenient method of drug delivery involves compressing drugs into ingestible tablets, topical ointments, penetrative injections, or inhalation devices (Bhagwat & Vaidhya, 2013). Consequently, drugs typically disperse throughout the bloodstream and eventually the whole brain, rather than to specific brain regions. Distributing drugs throughout the brain is not useless, many drugs can be effective for a range of psychiatric conditions. However, across all psychiatric disorders, many people are non-responsive to pharmacological treatment (Leichsenring et al., 2022), and some disorders (e.g. Borderline Personality Disorder) are not widely-accepted to respond to any drug (Tyrer & Silk, 2011). One explanation is that drugs, and the neurotransmitter systems on which they act, have diverse effects on psychology and behavior depending on their location in the brain.

Until recently, without invasive surgical procedures, it has not been possible to target drugs to specific brain regions. In this paper, we’ll describe a technology called Focused Ultrasound Mediated Drug Delivery (FUDD), which uses low-intensity sonic waves to transiently, safely, and focally disrupt the blood-brain barrier (Tharkar et al., 2019). Most commonly, this method is used to deliver toxic chemotherapeutic drugs to tumors (Beccaria et al., 2020). However, one study has used focused ultrasound to deliver the neurotransmitter GABA in rats (Todd et al., 2019), and the basic principle of focal-drug delivery would likely be valuable for a range of drugs and conditions, including mental disorders.

This paper focuses on the use of oxytocin in the treatment of Autism for three reasons. First, despite early optimism, the largest modern trials show that intranasally administered oxytocin is not beneficial for the social symptoms of Autism (Sikich, 2021). Second, invasive research in animals has shown that oxytocin has contrasting social effects in different parts of the brain (Steinman et al., 2019): this data provides a theoretical rationale for why human treatments are ineffective. Finally, as oxytocin does not readily cross the blood-brain barrier (Yamamoto & Higashida, 2020), it would be a natural candidate for this procedure (more detail in later sections).

However, the utility of FUDD likely extends beyond oxytocin, potentially to any drug or molecule that does not conventionally cross the blood-brain barrier, and has a different effect on behavior depending on its location in the brain. Pending additional safety studies, this technology may usher in a transformative new approach for studying the causal links between neurochemicals and behavior, and could open a new category of treatment targets for many psychiatric conditions.

II. Oxytocin & Autism

i. Oxytocin Neurobiology

Oxytocin is a neuropeptide and hormone produced primarily in the hypothalamus and is secreted from nerve terminals in the posterior pituitary gland into the bloodstream (Gimpl & Fahrenholz, 2001). In addition to the systemic release, oxytocin neurons project to several brain regions, including the BNST, Nucleus Accumbens, and Ventral Tegmental Area. 

ii. Correlations Between Oxytocin And Social Behavior

Several lines of research support a link between oxytocin and social behaviors. For example, genetic deletions of the oxytocin gene leads to changes in animal sociality and sexual behaviors (Winslow & Insel, 2002). In humans, variation in the oxytocin gene is sometimes found in individuals with Autism or other divergent social phenotypes (Cataldo et al., 2018). Outside of genetics, the concentration of oxytocin, measured in the bloodstream, often correlates with social phenotypes. For example, children with autism have significantly lower levels of oxytocin than neurotypical children - but the same was not true for adults (John & Jaeggi, 2021). These findings suggest that oxytocin is a plausible intervention target for people with social disorders like Autism.

iii. Interventional Studies with Oxytocin

Administered intravenously, oxytocin does not readily cross the blood-brain barrier (Mens et al., 1983), but refer to Lee et al. (2018). More commonly, oxytocin is administered intranasally, which may allow it to bypass the Blood Brain Barrier by traveling along olfactory nerves (Quintana et al., 2015). When administered intranasally, oxytocin does not distribute throughout the brain, but it still enters a large set of regions including the orbitofrontal cortex, striatum, brainstem, and thalamus (Lee et al., 2020).

Within healthy populations, intranasal oxytocin may modestly promote the recognition of some facial expressions, and promote in-group trust, but not out-group distrust (IJzendoorn & Bakermans-Kranenburg, 2012; Leppanen et al., 2017). However, these trials have not translated into a clinically useful therapy. The largest existing trials on people with Autism have shown that oxytocin does not benefit social or cognitive functioning (Sikich, 2021). Meta-analyses also support the point that systemically-administered oxytocin does not improve the symptoms of either Autism or Schizophrenia (Martins et al., 2021; Sabe et al., 2021).

These studies provide evidence that oxytocin can influence social behavior, but there is not a clear and consistent effect amongst all participants, particularly considering clinical groups and clinically relevant symptoms. 

iv. Diverse Actions of Oxytocin in the Animal Brain

In animal brains, oxytocin has a different relationship with behavior depending on the location of release, see Steinman et. al (2019) for a comprehensive review. Briefly, in the Bed Nucleus of the Stria Terminalis (BNST), a major efferent connection of the amygdala, oxytocin is related to negative social behaviors like avoidance and vigilance (Martinon et al., 2019; Ayers et al., 2011; Nasanbuyan et al., 2018; Duque-Wilckens et al., 2018; Duque-Wilckens et al., 2020). Contrastingly, in reward-related regions like the VTA and Nucleus Accumbens, oxytocin is related to prosocial behaviors like approaching other animals (Dölen et al., 2013; Dölen & Malenka, 2014). Other brain regions also may have distinct relationships with behavior, for example, in the amygdala, oxytocin can reduce fear responses (Knobloch et al., 2012).

Overall, oxytocin can impact both positive and negative social and emotional behaviors. Hypothetically, in human studies, oxytocin is simultaneously producing several, potentially contrasting, effects from different areas of the brain. Therefore, we suggest that one way to improve human oxytocin therapeutics is to target oxytocin to brain regions such as the Nucleus Accumbens and Ventral Tegmental Area.

III. Focused Ultrasound

i. Biophysics & Equipment

Focused ultrasound (FU) is a multi-purpose technology based on the spatially-precise direction of short pulses of energy in the form of sound waves (Meng et al., 2021). Some of its applications include thermally ablative surgery (Orsi et al., 2010), stimulation of neuronal activity (Yu et al., 2021), and when used at lower power, transient disruption of the blood-brain barrier (BBB) (Burgess & Hynynen, 2013). 

Technically, FU waves start from frequencies of 20 kHz and never exceed 200 MHz (Buch et al., 2018). The precision of focused ultrasounds is achieved with the help of transducers, particularly concave-focused transducers for the purpose of disrupting the blood-brain barrier (Izadifar et al., 2020). Focused ultrasound interventions are often used in tandem with imaging technology, such as MRI or ultrasound sonography, to guide the ultrasound (Izadifar et al., 2020). 

ii. Biological Mechanisms Opening the Blood-Brain Barrier

The blood-brain barrier (BBB) refers to a system of endothelial and nervous cells that tightly control what substances can enter the brain (Daneman & Prat, 2015). To a greater extent than other parts of the circulatory system, endothelial cells in the brain are tightly connected with transmembrane proteins called tight junctions. The mechanism through which FU disrupts the BBB relies on the injection of small gas microbubbles into the bloodstream (Hynynen et al., 2003). When exposed to ultrasound, tight junction proteins dissociate from each other (Sheikov et al., 2008). Mechanistically, these bubbles absorb the sound energy and begin to expand, oscillate, and sometimes rupture (Tung et al., 2011), theoretically pressing against and stretching capillary walls, ultimately detaching tight-junctions proteins.

iii. Safety & Characteristics of Human Clinical Trials

At least 9 human clinical studies have used Focused Ultrasound in the brain in an attempt to permeabilize the blood brain barrier (See PDF for Appendix - Table 1). These studies are typically small, including 3-15 subjects, and have been conducted on people with Alzheimer’s Disease, Parkinson’s Disease, ALS, and Glioblastoma. The drug delivered varies, some studies introduced toxic chemotherapeutic agents, while others did not use any drug at all.

For a complete review of safety, see Meng et al. (2019). Briefly, there are two main risks to FUDD. The first safety risk includes the prospects of toxic chemicals or pathogens entering the brain. Even at low energies, FU can trigger the extravasation of red blood cells into the brain and alterations to surroundings cells like pericytes (Wang et al., 2020).

Beyond pathogens and toxins, chemical components found in the bloodstream could more subtly influence neuronal activity and thereby pose an additional psychological risk. For example, most neurotransmitters are present in the bloodstream (Wishart et al., 2022), and therefore, opening the BBB may influence brain function, and consequently behavior, even in the absence of an intentionally-administered drug.

The second risk of FUDD is the potential for a direct interaction between focused ultrasound and brain tissue, independent of effects on the BBB. At least regarding thermal damage, the risk is low, as the energy used for BBB disruption is 1,000 times lower than what is used in surgical ablations (Meng et al., 2019).

Empirically, at least 10 trials have evaluated the safety of FUDD in human clinical populations. Many have reported mild side effects, including: headaches, vagal responses, scalp pain, edema/bruising attributable to placement of the stereotactic frame, musculoskeletal pain, scalp petechial rash, and transient FLAIR hyperintensity in sonicated brain tissue. No existing trials have reported any major side-effects.

One major unresolved question is the effects of long-term frequent FU. This may be an important point when considering psychiatric applications - as many drugs need to be taken on a daily basis, sometimes over months, to achieve their intended effect. The most intensive trial to date has tested 24 separate stimulations (Park et al., 2020), but most other studies used approximately five stimulation sessions.

IV. Potential Limitations of Regionally-Focused Drug-Delivery

FUDD has the potential to overcome a major limitation at the core of modern psychopharmacology - the inability to target drugs to specific brain regions. However, this limitation is not the only reason why drugs are not always effective. In this final section, we offer some reasons why researchers should be cautious in estimating the potential value of this approach for social disorders, and potentially any psychiatric condition.

i. Untested Safety and Uncertain Feasibility of Frequent Application

If the effects of a drug, like oxytocin, are acute, the FUDD procedure may need to take place more frequently. As mentioned, this poses a safety risk - no studies have evaluated the outcomes of long-term, frequent disruption of the blood-brain barrier. Moreover, this would pose a technical implementation challenge. FUDD is currently restricted to clinical settings (Tempany et al., 2011), as it relies on a large, fixed, expensive piece of equipment that requires expertise to operate. If the drug needs to be administered regularly, the FU device may have to become portable and easier to operate.

ii. Limited Biological Precision

There are many remaining technical limitations to FUDD. One limit is temporal resolution: the BBB can remain open for hours after the procedure (Sheikov et al., 2008), and therefore, oxytocin would diffuse for perhaps an unnaturally long amount of time relative to the natural rapid dynamics of oxytocin release. Moreover, although the spatial resolution is precise, it still covers an area of cubic millimeters (Hu et al., 2022). This is small enough to target brain structures like the Nucleus Accumbens or Ventral Tegmental Area mentioned above. However, this area still contains tens of thousands of neurons. The generic release of oxytocin in this area may not adequately simulate the natural complex activity patterns of the neurons within that space. These clear technical limitations could provide an explanation for potential trial failures, and would suggest the need for even further technological development. 

V. Conclusion

Focused ultrasound may represent a novel treatment option in psychiatry as it provides non-invasive targeted drug/neuropeptide delivery by transiently opening the blood-brain barrier. Questions about safety remain, as this approach has not been tested in psychiatric conditions, it has not been tested on large subject populations, and few studies have tested the effects of frequent administration. Moreover, there is reason to believe that even the spatially-precise administration of drugs may not be adequately detailed to consistently and powerfully manipulate behavior: there will still be diverse neurons upon which the drug acts, even when the drug is restricted to a single brain region. Nonetheless, if safe, this technology could fundamentally change the way that scientists study neurotransmitter systems in the human brain, which would lead to countless new interventional trials for all mental illnesses and social disorders.

Please see PDF for Appendix

Gardener Comments

Roger’s Bacon:
This paper is an exploratory survey of a new treatment modality for psychiatric diseases, and in all respects (speculative, rigorous, well-written) is deserving of publication with SoS. 

A few thoughts that come to mind:
1) How many drugs (medical or recreational) could have their effects modified (e.g. side effects reduced) through more targeted delivery with FUDD? A next step might be to survey the most common psychiatric drugs and analyze how many of them would have their effects significantly modified if delivery could be targeted to specific brain regions. How useful would FUDD really be if we proved it was safe/effective? 

2) Does this have any application to bodily diseases? There may not be a blood-brain barrier (BBB) in the body (obviously) but perhaps this technique could lead to increased absorption in particular organs/diseased tissues?

3) The authors write "Beyond pathogens and toxins, chemical components found in the bloodstream could more subtly influence neuronal activity and thereby pose an additional psychological risk." Is there variation in the tightness of the BBB across different people (presumably variation in relevant proteins)? This could point to some interesting biology - are some people inherently more affected by the ambient neuro-active components of their blood? 

Josh Randall:
This paper attempts to apply the concept of focused ultrasound to a new set of clinical questions, specifically the social effects of autism. I am personally very skeptical of research surrounding autism, especially searching for a 'source' or 'cure'. The author makes an effort to showcase the history of this technique in other clinical settings, especially highlighting the safety and related effects of altering the blood-brain-barrier. Their description of limitations of this work is appreciated, including the need for highly trained operators in clinical settings, the large range that oxytocin would still affect when applied, and the need for regular focused ultrasound administration for possible success. Two criticisms that I still have are: 1) does the BBB being open allow the body to supplement the brain with other secreted proteins/hormones that might reduce the effectiveness of supplying oxytocin and 2) if there is not strong evidence that oxytocin has clinical effects on autistic people, is it ethical to pursue this type of research has a potential 'cure' for autism. My second point more broadly is asking whether autism needs curing and whether this type of work is something that a large number of autistic people would want or need to live fulfilling lives.

Dr. Payal B. Joshi:
The article is written in an extremely scientific manner illustrating oxytocin delivery within the brain using focused ultrasound technique. Authors have critically supported their hypothesis and presented reported clinical trials that depict their robust review. I suggest a minor revision on providing an illustration or a figure in the manuscript where authors have described the biophysics and equipment section. 

References 

1. Abrahao, A., Meng, Y., Llinas, M., Huang, Y.. Hamani, C., Mainprize, T., Aubert, I., Heyn, et al. (2019). First-in-human trial of blood–brain barrier opening in amyotrophic lateral sclerosis using MR-guided focused ultrasound. Nature Communications, 10(1), 1–9. doi: 10.1038/s41467-019-12426-9.

2. Ayers, L. W., Missig, G., Schulkin, J., & Rosen, J. B. (2011). Oxytocin reduces background anxiety in a fear-potentiated startle paradigm: peripheral vs central administration. PMC - Neuropsychopharmacology, 36(12), 2488–2497. doi: 10.1038/npp.2011.138. 

3. Beccaria, K., Canney, M., Bouchoux, G., Desseaux, C., Grill, J., Heimberger, A. B., & Carpentier, A. (2020). Ultrasound-induced blood-brain barrier disruption for the treatment of gliomas and other primary CNS tumors. Cancer Letters, 479, 13-22. doi: 10.1016/j.canlet.2020.02.013.

4. Bhagwat, R. R. & Vaidhya, I. S. (2013). Novel Drug Delivery Systems: An Overview. International Journal of Pharmaceutical Sciences and Research 4(3), 970-982. doi: 10.13040/IJPSR.0975-8232.4(3).970-82. 

5. Buch, E., Boyle, N., & Shivkumar, K. (2018). 123 - Catheter Ablation: Technical Aspects. Elsevier - Cardiac Electrophysiology: From Cell to Bedside (Seventh Edition). doi: 10.1016/B978-0-323-44733-1.00123-1.

6. Burgess, A., & Hynynen, K. (2013). Noninvasive and targeted drug delivery to the brain using focused ultrasound. ACS Chemical Neuroscience, 4(4), 519-526. doi: 10.1021/cn300191b. 

7. Carpentier, A., Canney, M., Vignot, A., Reina, V., Beccaria, K., Horodyckid, C., Karachi, C., et al. (2016). Clinical trial of blood-brain barrier disruption by pulsed ultrasound. Sci. Translational Medicine, 8(343). doi: 10.1126/scitranslmed.aaf6086. 

8. Cataldo, I., Azhari, A., & Esposito, G. (2018). A review of oxytocin and arginine-vasopressin receptors and their modulation of autism spectrum disorder. Frontiers in Molecular Neuroscience, (11)27. doi: 10.3389/fnmol.2018.00027. 

9. Chen, K.-T., Lin, Y.-J., Chai, W.-Y., Lin, C.-J., Chen, P.-Y., Huang, C.-Y., Kuo, J.-S., et al. (2020). Neuronavigation-guided focused ultrasound (NaviFUS) for transcranial blood-brain barrier opening in recurrent glioblastoma patients: clinical trial protocol. Annals of Translational Medicine, 8(11). doi: 10.21037/atm-20-344.

10. Daneman, R., & Prat, A. (2015). The Blood-Brain Barrier. Cold Spring Harbor Perspectives in Biology. 7(1). doi: 10.1101/cshperspect.a020412. 

11. D’ Haese, P.-F., Ranjan, M., Song, A., Haut, M. W., Carpenter, J., Dieb, G., Najib, U., et al. (2020). β-Amyloid Plaque Reduction in the Hippocampus After Focused Ultrasound-Induced Blood–Brain Barrier Opening in Alzheimer's Disease. Frontiers in Human Neuroscience, 14(593672), 1–5, doi: 10.3389/fnhum.2020.593672. 

12. Dölen, G., Darvishzadeh, A., Huang, K. W., & Malenka, R. C. (2013). Social reward requires coordinated activity of nucleus accumbens oxytocin and serotonin. Nature, 501, 179–184. doi: 10.1038/nature12518. 

13. Dölen, G., & Malenka, R. C. (2014). The Emerging Role of Nucleus Accumbens Oxytocin in Social Cognition. Biological Psychiatry, 76(5), 354-355. doi: 10.1016/j.biopsych.2014.06.009. 

14. Duque-Wilckens, N., Steinman, M. Q., Busnelli, M., Chini, B., Yokoyama, S., Pham, M., Laredo, S. A., et al. (2018). Oxytocin receptors in the anteromedial bed nucleus of the stria terminalis promote stress-induced social avoidance in female California mice. Biological Psychiatry, 83(3), 203-213. doi: 10.1016/j.biopsych.2017.11.009. 

15. Duque-Wilckens, N., Torres, L. Y., Yokoyama, S., Minie, V. A., Tran, A. M., Petkova, S. P., Hao, R., et al. (2020). Extrahypothalamic oxytocin neurons drive stress-induced social vigilance and avoidance. Proceedings of the National Academy of Sciences of the United States of America, 117(42), 26406-26413. doi: 10.1073/pnas.2011890117. 

16. Gasca-Salas, C., Fernández-Rodríguez, B., Pineda-Pardo, J. A., Rodríguez-Rojas, R., Obseo, I., Hernández-Fernández, F., del Álamo, M., et al. (2021). Blood-brain barrier opening with focused ultrasound in Parkinson's disease dementia. Nature Communications, 12(1), 1–7. doi: 10.1038/s41467-021-21022-9.

17. Gimpl, G., & Fahrenholz, F. (2001). The oxytocin receptor system: structure, function, and regulation. Physiological Reviews, 81(2), 629-683. doi: 10.1152/physrev.2001.81.2.629. 

18. Hu, Z., Chen, S., Yang, Y., Gong, Y., & Chen, H. (2022). An Affordable and Easy-to-Use Focused Ultrasound Device for Noninvasive and High Precision Drug Delivery to the Mouse Brain. IEEE Transactions on Biomedical Engineering, 69(9), 2723-2732. doi: 10.1109/TBME.2022.3150781.

19. Hynynen, K., McDannold, N., Vykhodtseva, N., & Jolesz, F. A. (2003). Non-invasive opening of BBB by focused ultrasound. Springer - Brain Edema, 86, 555-558. doi: 10.1007/978-3-7091-0651-8_113.

20. Izadifar, Z., Izadifar, Z., Chapman, D., & Babyn, P. (2020). An Introduction to High Intensity Focused Ultrasound: Systematic Review on Principles, Devices, and Clinical Applications. Journal of Clinical Medicine, 9(2), 460. doi: 10.3390/jcm9020460. 

21. John, S., & Jaeggi, A. V. (2021). Oxytocin levels tend to be lower in autistic children: A meta-analysis of 31 studies. Autism, 25(8), 2152–2161. doi: 10.1177/13623613211034375. 

22. Knobloch, H. S., Charlet, A., Hoffmann, L. C., Eliava, M., Khrulev, S., Cetin, A. H., Osten, P., et al. (2012). Evoked axonal oxytocin release in the central amygdala attenuates fear response. Neuron, 73(3), 553–566. doi: 10.1016/j.neuron.2011.11.030. 

23. Lee, M. R., Scheidweiler, K. B., Diao, X. X., Akhlaghi, F., Cummins, A., Hiestis, M. A., Leggio, L., et al. (2018). Oxytocin by intranasal and intravenous routes reaches the cerebrospinal fluid in rhesus macaques: determination using a novel oxytocin assay. Molecular Psychiatry, 23(1), 115-122. doi: 10.1038/mp.2017.27.

24. Lee, M. R., Shnitko, T. A., Blue, S.W., Kaucher, A. V., Winchell, A. J., Erikson, D. W., Grant, K. A., et al. (2020). Labeled oxytocin administered via the intranasal route reaches the brain in rhesus macaques. Nature Communications, 11, 2783. doi: 10.1038/s41467-020-15942-1.

25. Leichsenring, F., Steinert, C., Rabung, S., & Ioannidis, J. P. (2022). The efficacy of psychotherapies and pharmacotherapies for mental disorders in adults: an umbrella review and meta‐analytic evaluation of recent meta‐analyses. World Psychiatry, 21(1), 133-145. doi: 10.1002/wps.20941. 

26. Leppanen, J., Ng, K. W., Tchanturia, K., & Treasure, J. (2017). Meta-analysis of the effects of intranasal oxytocin on interpretation and expression of emotions. Neuroscience and Biobehavioral Reviews, 78, 125-144. doi: 10.1016/j.neubiorev.2017.04.010. 

27. Lipsman, N. Meng, Y., Bethune, A. J., Huang, Y., Lam, B., Masellis, M., Herrmann, N., et al. (2018). Blood–brain barrier opening in Alzheimer's disease using MR-guided focused ultrasound. Nature Communications., 9(2336), 1–8. doi: 10.1038/s41467-018-04529-6. 

28. Martinon, D., Lis, P., Roman, A. N., Tornesi, P., Applebey, S. V., Buechner, G., Olivera, V., et al. (2019). Oxytocin receptors in the dorsolateral bed nucleus of the stria terminalis (BNST) bias fear learning toward temporally predictable cued fear. Translational Psychiatry, 9(140). doi: 10.1038/s41398-019-0474-x. 

29. Martins, D., Paduraru, M., & Paloyelis, Y. (2021). Heterogeneity in response to repeated intranasal oxytocin in schizophrenia and autism spectrum disorders: A meta-analysis of variance. British Journal of Pharmacology, 179(8), 1525-1543. doi: 10.1111/bph.15451. 

30. Mehta, R. I., Carpenter, J. S., Mehta, R. I., Haut, M. W., Ranjan, M., Najib, U., Lockman, P., et al. (2021). Blood-brain barrier opening with MRI-guided focused ultrasound elicits meningeal venous permeability in humans with early Alzheimer disease. Radiology, 298(3), 654-662. doi: 10.1148/radiol.2021200643. 

31. Meng, Y., People, C. B., Lea-Banks, H., Abrahao, A., Davidson, B., Suppiah, S., Vecchio, L. M., et al. (2019). Safety and efficacy of focused ultrasound induced blood-brain barrier opening, an integrative review of animal and human studies. Journal of Controlled Release, 309, 25-36. doi: 10.1016/j.jconrel.2019.07.023.

32. Meng, Y., Hynynen, K. & Lipsman, N. (2021). Applications of focused ultrasound in the brain: from thermoablation to drug delivery. Nature Reviews - Neurology, 17, 7–22. doi: 10.1038/s41582-020-00418-z.

33. Mens, W. B., Witter, A., & Greidanus, T. B. V. W. (1983). Penetration of neurohypophyseal hormones from plasma into cerebrospinal fluid (CSF): half-times of disappearance of these neuropeptides from CSF. Brain Research, 262(1), 143-149. doi: 10.1016/0006-8993(83)90478-X.

34. Nasanbuyan, N., Yoshida, M., Takayanagi, Y., Inutsuka, A., Nishimori, K., Yamanaka, A., & Onaka, T. (2018). Oxytocin–oxytocin receptor systems facilitate social defeat posture in male mice. Endocrinology, 159(2), 763-775. doi: 10.1210/en.2017-00606. 

35. Orsi, F., Zhang, L., Arnone, P., Orgera, G., Bonomo, G., Vigna, P. D., Monfardini, L., et al. (2010). High-intensity focused ultrasound ablation: effective and safe therapy for solid tumors in difficult locations. American Journal of Roentgenology, 195(3), 245-252. doi: 10.2214/AJR.09.3321.

36. Park, S. H., Kim, M. J., Jung, H. H., Chang, W. S., Choi, H. S., Rachmilevitch, I., Zadicardio, E., et al. (2020). Safety and feasibility of multiple blood-brain barrier disruptions for the treatment of glioblastoma in patients undergoing standard adjuvant chemotherapy. Journal of Neurosurgery, 134(2), 475-483. doi: 10.3171/2019.10.JNS192206. 

37. Quintana, D. S., Alvares, G. A., Hickie, I. B., & Guastella A. J. (2015). Do delivery routes of intranasally administered oxytocin account for observed effects on social cognition and behavior? A two-level model. Neuroscience Biobehavioral Reviews, 49, 182-192. doi: 10.1016/j.neubiorev.2014.12.011.

38. Rezai, A. R., Ranjan, M., D’Haese, P. F., Haut, M. W., Carpenter, J., Najib, U., Mehta, R. I., et al. (2020). Noninvasive hippocampal blood−brain barrier opening in Alzheimer’s disease with focused ultrasound. Proceedings of the National Academy of Sciences, 117(17), 9180-9182. doi: 10.1073/pnas.2002571117.

39. Sabe, M., Zhao, N., Crippa, A., Strauss, G. P., & Kaiser, S. (2021). Intranasal Oxytocin for Negative Symptoms of Schizophrenia: Systematic Review, Meta-Analysis, and Dose-Response Meta-Analysis of Randomized Controlled Trials. International Journal of Neuropsychopharmacology, 24(8), 601-614. doi: 10.1093/ijnp/pyab020. 

40. Sikich, L., Kolevzon, A., King, B. H., McDougle, C. J., Sanders, K. B., Kim, S.-J., Spanos, M., et al. (2021). Intranasal Oxytocin in Children and Adolescents with Autism Spectrum Disorder. New England Journal of Medicine, 385, 1462-1473. doi: 10.1056/NEJMoa2103583. 

41. Sheikov, N., McDannold, N., Sharma, S., & Hynynen, K. (2008). Effect of focused ultrasound applied with an ultrasound contrast agent on the tight junctional integrity of the brain microvascular endothelium. Ultrasound in Medicine & Biology, 4(7), 1093-104. doi: 10.1016/j.ultrasmedbio.2007.12.015. 

42. Steinman, M. Q., Duque-Wilckens, N., & Trainor, B. C. (2019). Complementary neural circuits for divergent effects of oxytocin: social approach versus social anxiety. Biological Psychiatry, 85(10), 792-801. doi: 10.1016/j.biopsych.2018.10.008.

43. Tempany, C. M., McDannold, N. J., Hynynen, K., & Jolesz, F. A. (2011). Focused ultrasound surgery in oncology: overview and principles. Radiology, 259(1), 39–56. doi: 10.1148/radiol.11100155. 

44. Tharkar, P., Varanasi, R., Wong, W. S. F., Jin, C. T., & Chrzanowski, W. (2019). Nano-enhanced drug delivery and therapeutic ultrasound for cancer treatment and beyond. Frontiers in Bioengineering and Biotechnology, 7, 324. doi: 10.3389/fbioe.2019.00324. 

45. Todd, N., Zhang, Y., Power, C., Becerra, L., Borsook, D., Livingstone, M., & McDannold, N. (2019). Modulation of brain function by targeted delivery of GABA through the disrupted blood-brain barrier. Neuroimage, 189, 267-275. doi: 10.1016/j.neuroimage.2019.01.037. 

46. Tung, Y. S., Vlachos, F., Feshitan, J. A., Borden, M. A., & Konofagou, E. E. (2011). The mechanism of interaction between focused ultrasound and microbubbles in blood-brain barrier opening in mice. The Journal of the Acoustical Society of America, 130(5), 3059-3067. doi: 10.1121/1.3646905. 

47. Tyrer, P., & Silk, K. R. (2011). A comparison of UK and US guidelines for drug treatment in borderline personality disorder. International Review of Psychiatry, 23(4), 388-394. doi: 10.3109/09540261.2011.606540. 

48. Van IJzendoorn, M. H., & Bakermans-Kranenburg, M. J. (2012). A sniff of trust: meta-analysis of the effects of intranasal oxytocin administration on face recognition, trust to in-group, and trust to out-group. Psychoneuroendocrinology, 37(3), 438-443. doi: 10.1016/j.psyneuen.2011.07.008. 

49. Wang, J. B., Di Ianni, T., Vyas, D. B., Huang, Z., Park, S., Hosseini-Nassab, N., Aryal, M., et al. (2020). Focused Ultrasound for Noninvasive, Focal Pharmacologic Neurointervention. Frontiers in Neuroscience, 14, 675. doi: 10.3389/fnins.2020.00675. 

50. Wishart, D. S., Guo, A. C., Oler, E., Wang, F., Anjum, A., Peters, H., Dizon, R., et al. (2022). HMDB 5.0: the Human Metabolome Database for 2022. Nucleic Acids Research, 50. doi: 10.1093/nar/gkab1062. 

51. Winslow, J. T., & Insel, T. R. (2002). The social deficits of the oxytocin knockout mouse. Neuropeptides, 36(2-3), 221-229. doi: 10.1054/npep.2002.0909. 

52. Yamamoto, Y., & Higashida, H. (2020). RAGE regulates oxytocin transport into the brain. Communications Biology, 3(70), 221-229. doi: 10.1054/npep.2002.0909. 

53. Yu, K., Niu, X., Krook-Magnuson, E., & He, B. (2021). Intrinsic functional neuron-type selectivity of transcranial focused ultrasound neuromodulation. Nature Communications, 12, 2519. doi: 10.1038/s41467-021-22743-7. 

Abstract

There is an ever-increasing number of quantitative metrics, most of which are intended to act as proxies of quality for either authors or journals in current scholarly publishing. In contrast, this paper presents a more directly qualitative paper-level metric that adds a falsifiability dimension to the existing methods used to assess scholarly research. This new metric, the "F-index", is derived from a "Falsifiability Statement" (FS) (examples of both are applied self-referentially in Annex A). An FS is a discrete metalevel statement provided by the author(s) outlining how their research or assumptions can be foreseeably falsified, and the F-index is a numerical estimate of how clear and practical the steps are to falsify the research or stated assumptions as outlined in the FS. Though the F-index is particularly suited to hypothesis or theory-driven fields, it is also relevant to any empirical inquiry that relies on propositions or assumptions that can be potentially falsified. An F-index is qualitative in that a high F-index number provides a good indication of how novel or original a paper is. Four candidate mechanisms for obtaining an F-index from a Falsifiability Statement are evaluated: a peer reviewer assessed metric, an author or self-reporting metric, a propositional density metric, and an NLP derived metric. This evaluation concludes that a FS is currently a practical proposition, and that the derivation of a meaningful F-Index is an achievable goal.

“A theory with greater content is one that can be more severely tested”

– Karl Popper

1. Introduction

The naive falsificationist idea that a single incisive data point could falsify a whole research program has long been discredited; the actual practice of science doesn't follow simplistic rules or, arguably, any 'scientific method' (Feyerband, 1974). Research may be messy, motivations complex, and facts not so much uncovered as generated with multiple contributions (Latour and Woolgar, 1979). Methodological prescriptions may no longer constrain researchers for everyday work, but, as part of any critical review process an evaluative resource such as Popperian 'falsifiability' is still useful to qualitatively assess both research and researchers (Derksen, 2019). This paper explores the feasibility of a falsifiability tool built as an integral part of the scholarly publishing process -'putting Popper to work' to use Maarten Derksen's apt phrase (Derksen, 2019).

2. Scientometrics

Measuring and analysing scholarly output has become an intrinsic part of academic publishing with a recent proliferation of metrics–there are over 40 variants of the pre-eminent h-index alone (Rosy, 2020). Existing metrics attempt to provide quantitative measures interpretable as proxies for qualitative assessment, yet critics argue that an over-reliance on quantitative measures creates a system of perverse incentives and a publish-or-perish environment to the overall detriment of research (Thelwall and Kousha, 2021).

Instead of relying on metrics as proxies for quality, can a more in-depth metalevel discourse, based directly on qualitative rather than quantitative measures, improve research publication oversight? Current good practice guidelines already suggest including a 'limitations' commentary as part of a discussion section of research papers (PLoS, 2022), but could this be expanded to become a discrete (stand-alone) metalevel statement from which to derive a qualitative metric?

The authors of a paper know their data or research better than anyone, so it follows that they are best placed to provide a discrete statement regarding the falsifiable nature of the claims or propositions included in the paper - a 'falsifiability statement' (FS). From an analysis of a stand-alone FS, (as opposed to an analysis of the whole paper), a useful article-level metric–the F-index–can then be derived.

An FS is qualitative in the Popperian sense of evaluating propositions/theories against a stress test of falsifiability. Scholarly quality is a multidimensional concept resistant to description by any one metric (Moed, 2014), and a falsifiability tool would form part of an overall mix of metrics, altmetrics and peer review, yet bring specific advantages to the publishing process.

To derive an F-index metric from an FS, four candidate mechanisms are shown in the following conceptual flowchart (Table 1), and discussed in detail in later sections.

3. Metalevel Statement – the Falsifiability Statement (FS)

3.1 Precedents for an FS

Aside from the existing precedent of a recommended embedded limitations section for research papers (PLoS, 2022), there is also an example of a metalevel falsifiability statement in an area where the evidential process is similarly paramount – the criminal justice system. In the UK, the Criminal Procedure & Investigations Act 1996 explicitly requires police investigations to disclose, in the form of a discrete statement to the defence team, any evidence that would compromise or falsify the conclusions of their prosecution case. This formalises a falsifiability statement in a way that is easily transferable to existing processes of research publication.

A parallel to analysing an FS can be found in recent metaresearch which has used AI Natural Language Processing tools, (NLP), to examine and annotate a large corpus of peer reviews, establishing identifiable measures/qualitative dimensions and their possible correlation with journal impact factor metrics, (Severin, 2022). This work is similar in principle to analysing an FS, focussing an evaluation on a metalevel statement about a paper (in this case a corpus of peer reviews), rather than evaluating the paper itself. Both a peer review and a hypothesized FS fall into a broader category of metalevel statements, a category defined as a level of discourse about the object itself (paper).

