I agree that most AI questions are not about AI. However, it seems that in practice, most AI discussions seem to be about AI.
This entire piece is predicated on the arrival of transformative AI. Not necessarily in the narrower, more demanding sense it’s used in certain parts of the internet, but in a more relaxed sense: something that’s a fundamental sea change at least on par with, eg, electricity.
And this is where online debates get stuck in the mud. It seems we cannot clear this hurdle in the first place. The prevalence of the idea that AI is useless (vaporware, a stochastic parrot, overglorified autocomplete, a pure slop machine, an email-writing widget, etc) is both maddening and saddening.
It would be nice if mainstream discourse was about the questions Dan brings up. But in a remarkable Don’t-Look-Up-Like-fashion, you can’t get people to see what’s right in front of their eyes. Often the same groups who say climate change is an existential threat to humanity (despite the actual IPCC reports, despite “existential” having a different meaning, despite this unfolding over say the next 40 years or so) even as this is a position that requires a ton of social trust in experts, or plain technical and statistical expertise, engage in constant AI denialism.
I don’t want to turn this into a left-vs-right political debate. Neither side seems situationally aware (with the exception of people like Bernie Sanders). But having the difficult discussions with the Serious Adults In The Room presupposes the existence of those adults. Thus far they’re out buying milk from the store.
It would be nice, and normal, if a world in which AI did Fields Medal-level work in mathematics, wrote 75%+ of new code at Google, massively boosted cyberattack capabilities to the point where renowned experts like Alex Stamos are sounding the alarm, likely assisted writers in winning the Commonwealth Prize, wrote the majority of a paper published in a top philosophical journal like P&PA, operated humanoid robots that run faster and jump higher than world-record holding humans, designed viruses almost from scratch, and is already disrupting the early careers of white collar college graduates… was also a world in which people acted like this is what was going on.
But no. People are not even regular-AI-pilled, let alone transformative-AI-pilled.
I agree it's maddening and saddening, but eventually the vast majority of those people will be disabused (some more harshly than others) of their notions that it's all hype and marketing.
And then some will go sit around in the same bars that flat-earthers and "computers are a fad" folks go to, chatting with people who think radiation from transmission lines is killing their crops, wondering aloud why everyone else is soooo stupid that they just can't see that the entire world around them is merely an elaborate construction by evil AI trillionaires. So it goes.
So - it seems to me that many if the things that specialists have learned will continue to be true regardless of what happens with AI. AI won't change the Maxwell's equations or Quantum Mechanics. It won't change evolutionary psychology's insights into human nature. I think think you are mostly focused on the social sciencees in the sense that what sociologists end economists have learned about society so far won't be relevant in a society with autonomous AI. Maybe. We'll see. I am leas convinced of that than you are
Dan - I like your work and exploration. However quite often when I read the comments to your posts, I am swamped in academic jargon.
I hope that AI will make things simpler to understand for the layman & people without academic tendencies or background can become polymaths and philosophers in an epistemically richer society.
Based on current evidence, I am not optimistic. The marginal cost of producing ideas will trend to zero & based on current business models to monetise LLMs, there is limited incentive to promote simplicity.
Tempted to say the polymath you want is AI itself — unconstrained by disciplinary silos. It won't work. The whole literature on number, including the recent sociology of quantification, assumes that its object sits in the space of human reason — answerable, auditable — as number did even after it left the body: I can unpack a model, an index, and find the human under the surface of impersonality. With algorithms I can still unpack the training, the developers' motivations, and the rest. With an AI whose form is learned, not specified, that interrogation loses its object.
In discussions on specialization, I would suggest that it is better contrasted with "speciation" than "generalization" fresh out the gate. In fact, I would posit that what makes humans "special" at base is the ability to specialize WITHOUT (literal) speciation. Figuratively, this carries over to intellectual disciplines, where one could inquire whether the common obsessions of many of the most brilliant minds (multiverses, simulation theories, etc) constitutes runaway fisherian selection (confident assertion in abstruse territory as social signal of brilliance) and incurs risks of "speciation" (a growing lack of shared language, models, utility, etc).
Likewise, when AI experts speak in one breath about "alignment," and "AGI" in the next, it often seems they are glossing over the tradeoffs between "aligned" and "general" and/or the risks of "speciation" in either direction— extending in an aligned direction beyond what we can follow, or entrenching itself as the generalist that outcompetes other sources for widespread deference/dependence.
