Great piece. The overfitting framing made me think of the Rothera Antarctic study. 26 people from different countries spent six months with only each other for company, and their vowels drifted toward a shared blend, even though nobody noticed or intended it.
My hypothesis is that social media runs the same imitation mechanism at a much larger scale. Speech converges through exposure, and feeds expose billions of people to the same small set of voices, formats and trends, tuned by engagement metrics. It's a shared "station" with no walls, and your point about discriminators overfitting to their signals suggests the voices being imitated are themselves converging.
If that's right, the narrowing might go beyond taste and style to how people frame problems and express thoughts. I'd treat that last step as a hypothesis, not a finding. Rothera only showed convergence in pronunciation, and it was a small, isolated group. The open question is whether a shared exposure pool can produce the same effect without isolation.
It would also be testable: compare variance in vocabulary, framing or reasoning style between heavy feed users and people with more varied sources.
Originality is overrated, not because it might not be better than whatever exists, but we miss the effects of the social nature of our lives. Just as the essay discusses without ever naming it, we don't exist in a vacuum. It's a bit like the faux-capitalism of the world around us, the rate of return is more important than improvements to quality: the enshitification phenomena.
In terms of the preponderance of average, any number of explanations suit the narrative, and as haidt would agree, we'll find a reasonable explanation after the event and as psychology teaches, forget failures and errors, remembering only the good times.
What's actually curious and should give us optimism is the widespread retardation of the population using AI. Heterodox thinkers, like artists, will stand out more and perhaps their originality will return to be prized. On the other hand, the nefarious uses for AI should give us pause to reflect on what sort of societies are getting constructed, but unfortunately if history is any guide, we're incapable of honest self appraisal.
My own thinking is we've lost track of what's virtue and thus our ability to make judicious choices has been lost.
Dalla Riva's argument, which you mention, that overlong and restrictive IP is one hidden cause of cultural stagnation deserves more weight in the overfitting picture. Copyright terms of life plus 70 keep the whole twentieth-century catalogue locked up, so the incentive is to re-exploit a few safe properties instead of letting anyone recombine them. That's the in-distribution trap in legal form: the generator can only sample from what the rights holders already approved.
It also changes how I read the corporate consolidation story you cite from deWaard. A handful of studios and labels can only squeeze creative risk-taking because they own the catalogue. With a 15-year term, most of that catalogue would be free for anyone to adapt, so consolidation would matter far less. A monopoly over content mostly stops being a monopoly.
I wrote a piece on this from an Australian angle: what happens if a country cuts its copyright term to 15 years, closes the trademark loophole that keeps public-domain works locked up, and builds a remake industry on the freed catalogue. It's essentially a proposal to deliberately inject variance into a stagnant system. It also covers why the threats of retaliation sound worse than they are.
I'd love to hear whether you think loosening the legal supply of raw material would actually counter the overfitting, or whether the discriminators (algorithms, box-office signals) would just overfit to the new material too.
Great piece. The overfitting framing made me think of the Rothera Antarctic study. 26 people from different countries spent six months with only each other for company, and their vowels drifted toward a shared blend, even though nobody noticed or intended it.
My hypothesis is that social media runs the same imitation mechanism at a much larger scale. Speech converges through exposure, and feeds expose billions of people to the same small set of voices, formats and trends, tuned by engagement metrics. It's a shared "station" with no walls, and your point about discriminators overfitting to their signals suggests the voices being imitated are themselves converging.
If that's right, the narrowing might go beyond taste and style to how people frame problems and express thoughts. I'd treat that last step as a hypothesis, not a finding. Rothera only showed convergence in pronunciation, and it was a small, isolated group. The open question is whether a shared exposure pool can produce the same effect without isolation.
It would also be testable: compare variance in vocabulary, framing or reasoning style between heavy feed users and people with more varied sources.
Originality is overrated, not because it might not be better than whatever exists, but we miss the effects of the social nature of our lives. Just as the essay discusses without ever naming it, we don't exist in a vacuum. It's a bit like the faux-capitalism of the world around us, the rate of return is more important than improvements to quality: the enshitification phenomena.
In terms of the preponderance of average, any number of explanations suit the narrative, and as haidt would agree, we'll find a reasonable explanation after the event and as psychology teaches, forget failures and errors, remembering only the good times.
What's actually curious and should give us optimism is the widespread retardation of the population using AI. Heterodox thinkers, like artists, will stand out more and perhaps their originality will return to be prized. On the other hand, the nefarious uses for AI should give us pause to reflect on what sort of societies are getting constructed, but unfortunately if history is any guide, we're incapable of honest self appraisal.
My own thinking is we've lost track of what's virtue and thus our ability to make judicious choices has been lost.
Dalla Riva's argument, which you mention, that overlong and restrictive IP is one hidden cause of cultural stagnation deserves more weight in the overfitting picture. Copyright terms of life plus 70 keep the whole twentieth-century catalogue locked up, so the incentive is to re-exploit a few safe properties instead of letting anyone recombine them. That's the in-distribution trap in legal form: the generator can only sample from what the rights holders already approved.
It also changes how I read the corporate consolidation story you cite from deWaard. A handful of studios and labels can only squeeze creative risk-taking because they own the catalogue. With a 15-year term, most of that catalogue would be free for anyone to adapt, so consolidation would matter far less. A monopoly over content mostly stops being a monopoly.
I wrote a piece on this from an Australian angle: what happens if a country cuts its copyright term to 15 years, closes the trademark loophole that keeps public-domain works locked up, and builds a remake industry on the freed catalogue. It's essentially a proposal to deliberately inject variance into a stagnant system. It also covers why the threats of retaliation sound worse than they are.
The Australian Remake Renaissance: https://mdnadimahmed888222.substack.com/p/the-australian-remake-renaissance
I'd love to hear whether you think loosening the legal supply of raw material would actually counter the overfitting, or whether the discriminators (algorithms, box-office signals) would just overfit to the new material too.