So here's something that's popped in my mind a few times. I was reminded the other day when my MIL, a pre-school teacher, was commenting on a 3yo and casually mentioned how that was the age of the "affirmation of the self". I wonder if we know enough about how the self develops in young children to maybe do some analogy with stages or whatever.
It would also be interesting to see a good effort aimed at falsifying the EToC. Maybe a bunch of claims could have consequences for how we expect the evolution of AI to go?
My dad is actually a speech language pathologist for toddlers, and he and I have discussed this. It's about 15 months that kids can start really using "I," which is about the same time as they pass the mirror test. However, on the mirror test there are huge cultural differences. One study found that in Africa many did not pass the test until 8 years old, with other cultures like Peru or the Philippines somewhere in the middle. Maybe Western culture hyper-optimized for self? Maybe noise? idk, the researchers explained the data in Africa with the tendency to beat kids for doing anything wrong, so they just sort of froze when they were given a novel task.
Lots of debate about recapitulation, not super popular at this point
>It would also be interesting to see a good effort aimed at falsifying the EToC.
Any ideas about how to do that? I'm all ears
>Maybe a bunch of claims could have consequences for how we expect the evolution of AI to go?
One of the reason I put work into this is the reach goal is to make a contribution on how humans became intelligent and then apply that to AI (probably for safety). My background in ML makes that a viable path if people start taking EToC seriously.
Yeah, recapitulation is what I had in mind, but couldn't remember the term. It really would be nice if it happened to support all of this, even if only weakly, though.
The AI thing was my current best guess of such a thing. Seems to me like if the EToC is true then it is highly unlikely that "scale is all you need" approaches will result in self-aware AIs, even if they can result in AGI. So that could be a sort of prediction.
But it would be really neat if there were some sort of archeological record the theory would forbid. Maybe even just a quick census of how old any new finding of art/religion/whatever would need to be to falsify the theory would be great.
I'm also under the impression you went into things like the pronoun exploration expecting it might falsify this, though I'm not too sure. But basically it would be great if you could pull a stronger version of this. Sadly, I have no idea what that would look like in practice though. Maybe it's just not doable!
>Yeah, recapitulation is what I had in mind, but couldn't remember the term. It really would be nice if it happened to support all of this, even if only weakly, though.
Apparently imaginary friends are really similar to schizophrenia, and experienced to some extent by 50% of children.
>I'm also under the impression you went into things like the pronoun exploration expecting it might falsify this, though I'm not too sure.
I didn't realize that it was so hard to do, given the rest of the genome can be purchased for like $200. Now easier to check if the lack of diversity is due to selection (a prediction of EToC) or bottleneck. It's a big open question so it will be followed up on.
Funny to type that out, and appreciate that it's kind of out of my hands :)
Well, that does sound like a nice falsifiable prediction. Sadly not directly linking to self-awareness emerging, but still a very nice one (i.e., it can't falsify the whole thing, but can certainly refute the bigger picture focusing on women-men dynamics). That it's out of your hands only makes it better!
Something I’ve been thinking about from the methodology of reconstructing personality traits from word vectors is that it, at least in retrospect, makes sense that you would recover mostly “follows golden rule” from casual speech, as “can you trust this person” is a very large fraction of what any human cares about when interacting with another. I only really discuss personality as such when trying to introduce my friends to people I would think they like, but normatively it seems like we just say “yeah he’s a good guy/kid” when discussing a stranger. It’s been a while since I read the posts and I’m not sure if the surveys were designed to control for this effect or not
Which effect? The difference in how you treat character evaluations between strangers and friends? If the idea is that one uses a finer model when describing friends, it seems that also works into general language, no? Especially considering the models are trained (in part) on books. I'll bet many characters in Harry Potter could be placed on all of the Big Five. Even minor characters are described in more detail than good-bad.
Books are a good point, I was thinking about the Reddit text when I had this thought, I might need to reread the blog post to be sure, but I believe it worked off guessing what was speech about characters, and do we know how speech about characters is classified as such? I’ve never read potter but most of the fiction I do read illustrates characters’ personalities via their actions and internal monologue, not direct speech about them, would this be captured by the model as well?