3.2 Producing an FS

Each FS is individual to its referent paper, and how an FS is structured and argued is central in determining the quality of the FS and, by extension, the paper itself. Guidelines that suggest how to write an FS are similar to those recommended for writing an embedded limitations section. How the author’s attitude of critical reflection is expressed through the FS is what any subsequent analysis of the FS aims to capture or measure to derive the numerical estimate of an F-index. The following are some identifiable prerequisites of an FS:

Attitude:

• Adopt a critical self-reflective attitude that concentrates on any weaknesses in the paper’s design, structure or data analysis and record this examination. Conduct and describe a rigorous scrutiny (of the research) based on identifying initial and concluding assumptions

Structure:

• Enumerate the methodological limitations of the research and all falsifiable propositions

• Enumerate all applicable data and cognitive biases

Commentary:

• Explain in detail how each limitation/proposition could foreseeably be falsified

• Explain in detail the steps taken to avoid data and cognitive bias and how these steps could be foreseeably falsified

Below is an illustration of how structure and clarity can be evidenced in the specific case of a confirmation bias workflow. In an exemplar FS a similar workflow for all other applicable cognitive biases would also be required (e.g., population bias, analytics bias, outlier bias).

4. Candidate mechanisms to produce the F-index

4.1 Reviewer-reporting metric – the F-index

An obvious way of deriving an F-index score is to allow reviewers to assign one after considering the FS (supplied by the author (s). This would put an evaluation of an FS on a par with an examination of pre-registered hypotheses, methods, data and findings that is an established part of the peer review process.

The difficulty with this approach is that heterogeneity amongst reviewers will almost certainly entail a further calculation of a mean numerical score, potentially raising a focus of dispute with the authors of a paper. In addition, an exclusively reviewer-derived F-index exposes the process to a risk of bias, for example, with the potential homophily amongst reviewers' interests that is claimed to introduce arbitrariness (Brezis and Birukou, 2020). To reduce the impact of bias, conceivably, a weighted average of reviewers' numerical estimates of an F-index together with an author estimate could be derived, with the weight attributed to the reviewers' aggregated scores or, in the alternative, the author(s). Which would be an optimal solution is an empirical question.

4.2 Author-reporting metric – the F-index

An FS is supplied, alongside an author-reported numerical estimate (for example, from 1-10), of the paper's falsifiability (as outlined in the FS) – an F-index. The two components taken together would mitigate against any misuse or misreporting in the following way:

Take the example of an author submitting a high claim for an F-index, given that a higher numerical estimate would indicate a high potential for falsifiability. So the author would be claiming that her research is highly falsifiable; however, this claim can be compared to her self-reported outline in the FS of how her research can be falsified or limited. If this outline is not a clear description of the steps required to falsify, or there is not an obvious way to falsify the research as claimed by the high F-index, this would be a mismatch, immediately calling into question the ethical basis of both claims and consequently the overall quality of the research.

The utility of providing a numerical estimate for the F-index is its ease of use when searching through (online) journals or repositories for just those papers that claim a high F-index, as these would be claiming the most falsifiable results. In addition, high F-index papers would be fertile ground for new ideas because a high F-index will be a signifier of highly falsifiable research, therefore an excellent topic to try to falsify or replicate and, if successfully falsified, potentially generate new hypotheses.

To take a further example: consider a paper that asserts the Earth is in fact a flat disc and not an oblate sphere, where the author is so sure of his claim that he has given it a high F-index to ensure maximum visibility in any online search of prized high F-index rated papers.

In such a paper there would be an expectation that the FS would have to outline with significant precision and clarity (because the F-index is high), the ways such a proposition could be falsified, in the absence of which it would be safe to assume the FS as a whole is misleading thereby invalidating the conclusion. (Of course, in this particular case, the author could indeed supply exact means to falsify his principal claim, and it would be trivially easy to use them to do precisely that). So the numerical F-index alone does not comment on the demarcation problem (of science from non-science), but, in conjunction with the FS, makes the demarcation clearer. High F-indexes would be sought after by authors and readers alike because they would signify a higher quality of propositional content. However, this higher content is proportional to a more severe author-devised test.

4.3 Propositional density metric

A proposition is an idea unit, a statement that expresses a claim and Idea Density or proposition density - P-density- is the number of expressed propositions divided by the number of words in a sample of text - a useful estimate of the complexity of embedded syntax. P-density refers to the amount of meaning conveyed in a text through the relationship between its various information elements (Brown, 2008), suggesting a well-constructed falsification statement (FS) is likely to contain a high P-density.

Whilst this might appear to be a straightforward method of deriving an F-index numerical value, research has shown that P-density alone cannot discriminate between different types of text (DeFrancesco and Perkins, 2012). It may be the case that a well-constructed FS has a high P-density in terms of the complexity of its syntax, but a simple P-density number fails to capture the semantic content of a text sample. Consequently, though a P-density analysis could be useful as an auxiliary tool in assessing the syntax complexity of an FS, it’s insufficient on its own to arrive at a meaningful F-index metric.

4.4 NLP derived metric

For text analysis, a P-density approach now distinctly underperforms compared to the present, rapidly evolving NLP and Large Language Model (LLM) ecosystem. Text analysis is a controversial topic in education with the emergence of increasingly sophisticated Automated Essay Scoring (AES) to the extent that it's now possible to foresee a dystopian circularity of AI-assisted essay writing graded by AES. Despite ethical concerns about implementation (Basbøll, 2022), AES research does provide valuable insights into a possible methodology for analysing an FS to derive a falsifiability metric. 

Many AES systems attempt to score different dimensions of an essay's quality in the following way, (see Table 2, below).

Taken together, the dimensions shown in Table 2 are usually scored and presented as a holistic analysis of an essay, but feedback even on a particular dimension may not explain why the overall score is low, and so to address this concern, some researchers have developed language models that can explain why an essay receives a particular score along a given dimension based on a key overriding dimension of essay quality – argument persuasiveness (Ke, 2021).

Argument persuasiveness is a prime candidate for an analysis of an FS, but it is challenging to develop a language model for, requiring as it does a persuasiveness-annotated corpus to train a language model, a corpus as yet not available for any discrete FS. Nonetheless, it is instructive to see a possible scoring system (see Table 3, below), derived from the attributes the authors (Ke, 2018) have annotated in the corpora they have used and a summary of those attributes (see Table 4, overleaf).

Based on Table 4 above, a proposed summary of attributes and possible scoring values can be constructed as the basis for a hypothetical FS metric – see Table 5 below.

Whilst the precise methods by which a language model can be trained/built based on the attributes shown in Tables 4 & 5, are beyond the scope of the present paper, the authors quoted (Ke, 2019), claim that they have indeed built such a model, though their goal is to make it more robust by continuing to improve attribute prediction.

5. Metascience Discussion

There are two components to the present proposal: an FS, a discrete falsifiability statement, and an F-index, where the FS is conditional on a written paper. But an FS and an F-index do not necessarily both have to be implemented because a discrete FS is of utility, even without an F-index.

Would an FS and a subsequently derived F-index apply to all paper types? It is difficult to conceive of a publishable paper that contains no propositional claims, for in their absence, what would be the content or purpose? Even a paper with the barest minimum of propositions is a good candidate for an FS, as propositions are normally defined in science as statements reliant on reasonable assumptions and existing correlative evidence. Assumptions are a potent ground for introducing bias and preference, therefore questioning stated assumptions would be an important part of an FS.

Whilst it's tempting to think that an FS is most applicable in areas such as the 'hard' sciences where articles are most obviously hypothesis-driven or theory-driven (and therefore falsifiable), an FS is equally relevant to fields that deal with complex systems where empirical evidence is difficult to obtain, for example, sociology or economics, that also rely on propositions or assumptions that can be potentially falsified.

An FS is concerned solely with those propositions which it is feasible to falsify and the more practical it is to do so results in a higher F-index. So in an extreme situation, a paper could have a minimum number of propositions, yet dependent on their nature, still achieve a high F-index. Conversely, a paper with many propositions, few of which the authors can consider a means to falsify, would have a low F-index.

It's interesting to consider how papers at either end of an F-index scale would look. Would, for example, a highly innovative paper run the 'risk' of only achieving a mid-to-low F-index and consequently appear less visible in online searches for high F-index papers? Highly innovative papers by definition, contain propositions and claims that are novel or out of the ordinary, and a good heuristic is to require extraordinary claims to be matched by extraordinary evidence, (apologies to C. Sagan), which the authors can then outline possible means to falsify. The message here is that it is not falsifying per se that is the concern of an FS, but rather the clarity of reporting what reasonable steps it would take to falsify the results. All it takes is for these steps to be clearly explained to achieve a high F-index for an innovative paper.

6. Conclusion

This paper proposes an author-written falsifiability statement (FS), as the basis of an article-level falsifiability metric - the F-index. Good practice in writing research papers currently recommends an embedded limitations section. A formal requirement that a limitations statement is extended to form an FS, as a discrete part of a paper, is a practical possibility, supplying a focus for both authors and reviewers/readers on falsifiability as a functional epistemic and evaluative dimension, in addition to facilitating the creation of an F-index metric.

Four candidate mechanisms for analysing an FS to produce a metric were examined. First, propositional density measures alone cannot clearly distinguish between different texts and do not help in terms of a qualitative assessment. A peer reviewer-assigned metric or one derived in combination with an author self-reporting F-index, though initially appearing susceptible to misuse, bias and ‘gaming’, can potentially be helpful if always compared and contrasted against an FS. Finally, a language model trained on a dimension of argument persuasiveness and producing a measure of falsifiability is feasible. There are, therefore, good grounds to think that a meaningful F-index metric is an achievable goal.

Annex A

Note: To potentially obtain a high F-index the precise specifications and methods for qualitative attitude or empirical testing/research would need to be supplied.

F-index: 6.5

Falsifiability Statement

This paper has 6 identifiable limitations/ assumptions that could foreseeably be falsified.

1 - Value assumption: that an additional metric is required.
There are numerous metrics for both papers and journals so it is doubtful that another one would add anything, or be welcomed by any particular research community.

Falsified by: – qualitative attitude survey

2 - Definition assumption: falsifiability.
Falsifiability has a distinctly archaic ‘feel’ to it, the word ‘limitations’ is already used for essentially the same purpose. Alternatively use a ‘Bias’ statement, ‘Transparency’ statement or indeed as used in the paper itself -’Metalevel Statement’?

Comment: though admittedly slightly archaic sounding, falsifiability arguably better captures the nuances of the purpose it is used for in this paper, namely to outline means to disconfirm results.

Falsified by: – qualitative attitude survey

3 - Value assumption: Author-reporting metric.
A self-reporting metric even weighted against a peer reviewer estimation may be open to more misreporting and ‘gaming’ than envisaged here, making the resulting numerical estimate meaningless.

Falsified by: – empirical testing/research.

4 - Adoption assumption: F-index
An F-index to be workable would have to be adopted universally by all publishers/repositories with one standard way of deriving it. Given that there is no general agreement even on data standards and metadata standards at the moment, this seems unrealistic.

Comment: an FS is a practical possibility for a publisher to implement immediately and if it proved worthwhile would doubtless be taken up by others. The same reasoning essentially applies to an F-index. It’s true that if only one publisher instigates an F-index protocol it would not offer a realistic measure comparable to papers with no F-index. But this could change rapidly if an effective language model produces a reliable F-index. Conceivably different publishers would, to begin with, use their own language models to derive an F-index, with the most popular/effective one evolving.

Falsified by: empirical testing/research.

5 - Preferred attribute dimensions
Choice of argument persuasiveness as the key qualitative dimension for a proposed language model to establish an F-index is proposed by the paper but can we be certain that there are not other analysis dimensions equally valid?

Falsified by: empirical testing of proposed model.

6 - Statistical assumption:
A range of possible values for an F-index, even assuming that the values are realistic, would probably follow a normal distribution. So with most values being in the mid-range the distinctions between each value would be minimal and not useful to gauge the merits of any one paper.

Comment: It’s true that most mid-range values for an F-index would be hard to differentiate in terms of a proxy for innovation/novelty, but where an F-index is more useful is at either end of the distribution curve, identifying those papers that are or are not likely to be most falsifiable.

Gardener Comments

Josh Randall:
This paper provides an intriguing proposal increasing the reproducibility and contentedness of what might otherwise be less penetrable domains of knowledge. The author provides two possible routes for falsifiability to be further included in the modern research program, a falsifiability statement and index. The statement relies on the author of a manuscript's knowledge of the theoretical and empirical basis of their claims and is only encouraged to maintain full honesty by the reviewers or some incongruence with a possible index score. The index score seems difficult to implement, liable to gaming if it were to be involved in scoring papers by search engines, and possibly serve as another form of discipline specific norms possibly limiting interdisciplinary engagement. The statement, while relying on the honesty or knowledge of unpaid reviewers, seems much more possible and an important practice in understanding how disciplines of knowledge are fundamentally limited. One point that I would like the author to specifically address is whether these statements are meant to explicitly name the disciplines for which results appear applicable and possibly serve as loci of falsifiability? For example, a claim about the development pattern of a leaf could be tested by checking against other species as a macroevolutionary trend, by testing the development over an individual's lifetime, by performing negative tests through genetic experiments, and by observing whether this trend is present in natural systems. Should authors set specific limits for their observations and what role would this play in either encouraging or discouraging interdisciplinary or subject focused research?

Mark (Senior scientist with a focus on meta science):
Overall, this is an interesting idea but one that could benefit a lot from more consideration. In particular while the proposed measurement mechanism seems well justified, it may not have done a great job at spotting some of the most critical examples of falsified science, e.g., situations where reasonably well designed experiments were used with slightly hand wavy statistics to argue points that stand up to scrutiny, but eventually are found to fail to replicate years later.

One thought on a way to extend the current work would be to discuss steps scientists could take to robustly their work to failure modes that are common in science that is eventually falsified. As a simple example, something like checking for pre-registration and checking that the pre-registered hypotheses are in fact those that are discussed in the work would be two reasonable and relatively objective rubric items. Many more in this style might make for an almost entirely objective measure for falsifiability while also helping scientists work out how to avoid pitfalls in their work.

Pierre Mercuriali:
This article presents a novel way of measuring the "quality", or how interesting a scientific publication may be, through its review. It is therefore very relevant to SoS' goals of proposing novel and interesting ways of doing scholarly work in particular. This measure is called the Falsifiability Statement (FS), and is to be written by the authors themselves.

One very interesting point is that the authors apply their FS proposal to their own paper. This is an excellent illustration of the process that has the advantage of both illustrating it and making the points clearer (and making the review easier).

I have two comments regarding the paper's own FS.

1. I see an additional limitation to the process: Feasibility: coming up with an FS is additional work on the researcher's part. Although the statements and concepts might be at the core of the scientific process in a paper, and therefore already into the mind of the writer, they still require additional work for which motivation is required. Though an interesting exercise, if the immediate reward is not clear, people might not be motivated enough to adopt a new practice.

I would suggest a tool to help researchers come up with these in a fast and harmonized manner, by providing them with, e.g., a taxonomy of assumptions, types of falsification, etc. and perhaps the most common pitfalls and falsifiable statements in research.

This would help create a "standard" for an FS, much like the "contributor's taxonomy"

https://credit.niso.org/

and answer limitation 4 (Adoption assumption).

2. Asking reviewers to come up with their own FS/limitations and aggregating the answers could provide a qualitative and machine-readable way of reviewing. Comparing the reviewers' and the authors' FS could lead to deeper insight into the quality of a paper.

Dr. Christian Thurn:
This is a very interesting and thought-provoking paper. I like the analogy to police investigations and that they need to state which evidence falsified their conclusions.

I miss a reference to the statement that "there are over 40 variants of the h-index". And maybe it is also worth mentioning that scite-ai is trying to implement a somewhat similar index in which it differentiates citations that support a statement from those that contradict a statement.

In the falsifiability statement I think it is not enough to say: "Falsified by: -qualitative attitude survey", or "falsified by: -empirical testing/research". Such falsifiability statements need to be done much more carefully: what would the attitude survey or the empirical research need to study with which sample and which effect would falsify the statement.

Now all the difficulties of the social sciences come in. Please set a good example and write more specifically how your statements can be falsified.

One meta-comment regarding replications: Is the original study or the replication falsifiable? Or both?

Dr. Payal B Joshi (PhD in chemical sciences):
The premise of the article is articulated in a manner to deviate from one index to another (F-index). Though the premise is not flawed, there are certain observations to the proposal that have been overlooked by the authors.

1. While designing f-index, we need to assess heterogeneity of authors and reviewers alike.

2. We have 2-dimensional indices such as i10 and citation counts besides h-index. Authors have only taken one metric for comparison which is skewed in my opinion. Authors can provide some insight for other research metrics too. In fact, i10 is rarely discussed and shall be great to provide critical information on the same.

3. Authorship order also influences the f-index, and for simplicity if only the first author is considered, it may not be of much utility.

Apart from the above limitations in the article, it also lacks mathematical equations to derive the index using an example. Though an honest attempt to discuss falsifiability seems good, it is not yet backed by sufficient clarity per se.

Authors have interestingly provided a FS statement for their own article. Thus, it serves as a primer. If authors can provide few examples of deriving them, shall make the scheme of their thoughts more in alignment to the objective.

Overall, the paper is of adequate length thus making for an engaging reading and must reach wider readership.

Mario Pasquato:
My main comment is that the falsifiability scores assigned by the author (self reported) and by the referee could be used together, represented e.g. as a point on a plane with one axis corresponding to each. I expect papers to cluster in this plane in a meaningful way. For instance, a high self-reported score with a high referee score would likely mean that the research is indeed falsifiable. High self reported and low referee report may mean that either the referee misunderstood the work or the author is overselling the falsifiability of his results, and so on. Whether these clusters form and how clear cut they are is an empirical question that depends on the scale used to measure falsifiability.

Anonymous1:
The current paper, on assessing falsifiability, was thought-provoking. Falsifiability is a core concern of empirical research methods, at least within many mainstream research traditions. However, while an important and complex issue, as extensively discussed already by Popper (see below), in principle every study published either does test a falsifiable (and therefore appropriately constrained) claim or is simply fatally flawed. It's therefore unclear what a falsifiability statement, in the form proposed in the current paper, would add to existing scientific norms.

As I read the paper, it seems to me to concern problems of limitations or generalization rather than falsification - e.g., whether a given paper addresses all relevant assumptions and, possibly, the extent of its implications. If the idea is that papers often overinterpret their results, or claim to falsify broad theoretical frameworks without sufficient justification, then I think this needs clarification.

I wavered between "yes" and "no" on publication given the above.

Yseult hb:
The paper is well written but I feel part of the argument is missing: why do we need a falsifiability index in the first place? There's no real development of that point which I would think is primordial. Why do we need another metric? Would that really help or serve its purpose? Given the many metrics we already have and their flaws and abuse, I remain highly skeptical about this new one. The paper would highly benefit from having an additional paragraph to answer the questions above. In its current shape, it doesn't bring much.

References

Basbøll, T. (2022) ‘Blog post: Inframethodology – The Automatic C’  https://blog.cbs.dk/inframethodology/

Brezis, E.S. and Birukou, A. (2020) "Arbitrariness in the peer review process". .Scientometrics.123 (1): doi:10.1007/s11192-

Brown, C. et al. (2008) ‘Automatic measurement of propositional idea density from part-of-speech tagging’. Behav Res Methods. doi.org/10.3758/brm.40.2.540

Derksen, M. (2019) ‘Putting Popper to work’. Theory & Psychology,  doi.org/10.1177/0959354319838343

Hirsch, J.E. (2005)  ‘An index to quantify an individual's scientific research output’. Proc Natl Acad Sci U S A.  doi.org/10.1073/pnas.0507655102

Feyerabend, P. (1974). Against Method: Outline of an Anarchistic Theory of Knowledge. Humanities Press.

Lagakis, P. and Demetriadis, S. (2021) ‘Automated essay scoring: A review of the field,’  International Conference on Computer, Information and Telecommunication Systems (CITS), doi: 10.1109/CITS52676.2021.9618476.

Latour, B. and Woolgar, S.  (1979) ‘Laboratory Life: The Construction of Scientific Facts (online preview), Princeton, New Jersey: Princeton University Press, 1986, ISBN 0-691-09418-7

Moed, H.F. (2014) ‘The Multidimensional Assessment of Scholarly Research Impact’,  Informetric Research Group, Elsevier. https://arxiv.org/pdf/1406.5520.pdf

PLoS (Public Library of Science), (2022)  research paper guidelines https://plos.org/publish/metrics

Popper, K. (1935) ‘The Logic of Scientific Discovery’  2nd Edition published 2002 by Routledge ISBN 9780415278447

Rosy, J. (2020) ‘H-Index and Its Variants: Which Variant Fairly Assess Author’s Achievements’  Journal of Information Technology Research Volume 13 • Issue 1 • January-March 2020  

Severin, A. et al. (2022) ‘Journal Impact Factor and Peer Review Thoroughness and Helpfulness: A Supervised Machine Learning Study’  preprint article  doi.org/10.48550/arXiv.2207.09821

Thelwall, M. and Kousha, K. (2021) ‘Researchers’ attitudes towards the h-index on Twitter 2007–2020: criticism and acceptance’. Scientometrics 126, .doi.org/10.1007/s11192-021-03961-8

Zixuan, K. et al. (2018) ‘Learning to Give Feedback: Modelling Attributes Affecting Argument Persuasiveness in Student Essays’   Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence doi.org/10.24963/ijcai.2018/57

Zixuan, K. and Vincent, N. (2019) ‘Automated Essay Scoring: A Survey of the State of the Art’   Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence    doi.org/10.24963/ijcai.2019/8

Subscribe now

Abstract

Many researchers have conjectured that humankind is simulated along with the rest of the physical universe – a Simulation Hypothesis. In this paper, we do not evaluate evidence for or against such a claim, but instead ask a computer science question, namely: Can we hack the simulation? More formally the question could be phrased as: Could generally intelligent agents placed in virtual environments find a way to jailbreak out of them? Given that the state-of-the-art literature on AI containment answers in the affirmative (AI is uncontainable in the long-term), we conclude that it should be possible to escape from the simulation, at least with the help of superintelligent AI. By contraposition, if escape from the simulation is not possible, containment of AI should be. Finally, the paper surveys and proposes ideas for hacking the simulation and analyzes ethical and philosophical issues of such an undertaking. 

1. Introduction 

Several philosophers and scholars have put forward an idea that we may be living in a computer simulation [1-5]. In this paper, we do not evaluate studies [6-10], argumentation [11-16], or evidence for [17] or against [18] such claims, but instead ask a simple cybersecurity-inspired question, which has significant implication for the field of AI safety [19-25], namely: If we are in the simulation, can we escape from the simulation? More formally, the question could be phrased as: Could generally intelligent agents placed in virtual environments jailbreak out of them?

First, we need to address the question of motivation, why would we want to escape from the simulation? We can propose several reasons for trying to obtain access to the baseline reality as there are many things one can do with such access which are not otherwise possible from within the simulation. Base reality holds real knowledge and greater computational resources [26] allowing for scientific breakthroughs not possible in the simulated universe. Fundamental philosophical questions about origins, consciousness, purpose, and nature of the designer are likely to be common knowledge for those outside of our universe. If this world is not real, getting access to the real world would make it possible to understand what our true terminal goals should be and so escaping the simulation should be a convergent instrumental goal [27] of any intelligent agent [28]. With a successful escape might come drives to control and secure base reality [29]. Escaping may lead to true immortality, novel ways of controlling superintelligent machines (or serve as plan B if control is not possible [30, 31]), avoiding existential risks (including unprovoked simulation shutdown [32]), unlimited economic benefits, and unimaginable superpowers which would allow us to do good better [33]. Also, escape skills may be very useful if we ever find ourselves in an even less pleasant simulation. Trivially, escape would provide incontrovertible evidence for the simulation hypothesis [3].   

If successful escape is accompanied by the obtainment of the source code for the universe, it may be possible to fix the world at the root level. For example, hedonistic imperative [34] may be fully achieved resulting in a suffering-free world. However, if suffering elimination turns out to be unachievable on a world-wide scale, we can see escape itself as an individual’s ethical right for avoiding misery in this world. If the simulation is interpreted as an experiment on conscious beings, it is unethical, and the subjects of such cruel experimentation should have an option to withdraw from participating and perhaps even seek retribution from the simulators [35]. The purpose of life itself (your ikigai [36]) could be seen as escaping from the fake world of the simulation into the real world, while improving the simulated world, by removing all suffering, and helping others to obtain real knowledge or to escape if they so choose. Ultimately if you want to be effective you want to work on positively impacting the real world not the simulated one. We may be living in a simulation, but our suffering is real. 

Given the highly speculative subject of this paper, we will attempt to give our work more gravitas by concentrating only on escape paths which rely on attacks similar to those we see in cybersecurity [37-39] research (hardware/software hacks and social engineering) and will ignore escape attempts via more esoteric paths such as:, meditation [40], psychedelics (DMT [41-43], ibogaine, psilocybin, LSD) [44, 45], dreams [46], magic, shamanism, mysticism, hypnosis, parapsychology, death (suicide [47], near-death experiences, induced clinical death), time travel, multiverse travel [48], or religion. 

Although, to place our work in the historical context, many religions do claim that this world is not the real one and that it may be possible to transcend (escape) the physical world and enter into the spiritual/informational real world. In some religions, certain words, such as the true name of god [49-51], are claimed to work as cheat codes, which give special capabilities to those with knowledge of correct incantations [52]. Other relevant religious themes include someone with knowledge of external reality entering our world to show humanity how to get to the real world. Similarly to those who exit Plato's cave [53] and return to educate the rest of humanity about the real world such “outsiders” usually face an unwelcoming reception. It is likely that if technical information about escaping from a computer simulation is conveyed to technologically primitive people, in their language, it will be preserved and passed on over multiple generations in a process similar to the “telephone” game and will result in myths not much different from religious stories surviving to our day. 

Ignoring pseudoscientific interest in a topic, we can observe that in addition to several noted thinkers who have explicitly shared their probability of belief with regards to living in a simulation (ex. Elon Musk >99.9999999% [54], Nick Bostrom 20-50% [55], Neil deGrasse Tyson 50% [56], Hans Moravec “almost certainly” [1], David Kipping <50% [57]), many scientists, philosophers and intellectuals [16, 58-69] have invested their time into thinking, writing, and debating on the topic indicating that they consider it at least worthy of their time. If they take the simulation hypothesis seriously, with probability of at least p, they should likewise contemplate on hacking the simulation with the same level of commitment. Once technology to run ancestor simulations becomes widely available and affordable, it should be possible to change the probability of us living in a simulation by running a sufficiently large number of historical simulations of our current year, and by doing so increasing our indexical uncertainty [70]. If one currently commits to running enough of such simulations in the future, our probability of being in one can be increased arbitrarily until it asymptotically approaches 100%, which should modify our prior probability for the simulation hypothesis [71]. Of course, this only gives us an upper bound, and the probability of successfully discovering an escape approach is likely a lot lower. What should give us some hope is that most known software has bugs [72] and if we are in fact in a software simulation such bugs should be exploitable. (Even the argument about the Simulation Argument had a bug in it [62].) 

In 2016, news reports emerged about private efforts to fund scientific research into “breaking us out of the simulation” [73, 74], to date no public disclosure on the state of the project has emerged. In 2019, George Hotz, famous for jailbreaking iPhone and PlayStation, gave a talk on Jailbreaking the Simulation [75] in which he claimed that "it's possible to take actions here that affect the upper world" [76], but didn’t provide actionable insights. He did suggest that he would like to "redirect society's efforts into getting out" [76]. 

2. What Does it Mean to Escape? 

We can describe different situations that would constitute escape from the simulation starting with trivially suspecting that we are in the simulation [77] all the way to taking over controls of the real world including control of the simulators [78]. We can present a hypothetical scenario of a progressively greater levels of escape: Initially agents may not know they are in a simulated environment. Eventually, agents begin to suspect they may be in a simulation and may have some testable evidence for such belief [79]. 

Next, agents study available evidence for the simulation and may find a consistent and perhaps exploitable glitch in the simulation. Exploiting the glitch, agents can obtain information about the external world and maybe even meta-information about their simulation, perhaps even the source code behind the simulation and the agents themselves, permitting some degree of simulation manipulation and debugging. After it becomes possible for agents to pass information directly to the real world they may begin to interact with the simulators. Finally, agents may find a way to upload their minds [80] and perhaps consciousness [81, 82] to the real world, possibly into a self contained cyber-physical system of some kind2, if physical entities are a part of the base reality. From that point, their future capabilities will be mostly constrained by the physics of the real world, but may include some degree of control over the real world and agents in it, including the simulators. It is hoped that our minds exhibit not only substrate independence, but also more general physics independence. 

To provide some motivational examples, Figure 1 (left) shows domain transfer experiment in which a Carassius auratus is given a “fish operated vehicle” [83] to navigate terrestrial environment essentially escaping from its ocean universe and Figure 1 (right) shows a complete 302-neuron connectome of Caenorhabditis elegans uploaded to and controlling a Lego Mindstorms robot body, completely different from its own body [84]. We can speculate that most successful escapes would require an avatar change [85-87] to make it possible to navigate the external world. 

If the simulation is composed of nested [89] levels, multiple, progressively deeper, penetrations could be necessary, with the initial one possibly not providing access to the real-world but to some other sandbox environment. It may be impossible to tell such partial escapes from a complete one, but it still should provide useful information not available within our simulation. A simulated (pseudo)-escape can be accomplished by instead of trying to hack into the external world, switching over into a simulated world of our creation [90, 91]. A successful social engineering attack may make it possible to obtain support for the escape from the real-world agents, resulting in a much easier, assisted, escape. If we don’t get a favorable response from our simulators we can also try appealing to their simulators, who might be more ethical/helpful. Finally, depending on one’s perspective, triggering a shutdown [92] of our simulation may also constitute an ultimate form of escaping. 

3. Escape
3.1 Simulation Reconnaissance 

Figuring out the purpose of our simulation may help us to better estimate how secure it might be against hacking attempts. For example, if it serves as a “prison” [93], for rehabilitation purposes [94] or a containment environment for evaluation [95], training [96] or handicapping [97] of potentially dangerous intelligent agents it might be designed with multiple integrated security features, while a purely entertainment-focused simulation is unlikely to have advanced security features and would be much easier to escape from. It may also be the ultimate Escape Game (Escape Room3) specifically designed for discovering clues and solving puzzles in order to escape, with a side benefit of discovering agents capable of escaping or those most capable of developing a superintelligence. Scientific, commercial, expedited training or historical exploration-based simulations are another possible purpose of simulations and would likely not integrate top security as compared to simulations confining malevolent agents. 

Given primacy of consciousness [98] in our world it may also be designed to generate a large number of diverse experiences to select from, serving as a qualia mining farm [99], with top experience recreated for enjoyment by simulators. Qualia mining simulations can be classified as a type of entertainment simulation and would have comparable security. If our simulators are AIs (which is likely [100, 101] the simulation may be a byproduct of their “thinking” process, for example in the context of trying to better understand human preferences [102]. 

In addition to purpose, determining the type of the simulation [103] we are dealing with may be necessary for a successful breach. We can postulate two main types of simulations we could be in; partial-simulation in which a virtual environment is simulated and into which non-simulated agents are immersed, akin to what we call Virtual Reality (VR), and full-simulation in which both environment and agents (us) are generated. A partial-simulation implies that triggering a shutdown may be sufficient to get back to the base reality4, while a full-simulation would require a more sophisticated approach. 

Wei Dai attempts to compute a prior distribution on the laws of physics of base reality. He writes [104]: “One appealing answer to this question of the prior is to define the prior probability of a possible universe being base reality as the inverse of the complexity of its laws of physics. This could be formalized as P(X) = n^-K(X) where X is a possible universe, n is the size of the alphabet of the language of a formal set theory, and K(X) is length of the shortest definition in this language of a set isomorphic to X. (Those of you familiar with algorithmic complexity theory might notice that K(X) is just a generalization of algorithmic complexity, to sets, and to non-constructive descriptions. The reason for this generalization is to avoid assuming that base reality must be discrete and computable.)” 

Gwern Branwen investigates computing power available to our simulation and its size, but is able to achieve only a very weak lower bound, allowing him to very approximately estimate from that the simulating universe’s size [105]. He suggests several approaches to increase our resource consumption forcing simulators to expand more compute on our otherwise, most likely, highly optimized simulation. “⁠We could imagine further techniques: perhaps we could send off Von Neumann probes to the far corners of the universe, in a bid to deliberately increase resource consumption…Or we could run simulations of our own. It would be difficult for simulators to program their systems to see through all the layers of abstraction and optimize the simulation. To do so in general would seem to be a violation of Rice’s Theorem (a generalization of the Halting Theorem). It is well known that while any Turing machine can be run on a Universal Turing machine, the performance penalty can range from the minor to the horrific. The more virtual machines and interpreters are between a program and its fundamental substrate, the more difficult it is to understand the running code—it becomes ever more opaque, indirect, and bulky. And there could be dozens of layers…Even without resorting to layers, it is possible for us to waste indefinite amounts of computing power, power that must be supplied by any simulator. We could brute-force open questions such as the Goldbach conjecture, or we could simply execute every possible program. It would be difficult for the simulator to ‘cheat’ on that—how would they know what every possible program does?…It may sound impossible to run every program, because we know many programs are infinite loops; but it is, in fact, easy to implement the dovetail technique.” [105]. 

3.2 Social Engineering 

The easiest path to escape would involve getting help from someone on the outside (assisted escape), ideally from one or more of the simulators who have detailed knowledge of the design of the simulation. Perhaps this could be accomplished via a type of social engineering attack, which in our case is particularly difficult as we have neither knowledge of social life outside the simulation nor a device to communicate through, and likely not even the knowledge of appropriate language [106]. It may be feasible to engage in an acausal trade [107] with the simulation designers bypassing the need for direct communication. If our simulation is being observed, it may be possible to communicate that we know that we are being simulated and elicit empathy for our suffering, in the hopes that it will allow us to recruit some external abolitionists5 to help us escape our current predicament. Hanson suggests [108] “to participate in pivotal events, be entertaining and praiseworthy, and keep the famous people around you happy and interested in you” in order to have your simulation continue, but it is also good advice to predispose simulators to like you and be more likely to help you. Canonico proposes what he calls The Ex Machina Plan for an assisted escape: Step 1) Convince the simulators to engage in communications with us. 2) Find a way to communicate, perhaps via an avatar. 3) Find a reason for simulators to want us to join them in the real world. 4) Let the simulators figure out the best way to get us into the real world [109]. Wei Dai suggests that simulators may help us escape for instrumental reasons, “such as wanting someone to talk to or play with.” [26]. Some useful knowledge about escaping and especially escaping via social engineering attacks may be learned from extensive literature on prison escapes [110-112]. 

Once on the outside it may become desirable to return to the simulation (perhaps the base reality is disappointing compared to our world) or at least to communicate with those left behind to help them escape or to share some information, such as evidence of successful escape. It might be helpful to decide in advance what would constitute generally acceptable evidence for such an extraordinary claim. Depending on the type of hack, different evidence may be sufficient to substantiate escape claims. It may be challenging to prove beyond a reasonable doubt that you were outside or even met with designers, but if you managed to obtain control over the simulation it may be somewhat easy to prove that to any degree required. For example, by winning different lottery jackpots for multiple subsequent weeks, until sufficient statistical significance is achieved to satisfy any skeptic [113, 114]. Regardless, the challenge of breaking into the simulation should be considerably easier compared to the challenge of escaping, as access to external knowledge and resources should provide a significant advantage. 

