TFP doesn't depend on research so much as management practice. My impression is that there are lots of great new ideas being developed in universities and garages, but large corporations have their own priorities. In fact, it often pays for a company to suppress innovation. We saw this in the 1960s with IBM slowing progress in computing by using its massive corporate power. We saw this with Detroit in the 1960s and 1970s until the Japanese were well on track to take over the car market completely. We're seeing it now with companies like Google, Facebook and Apple maintaining walled gardens and using their corporate power to suppress new technology. Why isn't TFP going up? That's easy. The money isn't in raising TFP. The money is controlling the market and suppressing innovation.
Interesting article. I hadn't heard of the phrase "burden of knowledge" before but it is a nice label for an important problem. I see the problem in terms of making knowledge simpler which is obviously necessary if human knowledge is to continue to progress. Unfortunately the present academic system does not encourage simplicity. I have had an article rejected by an academic journal because it was too simple with the suggestion I try to make it more complicated! And if knowledge were simpler there might not be as much demand for university courses or the services of experts. Universities have an obvious interest in keeping things difficult. My article "If Knowledge Were Simpler We Would All Be Wiser" (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3911835) discusses this problem and makes a few suggestions (including outsourcing to technology like Macbooks).
Would you have the effort to do a ChatGPT summary of the key points and recommended methodologies from your paper and post it here? It sounds interesting but I am too lazy to read it.
An intriguing suggestion. I did try ChatGPT but it kept on complaining of an error. However, the section headings on pages 2-3 are intended as a sort of summary. There is a section entitled "Keep it short" - advice I have obviously not taken sufficiently seriously - and another on AI as an alternative to simplification.
Copernicus didn't get rid of epicycles, or even reduce their number. His heliocentric model had MORE epicycles than the latest Ptolemaic ones. What he abolished was the equant, because it went against his idea of building everything out of perfect circles. https://entitledtoanopinion.wordpress.com/2011/10/16/dont-rely-on-yglesias-for-kuhns-copernicus-revisionism/ His model was both complicated & inaccurate for that reason. It was Kepler's "refinement" of replacing circles with ellipses that actually explained the most accurate data the most simply.
I don't think special relativity displaced Newtonian mechanics for most practical calculations, and I'm pretty sure every physicist has to learn Newtonian mechanics first before later learning the relativistic modification necessary to handle speeds near that of light.
That's exactly right. The big TFP win of the heliocentric model was that it gave a basis for assigning weights to Mercury and Venus when calculating horoscopes. I wish I was making that up.
I'm skeptical that institutional depreciation would result in such nice curves. It would be affected by policies being enacted, and so should show noticeable bends here and there. If you break these graphs down by country, do they match up with anything in particular? Different countries have different institutional development, after all. It is my impression that the grant-hunting is particularly onerous in the US, for instance.
Maybe the car is just going downhill at a speed determined by gradient and wind resistance, and pressing the pedal just produces more noise and hot exhaust? It would be no surprise to find moral failures where there is a narrative of progress but no real way to make gains.
In other words, maybe development of technology is a process independent of the scale of research effort. More effort is just pumping against the bottleneck of natural timescales like maturation of tech, having an idea and communicating it to others, building, saving, depreciation, setting up experiments etc. Obviously human generation time, too! As long as you have some demographic margin of safety so that you don't lose technology to random drift, like the Tasmanians did, population size does not matter. Who could with a straight face maintain that with 10x the population, steam age would have lasted just a couple of decades?
People are not special precious pointwise fountainheads producing unique innovations in an exploding chain reaction. They are going to think about the same things, have the same experience and schooling and tools and do redundant work. Good ideas, viable ways forward from the state of the art at a given time, are finite.
An example to illustrate the idea that technology develops autonomously and humans are just a medium: Let's say I'm doing some carpentry and I'm trying to find a solution to some problem. I could ruminate about it in bed at 2 AM, but most progress happens when I walk into the toolshed and let the tools suggest their solutions. Most likely the first attempts have disappointing results, I have to try a few things. The available technology dominates the process. Adding more people and experience to the toolshed might produce a better idea quicker, but diminishing returns strike very fast.
There's also the truism in software development that adding more workers rarely makes things happen any faster. Same thing here?
I bet you could construct a saner model of progress than what Bloom and Jones used. Do not think it's hard to do.
I know next to nothing about AI but I think the above points also work against a knowledge explosion resulting from running massive numbers of LLMs (+"scaffolding") in parallel.
This paper from 2018 has a section at the end on scientific funding where they propose that giving every researcher the same amount of money would produce more new ideas than the current US system, which disproportionately rewards "superstars."
I'll say though wrt new paradigms absolving the burden of knowledge: This reminds me of the relativity of wrong by Asimov, as we get more and more accurate knowledge there will have to be time where we do have the burden of knowledge but there won't be any paradigm shifts on the horizon to absolve us of that burden, and I think we are past that point.
