It helps us to situate ourselves within a wider world, to conceive of ourselves as individuals, and to understand how our minds work.
I think it’s a lot more useful than that. Human’s entire shtick as an animal is technology: social transmission of useful techniques, skills, and approaches, which allows good ideas to be passed down from one generation to the next, so they can accumulate. Thus we’re mammals specifically adapted to be good at learning from the previous generation and teaching the next.
Some techniques can be learnt just by watching. But frequently the hard part isn’t as much what the teacher’s hands are doing as what’s going on in their head that allows them to figure out what to do next with their hands. So that makes algorithm-to-text-to-algorithm extremely important. Humans are evolved not just to do smart things, but to be able to serialize what they’re doing to spoken language, so another human learning from then can deserialize that back from language to learn a copy of the algorithm. So we are likely specifically adapted to be good at introspection: it’s one of our core species abilities, along with opposable thumbs and bipedalism: they all requirements for being really good at tool use.
I’m sympathetic with the idea that introspection largely serves a social purpose. I meant for this sort of thing to be included in ‘understanding how our minds work’. That said, I still find it rather mysterious that it evolved. Bad introspection doesn’t seem particularly useful, and I wouldn’t expect a leap from no introspection to good introspection in a single generation. Seems plausible that introspection is partly a spandrel of other evolved architectural traits.
I suspect bad introspection is still better than no introspection:
[Elder]: (chips flint)
[Watching youngster]: “You hit higher up this time — why?”
[Elder]: a) “I have no idea” b) “Uhh… the lump was bigger?” c) <cogent and detailed explanation of relevant portion of flint-chipping algorithm>
Answer b) provides marginally more training signal to the youngster than a), even if a lot less than c). Every little helps. Speed and accuracy of transferring skills to youngsters was load-bearingly adaptive.
I think it’s a lot more useful than that. Human’s entire shtick as an animal is technology: social transmission of useful techniques, skills, and approaches, which allows good ideas to be passed down from one generation to the next, so they can accumulate. Thus we’re mammals specifically adapted to be good at learning from the previous generation and teaching the next.
Some techniques can be learnt just by watching. But frequently the hard part isn’t as much what the teacher’s hands are doing as what’s going on in their head that allows them to figure out what to do next with their hands. So that makes algorithm-to-text-to-algorithm extremely important. Humans are evolved not just to do smart things, but to be able to serialize what they’re doing to spoken language, so another human learning from then can deserialize that back from language to learn a copy of the algorithm. So we are likely specifically adapted to be good at introspection: it’s one of our core species abilities, along with opposable thumbs and bipedalism: they all requirements for being really good at tool use.
I’m sympathetic with the idea that introspection largely serves a social purpose. I meant for this sort of thing to be included in ‘understanding how our minds work’. That said, I still find it rather mysterious that it evolved. Bad introspection doesn’t seem particularly useful, and I wouldn’t expect a leap from no introspection to good introspection in a single generation. Seems plausible that introspection is partly a spandrel of other evolved architectural traits.
I suspect bad introspection is still better than no introspection:
[Elder]:
(chips flint)
[Watching youngster]:
“You hit higher up this time — why?”
[Elder]:
a) “I have no idea”
b) “Uhh… the lump was bigger?”
c) <cogent and detailed explanation of relevant portion of flint-chipping algorithm>
Answer b) provides marginally more training signal to the youngster than a), even if a lot less than c). Every little helps. Speed and accuracy of transferring skills to youngsters was load-bearingly adaptive.