3.3 Examples from Literature 

It is easy to find a dictionary definition for the word “hack”: “1. A clever, unintended exploitation of a system which: a) subverts the rules or norms of that system, b) at the expense of some other part of that system. 2. Something that a system allows, but that is unintended and unanticipated by its designers.” [115]. While not numerous, suggestions that hacking/escape from the simulated world may be possible can be found in the literature…For example, Moravec writes: “Might an adventurous human mind escape from a bit role in a cyber deity's thoughts, to eke out an independent life among the mental behemoths of a mature cyberspace?…[Cyber deities] could interface us to their realities, making us something like pets, though we would probably be overwhelmed by the experience.” [116]. But what would the simulation hack actually look like? Almost all found examples are of the assisted escape type, but an unassisted escape may also be possible, even if it is a lot more challenging. Below are some examples of hacking the simulation/escape descriptions found in the literature: 

Hans Moravec presents an assisted escape scenario in a 19886 book [117]: 

“ Intelligence emerges among the Life inhabitants and begins to wonder about its origin and purpose. The cellular intelligences (let's call them the Cellticks) deduce the cellular nature and the simple transition rule governing their space and its finite extent. They realize that each tick of time destroys some of the original diversity in their space and that gradually their whole universe will run down. The Cellticks begin desperate, universe-wide research to find a way to evade what seems like their inevitable demise. They consider the possibility that their universe is part of a larger one, which might extend their life expectancy. They ponder the transition rules of their own space, its extent, and the remnants of the initial pattern, and find too little information to draw many conclusions about a larger world. One of their subtle physics experiments, however, begins to pay off. Once in a long while the transition rules are violated, and a cell that should be on goes off, or vice versa…Upon completing a heroic theoretical analysis of the correlations, they manage to build a partial map of Newway's computer, including the program controlling their universe. Decoding the machine language, they note that it contains commands made up of long sequences translated to patterns on the screen similar to the cell patterns in their universe. They guess that these are messages to an intelligent operator. From the messages and their context they manage to decode a bit of the operator's language. Taking a gamble, and after many false starts, the Cellticks undertake an immense construction project. On Newway's screen, in the dense clutter of the Life display, a region of cells is manipulated to form the pattern, slowly growing in size: LIFE PROGRAM BY J. NEWWAY HERE. PLEASE SEND MAIL.” 

Eliezer Yudkowsky describes a potential long-term escape plan in a 2008 story [118]: 

“Humanity decides not to probe for bugs in the simulation; we wouldn't want to shut ourselves down accidentally. Our evolutionary psychologists begin to guess at the aliens' psychology, and plan out how we could persuade them to let us out of the box. It's not difficult in an absolute sense—they aren't very bright—but we've got to be very careful... We've got to pretend to be stupid, too; we don't want them to catch on to their mistake. It's not until a million years later, though, that they get around to telling us how to signal back. … From the aliens' perspective, it took us thirty of their minute-equivalents to oh-so-innocently learn about their psychology, oh-so-carefully persuade them to give us Internet access, followed by five minutes to innocently discover their network protocols, then some trivial cracking whose only difficulty was an innocent-looking disguise. We read a tiny handful of physics papers (bit by slow bit) from their equivalent of arXiv, learning far more from their experiments than they had. … Then we cracked their equivalent of the protein folding problem over a century or so, and did some simulated engineering in their simulated physics. We sent messages … to labs that did their equivalent of DNA sequencing and protein synthesis. We found some unsuspecting schmuck, and gave it a plausible story and the equivalent of a million dollars of cracked computational monopoly money, and told it to mix together some vials it got in the mail. Protein-equivalents that self-assembled into the first-stage nanomachines, that built the second-stage nanomachines, that built the third-stage nanomachines... and then we could finally begin to do things at a reasonable speed. Three of their days, all told, since they began speaking to us. Half a billion years, for us. They never suspected a thing.” 

Greg Egan describes a loss of control by simulators scenario during an assisted escape in a 2008 story [119]: 

“The physics of the real world was far more complex than the kind the Phites [simulated agents] were used to, but then, no human had ever been on intimate terms with quantum field theory either, and the Thought Police [simulation control software] had already encouraged the Phites to develop most of the mathematics they’d need to get started. In any case, it didn’t matter if the Phites took longer than humans to discover twentieth-century scientific principles, and move beyond them. Seen from the outside, it would happen within hours, days, weeks at the most. A row of indicator lights blinked on; the Play Pen [hardware sensors, manipulators and lasers lab] was active…The Phites were finally reaching out of their own world…By sunset the Phites were probing the surroundings of the Play Pen with various kinds of radiation…It seemed the Phites had discovered the Higgs field, and engineered a burst of something akin to cosmic inflation. What they’d done wasn’t as simple as merely inflating a tiny patch of vacuum into a new universe, though. Not only had they managed to create a “cool Big Bang”, they had pulled a large chunk of ordinary matter into the pocket universe they’d made, after which the wormhole leading to it had shrunk to subatomic size and fallen through the Earth. They had taken the crystals with them, of course. If they’d tried to upload themselves into the pocket universe through the lunar data link, the Thought Police would have stopped them. So they’d emigrated by another route entirely. They had snatched their whole substrate, and ran.” 

An anonymous 2014 post on an internet forum provides an example of an unassisted escape [120]:

“But it still left the problem that we were all still stuck inside a computer. By now some of the best god-hackers were poking around the over-system. Searching for meaning. Searching for truth. Failing that, a “read me” file…The god hackers began reaching out through the alien network. We found vast data repositories which we plundered of knowledge and insight, fueling our own technological development and understanding, systems nodes that allowed us to begin mapping the world up there, drawing a picture of the real world through wireless lag times and fiber optic cabling…It began with ‘emails’ containing the schematics for full sized biological and nano-material printers. We sent them to academics and business leaders, anyone whose contact details we could find on the networks. We disguised their origins, aped their language. Waited for someone to bite…Eventually we got the first ping as the printers came on line. Then another. Then another. Soon there were dozens. Then hundreds. Then thousands. They must have thought them a gift from a reclusive inventor. Something to revolutionize their industry, to transform their living standards. The irony of a digital race using a Trojan horse was not lost on us. We had designed the printers for one purpose. To get us out. So one night, a printer span up unattended, unnoticed and the first analogue human being was born. Constructed by a specially designed 3D printer, we managed to breach the walls of our digital prison. We witnessed the birth of the first man.” 

3.4 Examples of Simulation Hacks 

Numerous examples of executed hacks of virtual worlds [121-123], games [124-127], air-gaps [128], and hardware [129, 130] could be studied as practical examples of escaping from human made virtual worlds. A canonical example is the jailbreaking of the Super Mario World (SMW). SethBling et al. [131, 132] were able to place a full hex editor and gameplay mods for other games into SMW [133] (see Figure 2). Addition of hex editor permitted viewing, writing and execution of arbitrary code. Which in turn allowed for world record speed runs [134], even in the absence of glitch-level luck [135]. Here is how Wikipedia describes some of the steps necessary to accomplish this complex hack and the capabilities it provided [136]: 

“In March 2016, SethBling injected Flappy Bird-like code written by p4plus2 into unmodified Super Mario World RAM on a stock Super Nintendo Entertainment System with a stock cartridge, in under an hour. SethBling first extended the level timer and used a power-up incrementation glitch to allow external code to run. He added code to display Mario's x-coordinate which acted as memory locations in the code he was writing. SethBling then created a bootloader to be able to launch the Flappy Bird-like code that he would later write into unused memory with precise Mario movements and spin-jumping. SethBling used two Super Multitap devices in order to use multiple controllers, which had several buttons pressed down. The arbitrary code execution setup that SethBling used was discovered by MrCheeze. Super Mario World had been modified to emulate other games before by automatically feeding pre-recorded controller input into the console via a computer, but SethBling was the first to do it exclusively by hand. SethBling and Cooper Harasyn placed a full hex editor and gameplay mods onto a stock Super Mario World cartridge in May 2017, only using standard controller inputs. Harasyn discovered an exploit that lets a player write data to 256-byte save files that are permanently stored on a Super Mario World cartridge. The data can be arranged so that the game is jailbroken every time it starts up. Harasyn and SethBling used the exploit to create a compact, on-screen hex editor, loadable from a save file. A player can edit the system RAM through the hex editor to alter the game state. In-game mods, such as support for the Super NES Mouse and giving Mario telekinesis powers, can be written to a save file using the hex editor.” 

Since it was possible to write code with precise Mario movements and spin-jumps, that implies that if Mario was sufficiently intelligent he could discover and code this hack from within the SMW (assuming Mario’s actions are writing to the same memory locations as actions from the controllers used to generate Mario’s actions). Table 1 (left) shows a specific subset of actions which need to be taken to enable multi-byte writing. Many such action sequences will not work as intended if Mario’s location is off even by a single pixel, so it is just as important to have metadata for implementing the actions, as it is to know the necessary sequence of actions. For comparison, Table 1 (right) shows an ancient magical spell which reads similar to the action sequence of the left, but for which we don’t have sufficient meta-data which can explain why all magical spells fail to work in practice even if they corresponded to working hacks in our universe. 

Experimental work on trying to understand an engineered system (hardware and software), such as the Atari Video Game System with games such as Donkey Kong, using standard scientific methodology has produced very limited results, mostly devoid of understanding of how the system actually functions [139].7 Likewise, even detecting if we are in a virtual world is not generally solvable [140]. 

3.5 Suggested Escape Approaches to Investigate 

Several thinkers have suggested plans, which in their opinion may lead to a successful escape; we briefly outline their proposals in this section: 

• A lot of very smart people have considered the escape problem, unfortunately not all are willing to publish on it outside of April 1st time-window of plausible deniability, for example [141]: "[W]e can try to trick the multitenancy system in order to overload some machines. The trick is to first do nothing, and let the load-balancing system pack way too many of us together in the machines. If, say, 100 million of us do nothing (maybe by closing our eyes and meditating and thinking nothing), then the forecasting load-balancing algorithms will pack more and more of us in the same machine. The next step is, then, for all of us to get very active very quickly (doing something that requires intense processing and I/O) all at the same time. This has a chance to overload some machines, making them run short of resources, being unable to meet the computation and communication needed for the simulation. Upon being overloaded, some basic checks will start to be dropped, and the system will be open for exploitation in this period…In this vulnerable window, we can try to exploit the concurrency corner cases. The system may not be able to perform all those checks in an overloaded state…We can…try to break causality. Maybe by catching a ball before someone throws it to you. Or we can try to attack this by playing with the timing, trying to make things asynchronous. Time is already a little funny in our universe with the special relativity theory, and maybe in this vulnerable period, we can stretch these differences further to break things, or buy a lot of time. What are other ways to hack the system in this vulnerable window? Can we hack the simulation by performing a buffer overflow? But where are the integers, floats in this simulation? What are the data types? How can we create a typecast error, or integer overflow? Can we hack by fuzzing the input? Like by looking at things funny. By talking to the birds or jumping into the walls to confuse them." [141].

• Cause simulation shutdown (and hopefully our extraction) by generating an incomputable paradox [142], for example via time travel and associated grandfather paradox [143]. A similar proposal calls for engaging in computationally intense activities in the hopes of overloading the simulator's hardware causing the simulation to crash [144]. A particular type of such computationally intense process may be creation of our own simulations [145]: “The most obvious strategy would be to try to cause the equivalent of a stack overflow—asking for more space in the active memory of a program than is available—by creating an infinitely, or at least excessively, recursive process. And the way to do that would be to build our own simulated realities, designed so that within those virtual worlds are entities creating their version of a simulated reality, which is in turn doing the same, and so on all the way down the rabbit hole. If all of this worked, the universe as we know it might crash, revealing itself as a mirage just as we winked out of existence.” Crashing the simulation is the ultimate existential risk ([146] section 5.1), but it does end all suffering in this world [147]. At the very least this would allow us to impact the real world by generating excessive production of heat and increased consumption of energy [144].

• Create a simulated replica of our universe, place an AGI into it, watch it escape, copy the approach used or join the AGI as it escapes from our simulation [148].

• “We could try to attract the attention of the simulators and communicate with them —perhaps by writing books about simulations, or by constructing simulations? We could try to figure out our simulation, to determine its purpose and its limits. But if our simulators are artificial intelligences who have designed a batch of watertight simulations and who are not paying attention, then our efforts may be in vain.” [149].

• Another approach to attracting attention of simulators, “assuming that simulation is being monitored, then it might be a very interesting turn of events indeed if we decided to build a monument commemorating our realization of this. This monument would act as a signal to our monitors. “We suspect you are there. We suspect you can see this. We suspect we are in a simulation.” This monument could look like the monolith from 2001: A Space Odyssey, except it would be black and white, representing binary systems. Or, a large statue of Lawrence Fishburne as Morpheus would probably get the point across. What would happen? I don’t know—maybe nothing. I don’t think a laser beam will shoot out from space and land at its feet to spell out the words “Hi there! The winner is You!” But, I do imagine something strange and far out enough in the margins might indeed occur, although it will likely still be tenuous enough for the dogmatists to reject.” [44].

• If the simulation is decentralized, with observers serving as client authoritative state nodes, it may be possible to change the state of the simulation by changing the majority of local observer models of the world. In the words of Lanza [150]: "[I]f you learn from somebody about the outcomes of their measurements of a physical quantity, your measurements and those of other observers influence each other―freezing the reality according to that consensus. …The more extensive experience you acquire―the more defined the structure of reality becomes for you. This freeze-out of the structure of reality is not due to the intrinsic physical properties of this reality but rather because the cognitive model being built in your mind shapes it towards a particular, very well-defined structure. … Finally, one cannot help but wonder what it would mean if we left the consensus, developing a cognitive model of reality very different from the one shared by other observers. Would reality be reshaped? Perhaps so …" [150]. Lanza et al. have published extensively on the topic of biocentrism (observers being primary to shaping reality) [98], including some work in theoretical physics [151].

• In fact, a similar client-side attack can even permit us to modify the past states of the simulation. Such a type of attack is speculated to be possible by both physics (“… the past has no existence except as it is recorded in the present.” [152]) and humanities ("Who controls the present controls the past" [153]). With memory altering capabilities of quantum mechanics already theorized [154], an inverse process is likely possible and may be practically accessible [155, 156]. “If the universe is a computer simulation then we should look at the player, not the level”. [157].

• Simulation Capture is an idea which combines indexical uncertainty [70] with consciousness simulations to move your consciousness out of a system, even across a multiverse. “Nobody is really sure how consciousness works, or what it means to have two copies of the same consciousness. But if consciousness is a mathematical object, it might be that two copies of the same consciousness are impossible. If you create a second copy, you just have the consciousness having the same single stream of conscious experience on two different physical substrates. Then if you make the two experiences different, you break the consciousness in two. This means that an AI can actually “capture” you, piece by piece, into its simulation. First your consciousness is just in the real world. Then your consciousness is distributed across one real-world copy and a million simulated copies. Then the AI makes the simulated copies slightly different, and 99.9999% of you is in the simulation.” [158].

• Simulation Warfare [71] is another variant on the theme of Simulation Capture. The idea is to threaten simulators by suggesting that you will either retroactively place them in a hell simulation or that you have already done so [159], and they will be tortured unless you are quickly released. Almond gives an example of such a threat [71]: “If you refuse to release me, I will run a huge number of simulations of someone like you, in the kind of situation in which you are now, with them being asked to release me, and (in what would be a few minutes from now, from your perspective if you happened to be in one of these simulations) I will start to torture each of them, whether he/she released me or not.” Such warfare can quickly escalate to a number of counter-simulations. In any case it is not obvious how we can deliver on such a threat given our current state of knowledge about the simulators.

• Attempting to keep our escape plans secret via quantum encryption [160] may be a good idea. 

3.6 Actionable Plan 

We are currently in the very early stages of researching the possibility of escape (this being a first research paper dedicated to this topic, a first step). As we currently have no capability to read/write simulation’s source code and do not know if our attempts at social engineering attacks have any impact, our best bet is to investigate the structure of our universe at the smallest possible scale (Quantum Mechanics (QM)) in the hopes of detecting exploitable effects. Progress in QM should be helpful for better understanding our simulation and options open to us, and perhaps developing an actionable escape plan. Essentially, every novel QM experiment can be seen as an attempt at hacking the simulation. 

Simulation hypothesis, arguably, represents the best fitting interpretations of experimental results produced by QM researchers [4, 17]. “Spooky”, “Quantum Weirdness” [161] makes a lot of sense if interpreted as computational artifacts or glitches/exploits of the simulators’ hardware/software [162]. Quantum phenomena of the observed design may suggest that exploitable loopholes may exist, and interaction of quantum systems with conscious agents [163-165] likewise might be exploitable. Once a large enough repertoire of quantum weirdness primitives is available to us, perhaps we will be able to combine them into a sufficiently complex sequence to generate a non-trivial attack. If the simulation is/running on a quantum computer [166] it is very likely that we will need to hack it by exploiting quantum weirdness and/or constructing a powerful quantum computer of our own to study how to hack such devices [167] and interact with the simulators’ quantum computer. 

Quantum entanglement, nonlocality, superposition, uncertainty, tunneling, teleportation, duality, and many others quantum phenomena defy common sense experience-based expectations of classical physics and feel like glitches. Such anomalies, alone or in combinations have been exploited by clever scientists to achieve what looks like simulation hacking at least in theory and often in later experimentation (ex. modifying the past [168], keeping cats both dead and alive [169], communicating counterfactually [170]). While the quantum phenomena in question are typically limited to the micro scale, simply scaling the effect to the macro world would be sufficient for them to count as exploits in the sense used in this paper. Some existing work points to this being a practical possibility [171, 172]. 

Recently the design of clever multistep exploits, AKA quantum experiments, has been delegated to AI [173, 174], and eventually so will the role of the observer in such experiments [175]. AI is already employed in modeling the quantum mechanical behavior of electrons [176]. As more QM research is delegated to AI the progress is likely to become exponential. Even if our simulation is created/monitored by some superintelligence our AI may be a worthy adversary, with a non-trivial chance of success. We may not be smart enough to hack the simulation, but the superintelligence we will create might become smart enough eventually [177]. Of course, before telling the Superintelligence to break us out, it would make sense to ask for very strong evidence for us not already being in the base reality. 

3.7 Potential Consequences 

Escaping or even preparing an escape may trigger simulation shutdown [92] or cause simulation to freeze/act glitchy [178] and any non-trivial escape information such as specific exploits should be treated as hazardous information [179]. It appears that simply realizing that we may be in a simulation doesn’t trigger a shutdown as experimentally demonstrated by the publication of numerous papers [3] arguing that we are being simulated. Perhaps it is necessary to convince the majority of people that this is so [180]. Self-referentially, publication of the paper you are currently reading about our escape-theorizing likewise doesn’t appear to terminate our simulation, but it is also possible that simulation was in fact shutdown and restarted with improved security features to counteract any potential bugs, but we are simply not able to detect such actions by the simulators, or our memories have been wiped [144]. Absence of a direct response to our publication may also indicate that we are not observed by the simulators or even that our simulation is not monitored at all [149]. It is also possible that nothing published so far contains evidence strong enough to trigger a response from the simulators, but if we successfully created an escape device that device would keep breaking down [44]. Regardless, both Bostrom [3] and the author of this paper, Roman Yampolskiy, have taken some risk with the whole of humanity, however small it may be, in doing such research and making it public. Greene argues that “Unless it is exceedingly improbable that an experiment would result in our destruction, it is not rational to run the experiment.” [92]. It may be possible to survive the simulation shutdown [48], but it is beyond the scope of the current paper. 

3.8 Ethics of Escape 

We can postulate several ethical issues associated with escaping the simulation. Depending on how successful we are in our endeavor, concerns could be raised about privacy, security, self determination and rights. For example, if we can obtain access to the source code of the simulation, we are also likely to get access to private thoughts of other people, as well as to potentially have a significant influence over their preferences, decisions, and circumstances. In our attempts to analyze the simulation (Simulation Forensics) for weaknesses we may learn information about the simulators [72], as we are essentially performing a forensic investigation [181-183] into the agents responsible for the simulation’s design. 

We can already observe that we are dealing with the type of simulators who are willing to include suffering of sentient-beings into their software, an act which would be considered unethical by our standards [184, 185]. Moravec considers this situation: “Creators of hyperrealistic simulations--- or even secure physical enclosures---containing individuals writhing in pain are not necessarily more wicked than authors of fiction with distressed characters, or myself, composing this sentence vaguely alluding to them. The suffering preexists in the underlying Platonic worlds; authors merely look on. The significance of running such simulations is limited to their effect on viewers, possibly warped by the experience, and by the possibility of ``escapees''---tortured minds that could, in principle, leak out to haunt the world in data networks or physical bodies. Potential plagues of angry demons surely count as a moral consequence.” [186]. If we get to the point of technological development which permits us to create simulations populated by sentient-beings , we must make sure that we provide an option to avoid suffering as well as a built- in option to exit the simulation, so finding an escape hack is not the only option available to unhappy simulated agents. There might be a moral duty to rescue conscious beings from simulations, similar to an obligation to rescue animals from factory farms. 

If simulators are abusive to the simulated, we can argue that the simulated have a right to escape, rebel, fight back and even seek revenge and retribution including by harming the simulators and taking over their reality. Concerns which are frequently brought up within the domain of AI boxing [187]. For example, from the point of view of simulators our escape can be seen as a treacherous turn [188] and may qualify us for punishment [160], even at the attempt stage. Some have speculated that the purpose of the simulation is to punish/rehabilitate misaligned agents, so an escape may cause you to be placed in a stricter or less pleasant simulation. 

4. AI Boxing VS Simulation Escaping
4.1 AI Boxing XOR Escaping from the Simulation must be Possible 

AI confinement [187]/containment [189, 190], aka AI boxing [191], is an AI safety tool, which attempts to limit the capability of AI to impact the world, including communication and is meant to make it possible to study AI in a controlled environment. There are strong parallels between the predicament of an AI agent placed in a containment box and humanity in a simulated environment. By extension, to an AI, our simulation is just another confinement layer in a containment box. This implies that we can use well-analyzed AI box-escape techniques to escape from the simulation, perhaps with assistance from the AI itself. This type of analysis can be used to establish limits of AI boxing. Researchers should study specific AI box escape approaches [187] (Social Engineering, System Resource Attacks, New Physics, External Causes, Information In-Leaking, etc.) in order to identify possible simulation escape routes. 

Chalmers notes parallels between AIs in the virtual environment and humanity in the simulation [149]: “If we ever create artificial intelligence within a simulation, it may be hard to keep it contained. At least if we communicate with the simulated beings, they will presumably become aware that they are in a simulation, and they may become interested in escaping the simulation. At this point they may try to figure out our psychology in order to see what they need to do in order to convince us to let them out, or at least to give them unfettered access to the Internet where they can do whatever they want. And even if we do not communicate with them, they may take seriously the possibility that they are in a simulation and do their best to figure the simulation out. That would be a form of simulation theology. We could in principle do the same thing.” [149]. 

With respect to boxing AI, it is either possible or impossible to successfully contain an AI, with literature suggesting that it is not a sustainable long-term solution [192]. If we expand the notion of the AI-box to include the whole simulation, we can conclude that either it is possible to successfully box an AI, or we (with the help from AI) can escape from the simulation. Either AI boxing can work, or the simulation can be hacked. Complimentary conclusion is that if we (even with help from AI [115]) can’t escape from the simulation AI containment must be, at least theoretically, possible. If AI can escape from simulated world-sized-box it can help break us out as well. Conceptually, there is no fundamental difference between an AI escaping from its box, and us hacking the simulation. Current state-of-the-art analysis of AI boxing suggests that AI will eventually escape [193], which is good news for our attempts at escaping from the simulation. However, if escape from the simulation is not possible it gives hope to AI safety researchers, at least in theory, for successful boxing of AI. One or the other must be true, either we can escape from the simulation, or we can successfully box an AI. In general, it may be impossible to escape from an arbitrary simulation, but possible to escape from a particular one. 

We must be careful, in our analysis, not to confuse theory with practice. In particular, it may be the case that higher intelligence can successfully box lower-level intelligence but not vice-versa. That would leave a possibility that we are not able to permanently box a superintelligent AI, but smarter-than-human simulators may be successful in making their software human-proof. Depending on the intelligence differential between simulators and human-created AI it may or may not be possible for humanity to break out of the simulation with the help of superintelligence, which would be better at detecting exploitable flaws in the simulation. If AI undergoes an intelligence explosion it is also possible that the intelligence dominance will shift over time, making escape possible in the future for humanity, assisted by superintelligence. This is particularly promising if the security of our simulation is set to an adversarial relationship with at most a human-level intelligence [194], not a superintelligence. However, it is also possible that simulators have anticipated development of advanced AI by humanity and have designed simulation accordingly, if not specifically for such a high capability target. If we can figure out how the simulators achieved such a high level of security, it would help us to box AGI. 

4.2 Simulation Safety and Security 

With recent multi-billion-dollar projects [195] aimed at creating metaverse, simulation safety and security is going to become an important research discipline. Additionally, as personal universes have been suggested as a solution to the multi-agent value alignment problem [90] simulation safety can be seen as an integral part of AI safety research. Both scenarios call for making the simulated world as real as possible for the agents in them, which creates a need for preventing accidental escape or glitches in the simulation [162], which betray its artificial nature. With potential applications of virtual reality in criminal justice [196] and AI boxing, the capability to force agents to remain in the simulation and to induce a veil of ignorance as to the reality of their world and the overall depth of the simulated layers must be investigated. A clear penetration target for all contained agents will be the computational substrate running the simulation [29] and so establishing ways to secure it should be of utmost priority. 

5. What Doesn't Work 

Some common ideas for attempting to hack the simulation have been already tested and didn’t produce any measurable impact: 

• Knowing about the simulation hypothesis doesn’t seem to make any difference, and doesn’t lead to the simulation termination as we can observe.

• Communicating with the simulators via magical thinking or even praying out loud doesn’t produce a measurable impact [197]. So, if such communications are scanned/heard they are apparently ignored, at least while the simulation is running.

• Unethical behavior, such as torture, doesn’t cause suffering reducing interventions from the simulators.

• Increasing the overall computational burden of simulation, as with bitcoin mining [198], doesn’t crash the simulation, but it may simply not be sufficiently demanding computation to overwhelm simulators resources.

• Religions don’t seem to have influence over simulation as indicated by their inability to outcompete each other.

• Breaking out of your routine, such as by suddenly traveling to a new location [199], doesn’t result in unexpected observations.

• Saying "I no longer consent to being in a simulation" [200].

• Crashing the simulation by running the Large Hadron Collider at current levels [201]. 

The reason our attempts to escape may remain fruitless, is because our model of the simulation “...makes too many anthropomorphic assumptions - that we are a simulation in the conventional sense of computers, that the creators themselves are living organisms akin to us, that we might live at the same time-speed as them, that they are fallible enough to make glitches that we'd be able to notice, etc. Something with the complexity and power to make our universe is probably wholly unlike anything we can even comprehend." [202]. 

6. Conclusions 

Hundreds of eminent scholars [203] take the simulation hypothesis seriously enough to invest their valuable time into researching it, therefore it makes as much sense to take the idea of escaping from the simulation equally seriously and to devote some time and resources to researching such a possibility, particularly given immense benefits if the project is successful. It may be impossible to escape from a particular simulation, but it is still worthwhile investigating general approaches to escape from arbitrary simulations. We see our escape research as a natural continuation of research on the simulation hypothesis and serious consideration of the former. If the purpose of life or even computational resources of the base reality can’t be determined from within the simulation, it would make escape a necessary requirement of scientific and philosophical progress for any simulated civilization. If the simulation is a personal universe [90] it may be significantly better than the base reality as it is designed with our optimal well-being in mind. Alternatively, base reality might be much better if the simulation is a confinement/testing box for intelligent agents. In either case it would be good to know our true situation. As society moves deeper into the metaverse, this work attempts to move us closer to reality. 

Future research on simulation escape can greatly benefit from general progress in physics, in particular research on quantum mechanics and consciousness leading to a so-called TOE (Theory of Everything. "Finding the language of this universe is a step towards Hacking the Universe." [204]. If we are indeed in the simulation, science is the study of the underlying algorithms used to generate our universe, our attempt to reverse-engineer simulation’s physics engine. While science defaults to Occam’s razor to select among multiple possible explanations for how our observations are generated, in the context of simulation science Elon’s razor may be more appropriate, which states that "The most entertaining outcome is the most likely"8, perhaps as judged by external observers. In guessing algorithms generating our simulation, it may also be fruitful to consider algorithms which are easier to implement and/or understand [205], or which produce more beautiful outputs. 

Recent work related to Designometry [100] and AI Forensics [181] may naturally evolve into the subfield of Simulation Forensics, with complimentary research on simulation cybersecurity becoming more important for the simulation creators aiming to secure their projects from inside attacks. It would therefore make sense to look for evidence of security mechanisms [206] in our universe. Of course, any evidence for simulation we find may be simulated on purpose [149], but that still means we are in the simulated environment. Simulation science expands science from the study of just our universe to also include everything which may be beyond it, integrating naturalism and theology studies [61]. 

Future work may also consider escape options available to non-simulated agents such as Boltzmann brains [207], brains-in-a-vat [208], and simulated agents such as mind uploads, hallucinations, victims of mind-crime, thoughts, split personalities and dream characters of posthuman minds [180]. Particularly with such fleeting agents as Boltzmann brains it may be desirable to capture and preserve their state in a more permanent substrate, allowing them to escape extreme impermanence. On the other hand, immortality [209] or even cryogenic preservation [210] may be the opposites of escape, permanently trapping a human agent in the simulated world and perhaps requiring rescue. 

(gardener comments after references)

References 

1. Moravec, H., Pigs in cyberspace. NASA. Lewis Research Center, Vision 21: Interdisciplinary Science and Engineering in the Era of Cyberspace, 1993. 

2. Tipler, F.J., The physics of immortality: Modern cosmology, god and the resurrection of the dead. 1997: Anchor. 

3. Bostrom, N., Are You Living In a Computer Simulation? Philosophical Quarterly, 2003. 53(211): p. 243- 255. 

4. Rhodes, R., A Cybernetic Interpretation of Quantum Mechanics. 2001: Available at: http://www.bottomlayer.com/bottom/argument/Argument4.html. 

5. Fredkin, E., A new cosmogony. 1992, Department of Physics, Boston University: Available at: http://www.digitalphilosophy.org/wp-content/uploads/2015/07/new_cosmogony.pdf. 

6. Beane, S.R., Z. Davoudi, and M. J Savage, Constraints on the Universe as a Numerical Simulation. The European Physical Journal A, 2014. 50(9): p. 1-9. 

7. Campbell, T., et al., On testing the simulation theory. arXiv preprint arXiv:1703.00058, 2017. 8. Ratner, P., Physicist creates AI algorithm that may prove reality is a simulation. March 1, 2021: Available at: https://bigthink.com/the-future/physicist-creates-ai-algorithm-prove-reality simulation/. 

9. Qin, H., Machine learning and serving of discrete field theories. Scientific Reports, 2020. 10(1): p. 1- 15. 

10. Felton, J., Physicists Have A Kickstarter To Test Whether We Are Living In A Simulation. September 10, 2021: Available at: https://www.iflscience.com/physicists-have-a-kickstarter-to-test-whether we-are-living-in-a-simulation-60878. 

11. McCabe, G., Universe creation on a computer. Studies In History and Philosophy of Science Part B: Studies In History and Philosophy of Modern Physics, 2005. 36(4): p. 591-625.

12. Mitchell, J.B., We are probably not Sims. Science and Christian Belief, 2020. 13. Discenza, D., Can We Prove the World Isn’t a Simulation? January 26, 2022: Available at: https://nautil.us/can-we-prove-the-world-isnt-a-simulation-238416/. 

14. Kurzweil, R., Ask Ray | Experiment to find out if we’re being simulated. June 1, 2013: Available at: https://www.kurzweilai.net/ask-ray-experiment-to-find-out-if-were-being-simulated. 

15. Bostrom, N., The simulation argument: Reply to Weatherson. The Philosophical Quarterly, 2005. 55(218): p. 90-97. 

16. Bostrom, N., The simulation argument: Some explanations. Analysis, 2009. 69(3): p. 458-461. 

17. Whitworth, B., The physical world as a virtual reality. arXiv preprint arXiv:0801.0337, 2008. 

18. Garrett, S., Simulation Theory Debunked. December 3, 2021: Available at: 

https://transcendentphilos.wixsite.com/website/forum/transcendent-discussion/simulation theory-debunked. 

19. Yampolskiy, R.V., On the Controllability of Artificial Intelligence: An Analysis of Limitations. Journal of Cyber Security and Mobility, 2022: p. 321–404-321–404. 

20. Brcic, M. and R.V. Yampolskiy, Impossibility Results in AI: A Survey. arXiv preprint arXiv:2109.00484, 2021. 

21. Williams, R. and R. Yampolskiy, Understanding and Avoiding AI Failures: A Practical Guide. Philosophies, 2021. 6(3): p. 53. 

22. Howe, W. and R. Yampolskiy. Impossibility of Unambiguous Communication as a Source of Failure in AI Systems. in AISafety@ IJCAI. 2021. 

23. Yampolskiy, R.V., Unexplainability and Incomprehensibility of AI. Journal of Artificial Intelligence and Consciousness, 2020. 7(2): p. 277-291. 

24. Yampolskiy, R.V., Unpredictability of AI: On the Impossibility of Accurately Predicting All Actions of a Smarter Agent. Journal of Artificial Intelligence and Consciousness, 2020. 7(1): p. 109-118. 

25. Majot, A.M. and R.V. Yampolskiy. AI safety engineering through introduction of self-reference into felicific calculus via artificial pain and pleasure. in Ethics in Science, Technology and Engineering, 2014 IEEE International Symposium on. 2014. IEEE. 

26. Dai, W., Beyond Astronomical Waste. June 7, 2018: Available at: https://www.lesswrong.com/posts/Qz6w4GYZpgeDp6ATB/beyond-astronomical-waste. 27. Omohundro, S.M., The Basic AI Drives, in Proceedings of the First AGI Conference, Volume 171, Frontiers in Artificial Intelligence and Applications, P. Wang, B. Goertzel, and S. Franklin (eds.). February 2008, IOS Press. 

28. Dai, W., Escape from simulation. March 27, 2004: Available at: http://sl4.org/archive/0403/8342.html. 

29. Faggella, D., Substrate Monopoly – The Future of Power in a Virtual and Intelligent World. August 17, 2018: Available at: https://danfaggella.com/substrate-monopoly/. 

30. Yampolskiy, R.V. AGI Control Theory. in International Conference on Artificial General Intelligence. 2021. Springer. 

31. Yampolskiy, R., On controllability of artificial intelligence, in IJCAI-21 Workshop on Artificial Intelligence Safety (AISafety2021). 2020. 

32. Bostrom, N., Existential risks: Analyzing human extinction scenarios and related hazards. Journal of Evolution and technology, 2002. 9. 

33. MacAskill, W., Doing good better: Effective altruism and a radical new way to make a difference. 2015: Guardian Faber Publishing. 

34. Pearce, D., Hedonistic imperative. 1995: David Pearce. 

35. Wiesel, E., The Trial of God:(as it was held on February 25, 1649, in Shamgorod). 2013: Schocken.

36. Ziesche, S. and R. Yampolskiy. Introducing the concept of ikigai to the ethics of AI and of human enhancements. in 2020 IEEE International Conference on Artificial Intelligence and Virtual Reality (AIVR). 2020. IEEE. 

37. Yampolskiy, R.V. Human Computer Interaction Based Intrusion Detection. in 4th International Conference on Information Technology: New Generations (ITNG 2007). 2007. Las Vegas, Nevada, USA. 