TFP doesn't depend on research so much as management practice. My impression is that there are lots of great new ideas being developed in universities and garages, but large corporations have their own priorities. In fact, it often pays for a company to suppress innovation. We saw this in the 1960s with IBM slowing progress in computing by using its massive corporate power. We saw this with Detroit in the 1960s and 1970s until the Japanese were well on track to take over the car market completely. We're seeing it now with companies like Google, Facebook and Apple maintaining walled gardens and using their corporate power to suppress new technology. Why isn't TFP going up? That's easy. The money isn't in raising TFP. The money is controlling the market and suppressing innovation.
Interesting article. I hadn't heard of the phrase "burden of knowledge" before but it is a nice label for an important problem. I see the problem in terms of making knowledge simpler which is obviously necessary if human knowledge is to continue to progress. Unfortunately the present academic system does not encourage simplicity. I have had an article rejected by an academic journal because it was too simple with the suggestion I try to make it more complicated! And if knowledge were simpler there might not be as much demand for university courses or the services of experts. Universities have an obvious interest in keeping things difficult. My article "If Knowledge Were Simpler We Would All Be Wiser" (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3911835) discusses this problem and makes a few suggestions (including outsourcing to technology like Macbooks).
Would you have the effort to do a ChatGPT summary of the key points and recommended methodologies from your paper and post it here? It sounds interesting but I am too lazy to read it.
An intriguing suggestion. I did try ChatGPT but it kept on complaining of an error. However, the section headings on pages 2-3 are intended as a sort of summary. There is a section entitled "Keep it short" - advice I have obviously not taken sufficiently seriously - and another on AI as an alternative to simplification.
Copernicus didn't get rid of epicycles, or even reduce their number. His heliocentric model had MORE epicycles than the latest Ptolemaic ones. What he abolished was the equant, because it went against his idea of building everything out of perfect circles. https://entitledtoanopinion.wordpress.com/2011/10/16/dont-rely-on-yglesias-for-kuhns-copernicus-revisionism/ His model was both complicated & inaccurate for that reason. It was Kepler's "refinement" of replacing circles with ellipses that actually explained the most accurate data the most simply.
I don't think special relativity displaced Newtonian mechanics for most practical calculations, and I'm pretty sure every physicist has to learn Newtonian mechanics first before later learning the relativistic modification necessary to handle speeds near that of light.
That's exactly right. The big TFP win of the heliocentric model was that it gave a basis for assigning weights to Mercury and Venus when calculating horoscopes. I wish I was making that up.
I'm skeptical that institutional depreciation would result in such nice curves. It would be affected by policies being enacted, and so should show noticeable bends here and there. If you break these graphs down by country, do they match up with anything in particular? Different countries have different institutional development, after all. It is my impression that the grant-hunting is particularly onerous in the US, for instance.
I agree with the titular point.
Anyway, here's a wild idea:
Maybe the car is just going downhill at a speed determined by gradient and wind resistance, and pressing the pedal just produces more noise and hot exhaust? It would be no surprise to find moral failures where there is a narrative of progress but no real way to make gains.
In other words, maybe development of technology is a process independent of the scale of research effort. More effort is just pumping against the bottleneck of natural timescales like maturation of tech, having an idea and communicating it to others, building, saving, depreciation, setting up experiments etc. Obviously human generation time, too! As long as you have some demographic margin of safety so that you don't lose technology to random drift, like the Tasmanians did, population size does not matter. Who could with a straight face maintain that with 10x the population, steam age would have lasted just a couple of decades?
People are not special precious pointwise fountainheads producing unique innovations in an exploding chain reaction. They are going to think about the same things, have the same experience and schooling and tools and do redundant work. Good ideas, viable ways forward from the state of the art at a given time, are finite.
An example to illustrate the idea that technology develops autonomously and humans are just a medium: Let's say I'm doing some carpentry and I'm trying to find a solution to some problem. I could ruminate about it in bed at 2 AM, but most progress happens when I walk into the toolshed and let the tools suggest their solutions. Most likely the first attempts have disappointing results, I have to try a few things. The available technology dominates the process. Adding more people and experience to the toolshed might produce a better idea quicker, but diminishing returns strike very fast.
There's also the truism in software development that adding more workers rarely makes things happen any faster. Same thing here?
I bet you could construct a saner model of progress than what Bloom and Jones used. Do not think it's hard to do.
I know next to nothing about AI but I think the above points also work against a knowledge explosion resulting from running massive numbers of LLMs (+"scaffolding") in parallel.
> Maybe the car is just going downhill at a speed determined by gradient and wind resistance
Culture and ideology (ie: scientism, fundamentalism, insularity) provide substantial resistance.
> People are not special precious pointwise fountainheads producing unique innovations in an exploding chain reaction.
Not yet anyways...but there was a time not that long ago where we weren't jetsetting around the world for relatively paltry sums of money.
Oh how I enjoy watching Humans try to figure out what is going on from within the system they exist within. Good luck boys!!
This paper from 2018 has a section at the end on scientific funding where they propose that giving every researcher the same amount of money would produce more new ideas than the current US system, which disproportionately rewards "superstars."
http://www.pluchino.it/talent-vs-luck.html
I'll say though wrt new paradigms absolving the burden of knowledge: This reminds me of the relativity of wrong by Asimov, as we get more and more accurate knowledge there will have to be time where we do have the burden of knowledge but there won't be any paradigm shifts on the horizon to absolve us of that burden, and I think we are past that point.