Thought provoking as always Sean. My immediate, cheeky thought -- if LLMs could somehow help me to sort through the bevy of research being produced on AI and human cognition that would be super freakin' helpful, as I feel like I'm bookmarking six studies a day that I'll never get around to reading.
Slightly more substantively, your post has me pondering the nature of scientific inquiry and in particular, scientific revolutions in the Kuhnian sense, the moments when our paradigms shift. The approach you sketch out here is highly empirical, and no shame in that. But Kuhn observed that "what occurs during a scientific revolution is not fully reducible to a reinterpretation of individual and stable data." He suggested, with some force I think, that revolutions come when scientists engage in "thought experiments" that create new metaphors for understanding the world differently.
Can LLMs help with this? Maybe! I certainly can imagine them helping spur new "thought experiments" in scientists. But just as I'm concerned that art and literature may be impoverished by the creation of this new tool that is built to generate something that "looks like" a sort of averaging of what's already been created, in my darker imaginings I can envision a future where this technology locks us into the scientific paradigm of the present.
But hopefully you prove me wrong and become the modern Einstein of the mind.
Your concern about locking us into the scientific paradigm of the present is sort of what I was trying to gesture at in the section about LLMs "terraforming" the scientific process, and relates to another, much older post I wrote about whether LLMs might change language (https://seantrott.substack.com/p/could-language-models-change-language). I think calcifying certain practices into place is a real danger, and it is conceivable that LLMs are way that calcification might arise.
If it's okay with you, I might add a footnote to that section pointing out the explicit connection to Kuhn (crediting you of course)? I think that's a really nice link.
Thought provoking as always Sean. My immediate, cheeky thought -- if LLMs could somehow help me to sort through the bevy of research being produced on AI and human cognition that would be super freakin' helpful, as I feel like I'm bookmarking six studies a day that I'll never get around to reading.
Slightly more substantively, your post has me pondering the nature of scientific inquiry and in particular, scientific revolutions in the Kuhnian sense, the moments when our paradigms shift. The approach you sketch out here is highly empirical, and no shame in that. But Kuhn observed that "what occurs during a scientific revolution is not fully reducible to a reinterpretation of individual and stable data." He suggested, with some force I think, that revolutions come when scientists engage in "thought experiments" that create new metaphors for understanding the world differently.
Can LLMs help with this? Maybe! I certainly can imagine them helping spur new "thought experiments" in scientists. But just as I'm concerned that art and literature may be impoverished by the creation of this new tool that is built to generate something that "looks like" a sort of averaging of what's already been created, in my darker imaginings I can envision a future where this technology locks us into the scientific paradigm of the present.
But hopefully you prove me wrong and become the modern Einstein of the mind.
Great point!
Your concern about locking us into the scientific paradigm of the present is sort of what I was trying to gesture at in the section about LLMs "terraforming" the scientific process, and relates to another, much older post I wrote about whether LLMs might change language (https://seantrott.substack.com/p/could-language-models-change-language). I think calcifying certain practices into place is a real danger, and it is conceivable that LLMs are way that calcification might arise.
If it's okay with you, I might add a footnote to that section pointing out the explicit connection to Kuhn (crediting you of course)? I think that's a really nice link.
I'd be honored!
Thanks for the conceptual connection! I think it's probably worthy of its own post but I wanted to add a reference here too (see the new footnote).