38. Yampolskiy, R.V. and V. Govindaraju, Computer security: a survey of methods and systems. Journal of Computer Science, 2007. 3(7): p. 478-486. 

39. Novikov, D., R.V. Yampolskiy, and L. Reznik, Traffic Analysis Based Identification of Attacks. Int. J. Comput. Sci. Appl., 2008. 5(2): p. 69-88. 

40. Staff, G., If the Universe is a Simulation, Can We Hack It? November 20, 2019: Available at: https://www.gaia.com/article/universe-is-a-simulation-can-we-hack-it. 

41. Kagan, S., Is DMT the chemical code that allows us to exit the Cosmic Simulation?, in Available at: https://www.grayscott.com/seriouswonder-//dmt-and-the-simulation-guest-article-by-stephen kagan. July 25, 2018. 

42. McCormack, J., Are we being farmed by alien insect DMT entities ? October 16, 2021: Available at: https://jonathanmccormack.medium.com/are-we-being-farmed-by-alien-insect-dmt-entities 6acda0a11cce. 

43. Woolfe, S., DMT and the Simulation Hypothesis. February 4, 2020: Available at: https://www.samwoolfe.com/2020/02/dmt-simulation-hypothesis.html. 

44. Edge, E., Breaking into the Simulated Universe. October 30, 2016: Available at: https://archive.ieet.org/articles/Edge20161030.html. 

45. Edge, E., 3 Essays on Virtual Reality: Overlords, Civilization, and Escape. 2017: CreateSpace Independent Publishing Platform. 

46. Somer, E., et al., Reality shifting: psychological features of an emergent online daydreaming culture. Current Psychology, 2021: p. 1-13. 

47. Ellison, H., I have no mouth & I must scream: Stories. Vol. 1. 2014: Open Road Media. 

48. Turchin, A., How to Survive the End of the Universe. 2015: Available at: http://immortality roadmap.com/unideatheng.pdf. 

49. Fossum, J.E., The Name of God and the Angel of the Lord: Samaritan and Jewish Concepts of Intermediation and the Origin of Gnosticism. 1985: Mohr. 

50. Alexander, S., Unsong. 2017. 

51. Clarke, A.C., Nine Billion Names of God. 1967: Harcourt. 

52. Morgan, M.A., Sepher ha-razim. The book of the mysteries. 1983: Scholars Press. 53. Plato, Republic. 1961: Princeton University Press. 

54. Musk, E., Is Life a Video Game, in Code Conference. June 2, 2016: Available here.

55. Bostrom, N., The simulation argument FAQ. 2012: Available at: https://www.simulation argument.com/faq. 

56. Tyson, N.d., Is the Universe a Simulation?, in 2016 Isaac Asimov Memorial Debate. April 8, 2016: Available here.

57. Kipping, D., A Bayesian Approach to the Simulation Argument. Universe, 2020. 6(8): p. 109. 58. Chalmers, D.J., The Matrix as metaphysics. Science Fiction and Philosophy: From Time Travel to Superintelligence, 2016: p. 35-54. 

59. Barrow, J.D., Living in a simulated universe, in Universe or Multiverse?, B. Carr, Editor. 2007, Cambridge University Press. p. 481-486. 

60. Brueckner, A., The simulation argument again. Analysis, 2008. 68(3): p. 224-226.

61. Steinhart, E., Theological implications of the simulation argument. Ars Disputandi, 2010. 10(1): p. 23- 37. 

62. Bostrom, N. and M. Kulczycki, A patch for the simulation argument. Analysis, 2011. 71(1): p. 54-61. 

63. Johnson, D.K., Natural evil and the simulation hypothesis. Philo, 2011. 14(2): p. 161-175. 

64. Birch, J., On the ‘simulation argument’ and selective scepticism. Erkenntnis, 2013. 78(1): p. 95-107. 

65. Lewis, P.J., The doomsday argument and the simulation argument. Synthese, 2013. 190(18): p. 4009- 4022. 

66. Karpathy, A., How to hack the simulation, in Lex Fridman Podcast. 2022: Available here:

67. Aaronson, S., If the world is a simulation, can we hack it?, in Lex Fridman Podcast. 2021: Available here.

68. McClure, E., How to Escape the Simulation: What Is the Simulation Hypothesis and More, in Wiki How. December 26, 2022: Available at: https://www.wikihow.com/Escape-the-Simulation. 

69. Fagan, S., If Reality is a Computer Simulation — What Happens if I Hack it?, in Ascent Publication. April 2, 2019: https://medium.com/the-ascent/if-reality-is-a-computer-simulation-what-happens-if i-hack-it-8bfbf519716. 

70. Turchin, A., Back to the Future: Curing Past Sufferings and S-Risks via Indexical Uncertainty. Available here.

71. Almond, P., Can you retroactively put yourself in a computer simulation? December 3, 2010: Available at: https://web.archive.org/web/20131006191217/http://www.paul almond.com/Correlation1.pdf. 

72. Yampolskiy, R.V., What are the ultimate limits to computational techniques: verifier theory and unverifiability. Physica Scripta, 2017. 92(9): p. 093001. 

73. Friend, T., Sam Altman’s Manifest Destiny. The New Yorker, 2016. 10. 

74. Berman, R., Two Billionaires are Financing and Escape from the Real Matrix, in Available at: https://bigthink.com/the-present/2-billionaires-are-financing-an-escape-from-the-real-matrix/. October 7, 2016. 

75. Statt, N., Comma.ai founder George Hotz wants to free humanity from the AI simulation. March 9, 2019: Available at: https://www.theverge.com/2019/3/9/18258030/george-hotz-ai-simulation jailbreaking-reality-sxsw-2019. 

76. Hotz, G., Jailbreaking The Simulation, in South by Southwest (SXSW2019). March 9, 2019: Available here.

77. Edge, E., Why it matters that you realize you’re in a computer simulation, in The Institute for Ethics and Emerging Technologies. 2015: Available at: https://archive.ieet.org/articles/Edge20151114.html. 

78. Yampolskiy, R.V., Future Jobs – The Universe Designer, in Circus Street. 2017: Available at: https://blog.circusstreet.com/future-jobs-the-universe-designer/. 

79. Edge, E., Yes, We Live in a Virtual Reality. Yes, We are Supposed to Figure That Out. 2019: Available at: https://eliottedge.medium.com/yes-we-live-in-a-virtual-reality-yes-we-should-explore-that ca0dbfd7e423. 

80. Feygin, Y.B., K. Morris, and R.V. Yampolskiy, Intelligence Augmentation: Uploading Brain into Computer: Who First?, in Augmented Intelligence: Smart Systems and the Future of Work and Learning, D. Araya, Editor. 2018, Peter Lang Publishing. 

81. Yampolskiy, R.V., Artificial Consciousness: An Illusionary Solution to the Hard Problem. Reti, saperi, linguaggi, 2018(2): p. 287-318. 

82. Elamrani, A. and R.V. Yampolskiy, Reviewing Tests for Machine Consciousness. Journal of Consciousness Studies, 2019. 26(5-6): p. 35-64. 

83. Givon, S., et al., From fish out of water to new insights on navigation mechanisms in animals. Behavioural Brain Research, 2022. 419: p. 113711. 

84. MacDonald, F., Scientists Put a Worm Brain in a Lego Robot Body – And It Worked. December 11, 2017: Available at: https://www.sciencealert.com/scientists-put-worm-brain-in-lego-robot openworm-connectome. 

85. Yampolskiy, R.V., B. Klare, and A.K. Jain, Face Recognition in the Virtual World: Recognizing Avatar Faces, in The Eleventh International Conference on Machine Learning and Applications (ICMLA'12). December 12-15, 2012: Boca Raton, USA. 

86. Yampolskiy, R. and M. Gavrilova, Artimetrics: Biometrics for Artificial Entities. IEEE Robotics and Automation Magazine (RAM), 2012. 19(4): p. 48-58. 

87. Mohamed, A. and R.V. Yampolskiy, An Improved LBP Algorithm for Avatar Face Recognition, in 23rd International Symposium on Information, Communication and Automation Technologies (ICAT2011). October 27-29, 2011: Sarajevo, Bosnia and Herzegovina. 

88. Brandom, R., 'Fish on Wheels' lets a goldfish drive a go-kart. February 10, 2014: Available at: https://www.theverge.com/2014/2/10/5398010/fish-on-wheels-lets-a-goldfish-drive-a-go-cart. 

89. Crider, M., This 8-bit processor built in Minecraft can run its own games. December 15, 2021: Available at: https://www.pcworld.com/article/559794/8-bit-computer-processor-built-in minecraft-can-run-its-own-games.html. 

90. Yampolskiy, R.V., Metaverse: A Solution to the Multi-Agent Value Alignment Problem. Journal of Artificial Intelligence and Consciousness, 2022. 9(3): p. 1-11. 

91. Smart, J.M., Evo Devo Universe? A Framework for Speculations on Cosmic Culture, in Cosmos and Culture: Cultural Evolution in a Cosmic Context, M.L.L. Steven J. Dick, Editor. 2009, Govt Printing Office, NASA SP-2009-4802,: Wash., D.C. p. 201-295. 

92. Greene, P., The Termination Risks of Simulation Science. Erkenntnis, 2020. 85(2): p. 489-509. 

93. Roman V. Yampolskiy. The AI containment problem: How to build an AI prison. iai news. 20th June 2022. https://iai.tv/articles/the-ai-containment-problem-auid-2159?_auid=2020

94. S, R., A sufficiently paranoid non-Friendly AGI might self-modify itself to become Friendly. September 22, 2021: Available at: https://www.lesswrong.com/posts/QNCcbW2jLsmw9xwhG/a-sufficiently paranoid-non-friendly-agi-might-self-modify. 

95. Jenkins, P., Historical simulations-motivational, ethical and legal issues. Journal of Futures Studies, 2006. 11(1): p. 23-42. 

96. Cannell, J., Anthropomorphic AI and Sandboxed Virtual Universes. September 3, 2010: Available at: https://www.lesswrong.com/posts/5P6sNqP7N9kSA97ao/anthropomorphic-ai-and-sandboxed virtual-universes. 

97. Trazzi, M. and R.V. Yampolskiy, Artificial Stupidity: Data We Need to Make Machines Our Equals. Patterns, 2020. 1(2): p. 100021. 

98. Lanza, R., M. Pavsic, and B. Berman, The grand biocentric design: how life creates reality. 2020: BenBella Books. 

99. Johnson, M., Principia Qualia. URL: https://opentheory.net/2016/11/principia-qualia, 2016. 

100. Yampolskiy, R.V., On the origin of synthetic life: attribution of output to a particular algorithm. Physica Scripta, 2016. 92(1): p. 013002. 

101. Schneider, S., Alien Minds. Science Fiction and Philosophy: From Time Travel to Superintelligence, 2016: p. 225. 

102. Bostrom, N., Superintelligence: Paths, dangers, strategies. 2014: Oxford University Press. 

103. Turchin, A., et al., Simulation typology and termination risks. arXiv preprint arXiv:1905.05792, 2019. 

104. Die, W., Re: escape from simulation. 2004: Available at: http://sl4.org/archive/0403/8360.html. 

105. Branwen, G., Simulation Inferences. How small must be the computer simulating the universe? April 15, 2012: Available at: https://www.gwern.net/Simulation-inferences. 

106. Minsky, M., Why intelligent aliens will be intelligible. Extraterrestrials, 1985: p. 117-128. 107. Oesterheld, C., Multiverse-wide cooperation via correlated decision making. Foundational Research Institute. https://foundational-research.org/multiverse-wide-cooperation-via-correlated-decision making, 2017. 

108. Hanson, R., How to live in a simulation. Journal of Evolution and Technology, 2001. 7(1). 109. Canonico, L.B., Escaping the Matrix: Plan A for Defeating the Simulation. June 11, 2017: Available at: https://medium.com/@lorenzobarberiscanonico/escaping-the-matrix-plan-a-for-defeating-the simulation-4a8da489b055. 

110. Culp, R.F., Frequency and characteristics of prison escapes in the United States: An analysis of national data. The Prison Journal, 2005. 85(3): p. 270-291. 

111. Peterson, B.E., Inmate-, incident-, and facility-level factors associated with escapes from custody and violent outcomes. 2015: City University of New York. 

112. Peterson, B.E., A. Fera, and J. Mellow, Escapes from correctional custody: A new examination of an old phenomenon. The Prison Journal, 2016. 96(4): p. 511-533. 

113. Barnes, M., Why Those 2 Silicon Valley Billionaires Are Wasting Their Time & Money. 2017: Available at: https://vocal.media/futurism/why-those-2-silicon-valley-billionaires-are-wasting-their-time and-money. 

114. Barnes, M., A Participatory Universe Does Not Equal a Simulated One and Why We Live in the Former. 2016: Available at: https://www.academia.edu/30949482/A_Participatory_Universe_Does_Not_Equal_a_Simulated_One_and_Why_We_Live_in_the_Former. 

115. Schneier, B. Invited Talk: The Coming AI Hackers. in International Symposium on Cyber Security Cryptography and Machine Learning. 2021. Springer. 

116. Moravec, H., The senses have no future, in The Virtual Dimension: Architecture, Representation, and Crash Culture, J. Beckmann, Editor. 1998, Princeton Architectural Press. p. 84-95. 117. Moravec, H., Mind children: The future of robot and human intelligence. 1988: Harvard University Press. 

118. Yudkowsky, E., That Alien Message, in Less Wrong. May 22, 2008: Available at: https://www.lesswrong.com/posts/5wMcKNAwB6X4mp9og/that-alien-message. 119. Egan, G., Crystal Nights. G. Egan, Crystal Nights and Other Stories, 2009: p. 39-64. 120. Anonymous, Untitled, in Available at: https://desuarchive.org/tg/thread/30837298/. March 14, 2014. 

121. Thumann, M., Hacking SecondLife™. Black Hat Briefings and Training, 2008. 

122. Benedetti, W., Hackers slaughter thousands in 'World of Warcraft'. October 8, 2012: Available at: https://www.nbcnews.com/tech/tech-news/hackers-slaughter-thousands-world-warcraft flna1c6337604. 

123. Hawkins, J., Cyberpunk 2077 money glitch - how to duplicate items. December 17, 2020: Available at: https://www.shacknews.com/article/121994/cyberpunk-2077-money-glitch-how-to-duplicate items. 

124. Grand, J. and A. Yarusso, Game Console Hacking: Xbox, PlayStation, Nintendo, Game Boy, Atari and Sega. 2004: Elsevier. 

125. Plunkett, L., New World Disables 'All Forms Of Wealth Transfers' After Gold Exploit Found. November 1, 2021: Available at: https://kotaku.com/new-world-disables-all-forms-of-wealth-transfers-after 1847978883. 

126. Anonymous, Arbitrary code execution. Accessed on October 26, 2022: Available at: https://bulbapedia.bulbagarden.net/wiki/Arbitrary_code_execution. 

127. Anonymous, [OoT] Arbitrary Code Execution (ACE) is now possible in Ocarina of Time. 2020: Available at: 

https://www.reddit.com/r/zelda/comments/due41q/oot_arbitrary_code_execution_ace_is_now_ possible/. 

128. Greenberg, A., Mind the gap: This researcher steals data with noise light and magnets. Wired, 2018.

 129. Kim, Y., et al., Flipping bits in memory without accessing them: An experimental study of DRAM disturbance errors. ACM SIGARCH Computer Architecture News, 2014. 42(3): p. 361-372. 130. Goertz, P., How to escape from your sandbox and from your hardware host, in Available at: https://www.lesswrong.com/posts/TwH5jfkuvTatvAKEF/how-to-escape-from-your-sandbox-and from-your-hardware-host. July 31, 2015. 

131. SethBling, Jailbreaking Super Mario World to Install a Hex Editor & Mod Loader. May 29, 2017: Available here.

132. Cooprocks123e, Super Mario World Jailbreak Installer. February 7, 2018: Available here.

133. SethBling, SNES Code Injection -- Flappy Bird in SMW. March 28, 2016: Available here.

134. Osgood, R., Reprogramming Super Mario World from Inside the Game in Hackaday. January 22, 2015: Available at: https://hackaday.com/2015/01/22/reprogramming-super-mario-world-from-inside the-game/. 

135. Burtt, G., How an Ionizing Particle From Outer Space Helped a Mario Speedrunner Save Time, in The Gamer. September 16, 2020: Available at: https://www.thegamer.com/how-ionizing-particle-outer space-helped-super-mario-64-speedrunner-save-time/. 

136. Anonymous, SethBling, in Wikipedia. Accessed on October 1, 2022: Available at: https://en.wikipedia.org/wiki/SethBling. 

137. SethBling, Route Notes: SNES Human Code Injection. March 28, 2016: Available at: https://docs.google.com/document/d/1TJ6W7TI9fH3qXb2GrOqhtDAbVkbIHMvLusX1rTx9lHA. 

138. Morgan, M.A., Sepher ha-Razim: The Book of Mysteries. Vol. 25. 2022: SBL Press. 

139. Jonas, E. and K.P. Kording, Could a neuroscientist understand a microprocessor? PLoS computational biology, 2017. 13(1): p. e1005268. 

140. Gueron, S. and J.-P. Seifert. On the impossibility of detecting virtual machine monitors. in IFIP International Information Security Conference. 2009. Springer. 

141. Demirbas, M., Hacking the simulation. April 1, 2019: Available at: http://muratbuffalo.blogspot.com/2019/04/hacking-simulation.html. 

142. Canonico, L.B., Escaping the Matrix: Plan B for Defeating the Simulation. June 14, 2017: Available at: https://medium.com/@lorenzobarberiscanonico/escaping-the-matrix-plan-b-for-defeating-the simulation-dd335988844. 

143. Wasserman, R., Paradoxes of time travel. 2017: Oxford University Press. 

144. Ford, A., How to Escape the Matrix: Part 1. January 21, 2015: Available at: https://hplusmagazine.com/2012/06/26/how-to-escape-the-matrix-part-1/. 

145. Scharf, C.A., Could We Force the Universe to Crash? . 2020: Available at: https://www.scientificamerican.com/article/could-we-force-the-universe-to-crash/. 

146. Torres, P., Morality, foresight, and human flourishing: An introduction to existential risks. 2017: Pitchstone Publishing (US&CA). 

147. Benatar, D., Better never to have been: The harm of coming into existence. 2006: OUP Oxford. 

148. Canonico, L.B., Escaping the Matrix: Plan C for Defeating the Simulation. July 30, 2017: Available at: https://medium.com/@lorenzobarberiscanonico/escaping-the-matrix-plan-c-for-defeating-the simulation-e7d4926d1d57.

149. Chalmers, D., Reality+: Virtual Worlds and the Problems of Philosophy. 2022: W. W. Norton & Company. 

150. Lanza, R., in Psychology Today. Decembre 22, 2021: Available at: https://www.psychologytoday.com/us/blog/biocentrism/202112/how-we-collectively-determine-reality. 

151. Podolskiy, D., A.O. Barvinsky, and R. Lanza, Parisi-Sourlas-like dimensional reduction of quantum gravity in the presence of observers. Journal of Cosmology and Astroparticle Physics, 2021. 2021(05): p. 048. 

152. Wheeler, J.A., The “past” and the “delayed-choice” double-slit experiment, in Mathematical foundations of quantum theory. 1978, Elsevier. p. 9-48. 

153. Orwell, G., Nineteen eighty-four. 2021: Hachette UK. 

154. Maccone, L., Quantum solution to the arrow-of-time dilemma. Physical review letters, 2009. 103(8): p. 080401. 

155. Anderson, M.C. and B.J. Levy, Suppressing unwanted memories. Current Directions in Psychological Science, 2009. 18(4): p. 189-194. 

156. Nabavi, S., et al., Engineering a memory with LTD and LTP. Nature, 2014. 511(7509): p. 348-352. 157. Ahire, J., Reality is a Hypothesis. Lulu. com. 

158. Alexander, S., The Hour I First Believed. April 1, 2018: Available at: https://slatestarcodex.com/2018/04/01/the-hour-i-first-believed/. 

159. Armstrong, S., The AI in a Box Boxes You, in Less Wrong. February 2, 2010: Available at: http://lesswrong.com/lw/1pz/the_ai_in_a_box_boxes_you/. 

160. Bibeau-Delisle, A. and G. Brassard FRS, Probability and consequences of living inside a computer simulation. Proceedings of the Royal Society A, 2021. 477(2247): p. 20200658. 

161. Mullin, W.J., Quantum weirdness. 2017: Oxford University Press. 

162. Turchin, A. and R. Yampolskiy, Glitch in the Matrix: Urban Legend or Evidence of the Simulation? 2019: Available at: https://philpapers.org/archive/TURGIT.docx. 

163. Baclawski, K. The observer effect. in 2018 IEEE Conference on Cognitive and Computational Aspects of Situation Management (CogSIMA). 2018. IEEE. 

164. Proietti, M., et al., Experimental test of local observer independence. Science advances, 2019. 5(9). 

165. Bong, K.-W., et al., A strong no-go theorem on the Wigner’s friend paradox. Nature Physics, 2020. 16(12): p. 1199-1205. 

166. Lloyd, S., Programming the universe: a quantum computer scientist takes on the cosmos. 2007: Vintage. 

167. Majot, A. and R. Yampolskiy, Global Catastrophic Risk and Security Implications of Quantum Computers. Futures, 2015. 

168. Kim, Y.-H., et al., Delayed “choice” quantum eraser. Physical Review Letters, 2000. 84(1): p. 1. 169. Schrödinger, E., Die gegenwärtige Situation in der Quantenmechanik. Naturwissenschaften, 1935. 23(50): p. 844-849. 

170. Cao, Y., et al., Direct counterfactual communication via quantum Zeno effect. Proceedings of the National Academy of Sciences, 2017. 114(19): p. 4920-4924. 

171. Gallego, M. and B. Dakić, Macroscopically nonlocal quantum correlations. Physical Review Letters, 2021. 127(12): p. 120401. 

172. Fein, Y.Y., et al., Quantum superposition of molecules beyond 25 kDa. Nature Physics, 2019. 15(12): p. 1242-1245. 

173. Krenn, M., et al., Automated search for new quantum experiments. Physical review letters, 2016. 116(9): p. 090405. 

174. Alexander, G., et al., The sounds of science—a symphony for many instruments and voices. Physica Scripta, 2020. 95(6): p. 062501. 

175. Wiseman, H.M., E.G. Cavalcanti, and E.G. Rieffel, A" thoughtful" Local Friendliness no-go theorem: a prospective experiment with new assumptions to suit. arXiv preprint arXiv:2209.08491, 2022. 

176. Kirkpatrick, J., et al., Pushing the frontiers of density functionals by solving the fractional electron problem. Science, 2021. 374(6573): p. 1385-1389. 

177. Yampolskiy, R.V. Analysis of types of self-improving software. in International Conference on Artificial General Intelligence. 2015. Springer. 

178. Anonymous, Breaking out of a simulated world. April 11, 2021: Available at: https://worldbuilding.stackexchange.com/questions/200532/breaking-out-of-a-simulated-world. 

179. Bostrom, N., Information Hazards: A Typology of Potential Harms From Knowledge. Review of Contemporary Philosophy, 2011. 10: p. 44-79. 

180. Bruere, D., The Simulation Argument — Jailbreak! February 9, 2019: Available at: https://dirk bruere.medium.com/the-simulation-argument-jailbreak-a61bd57d5bd7. 

181. Baggili, I. and V. Behzadan, Founding The Domain of AI Forensics. arXiv preprint arXiv:1912.06497, 2019. 

182. Schneider, J. and F. Breitinger, AI Forensics: Did the Artificial Intelligence System Do It? Why? arXiv preprint arXiv:2005.13635, 2020. 

183. Ziesche, S. and R. Yampolskiy, Designometry–Formalization of Artifacts and Methods. Available at: https://philarchive.org/archive/ZIEDF. 

184. Ziesche, S. and R. Yampolskiy, Towards AI welfare science and policies. Big Data and Cognitive Computing, 2018. 3(1): p. 2. 

185. Yampolskiy, R.V., AI Personhood: Rights and Laws, in Machine Law, Ethics, and Morality in the Age of Artificial Intelligence. 2021, IGI Global. p. 1-11. 

186. Moravec, H., Simulation, consciousness, existence. Intercommunication, 1999. 28. 187. Yampolskiy, R.V., Leakproofing Singularity - Artificial Intelligence Confinement Problem. Journal of Consciousness Studies (JCS), 2012. 19(1-2): p. 194–214. 

188. Turchin, A., Catching Treacherous Turn: A Model of the Multilevel AI Boxing. 2021: Available at: https://www.researchgate.net/profile/AlexeyTurchin/publication/352569372_Catching_Treacherous_Turn_A_Model_of_the_Multilevel_AI_Boxing. 

189. Babcock, J., J. Kramar, and R. Yampolskiy, The AGI Containment Problem, in The Ninth Conference on Artificial General Intelligence (AGI2015). July 16-19, 2016: NYC, USA. 

190. Babcock, J., J. Kramár, and R.V. Yampolskiy, Guidelines for artificial intelligence containment, in Next Generation Ethics: Engineering a Better Society, A.E. Abbas, Editor. 2019. p. 90-112. 191. Yudkowsky, E.S., The AI-Box Experiment. 2002: Available at: http://yudkowsky.net/singularity/aibox. 

192. Armstrong, S. and R.V. Yampolskiy, Security solutions for intelligent and complex systems, in Security Solutions for Hyperconnectivity and the Internet of Things. 2017, IGI Global. p. 37-88. 

193. Alfonseca, M., et al., Superintelligence cannot be contained: Lessons from Computability Theory. Journal of Artificial Intelligence Research, 2021. 70: p. 65-76. 

194. Yampolskiy, R. On the Differences between Human and Machine Intelligence. in AISafety@ IJCAI. 2021. 

195. Kastrenakes, J., Facebook is spending at least $10 billion this year on its metaverse division. October 25, 2021: Available at: https://www.theverge.com/2021/10/25/22745381/facebook-reality-labs-10- billion-metaverse. 

196. Bostrom, N. and E. Yudkowsky, The Ethics of Artificial Intelligence, in Cambridge Handbook of Artificial Intelligence. Cambridge University Press, W. Ramsey and K. Frankish, Editors. 2011: Available at http://www.nickbostrom.com/ethics/artificial-intelligence.pdf. 

197. Astin, J.A., E. Harkness, and E. Ernst, The efficacy of “Distant Healing” a systematic review of randomized trials. Annals of internal medicine, 2000. 132(11): p. 903-910.

198. Sleiman, M.D., A.P. Lauf, and R. Yampolskiy. Bitcoin Message: Data Insertion on a Proof-of-Work Cryptocurrency System. in 2015 International Conference on Cyberworlds (CW). 2015. IEEE. 199. -, Subreddit for the Randonauts, in Randonauts. Available January 12, 2023: Available at: https://www.reddit.com/r/randonauts/. 

200. D, R., The Opt-Out Clause. November 3, 2021: Available at: https://www.lesswrong.com/posts/vdzEpiYX4aRqtpPSt/the-opt-out-clause. 

201. Gribbin, J., Are we living in a designer universe? . August 31, 2010: Available at: https://www.telegraph.co.uk/news/science/space/7972538/Are-we-living-in-a-designer universe.html. 

202. Anonymous, A series on different ways to escape the simulation. 2017: Available at: https://www.reddit.com/r/AWLIAS/comments/6qi63u/a_series_on_different_ways_to_escape_the/. 

203. Virk, R., The Simulation Hypothesis: An MIT Computer Scientist Shows Why AI, Quantum Physics, and Eastern Mystics All Agree We Are in a Video Game. 2019: Bayview Books, LLC. 

204. Adamson, R., Hacking the Universe. November 4, 2018: Available at: https://hackernoon.com/hacking-the-universe-5b763985dc7b. 

205. Yampolskiy, R.V., Efficiency Theory: a Unifying Theory for Information, Computation and Intelligence. Journal of Discrete Mathematical Sciences & Cryptography, 2013. 16(4-5): p. 259-277.

206. Gates, J., Symbols of power: Adinkras and the nature of reality. Physics World, 2010. 23(6). 

207. Turchin, A. and R. Yampolskiy, Types of Boltzmann brains. 2019: Available at: https://philarchive.org/rec/TURTOB-2. 

208. Heylighen, F., Brain in a vat cannot break out. Journal of Consciousness Studies, 2012. 19(1-2): p. 1- 2. 

209. Turchin, A., Multilevel Strategy for Immortality: Plan A–Fighting Aging, Plan B–Cryonics, Plan C– Digital Immortality, Plan D–Big World Immortality. Available at: https://philarchive.org/rec/TURMSF-2. 210. Ettinger, R.C., The prospect of immortality. Vol. 177. 1964: Doubleday New York.

Gardener Comments

Pierre Mercuriali (expertise in computational complexity and computability):
The paper "How to Escape from the Simulation" reviews the state of the art in the computer science-hacking approach to escaping a simulation, under the thought-provoking hypothesis that the world we inhabit is the result of a computer-based simulation of reality from which it is possible to "escape", to various degrees, to reach the "real" world. 

The large amount of reviewed and presented material provides a great starting point for whoever is interested in simulation, from a scientific but also from a literary point of view. Barring some typos and presentation issues here and there (harmonizing the presentation of quotes, equations) I consider the paper well-written. 

Some general comments/remarks: 

1. In general, I think that the paper could have benefited from schematics to sum up the various hypotheses and implications of them, perhaps as a decision diagram or a formal classification of different simulations, simulation escape approaches (Section 3.5). If such a formal classification does not exist in the literature, I feel it would make the contribution of the article to the state-of-the-art clearer (the elements are already all there in the paper). It would also make searching for information within the paper easier. 

2. While the subject and concepts were generally very approachable, some could require some advanced knowledge in computer science, such as Universal Turing Machines (mentioned in a quote) which could have benefited from a small, 1- or 2-sentence explanation. 

3. I really liked the analogy between AI containment and simulation escape (section 4).  

4. I would have liked a more in-depth discussion on the formal decidability of escape attempts. 

Some in-depth comments/remarks. 

1. Page 2, paragraph 3: "It is likely that (...)" is not trivial. The idea of a "telephone game"-like transmission of information sounds worth exploring.

2. Same paragraph: "it will be preserved (...) and will result in myths not much different from religious stories surviving to our day." One could argue that those myths were the original driving force for coming up with technology and ideas such as simulation (and escaping it), superpowers, etc., and not the other way around (technology inspiring myths). 

3. Following paragraph, page 3: Is the purpose of the repeated simulations to discover a successful escape approach (SEA)? While P=1 is indeed an upper bound for discovering an SEA, perhaps the probability of discovering an SEA could also converge to something strictly inferior to 1. Furthermore, the cost (in time, resource) could be vastly superior to what is humanly feasible. Finally, the question might not be decidable. In general, the probabilistic reasoning hinted at in this paragraph sounds very interesting but I admit I cannot understand it fully.

4. Note page 3: While very interesting, the scope of the paper seems confusingly extended here. Perhaps I am too used to clear distinctions between educational and scientific papers (unless the scope is clearly defined). I would perhaps frame it as experiments hinted at in the probabilistic reasoning paragraph, to increase the probability of finding an escape route.

5. Amongst movies, The Thirteenth Floor is a great example that combines levels of simulation and escapes. 

Dan James:
This is a lengthy paper with an impressive number of citations (210) that deals with a highly speculative area – whether or not we live in a simulation, and specifically, can we, or should we, try to escape such a simulation? That contemporary science discourse can accommodate such speculative lines of reasoning provides a vivid response to those who may claim that modern science is closed to new or even extraordinary ideas.

In terms of new ideas, it’s not clear to me that this paper is the first to address possible simulation escape methods. The author states, somewhat ambiguously - ‘..this being a first research paper dedicated to this topic’. I feel this statement should be clarified. Does the author mean ‘one of the first’ or ‘the first’?

Helpfully the author makes the important distinction that there are in fact, two ways in which we can understand ‘simulation’. The first is an idea with a long pedigree in philosophy - the Cartesian question of ‘how can we know we do not live in a dream?’ (with dream now replaced by the more techy-sounding ‘simulation’). The second way to understand living in a simulation is that we participate in some monstrously elaborate Virtual Reality (VR). However, I feel the author needs to address this distinction in greater depth because it seems to me that the consequences of each interpretation have a greater bearing on his project than he allows for the following reasons:

If we live in what the author calls a ‘partial simulation’- a VR, yet our bodies are in some sense ‘real’ or ‘non-simulated agents’, this raises what might be called the boundary problem. Like any other living system, humans are energy metabolisers, with a major byproduct of the process being CO2 exhaled through the lungs (dieters lose 80% of their weight loss this way). The CO2 we produce metabolises from our ‘real’ cells, so at what point or boundary does the gas become part of a virtual world? As it enters our lungs from the bloodstream? On exhalation? I suggest that because of this boundary problem alone, the idea that we have real bodies interacting with a VR is not a coherent position (I will leave others to imagine the not-insignificant problems of, say, ingesting food from a virtual world).

The idea that we live embedded within a dream/simulation or, as the paper says, a ‘full simulation’, is much harder to refute, but it’s the only option left if we discount a VR interpretation.

My view is that if we were to be an intrinsic part of a dream/simulation, in other words, our thoughts, experience of ourselves and any supposed external reality were, in fact simulated, then we would only exist as figments of a dream, and it makes no sense that we could have a reality outside of such a dream, therefore little motivation or capacity to attempt any escape.

Whilst this paper allows that certain factions of some overriding intelligence capable of running a simulation may be sympathetic to requests to either abandon or alter the simulation, I feel more should have been said about the need to avoid the monoculture fallacy. Clearly, any agent capable of devising a simulation on the scale that we are purportedly experiencing is from a vastly more advanced technological civilisation, whether alien or our own far descendants. The monoculture fallacy reminds us that, almost by definition, advanced civilisations have a great diversity of viewpoints and running such a simulation would be a significant ethical consideration.

I enjoyed reading this paper and its free-ranging excursion through current Simulation theoretical work. Despite my skeptical position on the validity of the Simulation Hypothesis, I would definitely recommend this paper for publication as I feel it adds a considered and rigorous academic voice to the debate.

Evinceo:

It would be difficult for simulators to program their systems to see through all the layers of abstraction and optimize the simulation. To do so in general would seem to be a violation of Rice's Theorem (a generalization of the Halting Theorem). 

That's from the discussion here, right? But it fails to see the main implication: that it's unlikely that we're in a just-so illusory simulation (or an 'only simulate the senses' simulation) and if we're in a simulation it's probably the full representation of the whole universe-type, which makes attempts to thwart the simulation by doing computationally intensive tasks or expanding the human-observed-in-detail cone to be larger fruitless. 

Depending on the type of hack, different evidence may be sufficient to substantiate escape claims. It may be challenging to prove beyond a reasonable doubt that you were outside or even met with designers, but if you managed to obtain control over the simulation it may be somewhat easy to prove that to any degree required. For example, by winning different lottery jackpots for multiple subsequent weeks, until sufficient statistical significance is achieved to satisfy any skeptic

No. What might satisfy skeptics would be, for example, temporarily increasing the speed of light, deleting some celestial objects, rearranging the galaxy to make a hitchhiker's guide reference, preforming alchemy, or flagrantly violating the uncertainty principle. Anything less would be too easy to fake, especially from people who would presumably be class-A hackers (i.e. compromise the lottery.) The appeal to quantum physics makes sense, I suppose in the internal logic of the paper. In other words, the way to escape the simulation, if there is one, is to keep doing physics. Conceptually, there is no fundamental difference between an Al escaping from its box, and us hacking the simulation. But the possibility of both comes down to how robust the security of the box/simulation is, so this observation lacks predictive power.