I agree that most AI questions are not about AI. However, it seems that in practice, most AI discussions seem to be about AI.
This entire piece is predicated on the arrival of transformative AI. Not necessarily in the narrower, more demanding sense it’s used in certain parts of the internet, but in a more relaxed sense: something that’s a fundamental sea change at least on par with, eg, electricity.
And this is where online debates get stuck in the mud. It seems we cannot clear this hurdle in the first place. The prevalence of the idea that AI is useless (vaporware, a stochastic parrot, overglorified autocomplete, a pure slop machine, an email-writing widget, etc) is both maddening and saddening.
It would be nice if mainstream discourse was about the questions Dan brings up. But in a remarkable Don’t-Look-Up-Like-fashion, you can’t get people to see what’s right in front of their eyes. Often the same groups who say climate change is an existential threat to humanity (despite the actual IPCC reports, despite “existential” having a different meaning, despite this unfolding over say the next 40 years or so) even as this is a position that requires a ton of social trust in experts, or plain technical and statistical expertise, engage in constant AI denialism.
I don’t want to turn this into a left-vs-right political debate. Neither side seems situationally aware (with the exception of people like Bernie Sanders). But having the difficult discussions with the Serious Adults In The Room presupposes the existence of those adults. Thus far they’re out buying milk from the store.
It would be nice, and normal, if a world in which AI did Fields Medal-level work in mathematics, wrote 75%+ of new code at Google, massively boosted cyberattack capabilities to the point where renowned experts like Alex Stamos are sounding the alarm, likely assisted writers in winning the Commonwealth Prize, wrote the majority of a paper published in a top philosophical journal like P&PA, operated humanoid robots that run faster and jump higher than world-record holding humans, designed viruses almost from scratch, and is already disrupting the early careers of white collar college graduates… was also a world in which people acted like this is what was going on.
But no. People are not even regular-AI-pilled, let alone transformative-AI-pilled.
I agree it's maddening and saddening, but eventually the vast majority of those people will be disabused (some more harshly than others) of their notions that it's all hype and marketing.
And then some will go sit around in the same bars that flat-earthers and "computers are a fad" folks go to, chatting with people who think radiation from transmission lines is killing their crops, wondering aloud why everyone else is soooo stupid that they just can't see that the entire world around them is merely an elaborate construction by evil AI trillionaires. So it goes.
So - it seems to me that many if the things that specialists have learned will continue to be true regardless of what happens with AI. AI won't change the Maxwell's equations or Quantum Mechanics. It won't change evolutionary psychology's insights into human nature. I think think you are mostly focused on the social sciencees in the sense that what sociologists end economists have learned about society so far won't be relevant in a society with autonomous AI. Maybe. We'll see. I am leas convinced of that than you are
Dan - I like your work and exploration. However quite often when I read the comments to your posts, I am swamped in academic jargon.
I hope that AI will make things simpler to understand for the layman & people without academic tendencies or background can become polymaths and philosophers in an epistemically richer society.
Based on current evidence, I am not optimistic. The marginal cost of producing ideas will trend to zero & based on current business models to monetise LLMs, there is limited incentive to promote simplicity.
Tempted to say the polymath you want is AI itself — unconstrained by disciplinary silos. It won't work. The whole literature on number, including the recent sociology of quantification, assumes that its object sits in the space of human reason — answerable, auditable — as number did even after it left the body: I can unpack a model, an index, and find the human under the surface of impersonality. With algorithms I can still unpack the training, the developers' motivations, and the rest. With an AI whose form is learned, not specified, that interrogation loses its object.
In discussions on specialization, I would suggest that it is better contrasted with "speciation" than "generalization" fresh out the gate. In fact, I would posit that what makes humans "special" at base is the ability to specialize WITHOUT (literal) speciation. Figuratively, this carries over to intellectual disciplines, where one could inquire whether the common obsessions of many of the most brilliant minds (multiverses, simulation theories, etc) constitutes runaway fisherian selection (confident assertion in abstruse territory as social signal of brilliance) and incurs risks of "speciation" (a growing lack of shared language, models, utility, etc).
Likewise, when AI experts speak in one breath about "alignment," and "AGI" in the next, it often seems they are glossing over the tradeoffs between "aligned" and "general" and/or the risks of "speciation" in either direction— extending in an aligned direction beyond what we can follow, or entrenching itself as the generalist that outcompetes other sources for widespread deference/dependence.