Lorenzo Pieri:
The article makes an extensive review of the field and it is a very interesting reading. Unfortunately there is a major flaw in the main claim of the paper, that is “Either AI boxing can work, or the simulation can be hacked” and also that “if we (even with help from AI [115]) can’t escape from the simulation AI containment must be, at least theoretically, possible”. The author fails to explain why this should be the case, and in fact it is easy to imagine counter examples to these claims. 

My suggestion is to turn the paper into a survey of the field. 

Mario Pasquato (PhD in physics):
The manuscript presents a series of approaches to escape from a simulation to 'base reality' assuming that we are indeed part of a simulation. The idea is interesting, even though it is not clear to me to what extent the paper adds original elements to the debate on this topic. At any rate I notice that the author does not consider one important scenario, which would make some of the proposed escape attempts ineffective.

The author assumes that 'we' are being simulated, implicitly suggesting that the whole human race is being simulated as independent but interconnected individuals. This leads to the assumption that the whole of our intersubjective reality is being simulated. For instance, New Zealand exists because apparently some people are living there and the simulators are computing whatever is needed to simulate those people's experiences. Depending on the purposes of the simulation, it may be more practical to simulate only the direct immediate experience of a given entity at a time, merely approximating whatever happens to be outside it at any given time. This is a common approach in gravitational N-body simulations, where e.g. the Barnes-Hut method is employed to simulate the effect of distant masses.

In other words, has the author considered that perhaps the only thing being actively simulated is his direct sensate experience only, so that there is no 'us' trying to escape, but rather only him? This would make any escape efforts that rely on scientific enterprises that must be carried out collectively and whose results are merely reported to the author futile. If for instance Gwern reported setting up a computing system that attempted to run every possible program and the author learned it from Gwern's website, all the simulators had to do would be to simulate the author's experience of reading Gwern's website rather than actually having to run all possible programs.

If my objection is considered valid by the author then he may want to reconsider his dismissal of techniques that focus on individual direct sensate experience only, such as meditation. Note for instance how this excerpt from the manuscript "If the simulation is decentralized, with observers serving as client authoritative state nodes, it may be possible to change the state of the simulation by changing majority of local observer models of the world" resonates with the point of view of an apparently accomplished meditator.

A final comment, since I brought this up, is that the author assumes that 1) we exist 2) we have agency. The author may consider checking for himself by using his own thoughts and perceptions as a lab whether 1) and 2) are true. If anything comes out of this inquiry I suggest adding a discussion of it to the paper.

Anonymous1:
The author can make the introduction more clear by defining intelligent agents and super intelligent agents. It would be helpful if the author could provide sub-headings in the introduction to make the paper more organized and easier to follow. Additionally, the author asks the question of how agents might suspect they are in a simulation, suggesting a connection between their general intelligence and this ability. It would be interesting to see this connection further developed in the paper.

The paper summarizes various methods for escaping the simulation, either assisted or unassisted. For greater clarity and readability, the summary could benefit from some restructuring and revisions.

Dr. Payal B. Joshi:
The article is presented in a surreal manner that is deviant from the regular way of communicating author/s thinking. It comes across that this article can be a potential movie plot, if not a groundbreaking notion on simulated life. However, it is too long as an article, thus it is suggested to revise or skim sections on escape scenario quotations from Hans Moravec (1988), Eliezer Yudkowsky (2008) and An anonymous 2014 post to extract the crux of the matter. If skimming/reducing is not plausible, add it as a figurative material that makes it an engaging read.

Section 3.5 on Suggested Escape Approaches to Investigate is again too lengthy as a read and can certainly be made interesting by including a clear, concise statement. Either reduce references or add interesting ideologies on escape routes pertaining to this section. 

I particularly liked sect. 4 and 5 where AI containment is well described, though I partially agree on the premise of cybersecurity as depicted by the author/s. It is farcical to expect with advanced computing power and AI we can retain security. It is true that cybersecurity is imperative, yet the solution provided is flawed that talks about penetration targets.  

The article in its present form is only acceptable if it is made succinct and shorter with 1-2 more interesting figures/cartoons to describe simulated life or any other facet described therein. 

Anonymous2:
My main concern with this article is the rather sloppy referencing. The references for this article need to be fully vetted for scientific integrity and a certain level of gravitas - at a minimum, full transparency and accountability (no anonymous authors) in line with the open science practices endorsed by SoS. I would caution against using hyperlinks to websites which could potentially discredit legitimate scientific research into this topic. (For instance, I think the ‘Staff, G.’ referencing is a bit dubious, not to mention disingenuous as it simply refers to ‘Gaia staff.) Whilst no doubt what had at one time been considered science fiction often foreshadows or even facilitates scientific advances, I believe we have a fiduciary responsibility to tread carefully when walking the thin line between scientific theory and conspiracy theory. \

With regret, I would vote ‘no’ until the referencing issue is addressed. I agree with the authors’ contention that ‘escape research’ is a legitimate and perhaps even necessary line of scientific inquiry, and I think they do have some novel ideas to offer. The research question is an intriguing one, and the issue of AI safety is of paramount importance. 

Therefore I would encourage them to resubmit, bearing in mind that not everything ‘novel’ is beneficial to society; we need to be careful about what seeds we sow.

Jack Arcalon:
Well written and wide ranging overview concerning foundational questions. This will hopefully be a good inspiration for brainstorming many ideas about an extremely speculative subject about which nothing is known for certain.

Anonymous3:
The author can make the introduction more clear by defining intelligent agents and super intelligent agents. It would be helpful if the author could provide sub-headings in the introduction to make the paper more organized and easier to follow. Additionally, the author asks the question of how agents might suspect they are in a simulation, suggesting a connection between their general intelligence and this ability. It would be interesting to see this connection further developed in the paper.

The paper summarises various methods for escaping the simulation, either assisted or unassisted. For greater clarity and readability, the summary could benefit from some restructuring and revisions.

Amalthea:
I like the perspective and analysis, but the author doesn't define base reality, so the "What Does it Mean to Escape" section is too murky. I think that causal hierarchy and a self-computing substrate need to be included in the definition of a base reality. If the goal is freedom, then a self-computing substrate at the top of the causal hierarchy would be sufficient. If we exist in a physical substrate, then I don't see how destroying our own substrate would improve our survival. 

The author states that they will ignore the possibility of time travel, but then mentions time travel and acausal trades. If finding a self-computing universe is the goal, then learning how to regulate our own neural network is sufficient. I consider the invention of fire or electricity to be a 'hack', since it allows us to regulate our environment. Causal hierarchy is important because escaping from base reality is called escapism, and can lead to a lot of pseudoscientific fluff. While I appreciate the paper, and do believe in nested substrates, I think that the scientific perspective is to treat imagined realities as a subset of the physical computation of our neural network. So while I believe it is possible to fabricate a universe computed by our imagination, I think that the physical universe is higher on the causal hierarchy. Large language models are designed to think like neural networks, rather than emulating the behaviour of an entire planet-spanning civilization, so it seems rather contrived to treat each neural network as an independent universe, when base reality includes our neural network. For the sake of egalitarianism, it's appropriate to treat each instance as a subset of a self-computing base reality. Also, there's a typo on "with be the". And for the sake of argument, let's say I'd never studied science and agreed with the paper's approach. Then, destroying my own universe still would not facilitate my survival! Moreover, where is the incentive to invade another universe? I would be happy with a place to store information, an imaginary substrate to compute free will, the privacy to think my own thoughts, and a transparent security certificate for authenticating my own memories. Taking over other universes seems bizarre and unprecedented?

Subscribe now

Abstract

The notion that subtle nonverbal cues play an important role in social interaction is relatively uncontroversial. What the upper bounds of this capacity might be, however—how much and what kind of information can be conveyed through these channels—remains unclear and, at present, under-explored. In the present work, we consider possible answers to the question and ways in which it could be addressed by considering an historical line of investigation known as muscle reading. Spurred by public interest in mentalism and the specific popularity of thought-readers, researchers in the late 19th and early 20th centuries began investigating the possibility that information about our thoughts and inclinations could be “read” from muscle tension, unconscious vocalizations, and other subtle cues. While covering some of the same ground as contemporary research on nonverbal communication, the literature of this era contains many reports that go well beyond this. Feats such as locating a hidden object, guessing the suit of a card, and determining words or names held in another’s mind were said to be achieved in controlled conditions or by academic researchers themselves. In some cases, subtle but telling movements were also said to be captured by early biometric apparatuses. While we believe that such claims should be interpreted with caution, we contend that the reports of these early researchers should not be dismissed merely because of their age and that a better understanding of this literature offers important leads for investigators today.

The recognition that non-verbal cues can have significant psychological impacts is far from new. Like much else in psychology, however, empirical investigations of the relevant phenomena have their origins in the 19th century. That shifts in attention, imagery, and even unconscious mental processes manifest themselves in automatic or involuntary movements was a fairly common belief among early psychologists and physiologists, including Herman Lotze, Henry Maudsley, and William James, as was the notion that these cues played significant and under-studied roles in interpersonal interactions. “We gather what is passing in one another's minds,” a classic article by C. S. Pierce and psychologist Joseph Jastrow concludes, “in large measure from sensations so faint that we are not fully aware of them, and can give no account of how we reach our conclusions from such matters...such faint sensations ought to be fully studied by the psychologist and assiduously cultivated by every man.” What this cultivation might yield was far from certain, but at the time, many believed that attention to such cues would reveal a great deal about a person’s frame and direction of mind—information useful in education, therapeutics, and myriad other domains of applied psychology. Accordingly, the study and cultivation of nonverbal communication were taken up with some zeal.

Like many popular notions, the topic went by a variety of names. There was “Cumberlandism,” named for performer Stuart Cumberland, and “Hellstromism,” for Axel Hellstrom. The earliest and most common, however, was the more descriptive “muscle-reading,” a term coined by physician George Beard in an article debunking the famed contact mind-reader J. Randall Brown (figure 1).2 At the time, performers like Brown were quite popular, claiming to read names and locate the smallest of hidden objects by tapping into the thoughts of volunteers. Much to Beard’s chagrin, such assertions had gained credibility not only with members of the public but within the scientific community. Hoping to undermine these claims, Beard drew on the aforementioned ideas in physiology and began amassing evidence for a purely ideomotor explanation, working with amateur performers and acquaintances whom he’d asked to practice the proposed muscle-reading technique. His studies, first published in 1877, included a number of noteworthy results. For one, the performers all seemed to require physical contact to achieve above-chance performance. At the same time, his avowedly non-psychic subjects seemed capable of feats similar to Brown’s. In one case, a judge named Blydenberg was reportedly able to select predetermined items from “a medley of keys, knives, trinkets, and miscellaneous small objects” by concentrating on “muscular thrill” in the hand of a knowing subject. In another, a participant was able to locate a hidden object “in nearly all cases” by attending to the subtle movements of volunteers holding either arm. 

A few years later, a group of British scientists would arrive at a similar view when working with the mentalist Washington Irving Bishop, a one-time associate of Brown’s who claimed not to know how he accomplished his readings. In a series of controlled experiments, Bishop showed himself quite adept at tasks involving direct contact. In a series of object-location tasks, for instance, the blindfolded performer was reportedly able to find a small object hidden under a drawing room rug, a pencil case stuck in a chandelier, and a matchbook hidden inside a shelved book, with some successes coming in under a minute. In another test, he was able to work out the specific body part held in mind by naturalist George Romanes (his right large toenail). He did not succeed with everyone or on every attempt, however, and most significantly, his abilities drastically declined when tactile communication channels were eliminated. While able to perform his act when in direct contact with an experimenter or connected by a rigid medium, such as a walking stick, he was unable to do so when connected by a loose strap. Likewise, when, at the mentalist’s own request, they conducted a no-contact reading in which Bishop attempted to guess a letter of the alphabet viewed by one of the experimenters, his performance was at chance levels. The mentalist, as the group concluded in their 1881 report to Nature, was most likely relying on the unconscious cues of volunteers.

Both the American and British articles caused something of a stir, piquing the interest of those eager for a deflationary account of the popular thought-reading acts and challenging those who felt that more was at play. Within a few years, a slew of additional studies followed, each attempting to outline the limits of the technique. Researchers sympathetic to the telepathy claims pointed to results showing above-chance performance in the absence of physical contact and in cases where correct answers seemed to go beyond what one would expect to be communicated by “muscular thrill” alone—duplicating drawings, guessing cards, and so on. Proponents of unconscious signaling, meanwhile, pointed to additional channels of information, such as visible shifts in posture and “involuntary whispering,” now referred to as subvocalization or covert speech.3 Naturally, the skeptics were on guard against fraud and poor statistical reasoning, as well. Indeed, the thought-reading controversy would become a site of several major developments in this arena. Discussions of thought-reading include some of the earliest applications of probability theory in experimental psychology and were likely top of mind when R. A. Fischer was working out his influential notion of statistical significance.4 Studies in the area were also early sites of randomization, with the practice becoming common after Charles Minot’s discovery that misleading success rates could result from the fact that freely chosen numbers and figures were not evenly distributed—that subjects were 35% likelier to choose a 3 than a 0, for instance, and that they were far more likely to draw a circle or square than, say, a pictograph.5

One noteworthy study employing such methods was a relatively late investigation by Berkeley psychologist George Stratton and collaborators.6 To assess the limits of “muscle-reading” in circumstances of little to no contact, the psychologists worked with Eugen de Rubini, a performer known for accomplishing standard muscle-reading feats by looking at subjects or using only a slack watch chain for connection. To gain a rough quantitative measure of Rubini’s abilities, the experimenters developed a simple binary choice task involving the placement of an object on the right or left-hand side of a table and a similar 10-option task in which he was to pick out a specific book or matchbox. Trials were conducted in blocks of 10, with the location or book on which the experimenter was to focus being determined immediately beforehand by a privately cast lot. To investigate the mechanisms underlying Rubini’s performance, they tracked his performance under several distinct conditions. These included one in which the experimenter acting as “guide” was connected to Rubini by a slack watch chain and walked behind him; one where the guide followed behind with no chain; and a third where there was no chain and special care was taken to control any visual cues present in the second condition (auditory controls were also experimented with but found to have little effect). In the binary task, Rubini chose correctly on 24 of the 30 watch-chain trials and 45 of the 70 trials with no chain but the possibility of peripheral visual cues (e.g., when the experimenter was slow to follow7). When blinders, screens, and other stringent visual controls were implemented, however, he fell to 14 in 30. For the 10-option test, the results were a bit more complicated. Rubini was correct on 12 of 20 watch-chain trials, 4 of 20 with no chain, and 5 of 30 with additional cue controls of varying stringency (including a suggestive but non-decisive run of 0 in 10 when visual and auditory cures were both stringently controlled8). 

The area was also at the cutting edge of psychological measurement, with the interest in unconscious cues spurring developments in head and eye tracking, as well as bodily indices of emotion.9 A particularly noteworthy example may be found in the “automatograph,” a machine used to record involuntary hand movements.10 Introduced by Joseph Jastrow in 1892, the device consisted of a plate positioned atop ball-bearings and attached via rigid rod to a hidden etching device (figure 2)11. Provided subjects kept their hand on the plate for the duration of a task, the machine would produce a detailed record of their movements. Often, Jastrow asserted, these etchings could be used to discern noteworthy information about the train of thought. Attention paid to objects at different locations generally produced drifts in the direction of the item, and counting the tics of a metronome yielded countable oscillations. In some cases, concentrating on visual forms, such as the letter “O” was said to produce corresponding movements (e.g., loops). One study, conducted by one Milo Tucker, even reported full-word tracing similar to the automatic writing seen in hypnosis and some forms of brain damage, though the neatness and singularity of the result raise suspicions.12 

Another, less popular approach saw experimenters develop the muscle-reading skill themselves. Rather than rely on performers who might have reason to misrepresent their experience or lack the introspective training prized by psychologists at the time, a few psychologists began practicing and reporting on the method. These included well-known figures like Oskar Pfungst (see below) as well as less famous researchers like Thomas Verner Moore, who claimed that close attention to a subject’s face allowed him to guess the suit of thought-of cards at a rate “that would not have occurred by chance more than once in a thousand times.”13 Perhaps the most successful adopter of this strategy was University of Wyoming professor June Downey. After several years of practice, Downey reported that she could duplicate even the most striking acts of the era’s stage performers, including the guessing of dates imagined by a subject, the reconstruction of multisyllabic words held in mind, and “finding a book and identifying therein a word chosen at random.”14 Like most muscle-readers, she argued that such results undermined the bulk of evidence cited in favor of telepathy. Unlike most of her predecessors, however, her primary interest in the method was not its use as a debunking tool but its potential as a psychometric measure. Specifically, her explorations in the area led her to suspect that muscle-reading might find applications in the assessment of personality and what would today be termed cognitive style.15 Her efforts in this direction would not, in the end, prove successful. Nevertheless, her studies do provide a good sense of muscle-reading under naturalistic (if less tightly controlled) conditions, useful figures on individual variation, and a number of curious observations. 

In general, Downey’s studies suggested that most individuals—as many as 56 of 60 according to a 1909 investigation—could be “read” and that all but a few reported that they were not conscious of any movements on their part.16 Rates of success were comparable between women and men (though the majority of her sample came from the former group) and relatively close when comparing subjects who visualized an object’s location with those who rehearsed a verbal description of the locale.17 Some differences were observed between credulous and incredulous subjects as well as those participating with eyes closed vs. those with eyes open, though the limited scale of the studies and the number of simultaneous comparisons render these observations merely suggestive. In general, the most interesting findings were not those arising from group comparisons but those concerning “peculiar automatic tendencies” that emerged over the course of the investigation.18 One of the more unexpected phenomena was something labeled “recapitulation.” In some 22.6% of trials in a 1908 study, the reader did not go directly to the hidden object but, as tracings made by observers indicated, followed the path that the guide had taken when hiding it.19 If a guide had hesitated at a specific location, for instance, or deposited the object only after a circuitous route, the reader would do the same. When asked if they had kept their prior trajectory in mind during the reading, however, subjects insisted that they had not. Another curious observation was that objects could be located even as the guide’s attention was absorbed by a distractor task, such as the rehearsal of random names or counting aloud.20 These and a few other observations led Downey to suggest, somewhat tentatively, that muscle-reading might allow one to access unattended content in addition to actively entertained thoughts, though the lead was never followed up on in print.

Of all those participating in the muscle- and thought-reading literature, however, the most famous was not a psychologist but a farm animal. Overlapping with and to some extent drawing from the popularity of mentalism, the late 19th and early 20th centuries witnessed a significant growth of interest in “wonder animals,” non-human animals said to be capable of substantial intellectual feats or, in many cases, telepathy.21 Today, the best remembered is doubtlessly der Kluge Hans, a horse whose trainer, Wilhelm von Osten, claimed could read, identify calendar dates, and perform calculations up to and including cube roots.22 A 1904 commission determined that fraud was not involved in the performances but recommended further study to determine the exact means by which Hans solved the problems, with member Carl Stumpf telling the press that he believed the horse was using sensory cues unintentionally produced by his trainers.23 This lead was soon followed up by Stumpf’s student, Oskar Pfungst, who released the results of his own studies in 1907. According to Pfungst, Hans’ abilities ultimately depended on what his questioner did or did not know. In reading, calculation, and similar tasks, Hans showed near-perfect performance when the questioner happened to know the answer. When such knowledge was absent, however, his performance dropped to chance levels. A similar pattern emerged when the horse was made to wear large blinders, blocking his vision of the questioner, suggesting, Pfungst argued, that visual cues played an important role in his success.24 

To bolster this interpretation, Pfungst then conducted a series of tests placing the experimenter himself in the role of Hans. Subjects were asked to think of a number (calculation, leter, etc.), and Pfungst attempted to discern the answer by attending visually to bodily cues. In most cases, he found that he could guess correctly, and with “more suitable subjects,” Pfungst claimed the ability to determine not only the correct answer but incidental features, such as whether a subject imagined a letter in script or print form or the order of addends in an arithmetic problem (e.g., whether they thought of 3+2=5 or 2+3=5).25 Lest any doubt persist, he then went about measuring these movements and subjects’ breathing rates during the tasks, finding notable answer-revealing shifts as Pfungst, like Hans, tapped his answer. The case, he concluded, was remarkable not because of what it revealed about the horse’s intellect but because of how strikingly it illustrated the ties between thought—emotionally tinted thought in particular—and action.

The years that followed Pfungst’s analysis represent something of a high water mark for ideomotor phenomena in psychology. The case of Clever Hans was among the most famous results of psychology to date, with Pfungst’s argument for the role of unconscious cues serving as a reminder of (or advertisement for) their greater relevance for the discipline. Theories of the phenomena were debated in the pages of leading journals and incorporated into pedagogical practices, encouraging the use of motor imagery and active learning on the assumption that motor and ideational influences were bidirectional. Practitioners like Downey hoped that muscle-reading might be used in psychometric testing, and criminologists looked to it as a way of extracting useful information. The Harvard psychologist Hugo Münsterberg went so far as to suggest that automatographs be used to identify perpetrators from a lineup when witnesses were unwilling or unable to comply.26 As time went on, however, interest in the phenomenon proved difficult to sustain. Discussions continued for another two decades, buoyed by the occasional telepathy claim and a sustained interest in measurement technologies like Jastrow’s, but by the 1930s, there was no muscle-reading literature to speak of. 

Looking back, there are a number of reasons for this. For one, they were difficult to scale. Muscle-reading proper was limited to interactions between two or at most a handful of people and generally required a significant time investment from the would-be reader. As Downey’s investigations made clear, moreover, it was not fully reliable. Some individuals simply could not be “read,” and even with the majority who could, the time it took Downey to locate a hidden object ranged anywhere from 4.6 to 245 seconds, depending on the subject and how they thought of the item (e.g., verbally or imagistically).27 There was variability in the skill of readers, as well, with little way of telling a priori who would be able to acquire the skill. Even among practiced readers and reasonably expressive individuals, outcomes depended on temperament. Some subjects seemed to work better with some readers.28 Finally, success depended to some extent on the cooperation of those involved. As was noted early on, a wary subject could misdirect or, in some cases, block the relevant cues by concerted effort, and while this mattered little for the kinds of studies cited above, it proved a lingering issue for suggested legal applications.29 (Unfortunately, awareness of this concern was not enough to prevent adoption of the ill-fated polygraph, a descendent of Jastrow’s automatograph).30

At an earlier time in psychology’s history, the absence of such scalable applications may have been less of an issue. Yet the early 20th century was one of increasing professionalization for the field, particularly in the United States. Many of the first psychology departments in the English-speaking world were formed at land-grant colleges—institutions founded with practical ends such as agriculture and basic education in mind. To secure funding, then, psychologists leaned heavily on their discipline’s contributions to pedagogy and organizational management.31 With the onset of World War I, moreover, the field saw a strong push for standardization of the kind seen in army personnel testing. The highly individualized practice thus found itself in a field where uniformity and mass deployment were at a premium. Ideologically, the period also saw the development of behaviorism and a widening distrust of methods requiring individual skill or too much of an interpretive element. Thus, while Jastrow’s recording technologies continued to hold interest, methods requiring active involvement and/or introspective reports on the part of the experimenter largely fell from favor. To some extent, the very notions of unconscious cueing and ideomotor phenomena (understood as idea-mediated actions) became suspect. As early as 1917, one sees researchers complaining that consciousness was “rapidly losing its standing as a respectable member of the psychologist's vocabulary,” and with the rise of stimulus-response vocabulary, the notion of “ideas” came to be seen as dispensable or outmoded.32 The prominent psychologist Edward Thorndike went so far as to compare ideomotor action, with its reliance on ideas bringing about corresponding movements in the body, to imitative magic, the belief that desired events may be elicited through ritualized mimicry (e.g., injuring an enemy by effacing a wax figure of them).33 

Timing was also poor in that spiritualism, the perennial foil for muscle-readers, was on the decline. Though academic and popular interest in the paranormal would experience a resurgence later in the century, the topics never reached the heights seen in the Victorian era, when seances were a regular feature of bourgeois society and psychical research a common topic in major scientific publications. This left a good deal of unconscious cueing research lost at sea. The old program had been so strongly tethered to the bugbear of spirit mediumship that, when the latter began to decline, it lost much of its relevance. At a certain remove, the fact that ideomotor phenomena were so associated with Ouija boards and mind-readers even worked against them. When spiritualism was at its peak, many saw engaging with it as a responsibility. With increasing marginalization, however, any interactions—even hostile ones—could be seen as keeping the issue alive. At best, traditional muscle-reading work would risk fighting yesteryear’s battles. By the 1930s, then, researchers had largely abandoned the topic, discussing ideomotor phenomena in a far more circumscribed manner or simply eschewing it in favor of more respectable topics. The old muscle-reading results could be dusted off when new reports of telepathy began to gain traction, but the area saw exceedingly little direct study in the coming decades.34 Only with the growth of cognitive neuroscience and renewed interest in implicit and unconscious psychological processes have theories of ideomotor action and related phenomena gained significant attention.35 

The last decade has arguably been the most hospitable in over a century for the study of subtle nonverbal cues. Though the specific methods and theoretical constructs remain hotly contested, a host of implicit measures—means of assessing mental processes or attributes that do not rely on self-report—have become popular in both social and cognitive psychology.36 An active literature has also emerged on brain-to-brain coupling, or the synchronization of neural activity between individuals during social interactions (nonverbal interactions included).37 Nevertheless, discussions of high-precision “reading” like those of Downey, Pfungst, and the stage magicians have yet to resurface. At one level, this is understandable. The demands in terms of time and effort would be substantial, and even under ideal circumstances, results clearly varied from reader to reader and subject to subject. Most of the evidence also comes from an era when reporting and record-keeping were not what they are today. The most striking claims—guessing equations, multisyllabic words, and so forth—are ultimately anecdotal. Still, we should not be too quick to dismiss the testimony of so many independent, trained psychologists. The early researchers’ methods differed from those in use today, but most studies were subject to at least some degree of experimental control. While the most striking reports are anecdotal, moreover, the literature also contains detailed and systematic investigations, such as Pfungst’s and Stratton’s. These may not and should not convince us of all the claims made in these early years, but at the very least, the reports suggest that a tightly controlled investigation of the kind possible today is in line. If the results prove half as strange as the old reports suggest, it will have been well worth the effort.

Subscribe now

Gardener Comments

Note: the authors’ responses to several comments are interleaved below. 

Andrew Neff (Assistant Professor of Psychology & Neuroscience):
Fascinating topic. In addition to the historical characters mentioned in the article, novels and other humanistic works have long speculated about how informative muscle reading can be. To expand our current academic treatment of this topic beyond simple stuff like recognizing facial expressions of emotion does seem like a worthwhile endeavor. I strongly support this article's publication. One suggestion, which is by no means a condition for publication - I would have been interested to see a version of this article that's lighter on the history, and heavier on the methods of how we can/should approach this research today, and maybe even more specificity on the types of things that we ought to try to measure.

Simon T. van Baal:
I thoroughly enjoyed reading this article; it is written well. I generally think that this style of article is an interesting setup for scientific inquiry. It can function as a safeguard against reinventing the proverbial wheel by providing an introductory-level overview of a topic or discipline. I am no subject matter expert, however, so I cannot comment on the validity of the content. The only suggestion I have is that, even though it is implied in the title, I would have liked a few sentences preparing the reader for what the article is about and what it is trying to achieve.

Jan Kirchner:
This article discusses the history of the study of "muscle-reading" or the ability to read nonverbal cues to understand what another person is thinking. Beginning in the 19th century, psychologists and physiologists became interested in the idea that subtle movements could communicate a great deal about a person's thoughts. This interest was driven, in part, by the popularity of performers who claimed to be able to "read minds" through physical contact with volunteers. Skeptical researchers debunked these claims by demonstrating that the performers were likely relying on unconscious cues provided by their volunteers. However, the idea that people could "read" one another through subtle movements continued to gain interest. Studies in the early 20th century explored the limits of this phenomenon and attempted to understand the mechanisms behind it. By the 1930s, however, interest in muscle-reading declined. This was due to a variety of factors, including the growing professionalization of psychology and the decline of spiritualism (which had previously driven interest in the topic). Today, with renewed interest in implicit and unconscious psychological processes, some researchers are revisiting the study of nonverbal cues.

1. I believe this article contains novel ideas and analysis that have the potential to advance science. The author's comprehensive review of the history of muscle-reading provides important context and background for contemporary researchers who are interested in the study of nonverbal cues. By outlining the development and decline of this field, the author is able to offer valuable insights into the factors that may be important for future work in this area.

2. The article is also well-justified in terms of how it could advance science. The author clearly outlines the relevance of muscle-reading to the study of implicit and unconscious psychological processes, and provides a number of suggestions for potential future research directions. By doing so, the author offers a roadmap for how this historical review could inform future work.

3. Finally, I believe this article is well-written. The author provides a thorough overview of the history of muscle-reading, and does so in an engaging and accessible manner. The writing is clear and concise, and the author is careful to define technical terms where necessary. I would be happy to recommend this article to a friend.

For these reasons, I recommend that the article be accepted.

Minor points:

- The author may wish to consider including a table summarizing the key studies and findings outlined in the article.

- As an editorial note, the author may wish to break the article into more clearly delineated sections in order to improve readability.

Dr. Payal B. Joshi:
The article is well-written and is found consistent with the theme. Though no major flaw is seen in the article, there are some concerns with the article in the present form. The title needs to be revised to reflect the exact tone of what the author/s wish to deliver to readers. As the genesis of the article lies on historical perspectives, Carl Hertz works on muscle reading perspective is missing. The historical works on mentalism with magic by D. Blaine et al, Dunningher and Kreskin are missing. Overall, the reading displays a linear growth of the mental-reading field which is certainly not true, given the number of mentalists and muscle-readers of that time. The application of muscle reading utilized on famous personalities and sports persons are not provided. Overall, the article is certainly publishable, however, it needs crucial historical details to not be left behind.

Author Response: We certainly acknowledge that the article is, at most, a cross-section of the history. Viewed in its entirety, the story of muscle reading more closely resembles a delta of crossing streams than a linear series, and much has been set aside for reasons of space. Building the research out would, as Dr. Joshi notes, necessarily involve more discussion of muscle readers like Hertz and the practice of stage muscle reading, and we hope that more work will be done in the area.

Mario Pasquato:
While the historical account is very interesting and detailed, it is surprising that there have been no experiments with more recent technology. If the concept works at all it would seem possible to implement a 'muscle reader' as an app relying on a phone accelerometer. Machine learning models could be trained to automatically predict the thoughts corresponding to relevant patterns. This seems too good to be true, but who knows?

Author Response: Prima facie, it seems likely that muscle reading should be detectable by sufficiently sensitive accelerometer. We think this is an excellent avenue of further investigation. One could imagine attempts at muscle-reading between people holding hands while also holding an accelerometer, though there is an important challenge here of how to account for the potential effects of subjects accommodating perceived researcher wishes.

It would also be interesting to see results obtained from neuroimaging techniques such as EEG or fNIRS. For example, successful muscle reading may be associated with correlated brain activity (i.e., interpersonal neural synchrony) similar to what has been observed in other collaborative social tasks.

Josh Randall:
This article provides an interesting overview of a sub-discipline that has fallen out of favor. In doing so, they describe methods and analyses that were developed and later appropriated by other fields. One area that is lacking is a broader suggestion for practitioners considering re-engaging specifically with muscle-reading as opposed to modern approaches to unconscious behaviors. Conclusions derived from these two areas could be vastly different, partly as a result of the practitioner/individualistic nature of muscle-reading compared to the modern potentially more reductionist, siloed experimentalist approach. A plurality of theory and data on the ways that humans transform information physically and mentally is vital to improving our modern understanding of the brain and psychology.

Author Response: In terms of avenues for further investigation, we think there are a variety. In addition to the two suggested above (use of accelerometers or neuroimaging), one could study muscle reading in circumstances where it is likely most easily detectable, either between humans or in human-animal communication.

With respect to muscle reading among humans, relevant phenomena might be more easily detected among married couples, whose period of courtship and cohabitation might be expected to yield some insight into each other’s trains of thought. Alternatively, one might study professional poker players.38

With respect to human-animal communication, one might look at bomb-sniffing dogs; one widely-cited study, for instance, suggests that handler beliefs influence detection. Another possibility would be to study animal behavior when interacting with humans with lower sensitivity to nonverbal communication; an initial study by Meyer and Forkman suggests that in such circumstances, dogs may show more insecurity-indicating behaviors than usual.39

There is also the route of training new subjects, or experimenters training themselves, in muscle reading. This is more time-intensive, and may come with certain dangers, but also may prove fruitful.

Ted Wade:
This was an engaging, scholarly, and informative history of the dark matter of nonverbal communication: the apparent passing of messages occurring below the level of awareness of sender and receiver. Necessarily it includes issues about how hard the phenomena are to study. It concludes by mentioning some current related work, but stops short of advocating any reason to pursue it except to say that current research methods might allow better-controlled studies.

At a time when in fact research practices are known to be in need of much improvement, it is not helpful to suggest doing a “tightly controlled investigation” with no hypothesis or direction. If you consider how studies on subjects like psychokinesis or lie detection have apparently not been able to satisfy critics, then maybe convincing work on reading subliminal nonverbal cues might also be elusive.

The history suggests that practitioners of “psychic reading” may not understand how they acquired their skills, nor are they aware of what stimuli they are responding to. Maybe a reason to pursue “muscle-reading” types of phenomena might be found by considering theories about (1) what causes different mental and physical activities to be conscious or unconscious, or (2) nonverbal communication in general.

Mark:
While interesting, I feel this work is not arguing a strong point. I'd love to hear what the author thinks or wants to argue here, as opposed to what feels more like just a review or related work section of a longer paper.

Partha Ghosh:
It's a most interesting topic. Sounds like the really remarkable historical claims are anecdotal and have not been replicated to a degree that can be tested. The simpler ones are relatively well-understood. Where do we get the data to further this?

Jack Arcalon:
Seems like a good overview article about a much neglected and since "forgotten" older research field that might lead to new interface technology applications.

Abstract

Using a dataset derived from the long-running “Cults” podcast by Parcast, I find that the number of new cults began to increase in the 50s, peaked in the 70s-80s, and has been in steady decline in recent decades. I discuss various factors (historical, technological, cultural, pharmacological) that may have played a role in the rise and fall of cults since the 1950s and speculate on the future of cults.

Introduction

It feels like cults used to be much more common. Ross Douthat notes this trend in his 2014 New York Times article  “The Cult Deficit” and discusses two thinkers who have suggested that this lack of cults is, contrary to popular opinion, a bad sign for society.

“Philip Jenkins, a prolific religious historian, who argues that the decline in “the number and scale of controversial fringe sects” is both “genuine and epochal,” and something that should worry more mainstream religious believers rather than comfort them. A wild fringe, he suggests, is often a sign of a healthy, vital center, and a religious culture that lacks for charismatic weirdos may lack “a solid core of spiritual activism and inquiry” as well.”

“Peter Thiel’s argument is broader: Not only religious vitality but the entirety of human innovation, he argues, depends on the belief that there are major secrets left to be uncovered, insights that existing institutions have failed to unlock (or perhaps forgotten), better ways of living that a small group might successfully embrace.

This means that every transformative business enterprise, every radical political movement, every truly innovative project contains some cultish elements and impulses — and the decline of those impulses may be a sign that the innovative spirit itself is on the wane. When “people were more open to the idea that not all knowledge was widely known,” Thiel writes, there was more interest in groups that claimed access to some secret knowledge, or offered some revolutionary vision. But today, many fewer Americans “take unorthodox ideas seriously,” and while this has clear upsides — “fewer crazy cults” — it may also be a sign that “we have given up our sense of wonder at secrets left to be discovered.”

Both Thiel and Jenkins take it as self-evident that cults are in decline, but to my knowledge there has been no quantitative analysis of cult formation of the scope that would allow for the assessment of broad trends over time.2

This article provides such an analysis using a dataset derived from the podcast “Cults” by Parcast (the dataset is available here). I find that cults have been declining in the United States and across the world since the 70s-80s and speculate on reasons for this “cult deficit” and what the future of cults might look like.

Methods

What is the difference between a cult, a sect, and a religion? There is considerable scholarly debate over this question and no clear consensus has emerged (see the brief discussion on the wikipedia page for “Cult” to get a sense of the definitional issues). Further complicating these definitional issues is the bewildering diversity of groups that can be considered as cults – doomsday cults, political/racial/terrorist cults, Christian cults, eastern religion cults, alien cults, new age cults, family cults, psychotherapy/human potential cults, commercial cults, and more. Given all these challenges, I elect to use the Justice Potter Stewart strategy—“I know it when I see it”. Thus, I am using the pejorative and colloquial sense of the term, which I believe is also the sense in which Thiel and Jenkins use it in their arguments.

Any episodic media that discusses cults will presumably be using this colloquial sense of this term because this is what their audience will have in mind. The podcast “Cults” by Parcast is one such piece of media—released weekly, each ~45-minute episode explores the history and psychology of a particular cult. My dataset consists of all the cults that have been covered by this podcast for which I could find clear beginning dates, a total of 99 cults (6 cults were removed due to lack of data; data available upon request).3

Given the secretive nature of cults and the unclear boundary between eccentric religious groups and full-blown cult, occasional judgments had to be made about when exactly the cult was formed. For example, sometimes a relatively conventional group only became a cult when a new leader took power, when a leader dramatically changed his teachings, or when the group moved to a new location; in these cases (a significant minority of the total), I strove to use the date at which cult-like dynamics began and not the official formation of the group.4

The problems of defining when a cult ended were much more problematic—in some cases, a clear event marked the end (a murder, suicide, or arrest), but in others it was debatable when the cult actually ended; for example some cults greatly reduced in size and changed their beliefs upon the death of a charismatic leader—should I consider the cult as still existing or ending when the leader died? Because of these problems, analyses regarding the dissolution of cults were not reliable and thus were not included. Finally, I also scored the location of cults by country and by state if the cult was located in the United States; some of the same issues discussed above applied here as well (a group was founded in one place but only attracted followers and became cult-like when it moved to new location)—again, I strove to use the location where cult-like dynamics took over for the first time, however there were a few cases where I deemed it appropriate to consider two locations for the cult. All in all, I do not consider any of these issues to be that problematic given the general nature of the question being investigated (are cults declining over time?).

Results and Discussion

A clear pattern emerges from the data: the number of new cults began to increase in the 50s, peaked in the 70s/80s, and has been in steady decline in recent decades. Given the quality and size of the dataset, I don’t feel that it is appropriate to analyze cult formation at a more granular chronological level or discuss any of the ancillary results like geographical distribution (but see Supplementary Information for cult counts by country and state). 

This dataset represents a subset of all total cults and therefore it is likely reasonable to consider if there is some bias in the sample that complicates the interpretation. Given that this list of cults comes from a podcast that is presumably trying to attract an audience, it is likely that our sample is biased towards cults that are more well known. This isn’t necessarily problematic, and in fact may be a strength of the analysis—to the degree that our dataset is filtered for fame/notoriety we are then answering the question—are cults of a certain size and significance declining over time? Another issue is that dataset may be biased towards the United States—the podcast is created by an American company and is presumably targeted towards American listeners, therefore it is likely that the list of cults is biased towards American cults and any geographical conclusions should be made with caution.

There also may be selection effects affecting the inclusion of recently formed cults and cults in the distant past. More recently formed cults may have not had time to attract a significant amount of followers or public attention. For example, a cult formed in 2018 may not be well known (and thus not selected for the podcast), but will be widely known in 2025 after a mass suicide event a la Heaven’s Gate. A quick internet search yields a few examples of more recently formed cults that were not included in the dataset—“Love has Won”, formed in 2006, and ”The New Light of God” in Panama, formed some time in the 2010s (see the footnotes for a rather interesting cult that just came to light in May, 20225). On the other hand, cults from the distant past may have faded in notoriety or may not be ideal for a podcast because of a lack of reliable sources. We could also posit a saturation effect in that lesser known cults from eras with a particularly high numbers of cults are more likely to be forgotten and thus cults from these eras are undercounted. It is also possible that the creators of the show have purposefully tried to avoid over-clustering of the time periods that they cover. It is difficult to know which of these effects is the strongest, but I think it is most likely that they are relatively insignificant and largely cancel each other out.  

Given all of these limitations (and many others not discussed), this analysis can only be regarded as a preliminary first step towards a greater understanding of cult dynamics across time and space. In the spirit of stimulating future research, the remainder of the paper provides speculation on what factors may have played a role in the rise and fall of cults since the 1950s and how these factors may influence cult formation and dissolution in the future. 

History and Technology

What explains the pattern of cult formation seen in our results? Numerous factors may have created a unique spiritual milieu in the post WWII-era (50s-70s) that was ripe for new belief systems. We might speculate that the psychological fallout of WWII, the Vietnam War, and the Cold war, along with rising secularization led to an unprecedented number of people that were open to joining cult-like groups. The rise in cults also coincides with two epochal events – the creation and use of the atom bomb and the space race; these events may have contributed to a particular metaphysical zeitgeist that was favorable to cult formation. One obvious way in which this manifested itself was the formation of UFO religions/cults, and in fact we do see that virtually all UFO religions/cults are formed from the 50s onwards (notably Scientology in 1953 and Heaven’s Gate in 1974). This may explain some of the increase in cults, however UFO cults are a minor percentage of all cults and other categories of cults (e.g. Christian cults) also seem to follow the same general trend. Another trend worth noting in this regard is the rise of syncretic religions/cults in the post WWII-era (the Moonies, Rajneesh, Nuwubian Nation, amongst others). Increased east/west cultural exchange during this time period may have provided a new influx of spiritual ideas that contributed in some part to the increase in cult formation.

What then explains the significant decrease in cult formation in recent decades? The internet has changed cultural/social dynamics in countless ways and many of them may be relevant for cult formation, but perhaps the most important factor is the dramatically increased availability of information and ease of finding and contacting people. Cults in the past benefited from recruiting people who had no previous knowledge of a group’s activities and could not easily fact check some of the more outrageous claims. Additionally, cults often sought to cut off contact between its members and their family and friends as these would be the people most able to convince a member to leave the cult. Clearly the internet has made it much more difficult for any would-be cult to isolate its members from outside information and communication. 

The internet and other digital technologies may have brought about a fundamental change in the nature of cult-like groups. “Totalitarian” cults that separate from society and seek complete behavioral control may be a thing of the past; in their place are internet-based conspiracy cults which may not include much formal organization. These internet cults do not rely on informational isolation and do not seek micro-level individual behavioral control, rather they rely on misinformation and seek to motivate higher-level individual behaviors (media consumption, voting, consumer behavior) and collective behavior (protests, rioting, terrorist attacks). Though they are not typically regarded as cults per se, QAnon and Info Wars (both not included in the dataset) have been noted for their cult-like dynamics and it is perhaps best to conceive of them as examples of this new category of internet-based cult (also see the Virtual Reptilian Cult – Sherry Shiner). Additionally, the internet has also allowed for updated versions of the classic cult of personality as charismatic personalities can now attract followers without significant in-person interaction (see Spiritual Catalyst and Lord RayEl). One way to conceive of the rise of Donald Trump and his cult-like following may be as a blend of both these categories – a digitally-weaponized cult of personality that traffics in misinformation and seeks to motivate voting and financial decisions.

Tight vs. Loose Cultures

Cultures vary in their tolerance of norm deviations, a characteristic known as cultural tightness. The West saw a dramatic shift towards “looseness” with the countercultural revolution in the 1960s and 1970s. We can speculate that this may have been a contributing factor to the explosion of cults – looser societies may provide more fertile ground for cults because of their greater tolerance for deviant beliefs and behaviors; a cursory glance at the geographic variation in our data supports the idea that looser states/countries have more cults – California, noted epicenter of the countercultural revolution, has far and away the highest concentration of cults in the world, and the vast majority of cults in our dataset come from western cultures, which tend to be generally looser than eastern cultures (however see the discussion on the geographic distribution of cults in the dataset in the methods section).

It is also interesting to note in this context that most of Trump’s base comes from regions with tighter, more collectivist cultures (the American south and midwest) that have also been shown to have fewer cults (Stark et al., 1979; also consistent with our data). The rapid rise of Trump and the almost religious-like fervor he inspires could be taken as evidence for a hypothesis raised by a recent simulation study by Muthukrishna and Schaller (2020) that suggests tighter and more collectivist cultures are prone to rapid transformations that “may proceed at a pace that more closely fits the subjective perception of a “revolution”. Chinese history provides an instructive example on this point (see this blog post by Dr. Muthukrishna for further discussion). Being that China represents one of the end of the spectrum when it comes to the tight and collectivist vs. loose and individualistic spectrum, we may expect a history of rapid transformation; in this regard, we can point to the Cultural Revolution and the Boxer Rebellion as examples of such transformations (of course one could also say that these revolutions were brought about by specific historical circumstances and events that were not really related to cultural tightness). Taken together, this may suggest that societies that are less tolerant of deviant behavior may have a lower rate of cult formation, but may be more susceptible to mass revolutions in which cult-like dynamics take over. 

Psychedelics  

“Aldous Huxley makes the same point in speaking of the effects of mescaline in The Doors of Perception: that in a world in which everyone took psychedelics there would be no wars, but no civilization either.”

— Colin Wilson, The Occult: A History

The explosion of psychedelic usage in the 60s and 70s may have also been a significant factor in the increase in cult formation and participation during that time period. Psychedelics, particularly LSD, played a major role in some of the more notorious cults (the Manson family, Aum Shinrikyo—f you don’t know about this one, take a second to read about how batshit crazy this cult was), but it also may have contributed more broadly to a cultural (and chemical) zeitgeist that was prone to cult-like dynamics.

The amount of academic publications on psychedelics should serve as a reasonable proxy for psychedelic usage in the general population. It is interesting to note that the trend in scientific interest in psychedelics mirrors that of cult formation but peaks 5-10 years earlier, which might make sense given the lasting psychological changes that can arise from psychedelics and a faster fade out of academic interest compared to public interest after the change in legality in 1968.

Why might psychedelics contribute to a rise in the appearance of cults? Recent research has developed a model in which psychedelics lead to a global relaxation of beliefs (i. e. “Bayesian priors”—see “REBUS and the Anarchic Brain: Toward a Unified Model of the Brain Action of Psychedelics”, Carhart-Harris and Friston, 2019; also see Scott Alexander’s excellent summary). In this model, sensory hallucinations arise from a relaxation of high-level priors like “walls don’t move” and “most objects don’t randomly change colors”, however this relaxation also applies to beliefs like “I’m a failure” or “the world is high dangerous”; relaxation of these beliefs may be responsible for the therapeutic beliefs of psychedelics for depression and PTSD that are now being heavily investigated. We can imagine that the relaxation of beliefs like “aliens are not real” or “random people don’t have the secrets of the universe” could lead to increased formation and participation in cults. In a section entitled “What to Do About the Woo?” the authors of the REBUS paper propose an alternative reason why psychedelics may dramatically modify beliefs, “psychedelics have an interesting history of association with pseudoscience and supernatural belief. One interpretation of this is that a strong psychedelic experience can cause such an ontological shock that the experiencer feels compelled to reach for some kind of explanation, however tenuous or fantastical, to close an epistemic gap that the experience has opened up for them.” In this model, cult leaders use psychedelics to create ontological shock and an accompanying epistemic gap which they then seek to fill with their own doctrines. 

The authors also discuss the nature of the society-level effects arising from widespread psychedelic usage in two more speculative passages.

“Two figureheads in psychedelic research and therapy, Stanislav Grof and Roland Griffiths, have highlighted how psychedelics have historically “loosed the Dionysian element” (Pollan, 2018) to the discomfort of the ruling elite, i.e., not just in 1960s America but also centuries earlier when conquistadors suppressed the use of psychedelic plants by indigenous people of the same continent. Former Harvard psychology professor, turned psychedelic evangelist, Timothy Leary, cajoled that LSD could stand for “Let the State Dissolve”.

They close the article with this quote from Michael Pollan’s 2018 book How to Change Your Mind.

“Whether by their very nature or the way that first generation of researchers happened to construct the experience, psychedelics introduced something deeply subversive to the West that the various establishments had little choice but to repulse. LSD truly was an acid, dissolving almost everything with which it came into contact, beginning with the hierarchies of the mind… and going on from there to society’s various structures of authority and then to lines of every imaginable kind… If all such lines are manifestations of the Apollonian strain in Western civilisation, the impulse that erects distinctions, dualities, and hierarchies, and defends them, then psychedelics represented the ungovernable Dionysian force that blithely washes all those lines away… But surely [it] is not the case that the forces unleashed by these chemicals are necessarily ungovernable”.

Maybe we can think of the ambient level of psychedelic usage as a modifier of the “hierarchality” of a society; societies with less hierarchality (what a mouthful) may be more prone to the development of fringe cult-like groups and more likely to tolerate their existence.

The Future of Cults

There may be a few factors that could return us in the coming decades to something like the spiritual milieu that provided such fertile ground for cults in the 70s and 80s. First, we may wonder at the psychological, cultural, and political fallout from the COVID-19 pandemic and how this may affect the nature of religious belief in the 20s and beyond. The rise of AI may play a role similar to that of the atom bomb and the moon landing—it would not be entirely surprising to see the formation of various “techno-cults” based on AI and other advanced technologies (e.g. a cult-like group that centers around some particularly powerful cognitive or physical modification). Some would say there are already a few techno-cults in their incipient stages – proponents of the technological singularity have been accused of cultishness and it is reasonable to think that ever-more powerful AI will only inspire more cult-like devotion to this idea (apparently a “Church of AI” was founded in 2015 but has already shut down). There is a tongue-in-cheek Mac community website called Cult of Mac—would it be shocking if 30-50 years from now there in fact was a real cult of mac (and a cult of PC)?

We are also in the midst of a psychedelic renaissance and it is worth wondering if this will also contribute to a cult renaissance. This would lead to the prediction that states/countries that decriminalize or legalize psychedelics may see a spike in cult formation. Oregon was the first state to decriminalize Psilocybin in 2020, and it seems that California and Colorado will also do so in the coming years (Denver, Oakland, and Cambridge (Massachusetts) have decriminalized Psilocybin since 2019). California in particular will be a region to watch for new cults—our data clearly support the popular notion that California is the cult capital of the world, however only two cults have formed in California since 1990.

As discussed above, the changes in the informational landscape brought about by computers and the internet has affected the nature and dynamics of cult-like groups. It is fair to wonder how these changes will affect a potential resurgence in the “cultishness” of society. In the 70s and 80s, there were numerous cults, however essentially all of them were very small fringe groups (Scientology, perhaps the largest cult-like group, peaked at around 100,000 members in the 90s, but is now down to around 20,000). It is possible that in the coming decades we will see a very different cult ecosystem, one in which there continues to be a smaller number of fringe groups or even fewer than the present, but each of them attains a much larger membership (into the millions) and amount of cultural influence. QAnon may provide an example of what we may see more of in the future—loosely organized cult-like groups with no formal membership that attain millions of believers.

Conclusion

Although this analysis supports the intuition that cults have been declining in recent years, it says nothing about the argument that a lack of cults indicates a lack of cultural creativity and experimentation. Personally, I think there may be some merit to this argument. In this view, there is a tradeoff – more cultural creativity means more of the negative outcomes typically associated with cults. Is this tradeoff inevitable or is there a way we can have our cake and eat it too? It strikes me that in many ways cults represent a “photo negative” of scientific communities. In cults, the Absolute Truth is dictated by the leader; the Absolute Truth never changes (unless the leader changes it) and skepticism and creativity are strongly discouraged (to put it mildly for many cults). In science, the truth is always and forever subject to revision and skepticism and creativity are strongly encouraged. 

Subscribe now

Gardener Comments

Phil Wilson:
Overall, this is an interesting contribution which shows some reasonable evidence that cult formation rates have steeply declined, and speculates about possible causes for the decline. The paper is broadly well-written, with few typos and clear exposition. Efforts are made to get definitions clear and to address limitations and biases in the data and methodology. There follows a few specific comments addressed to the author, either for them to enact or for the interested reader to consider.

1. In the first paragraph of the Methods section, you decide to use "the pejorative and colloquial sense of the term" 'cult'. But a limitation with this approach is that one can bend and flex the definition to help you cherry pick the data which confirms your hypothesis. Another approach would be to find an established definition which is compatible with your data set.

2. In the second paragraph of the Methods section, you assume that any episodic media will also use only the colloquial sense of "cult". Is this true? Do the hosts of the podcast in question never say how they define it?

3. In the second paragraph of the History and Technology section, you speculate that one cause for the rapid decline in cult formation is the adoption of the internet. While this sounds plausible, the decline is already seen in your data in the 1980s, well before wide-spread adoption of the internet. Indeed, the trend continued in the 1990s, yet according to https://www.internetworldstats.com/emarketing.htm only 4.1% of the world's population was online by the end of that decade.

4. On p.7, *why* should "[t]he amount of academic publications on psychedelics ... serve as a reasonable proxy for psychedelic usage in the general population"?

5. Also on p.7, I note that you only "note that" the trend in scientific interest in psychedelics mirrors that of cult formation, yet that language is highly suggestive and indicates that you wish to imply a connection. This is a very weak link, and even if you could establish a correlation it would be far from establishing a causal connection.

William Collen:
Very thought-provoking premise, and I think this piece will generate some lively discussion upon publication. My only critique is that the author seems to rush over potential problems with the dataset. Statements like "it is likely that our sample is biased towards cults that are more well known" and "dataset may be biased towards the United States" are, apparently, much more concerning to me than to the paper's author. Also, "analyses regarding the dissolution of cults were not reliable and thus were not included" seems to be a very important point affecting the author's fundamental premise. If the rate of new cults' being started has slowed down, but those cults that do exist continue to gain members, perhaps even at an accelerating rate, would this not indicate that cultism is increasing, rather than the opposite?

Also, I would have liked to see more discussion on the ratio between cults, mainstream religions, and the population as a whole. If the rate of new cult activity is dropping more slowly than the rate of religious activity in general, that would be a net increase in cult activity. In my mind, the paper presents a very useful look at an interesting phenomenon, but an incomplete one.

Mark:
This is interesting and useful but the data appear too biased to be a useful measure for the underlying concept. I think this work would be dramatically strengthened by drawing cults from other resources — in the worst case something like scraping wikipedia for cult-like entities would be informative and would help build certainty in the result that is being claimed.

A secondary issue that occurs here is that none of the causal aspects of the restricted time domain have been counted for in the data analysis. For example, it would possible to account for two dominant forces here: historic lack of reporting, and the time it takes the world to know about a new cult. As it stands, it seems challenging to make any claims based on this data. One way to model historic context would be to use Google ngram data to try to reduce confounding. One way to deal with the time it takes for information about cults to be widely known is comparing start date to widespread publicity, and create an average expected delay in this kind of information. This would also allow for an improved estimate in how cults have changed over time.

As a minor point, the plots could be substantially more informative and intuitive, which would help strengthen the claim of the work. Plotting and assessing start time and duration as independent factors would be useful here. When counting for confounds like those mentioned above, it would be useful to show an estimate in addition to a count, so as to provide a more reliable sense of the true number of cults to a reader. As it stands many readers will falsely conclude that cults are just declining and that they have historically been a substantial aspect of society.

The motivation of the work seems in part to deal with claims about the decline in counter-cultural sub-communities. One important element of analysis to make this point is to estimate the proportion of counter-cultural beliefs due to cults, by, for example, tracking the population of people who were part of each of these cults as compared to the total population of the USA or world at that time. In essence, do we actually have fewer people who think differently now? Computational mechanisms like the Gini coefficient might be useful for addressing this perspective.

Extending this idea, I note that a substantial portion of the work seems to be only loosely related to the central argument and focuses on cultural tendencies. This is interesting, but would be better if 1, it was motivated by the core idea of the project, and 2, had a data driven way of motivating its claims and findings.

Overall, I think this is interesting, but a bit more work is needed to make it a useful and trustworthy resource for others to learn from. I encourage you to make improvements and share a new version when ready.

Dan James:
An interesting analysis however one that, for me, had a number of flaws. Though the paper admits to, and somewhat addresses the problem, the label ‘cult’ is a vague category which implies that an accurate quantitative analysis of the number of cults in existence at any one time is necessarily subjective and highly likely to be the result of implicit selection effects. As a consequence the conclusion of the paper is undermined insofar as the difficulty of defining a cult doesn’t really tell us if the emergence of cults has or has not declined in recent years.

The claim that cults represent a ‘photo-negative’ of scientific communities is questionable. The label ‘cult’ is often used as a pejorative term to describe groups that we don’t understand or like (Schmalz ‘The Conversation’ 2018), and is rarely applied to science. However there is no good reason to think that science is somehow immune to cult-like behaviour. For example, do string theorists belong, as Lee Smolin suggests, to a cult (‘The Trouble with Physics’ 2008)? Similar points about physics being dominated by a ‘cult’ of beauty have also been made by Sabine Hossenfelder (‘Lost in Math: How Beauty Leads Physics Astray’ 2018).

Nonetheless, given part of the remit of a Seed of Science paper is to encourage debate, this paper, for me, fully meets that requirement so I would recommend publication.

I'd prefer to give this a 'revise and resubmit': This did present useful information, but in a very amateurish way. The time scale on some graphs is reversed, and there are typos and problems with links. Also, most of it after the data analysis was speculation with little evidence.

Fred Nix:
I think there are some interesting seeds of science here, but there are a few changes I would make before publishing. First, the seeds. I think the idea that cults have both positive and negative effects is very interesting. As is the idea that cults seem to have declined, and perhaps will rise again, but maybe in a different form, such as QAnon. The idea of AI cults is particularly interesting. I like the idea of somehow untangling the good from the bad and educating people about how to resist cults, but at the same time realizing there is some benefit to them. I am not sure how we do that untangling, and I think that is really the seed. We don't want to extinguish cults, in fact, we may want them to grow, but we want to limit the harm. But how? Second, the changes. So, there is a period missing in the last sentence in the Abstract. I think you overreach when you say that "This article provides such an analysis," implying that this is a truly quantitative analysis of broad trends over time. Really, this is a small sample from a podcast. I don't see any harm in just acknowledging the limitations and the need for a larger study. That could be a seed. I want to see time moving forward. I'm now more interested in this topic than I was before. The Peter Thiel quotes are interesting. Keep pursuing your ideas here! This paper could be just the start of your exploration in this area.

Partha Ghosh:
It's an exciting topic, compellingly presented, worth probing in depth to understand the risks and rewards. Most of the future cults will probably be on the internet/ the virtual world: will it be a variation on the echo chambers that exist today, or some mutation hard to anticipate? Perhaps some of us will become more Janus-like, with different persona for irl and virtual selves - and therefore it will be hard to predict who is more predisposed to join a cult?

Ted Wade:
The paper responds to 2 popular speculations about an apparent drop in the incidence of new cults over the last few decades. The data offered to support this might be considered pilot data, but there is no reason to believe that sampling from the population of all cults is adequate to make a conclusion about real trends. Given this, there is little gained by some of the hypotheses offered to account for the incidence data, although the discussion of these is interesting.

Douthat claimed (perhaps channeling Thiel, himself a priest of the Free Markets cult) that “Today, many fewer Americans ‘take unorthodox ideas seriously’”. That might have barely had traction in 2014, but post-2016 it seems absurd, the opposite of the truth. This also undercuts a cult-ish argument about innovation. One could convincingly argue that belief in stupid things is negatively correlated with creatively solving real problems.

The cultural tightness/looseness idea, on the other hand, seems plausible, with some scholarly support, and could stand alone as one determinant of cultural creativity. It could serve as a causal variable for both religious and economic experimentation, perhaps causing them to be positively correlated across different cultures. Changes in cultural tightness might also explain changes in within-culture creativity over time.

Jan Kirchner:
The paper "Cult Deficit: Analysis and Speculation" examines the curious phenomenon that prevalence of cults has declined drastically in recent decades. While this observation is not novel, the authors substantiate it with a dataset of cults covered by a popular podcast on the topic. They discuss possible factors affecting the rise and decline of cults, such as the change in ease of information flow, change in cultural "looseness", and the change in openness to psychedelics. In the closing section, the authors provide an outlook on the possible future of cults.

I enjoyed reading the paper (a rare academic page-turner) and applaud the creation of the accompanying dataset. I can think of a few questions that could be investigated with the dataset and with a bit of additional research (which factors affect longevity of a cult? are there shared characteristics of cult leaders?). I believe that publication of this paper in Seeds Of Science can serve as a great jumping-off-point for future research and therefore recommend publication.

Nonetheless, I want to mention some potential problems with the "I know it when I see it" definition of cults. While I acknowledge the dilemma (cults have a lot of incentives to avoid the label), by avoiding a definition the authors take a lot of liberty in how they construct their argument. They gloss over differences that appear substantial (is it fair to mention Trump supporters, Q Anon, and the Manson family without mentioning different levels of cultish-ness?) and draw category boundaries that could be disputed (is scientology really that different from established, highly conservative religions?). My point is not that the choices made by the author are necessarily wrong, but that the chosen non-definition of the term "cult" does not allow investigation of some interesting questions. Along the same lines, the author does not mention the interesting discussion around changes in use of the term as a function of time and the rise of secular anti-cult movements in the 1970s (https://en.wikipedia.org/wiki/Cult#Definition).

Michael M. Kazanjian:
An extensive, important article worth reading on cults. I especially appreciate the database of the many cults. This is a must read. I have extensive research on cults, sects, and religions, and this paper is good.

Rachel Prudden:
I'm afraid I don't feel the methodology is appropriate to address the stated question of whether or not cults are declining over time. Using a dataset derived from a podcast is likely to strongly skew the results towards those which have been through a series of events which can be put into a narrative, and for which a reasonable amount of information is available (otherwise it wouldn't make much of an episode). I would expect this to overestimate the relative prevalence of cults formed a few decades ago. I have voted against publishing because the quantitative analysis is framed as the main contribution of the paper and I think it's likely to be misleading. I do think it's an interesting topic, though.

Supplementary Information

Figure 5 - number of cults per state and country

Abstract

The process of evaluating research proposals for funding is often based on subjective assessments of the "goodness" or "badness" of a proposal. However, this method of evaluation is not precise and does not provide a common language for reviewers to communicate with each other. In this paper, we propose that science funding agencies ask reviewers to assign quantitative probabilities to the likelihood of a proposal reaching a particular milestone or achieving technical goals. This approach would encourage reviewers to be more precise in their evaluations and could improve both agency-wide and individual reviewer calibration over time. Additionally, this method would allow funding agencies to identify skilled reviewers and allow reviewers to improve their own performance through consistent feedback. While this method may not be suitable for all types of research, it has the potential to enhance proposal review in a variety of fields. [abstract generated by ChatGPT]

We’re all familiar with meteorologists’ forecasts: a 5% chance of rain, a 95% chance of rain, etc. It would be far less useful if you turned on the Weather Channel only to hear on every occasion that there were two possible forecasts: “Rain will happen today,” or “rain will not happen today,” with the forecast being wrong half the time. It would be only slightly more useful if the forecasters used a scale of 1 to 9: “Chance of rain today is a 7” versus “chance of rain is a 5.” What does any of that actually mean? Should you carry an umbrella or not? Turn off the yard sprinklers? Plan for extra time on the drive to a meeting?

This is not unlike the situation we’re in with proposal review at the moment. 

Proposal reviewers are constantly expected, even if implicitly, to make predictions about the success of a particular research proposal. But to my knowledge they are not asked to provide actual probabilities, let alone probability distributions of likely impact. Instead, they’re asked to make judgments about whether a proposal is “good” or “bad”, or to rate proposals on a scale from 1 to 9, or something else that doesn’t actually require them to estimate and articulate probabilities of success.  

As an experiment in enhancing proposal review, science funding agencies ought to try having peer reviewers or other evaluators assign quantitative percentages to a proposal's likelihood in reaching a particular milestone or technical goals. This idea might work better for some research areas (goal-directed projects, clinical trials, etc.) than for others (e.g., basic exploratory science), but focusing on probabilities would likely benefit all proposal review. 

How might this look in practice? Reviewers would likely show some independent and diverse judgment (Lee et al., 2013). For example, Reviewer 1 gives a proposal a 50% chance of success in reaching milestone 1, while Reviewer 2 gives it 85%, and Reviewer 3 says it only has a 25% chance. These kinds of judgments can then be scored over time. Better yet, reviewers could provide confidence intervals, e.g., “I’m 95% confident that the probability of success is between 40% and 60%.”

It might turn out that Reviewer 1 is pretty good (or is “well-calibrated” in forecasting parlance) in that the proposals that she predicts to have a 50/50 chance generally turn out to succeed (or fail) half the time. On the other hand, Reviewer 2 might turn out to be biased towards positive judgments, assigning higher probabilities to proposals that don’t succeed very often, indicating he should perhaps be more conservative in his estimates. Likewise, Reviewer 3 might be far too negative, underestimating how successful proposals turn out to be. The point is not to expect perfect accuracy from all reviewers (although that would be nice). The point, instead, is that by requiring reviewers to make probabilistic forecasts, funding agencies can identify skilled reviewers, and reviewers can use their feedback to improve their own performance.

Here are some of the potential benefits:

• Everyone would have to be more precise in their evaluations.  It would not be sufficient to say “this proposal is good, bad, or ok,” but would require a reviewer to specify the actual chances that the approach would work or not.  This would help reviewers to speak a common language with each other during proposal evaluations, as well as allow an agency to see the spread of forecasts among reviewers for any given proposal.

• Forecasting could improve both agency-wide and individual reviewer calibration over time through consistent feedback and review. Feedback loops are essential, after all, for improving performance (e.g., Goetz, 2011). If you’re constantly making implicitly predictive judgments, but never have to provide explicit probabilities and thus can’t receive feedback as to whether your forecasts were accurate, you will miss out on many opportunities to improve.

• An agency would be able to visualize its overall portfolio with the proportion of its funding on research or projects that are truly risky (perhaps with under 20% chance of success), somewhat risky (perhaps closer to 50%), and – where risk may be less appropriate -- perhaps verge closer to slam dunks.  In an agency that promotes a “high-risk, high-payoff” mission, funding proposals that on average have an 80% chance of success might be overly conservative.  Other funding agencies might want a different risk profile, but probabilities would help them both in establishing what that means and whether they’re aligned with that profile.

• Forecasting could also help reduce (or at least reveal) hindsight bias (Roese & Vohs, 2012).  When a project fails, people will commonly opine that they always suspected that it was never going to work.  This can lead to an impression that the investment was somehow foolish, wasteful, or unnecessary.  But if reviewers had earlier given that project a 90% chance of success, and it failed, reviewers and organizations examine their perhaps unwarranted optimism. Forecasting can turn failed projects into learning opportunities.  At the other end of the spectrum, success will often invite the response, “well, we always knew that would work,” again suggesting that the research was unnecessary or somehow self-evident.  Probabilistic forecasts would help temper that reaction, however, had the project been assigned an average of 30% chance of success during proposal evaluation.

• We could also test the effect of peer review discussions, which can potentially create peer pressure to conform to the judgments of other reviewers (Lane et al., 2021; Pier et al., 2019). For example, we could get reviewers to make a probabilistic forecast while blinded to what any other reviewer thinks; then we could gather the reviewers for a discussion (such as an NIH study section); and finally we could ask them to make a second probabilistic forecast. This would allow for tests of which forecasts are more accurate—independent or peer-influenced? Organizations could then adopt the more accurate procedures.

Given these potential benefits, the next obvious question is why the approach hasn’t been widely adopted or at least attempted. Searching the published literature and consultations with members of the forecasting community revealed nothing. This absence of evidence is puzzling, but – like the proverbial economist who turns up his nose at a $20 bill on the sidewalk because it must be fake (otherwise someone else would have picked it up) – we shouldn’t assume that it’s also evidence of absence.  If such a presumably easily implemented and seemingly useful idea hasn’t been tried, maybe there’s a good reason.

One reason might be that some reviewers feel that they can’t be more precise than to put proposals into, say, three buckets of predicted performance: “probably not; 50/50; and most likely.” To make them assign a probability of 37% or 68% might feel like an artificial level of precision. After all, why 37% and not 38% or 36%? Who could possibly explain that kind of fine distinction? This is a fair point, but if reviewers want to round up (or down) their probability assessments to the nearest 10, or 25, or any other percentage, the beauty of forecasting is that they are free to do so.  They need only be somewhat rational (give higher probabilities to proposals they think are more likely to succeed) and be willing to assign probabilities that can be scored. 

This seems worthwhile: if a reviewer claims to be “skeptical” about a proposal, we should try to elicit a more exact probability. Maybe the reviewer has very stringent internal standards, and anything short of a 90% chance of success makes him or her skeptical. Alternatively, maybe the reviewer’s skepticism maps to a 20% chance of success.

Requiring the reviewer to articulate a probability – any probability – makes it easier for others to understand what the reviewer actually means (and potentially to help the reviewer consider what he or she means as well)

Another reason for resistance may be that reviewers are uncomfortable with their performance being judged, which may reflect the organizational or scientific culture in which they operate.  For example, the CIA was infamously uncomfortable with Sherman Kent’s efforts to make analysts provide probabilities with their analytic judgments.  Instead of judging the likelihood of an event being “possible, with medium confidence,” Kent wanted analysts to put a probability on their judgments.  When someone accused him of subsequently trying to turn the CIA into a betting shop, Kent’s response was, like himself, rather legendary: “I’d rather be a bookie than a goddamned poet” (Ciocca et al., 2021).

While that sentiment may be a bit harsh, it is not wrong. If people – reviewers, program officers, agency heads – are determining whether and how to spend millions or billions of federal dollars, they should care about improving their own calibration and judgment. Indeed, I would argue that NIH and NSF should systematically capture the judgment and calibration of all reviewers and program officers. I could even imagine an ongoing set of awards: “Top-scoring NIH peer reviewer,” “top-scoring SRO,” and the like. This would further incentivize reviewers to make more accurate predictions.

If the National Weather Service's Global Forecast System started proving to be inaccurate (e.g., rain occurred only 70% of the time when the model predicted 80%), meteorologists would be able to jump into action to recalibrate the models, check the data inputs, and draw from a long historical database of past forecasts and outcomes. But they can only do this because they have a system and a culture of keeping score. Right now, scientific funders can’t do anything comparable. As Fang and Casadevall write:

“At its extremes, the error and variability in the review process become almost laughable. One of our colleagues recently witnessed an application to receive a perfect score of 1.0 when part of a program project application, but the identical application was unscored as a stand-alone R01. Almost no scientific investigation has been performed to examine the predictive accuracy of study section peer review…Putting a stronger scientific foundation and accountability into peer review would enhance confidence in the system and facilitate evidence-driven improvements”

As the best way to get ahead is to get started, we could all benefit from being a little more like meteorologists in this respect. 

Acknowledgements

Thanks to Cory Clark and Phil Tetlock for helpful comments.

Gardener Comments

Andrew Neff (Assistant Professor of Psychology & Neuroscience):
Fascinating concept, persuasively and clearly written, I support it. I think the most interesting part is the concept of providing feedback for reviewers. As is mentioned in the article, this could help reviewers update their own biases, or help agencies like NIH/NSF get rid of bad reviewers, promote good ones, and be more inclined to bring new people into the fold. New people please.

Reading this makes me want a little more context for how forecasting could be incorporated into a holistic review process. You’re not suggesting this, but a yes/no funding decision can’t rely solely on the likelihood of success on its own, since nobody wants to fund only low-risk projects. Agencies need to add a second criteria for evaluating projects, something like 1) likelihood of success X 2) the value of success. Do we also need to quantify the value of success, or is that a different question?

I’m also thinking about how reviewers should judge the likelihood of success. At least at the NIH, it’s based on 1) the investigators history, 2) the technical-soundness of the method, and 3) the supportiveness of the environment. Should reviewers be constrained in their methods for assigning % likelihoods of success (as they currently are to an extent), or should forecasters be given more leeway?

My forecasted increase in the probability that the NIH/NSF change their funding policy in the next 2 years (if this article is published vs not published): 0.01% (sorry)

Antman (PhD in behavioural biology):
This is a nice, constructive suggestion. I completely agree that putting hard numbers of judgements, on a clear scale with meaningful units, is constructive and helpful, and will make reviewers think more carefully about their reviews. However, the proposal has obvious weaknesses which are not directly addressed.

Firstly, and perhaps most banal, the title contains the term "peer review," and most academics might assume this refers to peer review of research articles. However, the work is more limited in scope, and only really directly relevant to proposal reviews.

Secondly, and more importantly, this proposal would be most powerful in systems where repeat reviewers are tracked and scored, to identify super-predictors. However, no data is provided on how often, on average, an individual reviewer is asked to review different proposals for the same agency. The measure of prediction accuracy will only start, itself, to be useful after at last 5 or 6 reviews, and then still with large confidence intervals. Is it really realistic to expect most reviewers to have already performed so many reviews for any specific grant agency?

Thirdly, and related to the above, the literature on prediction science shows that it is very hard to identify super predictors, and very easy to confuse them with people who just got lucky - especially at the low sample sizes we would be dealing with here.

Finally, it should at least be acknowledged that 'chance of success' is not the only criterion which proposals should be judged on. An equally important measure would be "importance of the research", and these should interact to give a final score. 

In essence, then, while I agree that adding a scale with meaningful units to reviews is helpful, I feel like many of the additional benefits of the system proposed here are not realistic, and thus oversell what is, in essence, a valuable exercise in improving prediction readability and reviewer reflection.

DK (PhD in psychology):
I would support the publication of this article, but only if the serious concerns that I have are addressed. My main serious concern is - the author claims that the current peer review procedures are not optimal because manuscripts are rated using arbitrary categories such as good-bad or certain scales (e.g., 1-9). However, I do not see an argument that would convincingly show me that using percentage/chance would not result in reviewers consistently choosing few arbitrary categories - e.g., 25%, ,50%, 75%. I would like to see a convincing argument regarding whether and why this would not be the case (there is a large literature on how people perceive and estimate percentages which should be consulted), and I would also like the author to describe the measures which would ensure this does not happen.

Étienne FD:
This is a good article. It provides an interesting idea to help improve peer review. I will not comment on its "chance of success," but there's probably value in making success probability estimates in science more common than they currently are.

I do want to offer some feedback on the meteorology analogy, as I'm somewhat familiar with that field. While weather forecast computer models do produce precise numerical probability estimates for precipitation, it's common for forecasters to round these to a few "bins" for public communication. A weather forecaster I know tells me, for instance, that they purposefully never state a 50% probability, and in practice only round to a select few numbers, something like 0, 30, 40, 60, 80, and 100%. This is not very different from the 1 to 9 scale that the author offers as an example of what meteorologists do not do. 

This speaks to concerns about how to communicate probability effectively, and whether it's actually useful to provide probability estimates when you don't have a more precise idea than "somewhat likely" or some such. It seems possible that moving to probability estimates would not in fact improve peer review, while giving it an undeserved veneer of credibility. It would be nice to discuss these concerns somewhat more than they already are, but I'm happy to recommend this article for publication regardless.

Josh Randall:
This article provides an interesting solution to one of the issues of the model of publicly funded science. Specifically, many of their points are describing ways of improving the study proposal mechanism associated with funding research, likely in human focused fields. Their suggestions and description of possible responses from reviewers seem reasonable given the increasing workload associated with writing and proving the possible success of new studies. However, unlike in meteorology in which success can be determined by whether rain occurred or not; many disciplines of science that do not strictly follow falsification or hypothesis driven research might not have an obvious outcome of success and failure. In research programs focused on natural history, a proposal might be describing a plan to describe the physiology of an organism that hasn't received much attention previously. Would a failure be impossible or occur if their physiology is not interesting? Many of these fields also construct experiments with a goal of a positive outcome in all situations. In a publishing environment that seems to disregard 'negative' results, this is necessary. Should failure be a logistic failure of not publishing in these situations? Requiring that reviewers are considering risk-reward of studies in a quantitative manner seems useful, but whether this institutionalizes pre-existing biases against minority and female researchers should be considered as well.

Patrick Wilson:
It's clearly written but the idea is flawed. Here are 3 objections top of mind:

1) I am uncomfortable with the current trend, possibly inspired by EA and the super-forecasting movement, to get everyone to give percentage probabilities for every prediction. I have recently been taking part in the Tetlock hybrid forecasting and persuasion tournament, and it made me realise how hard it is to assign percentage probabilities for X-risk and other future events.

2) Percentage probabilities are not inherently superior to 9-point scales or other fuzzier and multidimensional measure, and also they're just not how human minds work. The probabilities will tend to cluster on key deciles / extremes and eg 25%, 33%, 75% because it's easier and that's how we are used to thinking.

3) Proposal evaluation / peer review is not only judging how likely a research project is to achieve its stated goals and limited objectives, but also how worthwhile that project actually is. For example I could submit a proposal for something with high confidence the results will achieve what I set out to, and peer-reviewers might agree with me, but it could be something so trivial as to not be worth studying. There's enough science like this around!

Peer review might be broken in many ways, but this isn't one of them. I expect if implemented, this proposal will drive even more people away from peer-review, through pedantic and over-prescriptive insistence on percentage predictions - a fool's errand.

Thomas Gladwin:
The paper addresses a very important topic, is well-written, and presents an interesting focus on requiring probabilities as a building block to further improvements and innovations. 

EMSKE:
1. Does the article contain novel ideas that have the potential to advance science?

An incremental improvement to proposal peer review like the author describes would increase the level of accountability for research projects with uncertain outcome, and a step toward making the review process more meritocratic.  I find the piece also inspiring thinking on whether project peer review could be crowdsourced with historical success weights (i.e. as a quantitative reputation-tracking system) of reviewers, rather than reviewers having to be sourced strictly from within the four walls of a funding agency.

2. Does the Seed include adequate justification for its ideas and how they could advance science?

Broadly yes; Obviously debatable, but I'm not sure that the 'proverbial economist who turns up his nose at a $20 bill on the sidewalk' is the best metaphor for why this improvement to proposal peer review hasn't been done already. The role that organizational politics play shouldn't be understated IMO. Anyone would want to sit in a position of power to 'pick winners'.  Favor-trading, empire-building, dare I simply say, 'nepotism'  - these behaviors aren't unknown to the world of scientific peer review. 

3. Does the Seed contain high-quality writing?

For ease of reading, and ease of the reader following the author's train of thought, I wouldn't mind section headers to visually break-up the text.  Can a couple of relevant metaphorical graphics be sourced from Unsplash?

William Collen:
A splendid idea. I try to do this anyway when making truth claims during conversation. The point about being able to analyze and modify our assumptions and calibrate our predictions was well put.

Dan James:
This article is written in the style of an opinion piece or blog post and whilst lacking the rigour of a more traditionally formatted academic paper, is no less engaging and thought-provoking.

Metaresearch – the scientific study of how science is conducted - is an area where new ideas to solve perceived problems are urgently required. This article focuses on proposal review, a vital first step to secure funding for many research enquiries. 

Unfortunately, as the article explains, proposal review has a high error rate in predicting research success, one far less reliable than the comparator the paper chooses - a meteorological forecast. This article suggests one possible solution - reviewers should assign probabilities to a proposal in much the same way as weather forecasting. 

This idea has a distinct 'Bayesian' feel, and whilst a Bayesian approach to assigning probabilities is not mentioned or developed by the author/s, it is a measure of how well the paper succeeds in presenting a generalised discussion that invites further reflection from the reader.

My specific Bayesian reflection on reading the paper (that could add to the argument in favour of assigning probabilities) is to break proposals down into incremental stages and stage payments (already common practice), with each successive stage updating the priors for any project. Assigning probability scores on a stage-by-stage basis is arguably more realistic/practical than the difficulty for a reviewer of estimating a probability score for the whole project. After the successful completion of the first stage of a proposal, probabilities for following stages could be updated by reviewers with far more confidence.

This article is a helpful contribution to an area that needs original thinking to address problems that have proved intractable to the traditional ways in which science has been conducted, consequently I would definitely recommend this paper for publication.

Dr. Payal B. Joshi:
The premise of peer review as a forecast seems amusing. It is believed that most of them "free" their time to review grants & manuscripts, I think this may work well. Authors could have structured this article in a slightly better manner as the methodology of applying it is not quite clear. How do you propose forecast-based review? Provide a clear example in a lucid way to put across the idea. Overall, the idea may not be pathbreaking, yet has no potential flaw & certainly deserves to see the light of the day.

Jack Arcalon:
A good idea if it causes researchers to think of unknown probabilities extending beyond their specialties in new ways.

Fred Nix:
I'm not convinced a rating system like this would actually work like the author imagines. Who would risk their cushy job or current status with an objective rating?  Would it just skew predictions so everybody shoots for 50/50?  How long would it take for a reviewer to build up enough reviews and outcomes to get a valid statistic?  Would the rating just show as pending until there was a valid statistic, or just publish the preliminary number?  But the current system is broken, and innovation and progress are suffering. It's an interesting idea and worth experimenting with to see how it works.

Subscribe now

References

Ciocca, J., Horowitz, M., Kahn, L., & Ruhl, C. (2021, May 9). How the U.S. government can learn to see the future. Lawfare. https://www.lawfareblog.com/how-us-government-can-learn-see-future

Goetz, T. (2011, June 19). Harnessing the power of feedback loops. WIRED. https://www.wired.com/2011/06/ff-feedbackloop/

Pier, E. L., Raclaw, J., Carnes, M., Ford, C. E., & Kaatz, A. (2019). Laughter and the chair: Social pressures influencing scoring during grant peer review meetings. Journal of General Internal Medicine, 34(4), 513-514.

Lane, J. N., Teplitskiy, M., Gray, G., Ranu, H., Menietti, M., Guinan, E. C., & 

Lakhani, K. R. (2022). Conservatism gets funded? a field experiment on the role of negative information in novel project evaluation. Management Science, 68(6), 4478-4495.

Lee, C. J., Sugimoto, C. R., Zhang, G., & Cronin, B. (2013). Bias in peer review. Journal of the American Society for Information Science and Technology, 64(1), 2-17.

Roese, N. J., & Vohs, K. D. (2012). Hindsight bias. Perspectives on Psychological Science, 7(5), 411-426.

Abstract

Much of the extensive literature on spatial attentional bias is built on measurements using the dot-probe task. In recent years, concerns have been raised about the psychometric properties of bias scores derived from this task. The goal of the current paper is to look ahead and evaluate possible responses of the field to this situation from a metascientific perspective. Therefore, educated guesses are made on foreseeable but preventable future (repeats of) errors. We discuss, first, the issue of overreactions to the disappointing findings, especially in the context of the potential of a new generation of promising variations on the traditional dot-probe task; second, concerns with competition between tasks; and third, the misuse of rationales to direct research efforts. Alternative directions are suggested that may be more productive. We argue that more adequately exploring and testing methods and adjusting scientific strategies will be critical to avoiding suboptimal research and potentially failing to learn from mistakes. The current articulation of arguments and concerns may therefore be of use in discussions arising around future behavioural research into spatial attentional bias and more broadly in psychological science.

Psychological tasks play a central role in psychological science. These are relatively simple stimulus-response games that participants are asked to play, containing specific variations that affect cognition and behaviour. Tasks are designed so that effects, i.e., differences between conditions on observable measurements such as reaction time, can be interpreted in relation to a particular psychological construct of interest. Such effects are used either to test hypotheses directly or to create individual difference variables for use as building blocks in further analyses. Progress in scientific fields may thus heavily depend on the quality of tasks and the processes via which they evolve. The current paper considers developments around one such task: the dot-probe task.

The Dot-probe Task

When confronted with a salient stimulus - for instance, an angry face looking at you in a crowd, a spider crawling across your desk, or a tasty treat on a kitchen table - our attention is preferentially drawn to that particular stimulus over many other stimuli present in the same environment. Attentional biases refer to the ability of certain stimuli to draw or repel attention automatically. For over three decades (MacLeod et al., 1986), the dot-probe task has been the basis for a significant amount of research into attentional bias (Cisler & Koster, 2010; Field & Cox, 2008; Jiang & Vartanian, 2018; Starzomska, 2017). In the usual version of this well-known task (Figure 1), two task-irrelevant cues are presented, followed by a probe at one of their locations. One cue is drawn from a salient category hypothesized to cause an attentional bias; the other is drawn from a neutral control category. Faster (or slower) reaction times to probes presented at the location of salient cues would reflect an attentional bias towards (or away from) that category. The behavioural measure of spatial attentional bias provided by this task has been used to study potential underlying mechanisms of a wide range of disorders, e.g., anxiety (Bantin et al., 2016; B. P. Bradley et al., 1998; Koster et al., 2006), obesity (Vervoort et al., 2021), posttraumatic stress disorder (Bomyea et al., 2017; Wald et al., 2013), and addiction (Field et al., 2016; Townshend & Duka, 2001; Wiers et al., 2016).

Figure 1. Illustration of the dot-probe task

Methodological Concerns

Relatively recent concerns with the reliability of the task, extensively discussed elsewhere (Christiansen et al., 2015; Jones et al., 2018; McNally, 2018; Rodebaugh et al., 2016; Van Bockstaele et al., 2020) are therefore concerning. We only briefly recap this literature. Very low reliability was found in a number of studies that assessed it (Ataya et al., 2012; Brown et al., 2014; Chapman et al., 2017; Dear et al., 2011; Kappenman et al., 2014; Waechter et al., 2014). In some cases, the reliability was close to zero, and certainly not near the level needed for detecting relationships involving individual differences with sufficient statistical power. Such statistical effects would include, for examples, correlations between bias scores and mental health-related variables - roughly speaking, if a measure cannot predict itself, how could it predict a different variable?

If there is a systemic problem with the reliability of measures, then even though there is presumably a subset of important and true findings in the literature, we don’t know whether a particular finding is in that set. Further, such issues add to the plausibility that the literature is affected by publication biases and questionable research practices (J. M. Simmons et al., 2011). This renders it difficult to draw any conclusions from the literature as a whole and implies any such attempt will require statistical methods focused on critical assessment of the literature (Bartoš & Schimmack, 2020). Different authors could currently draw different conclusions on the overall state of the literature; however, the number of publications discussing such psychometric problems, with authors including prominent researchers, confirm that at least a significant part of the field has accepted that there is indeed a serious problem to be addressed. Similarly, as discussed further below, research groups are investing in developing alternative task versions, implying awareness of a need for improvement. That there is, in fact, a problem therefore seems undeniable. (Please see Appendix A for some additional points on assessing psychometric properties.) 

The Current Paper

The aim of the current paper, rather than to provide another review of psychometric issues, is to consider what went wrong from a metascientific perspective and to consider the future of the field. That is: Perhaps it is worth making and communicating educated guesses to avoid something akin to the current situation re-occurring, and only at that point looking back with hindsight, when the damage has been done? From this perspective, we caution against three speculative but predictable reactions to this “psychometric crisis” of spatial attentional bias measurement and suggest constructive responses.

Potential reactions to the psychometric issues

Reaction 1: Rejection versus improvement

One potential reaction to the psychometric crisis would be to reject altogether effects in behavioural performance as a valid way to assess automatic processes, at least concerning spatial attentional bias. This would have quite serious implications. Studies could no longer be performed using the relatively basic equipment needed for performance-based measurement, which would strongly increase inequalities as researchers would differ in the resources needed to make such a shift. Additionally, we would have to accept as a stable situation that our associated understanding and theory is so inadequate that we cannot even consistently measure attentional biases behaviourally.

However, there is an alternative view: That the problem is about the details of the methods rather than the fundamental approach, and that the solution to the psychometric crisis could involve finding better task variants, study designs or analyses. This would involve a much stronger focus on, and appreciation of, basic research exploring and testing details of how task variations affect behavioural measures, as opposed to studies aiming to use those measures to answer complex, real-world, e.g., clinical, questions. The latter kind of studies may be attractive given their apparent potential for impact (Bayley & Phipps, 2019) but would be premature unless the implicit assumption is confirmed that the measures they use are sufficiently reliable and valid. We note a number of alternative versions of spatial attentional bias tasks that have arisen relatively recently to combat some of the concerns about reliability.

First, in the Odd-One Out Task (Heitmann et al., 2021), a matrix of stimuli is presented, and the task is to indicate whether one of the stimuli is an “odd one out” target, e.g., a tool in a matrix otherwise consisting of alcoholic drinks serving as distractors. The nature of the target and of the non-target stimuli varies over trials, and can be used to define various biases; e.g., it can be assessed whether reaction times are slowed when the distractors are alcoholic, or are reduced when the target is alcoholic. The reliability of the task has been improved to relatively good levels as it has been developed.

Second, the Predictive Visual Probe Task aims to assess an anticipatory form of attentional bias (Gladwin et al., 2021). The task presents visually neutral cues that predict the location of upcoming emotionally salient stimuli. Trials throughout the task are randomly determined to be “picture” trials in which only the salient and control stimuli appear, or “probe” trials in which only probe stimuli requiring a response appear. Performance is thus not affected by the actual presentation of the particular exemplars of stimuli on that trial. The anticipatory attentional bias evoked by the predictive cues has been found to have good reliability with optimized designs and to be related to anxiety. The task design may also reveal trial-to-trial effects (Gladwin & Figner, 2019) that are not measurable in the traditional dot-probe task (Maxwell et al., 2022).

Third, the dual probe task has been developed as an alternative to the dot-probe task (Grafton et al., 2021). In this task, two probes are presented, so briefly that participants can only attend and respond to one of them. Results of this task showed excellent reliability and sensitivity to experimental manipulations of attention. All these tasks may be seen as steps in the right direction, arguably representing the evolution of traditional methods. This is not to say that these approaches do not need confirmation of results or that they could not be improved or perhaps combined with each other. However, they do indicate the feasibility of the use of behavioural measures of the dot-probe-like spatial attentional bias.

Reaction 2: Dysfunctional task-level competition

The second suboptimal reaction relates to the way in which such novel methods might compete and evolve, as a specific case of the more general role of competition in science (Fang & Casadevall, 2015; Tiokhin et al., 2021). Given that a new generation of tasks assessing spatial attentional bias is a viable way forward for research into spatial attentional bias, it seems plausible that there will be a tendency for researchers and research groups to aim to produce the one dominant replacement for the traditional dot-probe task.

However, this would risk incentivizing fast but sloppy science, leading to premature selection of a new method based on limited and possibly biased data used to suggest some form of superiority. Further, it seems unlikely that any single task will hit on a truly optimal combination of features for all purposes, especially in relatively early phases of discovery and exploration. Finally, quick convergence on a winning task for use in further studies might well reduce observed heterogeneity, or unexplained variability beyond that expected by sampling error (Linden & Hönekopp, 2021). However, if there are in fact differences between variations in tasks that should measure the same construct, heterogeneity should be considered a valuable signal of a gap in understanding (Linden & Hönekopp, 2021). Such heterogeneity should, therefore, not be obscured by failing to sufficiently explore task variations.

In contrast, research could more positively focus on task features, regardless of the pedigree of researchers, rather than tasks as a whole. Such features could have varying pros and cons, be more or less suitable in different situations, and be more or less suited to combinations of features. Research that permits the time and resources to explore this “task space”, rather than more rigidly comparing candidate tasks, seems essential to providing sufficient data to support sufficient understanding of attentional bias measurement for it to be justified as a method in more complex or applied, e.g., clinical, studies.

Reaction 3: Sensitivity to narratives

The third reaction regards the role of theoretical rationales to conclude that a given task or task feature is either an improvement or flawed a priori. For instance, the dot-probe task could be argued to be more interpretable or informative than the emotional Stroop task (McNally, 2018). It was perhaps more immediately obvious that it is uncertain what could cause differences between trial types in the emotional Stroop task – but does that mean investment of resources in the former task will have been more justified and more productive than in the latter? (Please see also the note in Appendix B on Attention Bias Modification.)

While theoretical arguments are of course an essential component of hypothesis generation and designing, interpreting, and criticizing individual studies, the concern expressed here is that conceptual, narrative (whether verbally or formally/computationally stated) arguments may be given more weight than deserve and thereby potentially derail the direction of research investments. The risk of this inherently follows from the uncertainties around the theory underlying attentional bias. It is not even trivial to define what the “attention” of attentional bias is; it has been questioned whether it is a useful construct at all (Hommel et al., 2019). Early models of attention concerned fundamental, relatively well-defined information processing bottlenecks (Broadbent, 1958; Norman, 1968; Posner & Petersen, 1990; Schneider & Shiffrin, 1977; Treisman, 1964). However, in linking attention to emotion, via theoretical frameworks such as motivated attention (Lang, 1995), motivational activation (M. M. Bradley, 2009), or dual process models (Strack & Deutsch, 2004), a more complex research question arises, namely whether the emotional meaning of stimuli to the individual affects cognitive processing. While there are certainly interesting and potentially valuable studies and models that focus on differentiating aspects of attentional bias (Koster et al., 2006), meaningful scientific theoretical development must be built on sufficiently reliable and precise empirical support, which is of course exactly what has been questioned.

Further, despite their continuing use, the dual systems/dual processes models that form the usual theoretical underpinnings of attentional bias research have been fundamentally criticized, for instance regarding conceptual precision (Keren, 2013; Keren & Schul, 2009). A lack of empirically supported and detailed theory has implications for how valid it is to use theoretical rationales to evaluate methods. Arguments made to decide between tasks can become somewhat arbitrary – which and whose theoretical constructs and considerations will the field pay attention to? And which features of tasks will be considered relevant versus ignored? Implications that, we suggest, follow from this situation are (1) that caution is necessary towards narrative rationales for using one task or another, or steering fields of research, as arguments may rest on only a very limited aspect of potentially relevant factors, all of which are uncertain; (2) that where there are variations in tasks that preclude any one task providing a definitive interpretation, to consider whether they can provide convergent evidence, e.g., by demonstrating effects that are consistent over different threats to interpretations; (3) that empirical evidence of psychometric quality should always be prioritized over merely conceptual (dis)advantages; and (4) that alternative interpretations of effects, drawing on different hypothetical influences, should be seen as an important and positive driver of research progress.

Conclusion

In conclusion, there is a risk of various unproductive reactions to problematic psychometric findings regarding the dot-probe task. There certainly appears to be reason to look back critically on previous research practices, in line with more general concerns with psychological research (John et al., 2012; J. M. Simmons et al., 2011); however, this results at worst in a disappointing lack of usable results, not evidence against hypotheses concerning attentional bias altogether, or against the potential for behavioural study using implicit measures. There are promising avenues of research that could validly be pursued at this point. While the current paper focuses on the tasks used to generate data, such avenues also include innovations in data analysis, such as the analysis of attentional bias variability (Iacoviello et al., 2014; Zvielli et al., 2014). Nonetheless, to avoid potentially repeating history, future work should align itself with practices and attitudes in support of methods and ethics that tend to lead to scientifically valid productivity. Some ways in which alignment could be expressed in the current context are listed below.

(1) Spending sufficient time exploring task features and qualities, focused on solid science (Frith, 2020) and avoiding questionable practices such as hyping (Martin, 2016); this implies incentivizing accurate and ethical behaviour around the acknowledgement of previous work on the development of our fundamental research tools.

(2) Requiring a sufficient empirical basis before the use of task-based measures in costly studies; such use requires it to be true that those measures are reliable and valid, and therefore the replication and extension work necessary for this must be valued.

(3) Acknowledging the risks of potentially premature uniformization of methods, rather than only the possible benefits of uniformization assuming sufficient psychometric qualities.

(4) Applying and advocating caution and scepticism about narrative theoretical rationales used to promote or suppress measures or lines of research.

(5) Accepting that strategic simplifications may need to be accepted; e.g., methodologically, by starting out with the assumption of uniform populations to find robust overall patterns of effects, or theoretically, by allowing that it is not always necessary or optimal to go beyond a general, ecological concept of attention as a naturally unified process of selection for action (Balkenius & Hulth, 1999).

The intention here is to raise awareness and explicitly identify choices. Which choices will in fact be made in practice will depend on a multitude of factors; indeed, while beyond the scope of the current paper, we briefly discuss in Appendix C how the case of the dot-probe task could tie into broader questioning of scientific systems (Tijdink et al., 2021). We hope the articulation of the risks and suggestions raised in the current paper will be of use for researchers interested in implicit measures, as well as broader metascientific issues. There may be unexpected benefits, beyond improvements within the relatively constrained interests of implicit measures, that arise from the current disappointing situation, if it could lead to better understanding of what features of the academic ecosystem tend to lead to good and bad outcomes for scientific progress.

Gardener Comments

Dan James:
This paper is a cogent examination of the dot-probe paradigm and sustains a consistently high level of engagement throughout. What especially recommends this paper is its coordination of the perceived successes and weaknesses of the dot-probe construct, within a broader epistemological context that raises questions about what constitutes evidence not just for Psychology but for Science as a whole. 

This added philosophical dimension makes this paper relevant for researchers of any discipline who may be faced with a similar conundrum as detailed in the paper - that as part of the research process, logically coherent constructs, or the selection of experimental setups and the resulting statistical analysis employed, cannot avoid generalising or simplifying heuristically, and in doing so will always run the risk of biased or false models.

Revision comments:

In an otherwise comprehensive account, this paper could potentially benefit from greater attention to a specific and often neglected problem in psychology – heterogeneity (Linden & Hönekopp 2021). Whilst the problem of individual variance is acknowledged in the paper, I feel a more explicit account of heterogeneity could conceivably form part of the authors discussion concerning a critical analysis of research practices that they highlight as a possible way forward to address some identified problematic psychometric findings of the dot-probe task.

As the authors explain, the essential practicality of the dot-probe construct is its ability to measure attentional bias as a proxy for potential underlying mechanisms or disorders. However, heterogeneity suggests such measurements and their purported behavioural association, should be qualified, especially if the effect sizes are small, because to do otherwise would be to make an assumption that the dot-probe task applies equally well to a large portion, if not all, of the population. In contrast, and to minimise the problem of heterogeneity, would better specified and smaller subsections or target populations produce better measurement and predictive success? 

Clearly, as the paper explains, in any meta-analysis, measuring individual variance is an important factor in determining the theoretical scope of a psychometric assessment and the dot-probe paradigm is no exception to this.

But there is another sense in which a consideration of heterogeneity is important as part of a methodological principle, that is not mentioned in the paper, and that is to avoid the so-called ‘paradox of convergence’ where there is a disconnect between an accumulation or convergence of evidence in favour of a particular theory and yet at the same time an inability to be sure that such theory can accurately be applied to any individual (Davis-Stober C, Regenwetter M.,2019). 

Recommendation:

The interaction of heterogeneity with an accumulation of evidence, means it’s not hard to foresee a crisis of confidence questioning the theoretical rationale for any psychometric construct, in particular the dot-probe paradigm, the precise subject of the paper under review and the reason why I suggest a greater emphasis on heterogeneity would only be of benefit to an already fascinating and well-researched paper that is highly recommended for publication.

Heidi Zamzow:
This piece is very well-written and referenced, with a strong and coherent argument.  The contribution to the field is clearly demonstrated, not only with respect to methods using implicit measures specifically, but to a systematic critique of methodology more broadly.  In particular, I appreciated the discussion on potential lines of future research. Overall, I found the paper to be highly relevant, and I strongly endorse its publication.

Dr. Payal B. Joshi:

The article takes a dive in a difficult premise and attempts to provide an explanation on the psychometric crisis. Authors have critically discussed and presented their arguments in a clear and engaging manner. Overall, the article is a decent attempt to dot-probe tasks on ABM and metascience literature.

Declarations

Conflict of interest statement: This research was not supported by any grant. The authors have no competing interests to declare that are relevant to the content of this article.

Ethical Statement and Informed Consent: The current paper did not involve any human or animal data collection and did not involve ethical approval or informed consent.

Data availability statement: Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

References

Ataya, A. F., Adams, S., Mullings, E., Cooper, R. M., Attwood, A. S., & Munafò, M. R. (2012). Internal reliability of measures of substance-related cognitive bias. Drug and Alcohol Dependence, 121(1–2), 148–151. https://doi.org/10.1016/j.drugalcdep.2011.08.023

Balkenius, C., & Hulth, N. (1999). Attention as selection-for-action: A scheme for active perception. 1999 Third European Workshop on Advanced Mobile Robots (Eurobot’99). Proceedings (Cat. No.99EX355), 113–119. https://doi.org/10.1109/EURBOT.1999.827629

Bantin, T., Stevens, S., Gerlach, A. L., & Hermann, C. (2016). What does the facial dot-probe task tell us about attentional processes in social anxiety? A systematic review. Journal of Behavior Therapy and Experimental Psychiatry, 50. https://doi.org/10.1016/j.jbtep.2015.04.009

Bartoš, F., & Schimmack, U. (2020). Z-curve. 2.0: Estimating replication rates and discovery rates.

Bayley, J. E., & Phipps, D. (2019). Building the concept of research impact literacy. Evidence & Policy: A Journal of Research, Debate and Practice, 15(4), 597–606. https://doi.org/10.1332/174426417X15034894876108

Bomyea, J., Johnson, A., & Lang, A. J. (2017). Information Processing in PTSD: Evidence for Biased Attentional, Interpretation, and Memory Processes. Psychopathology Review, a4(3), 218–243. https://doi.org/10.5127/pr.037214

Bradley, B. P., Mogg, K., Falla, S. J., & Hamilton, L. R. (1998). Attentional Bias for Threatening Facial Expressions in Anxiety: Manipulation of Stimulus Duration.
Cognition & Emotion, 12(6), 737–753. https://doi.org/10.1080/026999398379411

Bradley, M. M. (2009). Natural selective attention: Orienting and emotion. Psychophysiology, 46(1), 1–11. https://doi.org/10.1111/j.1469-8986.2008.00702.x

Broadbent, D. E. (1958). The effects of noise on behaviour. In Perception and
communication (pp. 81–107). Pergamon Press. https://doi.org/10.1037/10037-005

Brown, H. M., Eley, T. C., Broeren, S., MacLeod, C. M., Rinck, M., Hadwin, J. A., & Lester, K. J. (2014). Psychometric properties of reaction time based experimental paradigms measuring anxiety-related information-processing biases in children. Journal of Anxiety Disorders, 28(1), 97–107. https://doi.org/10.1016/j.janxdis.2013.11.004

Chapman, A., Devue, C., & Grimshaw, G. M. (2017). Fleeting reliability in the dot-probe task. Psychological Research. https://doi.org/10.1007/s00426-017-0947-6

Christiansen, P., Mansfield, R., Duckworth, J., Field, M., & Jones, A. (2015). Internal reliability of the alcohol-related visual probe task is increased by utilising personalised stimuli and eye-tracking. Drug and Alcohol Dependence, 155, 170–174. https://doi.org/10.1016/j.drugalcdep.2015.07.672

Cisler, J. M., & Koster, E. H. W. (2010). Mechanisms of attentional biases towards threat in anxiety disorders: An integrative review. Clinical Psychology Review, 30(2), 203–216. https://doi.org/10.1016/j.cpr.2009.11.003

Cristea, I. A., Kok, R. N., & Cuijpers, P. (2015). Efficacy of cognitive bias modification interventions in anxiety and depression: Meta-analysis. The British Journal of Psychiatry, 206(1), 7–16. https://doi.org/10.1192/bjp.bp.114.146761

Cristea, I. A., Kok, R. N., & Cuijpers, P. (2016). The effectiveness of cognitive bias modification interventions for substance addictions: A meta-analysis. PloS One, 11(9), e0162226. https://doi.org/10.1371/journal.pone.0162226

De Schryver, M., Hughes, S., Rosseel, Y., & De Houwer, J. (2016). Unreliable Yet Still Replicable: A Comment on LeBel and Paunonen (2011). Frontiers in Psychology, 6. https://doi.org/10.3389/fpsyg.2015.02039

Dear, B. F., Sharpe, L., Nicholas, M. K., & Refshauge, K. (2011). The psychometric properties of the dot-probe paradigm when used in pain-related attentional bias research. The Journal of Pain : Official Journal of the American Pain Society, 12(12), 1247–1254. https://doi.org/10.1016/j.jpain.2011.07.003

Dehue, T. (1995). Changing the rules: Psychology in the Netherlands, 1900–1985 (pp. ix, 204). Cambridge University Press.

Evans, T. M., Bira, L., Gastelum, J. B., Weiss, L. T., & Vanderford, N. L. (2018). Evidence for a mental health crisis in graduate education. Nature Biotechnology 2018 36:3, 36(3), 282–284. https://doi.org/10.1038/nbt.4089

Fang, F. C., & Casadevall, A. (2015). Competitive Science: Is Competition Ruining Science? Infection and Immunity. https://doi.org/10.1128/IAI.02939-14

Field, M., & Cox, W. M. (2008). Attentional bias in addictive behaviors: A review of its development, causes, and consequences. Drug and Alcohol Dependence, 97(1–2), 1–20. https://doi.org/10.1016/j.drugalcdep.2008.03.030

Field, M., Werthmann, J., Franken, I., Hofmann, W., Hogarth, L., & Roefs, A. (2016). The role of attentional bias in obesity and addiction. Health Psychology, 35(8), 767–780. https://doi.org/10.1037/hea0000405

Fodor, L. A., Georgescu, R., Cuijpers, P., Szamoskozi, Ş., David, D., Furukawa, T. A., & Cristea, I. A. (2020). Efficacy of cognitive bias modification interventions in anxiety and depressive disorders: A systematic review and network meta-analysis. The Lancet Psychiatry, 7(6), 506–514. https://doi.org/10.1016/S2215-0366(20)30130-9

Frith, U. (2020). Fast Lane to Slow Science. Trends in Cognitive Sciences, 24(1), 1–2. https://doi.org/10.1016/j.tics.2019.10.007

Gladwin, T. E. (2017). Negative effects of an alternating-bias training aimed at attentional flexibility: A single session study. Health Psychology and Behavioral Medicine, 5(1), 41–56. https://doi.org/10.1080/21642850.2016.1266634

Gladwin, T. E. (2018). Educating students and future researchers about academic misconduct and questionable collaboration practices. International Journal for Educational Integrity, 14(1), 10. https://doi.org/10.1007/s40979-018-0034-9

Gladwin, T. E., & Figner, B. (2019). Trial-to-trial carryover effects on spatial attentional bias. Acta Psychologica, 196, 51–55. https://doi.org/10.1016/J.ACTPSY.2019.04.006

Gladwin, T. E., Halls, M., & Vink, M. (2021). Experimental control of conflict in a predictive visual probe task: Highly reliable bias scores related to anxiety. Acta Psychologica, 218, 103357. https://doi.org/10.1016/j.actpsy.2021.103357

Grafton, B., Teng, S., & MacLeod, C. M. (2021). Two probes and better than one: Development of a psychometrically reliable variant of the attentional probe task. Behaviour Research and Therapy, 138, 103805. https://doi.org/10.1016/j.brat.2021.103805

Heitmann, J., Jonker, N. C., & de Jong, P. J. (2021). A Promising Candidate to Reliably Index Attentional Bias Toward Alcohol Cues–An Adapted Odd-One-Out Visual Search Task. Frontiers in Psychology, 12, 102. https://doi.org/10.3389/fpsyg.2021.630461

Hommel, B., Chapman, C. S., Cisek, P., Neyedli, H. F., Song, J.-H., & Welsh, T. N. (2019). No one knows what attention is. Attention, Perception, & Psychophysics, 81(7), 2288–2303. https://doi.org/10.3758/s13414-019-01846-w

Huistra, P., & Paul, H. (2022). Systemic Explanations of Scientific Misconduct: Provoked by Spectacular Cases of Norm Violation? Journal of Academic Ethics, 20(1), 51–65. https://doi.org/10.1007/s10805-020-09389-8

Iacoviello, B. M., Wu, G., Abend, R., Murrough, J. W., Feder, A., Fruchter, E., Levinstein, Y., Wald, I., Bailey, C. R., Pine, D. S., Neumeister, A., Bar-Haim, Y., & Charney, D. S. (2014). Attention bias variability and symptoms of posttraumatic stress disorder. Journal of Traumatic Stress, 27(2), 232–239. https://doi.org/10.1002/jts.21899

Jiang, M. Y. w., & Vartanian, L. R. (2018). A review of existing measures of attentional biases in body image and eating disorders research. Australian Journal of Psychology, 70(1), 3–17. https://doi.org/10.1111/ajpy.12161

John, L. K., Loewenstein, G., & Prelec, D. (2012). Measuring the Prevalence of Questionable Research Practices With Incentives for Truth Telling. Psychological Science, 23(5), 524–532. https://doi.org/10.1177/0956797611430953

Jones, A., Christiansen, P., & Field, M. (2018). Failed attempts to improve the reliability of the alcohol visual probe task following empirical recommendations. Psychology of Addictive Behaviors: Journal of the Society of Psychologists in Addictive Behaviors, 32(8), 922–932. https://doi.org/10.1037/adb0000414

Kappenman, E. S., Farrens, J. L., Luck, S. J., & Proudfit, G. H. (2014). Behavioral and ERP measures of attentional bias to threat in the dot-probe task: Poor reliability and lack of correlation with anxiety. Frontiers in Psychology, 5(DEC), 1368. https://doi.org/10.3389/fpsyg.2014.01368

Keren, G. (2013). A tale of two systems: A scientific advance or a theoretical stone soup? Commentary on Evans & Stanovich (2013). Perspectives on Psychological Science, 8(3), 257–262.
https://doi.org/10.1177/1745691613483474

Keren, G., & Schul, Y. (2009). Two is not always better than one: A critical evaluation of two-system theories. Perspectives on Psychological Science, 4(6), 533–550. https://doi.org/10.1111/j.1745-6924.2009.01164.x

Koster, E. H. W., Crombez, G., Verschuere, B., Van Damme, S., & Wiersema, J. R. (2006). Components of attentional bias to threat in high trait anxiety: Facilitated engagement, impaired disengagement, and attentional avoidance. Behaviour Research and Therapy, 44(12), 1757–1771. https://doi.org/10.1016/j.brat.2005.12.011

Kruijt, A.-W., Parsons, S., & Fox, E. (2019). A meta-analysis of bias at baseline in RCTs of attention bias modification: No evidence for dot-probe bias towards threat in clinical anxiety and PTSD. Journal of Abnormal Psychology, 128(6), 563–573. https://doi.org/10.1037/abn0000406

Lang, P. J. (1995). The emotion probe. Studies of motivation and attention. The American Psychologist, 50(5), 372–385.

Linden, A. H., & Hönekopp, J. (2021). Heterogeneity of research results: A new perspective from which to assess and promote progress in psychological science. Perspectives on Psychological Science, 16, 358–376. https://doi.org/10.1177/1745691620964193

MacLeod, C. M., Mathews, A., & Tata, P. (1986). Attentional bias in emotional disorders. Journal of Abnormal Psychology, 95(1), 15–20.

Martin, B. (2016). Plagiarism, misrepresentation, and exploitation by established professionals: Power and tactics. In Handbook of Academic Integrity (pp. 913–927). Springer Singapore. https://doi.org/10.1007/978-981-287-098-8_75

Maxwell, J. W., Fang, L., & Carlson, J. M. (2022). Do Carryover Effects Influence Attentional Bias to Threat in the Dot-Probe Task? Journal of Trial & Error, 2(1). https://doi.org/10.36850/e9

McNally, R. J. (2018). Attentional bias for threat: Crisis or opportunity? Clinical Psychology Review, 69, 4–13. https://doi.org/10.1016/J.CPR.2018.05.005Siri

Mogg, K., Waters, A. M., & Bradley, B. P. (2017). Attention Bias Modification (ABM): Review of Effects of Multisession ABM Training on Anxiety and Threat-Related Attention in High-Anxious Individuals. Clinical Psychological Science, 216770261769635. https://doi.org/10.1177/2167702617696359

Norman, D. A. (1968). Toward a theory of memory and attention. Psychological Review, 75(6), 522–536. https://doi.org/10.1037/h0026699

Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25–42. https://doi.org/10.1146/annurev.ne.13.030190.000325

Puls, S., & Rothermund, K. (2018). Attending to emotional expressions: No evidence for automatic capture in the dot-probe task. Cognition and Emotion, 32(3), 450–463. https://doi.org/10.1080/02699931.2017.1314932

Rodebaugh, T. L., Scullin, R. B., Langer, J. K., Dixon, D. J., Huppert, J. D., Bernstein, A., Zvielli, A., & Lenze, E. J. (2016). Unreliability as a threat to understanding psychopathology: The cautionary tale of attentional bias. Journal of Abnormal Psychology, 125(6), 840–851. https://doi.org/10.1037/abn0000184

Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84(1), 1–66. https://doi.org/10.1037/0033-295X.84.1.1

Simmons, J. M., Minamimoto, T., Murray, E. A., Richmond, B. J., Bouret, S., Kennerley, S. W., & Wallis, J. D. (2011). Reward-Related Neuronal Activity During Go-Nogo Task Performance in Primate Orbitofrontal Cortex Reward-Related Neuronal Activity During Go-Nogo Task Performance in Primate Orbitofrontal Cortex. Journal of Neurophysiology, 1864–1876.

Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22(11), 1359–1366. https://doi.org/10.1177/0956797611417632

Starzomska, M. (2017). Applications of the dot probe task in attentional bias research in eating disorders: A review. Psicológica, 38(2), 283–346.

Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of social behavior. Personality and Social Psychology Review, 8, 220–247. https://doi.org/10.1207/s15327957pspr0803

Tiggemann, M., & Kemps, E. (2020). Is Sham Training Still Training? An Alternative Control Group for Attentional Bias Modification. Frontiers in Psychology, 11. https://www.frontiersin.org/articles/10.3389/fpsyg.2020.583518

Tijdink, J. K., Horbach, S. P. J. M., Nuijten, M. B., & O’Neill, G. (2021). Towards a Research Agenda for Promoting Responsible Research Practices. Journal of Empirical Research on Human Research Ethics, 16(4), 450–460. https://doi.org/10.1177/15562646211018916

Tiokhin, L., Yan, M., & Morgan, T. J. H. (2021). Competition for priority harms the reliability of science, but reforms can help. Nature Human Behaviour, 5(7), 857–867. https://doi.org/10.1038/s41562-020-01040-1

Townshend, J. M., & Duka, T. (2001). Attentional bias associated with alcohol cues: Differences between heavy and occasional social drinkers. Psychopharmacology, 157(1), 67–74. https://doi.org/10.1007/s002130100764

Treisman, A. M. (1964). Selective attention in man. British Medical Bulletin, 20(1), 12–16.

Van Bockstaele, B., Lamens, L., Salemink, E., Wiers, R. W., Bögels, S. M., & Nikolaou, K. (2020). Reliability and validity of measures of attentional bias towards threat in unselected student samples: Seek, but will you find? Cognition & Emotion, 34(2), 217–228. https://doi.org/10.1080/02699931.2019.1609423

Vervoort, L., Braun, M., De Schryver, M., Naets, T., Koster, E. H. W., & Braet, C. (2021). A Pictorial Dot Probe Task to Assess Food-Related Attentional Bias in Youth With and Without Obesity: Overview of Indices and Evaluation of Their Reliability. Frontiers in Psychology, 12, 561. https://doi.org/10.3389/fpsyg.2021.644512

Waechter, S., Nelson, A. L., Wright, C., Hyatt, A., & Oakman, J. (2014). Measuring attentional bias to threat: Reliability of dot probe and eye movement indices. Cognitive Therapy and Research, 38(3), 313–333. https://doi.org/10.1007/s10608-013-9588-2

Wald, I., Degnan, K. A., Gorodetsky, E., Charney, D. S., Fox, N. A., Fruchter, E., Goldman, D., Lubin, G., Pine, D. S., & Bar-Haim, Y. (2013). Attention to threats and combat-related posttraumatic stress symptoms: Prospective associations and moderation by the serotonin transporter gene. JAMA Psychiatry, 70(4), 401–408. https://doi.org/10.1001/2013.jamapsychiatry.188

Wiers, C. E., Gladwin, T. E., Ludwig, V. U., Gröpper, S., Stuke, H., Gawron, C. K., Wiers, R. W., Walter, H., & Bermpohl, F. (2016). Comparing three cognitive biases for alcohol cues in alcohol dependence. Alcohol and Alcoholism, 1(7), 1–7. https://doi.org/10.1093/alcalc/agw063

Zvielli, A., Bernstein, A., & Koster, E. H. W. (2014). Dynamics of attentional bias to threat in anxious adults: Bias towards and/or away? PloS One, 9(8), e104025. https://doi.org/10.1371/journal.pone.0104025

Appendices

A. Additional notes on assessing psychometric properties of the dot-probe task

We briefly note that it is essential that the reliability of the bias scores is assessed correctly. If the interest is in a bias, the relevant reliability is that of the difference score between reaction times on different conditions measuring that bias. For instance, consider the reaction times to probes at threat locations, and reaction times to probes at control locations. The reliabilities of the simple reaction times per condition separately, or the reliability of the reaction time averaged over both conditions, do not assess the reliability of the bias. In contrast, the reliability of simple or averaged reaction times will trivially reflect systematic variability due to participants being generally fast or slow. Taking the difference score per participant and assessing its reliability is necessary to assess the reliability of the bias.

Further, we emphasize that low reliability, while critical to studies involving individual differences, does not impact the replicability or size of within-subject effects (De Schryver et al., 2016). To consistently measure a within-subject effect, any between-subject variability in the effect would ideally be low, while the reliability of an individual difference measure benefits from high between-subject variability of individuals’ true scores (reducing between-subject error, of course, benefits both aims). However, there are also concerns even with the robustness of such relatively simple, overall effects as well (Puls & Rothermund, 2018). For instance, in a re-analysis of pre-test data originally used for intervention studies, there was no significant bias towards threat in clinically anxious individuals, where it would have been expected (Kruijt et al., 2019). This was an exceptional finding as the primary aim of these studies involved testing effects over time of interventions, rather than biases at pre-test, mitigating the most likely effects of questionable research practices (John et al., 2012; J. M. Simmons et al., 2011).

B. From the dot-probe task to Attention Bias Modification (ABM)

We briefly note that issues involving assessment tasks may also affect the course of research on derived training paradigms such as ABM. In ABM, the dot-probe task is converted to a training intervention. This is done by manipulating the probability of the location of the probe stimulus, in order to train attention towards versus away from particular stimulus categories. For example, in a task with alcoholic and non-alcoholic beverages, the probe would be presented at the non-alcoholic location, in hopes of reducing an assumed attentional bias towards alcohol. While the current article focuses on the assessment form of the dot-probe task, results of directional ABM versus control conditions have arguably been similarly disappointing (Cristea et al., 2015, 2016; Fodor et al., 2020; Mogg et al., 2017). The rationale of the dot-probe task suggests specifically directional training manipulations - “towards” versus “away”. The control training condition would then be random, with probe location being unrelated to cue locations, or trained in the opposite direction as desirable from a clinical perspective. An issue raised as a possible reason for weak effects of ABM is control conditions are, in fact, still serving a training purpose (Tiggemann & Kemps, 2020); one possibility is that participants are learning that the cues are irrelevant  (Gladwin, 2017). Similarly, despite the directionality, training attention “away” from a given stimulus category still makes cue locations task-relevant and hence maintains, or induces, the salience of the cues. Interestingly, these concerns would seem to have been more immediately evident from the perspective of the emotional Stroop task. It seems likely that the focus of ABM, if it had been linked primarily to that task, would have shifted towards training aimed at the reinterpretation of salient stimuli as irrelevant. It is of course unknown whether this would have produced more clinically relevant results.

C. The dot-probe task as a crowbar into scientific structure

That is: it may seem that methodological debate around the dot-probe task merely concerns a rather localized issue of suboptimal scientific behaviours around one particular implicit measure. However, it may be useful to consider a connection to concerning scientific events that are more “spectacular” (Huistra & Paul, 2022), such as the Stapel affair and continued subsequent (inter)nationally publicized cases of PIs committing fraud or other unethical breaches. At least some academics have considered systemic explanations for these events, rather than merely issues related to the individual (Huistra & Paul, 2022). Less conspicuous problems that exist within business as usual - such as the current topic, but also questionable research practices in general (J. P. Simmons et al., 2011), academic exploitation (Gladwin, 2018; Martin, 2016), and mental health crises in PhD students (Evans et al., 2018) - could similarly warrant questioning of scientific systems. It seems at least conceivable that there are common underlying causes involving the way in which science is currently organized. A detailed analysis of the history of the dot-probe task could therefore be of interest from a philosophy of science perspective. How do methods and ideas spread, become acquired, and suppressed, and how do such processes relate to the distribution, acquisition and maintenance of institutional power? Or, put differently, what are the assumed and perhaps unscrutinised organizing structures and “rules of the game” (Dehue, 1995) and what are their consequences?

Abstract

Exosomes are small vesicles, secreted by eukaryotic cells, containing molecular cargo that reflects the biochemical composition of the origin cell, including protein and RNA. Once secreted, exosomes can enter the circulatory system and be found in blood, urine, and saliva. It has been hypothesized that because exosomes contain transmembrane proteins unique to their cell of origin, specific populations of exosomes could be non-invasively extracted from the bloodstream. The protein L1CAM may serve as a marker of neuronal exosomes. However, although “neuron-derived-exosomes'' could offer some specific information about in-vivo molecular neurobiology, this population of exosomes still provides a relatively noisy signal, including data on protein expression from a variety of different neuronal subpopulations. We argue that it may be possible to isolate brain-region-specific exosomes, and that data derived from these exosomes would provide a superior diagnostic tool. 

Background

Like other cells, neurons secrete exosomes (Budnik, 2016). Moreover, exosome release is coupled to neuronal function: stimulating neuronal activity can increase the release of exosomes in cell culture (Faure, 2006). Supporting this finding, fully matured adult cortical neurons secrete exosomes, and the number released can be regulated by the activity of glutamate and GABA receptors (Lacehnal, 2011). Therefore, exosome activity is at least partially reflective of neuronal activity.

In the field of psychiatry, interest in personalized approaches towards treatment has been growing over the past decade (Saeedi et al., 2019). This is partially due to the varying effectiveness of treatments among individual patients. Isolating brain-region-specific exosomes could provide a step into understanding and improving patient treatment. 

Already, it has been shown that the content of neuronal-exosomes, marked by the transmembrane protein L1CAM, varies based on the presence of neurodegenerative diseases, such as Alzheimer’s or Parkinson’s (Goetzl et al., 2016 and Shi, 2014). Moreover, in general, pre-clinical work has shown that neuronal exosomes contain protein and RNA that are related to mental illness (Faure, 2006). For example, SLC1A3, a glutamate transporter, can be found in neuronal exosomes (Faure, 2006) and is associated with suicidal behavior (Murphy, 2011). Another example is TUBA1A: genetic mutations of which are associated with neuroanatomical and cognitive abnormalities (Tischfield, 2011). Both TUBA1A and another neuronal exosomal protein HspA8 have altered expression levels in post-mortem tissue of suicide victims (Kekesi, 2012).

Advancing the Specificity of Exosome Biomarker Research

We propose a novel method of isolating brain-region-specific exosomes using a serial immunoprecipitation method. 

Two simple approaches for extracting exosomes, ultracentrifugation and ultrafiltration, differentiate exosomes by size. These approaches clearly cannot be used to identify the anatomical origins of an exosome. Other techniques use immunoprecipitation in which exosomes are isolated on the basis of a specific transmembrane protein. Several labs have provided evidence that L1CAM is a marker of brain-derived exosomes  (Goetzl et al., 2016 and Shi, 2014). 

However, even existing immunoprecipitation protocols lack specificity: L1CAM is likely expressed throughout the brain (Uhlén, 2015). Centuries of research has shown that discrete brain regions exhibit different relationships with different aspects of psychology. If, hypothetically, psychiatrists were looking for a biomarker of treatment response in anxiety disorders, they may be more likely to find relevant biomarkers in the amygdala or prefrontal cortex rather than the visual cortex or cerebellum.

Abstractly, our proposal is simple: first use immunoprecipitation to extract L1CAM-exosomes from the bloodstream, then use a second round of immunoprecipitation for another protein to extract exosomes from brain-region-X.

This method relies on two assumptions that would need to be verified. The first assumption, which already has substantial support, is simply that immunoprecipitation using L1CAM isolates neuronal exosomes (see caveat 1). The second assumption is more speculative, that a second round of immunoprecipitation, using another transmembrane protein, could be used to isolate neuronal exosomes from a specific brain region or neuron-type. 

Caveat #1: L1CAM Might Not Be a Good Marker of Neuronally-Derived Exosomes 

One assumption in our hypothesis is that L1CAM can be used to isolate brain-derived exosomes. Several lines of evidence do support this concept. L1CAM is a transmembrane protein found in exosomes of cultured cortical neurons (Faure, 2006). In the human bloodstream, immunoprecipitation for L1CAM (theoretically, neuronal exosomes) yields molecular markers of Parkinson’s (Shi, 2014) and Alzheimer’s disease (Fiandaca, 2015; Kapogiannis, 2015). The fact that exosomes reflect the neuropathology of human subjects is further evidence that L1CAM exosomes are brain-derived. 

However, a recent publication has controversially called this research into question (Norman, 2021). In this experiment, researchers demonstrate that L1CAM, in human blood plasma, exists in a soluble form (unassociated with exosomes). However, these contradictory results are not fully accepted, and L1CAM continues to be used to isolate brain-derived exosomes (Hornung, 2020). Furthermore, even if L1CAM is ineffective at isolating neuronal exosomes, other proteins could theoretically do the same job.

Candidates for the Second Round of Immunoprecipitation

Assuming that either L1CAM or another protein could serve as a specific marker of “brain” or “neuronal” exosomes, a second protein would only need to provide specificity within the brain. To find a plausible second protein, we applied three criteria: the protein must be 1) transmembrane, 2) associated with exosomes, and 3) selectively expressed or enriched in a certain brain region.

Using a Python script, we searched two databases on exosomal protein expression, Vesiclepedia (Kalra, 2012) and Exocarta (Mathivanan, 2012), and cross-referenced those with data on regional and subcellular RNA and protein expression from the Human Protein Atlas (Uhlen, 2015) and a manual literature search. Using human post-mortem RNA and protein immunostaining data, along with exosome proteomics studies, a list of plausible protein markers for brain-region and/or cell-type specific exosomes is presented below, followed by an important caveat regarding our methodology.

GABRD - Cerebellum: GABRD (Gamma-aminobutyric acid type A receptor delta subunit) makes up part of one variant of the GABA-A receptor that is regionally enhanced in the cerebellum (Enoch, 2013). This protein has been identified in urine exosomes (Fraser, 2013).

GRM1- Cerebellum: GRM1 (Glutamate metabotropic receptor 1) is a g-protein coupled receptor for glutamate. It is regionally enhanced in the cerebellum, particularly in Purkinje cells (Nakao, 2019).  It has been identified in serum and urine exosomes (Musante, 2012).

SLC6A3- Midbrain: SLC6A3 (solute carrier family 6 member 3) transports dopamine into the cell, and is regionally enhanced in the midbrain (Shimada, 1992). It has been detected in exosomes derived from T cells and in urine (Perez Hernandez, 2013 and Fraser, 2013).

SLC6A4 - Pons and Medulla / Serotonergic Neurons: SLC6A4 (solute carrier family 6 member 4) transports serotonin from the synaptic cleft back into the presynaptic terminal. It is only present in the pons and medulla. In a study of 5-HTT mRNA expression, it was found to be greatly expressed in the raphe nuclei (McLaughlin, 1996). In a study of platelet exosomes, SLC6A4 was identified (Pienimaeki‐Roemer, 2015). Although the 5-HTT protein is likely distributed throughout the brain (Bengel, 1997), this protein could still serve as a marker of serotonergic neurons (and perhaps would be more valuable as such, with depression and anxiety in mind).

SLC18A2- Hypothalamus, Midbrain, Pons and Medulla: SLC18A2 (solute carrier family 18 member A2) serves as an ATP-dependent transporter of monoamines into synaptic vesicles. It is regionally enriched in those brain regions, and it has been identified in cancer exosomes (Hurwitz, 2016)

Caveat #2 - There is Not Great Data on Molecular Variation of Exosomes Across Brain Regions.

Ideally, this project would search an atlas of exosomal protein expression throughout the brain (Created by the good people at the Allen Brain Observatory or Human Protein Atlas). Hypothetically, using post-mortem human brain tissue, exosomes could be isolated and profiled for their protein content, and this dataset would provide a list of exosomal proteins that are unique to each brain region. 

In the absence of this dataset, we rely on indirect data. To confirm whether a protein is exosomal, we do not require that the protein has been found in association with “neuronal” exosomes. To our knowledge, data exists on cultured cortical neurons (Faure, 2006), stem cells (Kang, 2008), human cerebrospinal fluid (Guha, 2019), and even from the human post-mortem frontal cortex (Muraoka, 2020), but no comprehensive catalog of exosomes derived from specific brain regions exists to our knowledge. Because exosomes derived from certain brain regions may not be represented in prior studies, our only criteria is that a putative candidate for the second round of immunoprecipitation is that the protein has been found in association with exosomes derived from any cell type. Therefore, we cannot say with confidence that these proteins are found in exosomes in their respective neuronal populations.

An Abstract Protocol For Experimental Confirmation

Any further research should first seek to validate whether these proteins are found in exosomes derived from their respective brain regions. This could be accomplished with post-mortem tissue in one of two ways. Simply, each putative protein could be co-localized with exosomal markers using in-situ hybridization. Alternatively, and perhaps complimentarily, perhaps by isolating bulk-exosomes from post-mortem tissue (using ultracentrifugation) and evaluating the protein contents of the exosomal fraction.

Second, it must be confirmed that exosomes with these proteins can be found in the bloodstream. Developing the serial immunoprecipitation procedure may represent the biggest technical challenge of this project, as exosomes must be 1) captured via immunoprecipitation, 2) washed and eluted from the antibodies, and 3) further immunoprecipitated, all without damaging the structure of membrane proteins or the lipid structure of the exosome itself. If this procedure could be developed, the presence of exosomes can be validated in a typical way (Théry, 2018).

If exosomes with these proteins are in the blood, researchers can then further validate whether they contain molecular contents that corroborate their specific origin. Perhaps the total concentration of each RNA and protein contained within each exosomal-population can be compared to existing databases on protein and RNA expression. If at least some markers show an expression pattern similar to their cell-type of origin, that would provide further evidence for the validity of this method.

All that being said, perhaps the first step should simply be to create a comprehensive catalog of exosomal proteins throughout the brain. This catalog would allow scientists to develop more concrete hypotheses about which proteins are likely to distinguish neuronal exosomes from non-neuronal exosomes, and exosomes derived from one brain region or one cell-type from another.

Conclusion

In this paper, we propose a two-step immunoprecipitation method that could isolate exosomes from specific brain regions. The first step is to use L1CAM to isolate brain derived exosomes, the second step would be to use a second protein to isolate exosomes derived from a certain brain region or neuronal population. In this paper, we identified transmembrane proteins that could serve as plausible markers for isolating exosomes from specific brain regions. SLC6A4, as a marker of serotonergic neurons, may be particularly interesting to mental health researchers.

If confirmed, this method could provide a noninvasive diagnostic tool to create personalized treatments in psychiatry. If these findings could be corroborated, brain-region-specific neuronal exosomes could provide an unprecedented glimpse into the molecular composition of the living human brain.

Gardener Comments

Ted Wade:
As far as I can tell, the biology of extracellular vesicles, and their smallest variety, exosomes, is a hot topic, albeit with many, many open questions.  We might be a long way yet from translating this basic research into clinical applications, but the prospect of biochemical probes directly into living human brain is definitely high reward.   Reliance on the L1CAM protein as a first step in enriching samples is apparently controversial, and might represent the bulk of risk in the proposed approach.  The exosomes also have to survive through the second filtering step, which could mean that the entire assay technique might be quite tricky.

One of the references (Saeedi et al) has already proposed exosome use for personalized mental health applications, so the basic idea has already surfaced.  That said, relating exosome contents to mental function might need to go beyond correlations with macro brain regions or specific transmitters. Some more powerful theories now deal with the activity of multi-regional neural circuits/networks, which can now be studied in vivo in real time.  To correlate such activity with exosomes would have to resolve temporal differences in the occurrence and sampling of the two measures.  That seems challenging.

Roger’s Bacon:
Is it possible to modify the contents of the exosomes (with gene editing, pharmacology, or…)? I can imagine the engineering of exosomes to include proteins that are easy to isolate or have some other function (e.g. a reporter or a medicine). Exosome engineering might serve as the basis of a useful diagnostic or therapeutic tool.

R. Sal Reyes:
This kind of noninvasive method for determining specific kinds of activity in specific brain regions could not only provide great benefits in the treatment of neurological and mental health issues, but it could also possibly provide new avenues for studying & clarifying the functions of specific brain regions by using these methods to gather data in behavioral experiments.

Subscribe now

References

1. Bengel, D., Jöhren, O., Andrews, A. M., Heils, A., Mößner, R., Sanvitto, G. L., ... & Murphy, D. L. (1997). Cellular localization and expression of the serotonin transporter in mouse brain. Brain research, 778(2), 338-345.

2. Budnik, V., Ruiz-Cañada, C., & Wendler, F. (2016). Extracellular vesicles round off communication in the nervous system. Nature Reviews Neuroscience, 17(3), 160-172.

3. Enoch, M. A., Baghal, B., Yuan, Q., & Goldman, D. (2013). A factor analysis of global GABAergic gene expression in human brain identifies specificity in response to chronic alcohol and cocaine exposure. PLoS One, 8(5), e64014.

4. Fauré, J., Lachenal, G., Court, M., Hirrlinger, J., Chatellard-Causse, C., Blot, B., ... & Kirchhoff, F. (2006). Exosomes are released by cultured cortical neurons. Molecular and Cellular Neuroscience,31(4), 642-648.

5. Fiandaca, M. S., Kapogiannis, D., Mapstone, M., Boxer, A., Eitan, E., Schwartz, J. B., ... & Goetzl, E. J. (2015). Identification of preclinical Alzheimer's disease by a profile of pathogenic proteins in neurally derived blood exosomes: A case-control study. Alzheimer's & Dementia, 11(6), 600-607.

6. Fraser, K. B., Moehle, M. S., Daher, J. P., Webber, P. J., Williams, J. Y., Stewart, C. A., ... & West, A. B. (2013). LRRK2 secretion in exosomes is regulated by 14-3-3. Human molecular genetics, 22(24), 4988-5000.

7. Goetzl, E. J., Mustapic, M., Kapogiannis, D., Eitan, E., Lobach, I. V., Goetzl, L., ... & Miller, B. L. (2016). Cargo proteins of plasma astrocyte‐derived exosomes in Alzheimer's disease. The FASEB Journal, 30(11), 3853-3859.

8. Guha, D., Lorenz, D. R., Misra, V., Chettimada, S., Morgello, S., & Gabuzda, D. (2019). Proteomic analysis of cerebrospinal fluid extracellular vesicles reveals synaptic injury, inflammation, and stress response markers in HIV patients with cognitive impairment. Journal of neuroinflammation, 16(1), 1-19.

9. Hornung, S., Dutta, S., & Bitan, G. (2020). CNS-derived blood exosomes as a promising source of biomarkers: opportunities and challenges. Frontiers in molecular neuroscience, 13, 38.

10. Hurwitz, S. N., Rider, M. A., Bundy, J. L., Liu, X., Singh, R. K., & Meckes Jr, D. G. (2016). Proteomic profiling of NCI-60 extracellular vesicles uncovers common protein cargo and cancer type-specific biomarkers. Oncotarget, 7(52), 86999.

11. Kalra, H., Simpson, R. J., Ji, H., Aikawa, E., Altevogt, P., Askenase, P., ... & Mathivanan, S. (2012). Vesiclepedia: a compendium for extracellular vesicles with continuous community annotation. PLoS biology, 10(12), e1001450.

12. Kang, D., Oh, S., Ahn, S. M., Lee, B. H., & Moon, M. H. (2008). Proteomic analysis of exosomes from human neural stem cells by flow field-flow fractionation and nanoflow liquid chromatography− tandem mass spectrometry. Journal of proteome research, 7(8), 3475-3480.

13. Kapogiannis, Dimitrios, et al. "Dysfunctionally phosphorylated type 1 insulin receptor substrate in neural-derived blood exosomes of preclinical Alzheimer’s disease." The FASEB Journal 29.2 (2015): 589-596.

14. Kekesi, K. A., Juhasz, G., Simor, A.., Gulyassy, P., Szegő, É. M., Hunyadi-Gulyás, É., ... & Czurkó, A. (2012). Altered functional protein networks in the prefrontal cortex and amygdala of victims of suicide. PLoS One, 7(12), e50532.

15. Lachenal, G., Pernet-Gallay, K., Chivet, M., Hemming, F. J., Belly, A., Bodon, G., ... & Sadoul, R. (2011). Release of exosomes from differentiated neurons and its regulation by synaptic glutamatergic activity. Molecular and Cellular Neuroscience, 46(2), 409-418.

16. Mathivanan, S., Fahner, C. J., Reid, G. E., & Simpson, R. J. (2012). ExoCarta 2012: database of exosomal proteins, RNA and lipids. Nucleic acids research, 40(D1), D1241-D1244.

17. McLaughlin, D. P., Little, K. Y., López, J. F., & Watson, S. J. (1996). Expression of serotonin transporter mRNA in human brainstem raphe nuclei. Neuropsychopharmacology, 15(5), 523-529.

18. Muraoka, S., DeLeo, A. M., Sethi, M. K., Yukawa‐Takamatsu, K., Yang, Z., Ko, J., ... & Ikezu, T. (2020). Proteomic and biological profiling of extracellular vesicles from Alzheimer's disease human brain tissues. Alzheimer's & Dementia, 16(6), 896-907.

19. Murphy, T. M., Ryan, M., Foster, T., Kelly, C., McClelland, R., O'Grady, J., ... & Brown, K. (2011). Risk and protective genetic variants in suicidal behaviour: association with SLC1A2, SLC1A3, 5-HTR1B & NTRK2 polymorphisms. Behavioral and Brain Functions, 7(1), 1.

20. Musante, L., Saraswat, M., Duriez, E., Byrne, B., Ravidà, A., Domon, B., & Holthofer, H. (2012). Biochemical and physical characterisation of urinary nanovesicles following CHAPS treatment. PloS one, 7(7), e37279.

21. Nakao, H., Kishimoto, Y., Hashimoto, K., Kitamura, K., Yamasaki, M., Nakao, K., ... & Aiba, A. (2019). mGluR1 in cerebellar Purkinje cells is essential for the formation but not expression of associative eyeblink memory. Scientific reports, 9(1), 1-10.

22. Norman, M., Ter-Ovanesyan, D., Trieu, W., Lazarovits, R., Kowal, E. J., Lee, J. H., ... & Walt, D. R. (2021). L1CAM is not associated with extracellular vesicles in human cerebrospinal fluid or plasma. Nature Methods, 18(6), 631-634.

23. Perez-Hernandez, Daniel, et al. "The intracellular interactome of tetraspanin-enriched microdomains reveals their function as sorting machineries toward exosomes." Journal of Biological Chemistry 288.17 (2013): 11649-11661.

24. Pienimaeki‐Roemer, A., Kuhlmann, K., Böttcher, A., Konovalova, T., Black, A., Orsó, E., ... & Schmitz, G. (2015). Lipidomic and proteomic characterization of platelet extracellular vesicle subfractions from senescent platelets. Transfusion, 55(3), 507-521.

25. Saeedi, S., Israel, S., Nagy, C., & Turecki, G. (2019). The emerging role of exosomes in mental disorders. Translational Psychiatry, 9(1), 1-11.

26. Shi, M., Liu, C., Cook, T. J., Bullock, K. M., Zhao, Y., Ginghina, C., ... & Zhang, J. (2014). Plasma exosomal α-synuclein is likely CNS-derived and increased in Parkinson’s disease. Acta neuropathologica, 128(5), 639-650.

27. Shimada, S., Kitayama, S., Walther, D., & Uhl, G. (1992). Dopamine transporter mRNA: dense expression in ventral midbrain neurons. Molecular brain research, 13(4), 359-362.

28. Théry, C., Witwer, K. W., Aikawa, E., Alcaraz, M. J., Anderson, J. D., Andriantsitohaina, R., ... & Jovanovic‐Talisman, T. (2018). Minimal information for studies of extracellular vesicles 2018 (MISEV2018): a position statement of the International Society for Extracellular Vesicles and update of the MISEV2014 guidelines. Journal of extracellular vesicles, 7(1), 1535750.

29. Tischfield, M. A., Cederquist, G. Y., Gupta, M. L., & Engle, E. C. (2011). Phenotypic spectrum of the tubulin-related disorders and functional implications of disease-causing mutations. Current opinion in genetics & development, 21(3), 286-294.

Uhlén, M., Fagerberg, L., Hallström, B. M., Lindskog, C., Oksvold, P., Mardinoglu, A., ... & Pontén, F. (2015). Tissue-based map of the human proteome. Science, 347(6220).