The idea is that most speech about people (that I’m privy to) is just “are they trustworthy” and outside of the context of more intimate settings we seem to have something a norm against speaking about personality directly, at least where I’m from in the southern us, and this would likely also be reflected in character’s speech in Anglo novels, so unless there are long gossip sessions being captured in the text most direct speech about characters should contain roughly the same information, at least in my model of it
The point is not even necessarily that I use a finer model when discussing friends, but that there are only a handful of circumstances at all where I directly speak about what I might consider the most salient aspects of my friends’ personalities, often because those aspects are a given or immediately obvious to anyone who meets them
>The idea is that most speech about people (that I’m privy to) is just “are they trustworthy” and outside of the context of more intimate settings we seem to have something a norm against speaking about personality directly
I'd translate this as the relative importance of various factors. The Big Five obscures this, but there is a huge range in how important different personality factors are. Factor one (the PFP/GFP) is 30-80% of the personality information, depending on preprocessing decisions. You are saying "in some situations, all we really discuss is GFP," which the data support. However it also supports the idea that there are other factors. In addition to data from word vectors and surveys, the existence of hundreds of personality adjectives is evidence that we do indeed spend a lot of time discussing others in fairly fine detail.
>most of the fiction I do read illustrates characters’ personalities via their actions and internal monologue, not direct speech about them, would this be captured by the model as well?
Yeah, this is captured by word vectors. Which makes less true that:
>there are only a handful of circumstances at all where I directly speak about what I might consider the most salient aspects of my friends’ personalities
The part of my brain responsible for not wanting to be wrong is trying to wrangle this into some evolutionary thing where all speech is affected by this process subconsciously but that’s not really relevant
All of this is really downstream of the fact that I want to render this information into a model that I might find useful on e.g. a dating app (all of the problems therein notwithstanding), in an alternate universe where the pfp is the standard personality test, a bio saying “high golden rule high dynamism” wouldn’t really tell me much about their likely interest profile or conversation style, which is what I’m mostly *using* a model of personality for. Were some of the more minor factors likely to include this information? I’d be interested in trying to develop this into something use for myself if the loadings can be exported readily
For that I'd say the Big Five is good enough (for government work). There probably is something that relates that to "communication styles" though not sure how that would translate to the apps
Interesting stuff! Quick question, when you write:
>The New Testament rewrites this story and answers what to do about it. EToC holds that the creation story in Genesis is a real memory of events 10,000 years ago. From there, it stands to reason that the New Testament writers 1,000 years later...
Is this timeline correct, are you saying the New Testament was started 9,000 years ago? Or is this a typo?
oof, reading over that is certainly not very readable, editing now...
I meant that The Fall may have happened ~10 kya, and was written down 3.5 kya in Genesis. If esoteric knowledge could last 6.5kya in oral traditions, then it's not much a leap to say that the New Testament written 1,500 years later is riffing off the same real knowledge of The Fall. The Gospels very much an answer to the Fall in the Old Testament.
Writing the post I listened to Margarita's at the Mall by Purple Mountains on repeat. The chorus asks "How long can a world go on under such a subtle God?" which is a pretty good summary of the Lamentations portion of the Old Testament. Fittingly written by the frontman for Silver Jews (himself Jewish).
So here's something that's popped in my mind a few times. I was reminded the other day when my MIL, a pre-school teacher, was commenting on a 3yo and casually mentioned how that was the age of the "affirmation of the self". I wonder if we know enough about how the self develops in young children to maybe do some analogy with stages or whatever.
It would also be interesting to see a good effort aimed at falsifying the EToC. Maybe a bunch of claims could have consequences for how we expect the evolution of AI to go?
Jose, always enjoy your comments!
My dad is actually a speech language pathologist for toddlers, and he and I have discussed this. It's about 15 months that kids can start really using "I," which is about the same time as they pass the mirror test. However, on the mirror test there are huge cultural differences. One study found that in Africa many did not pass the test until 8 years old, with other cultures like Peru or the Philippines somewhere in the middle. Maybe Western culture hyper-optimized for self? Maybe noise? idk, the researchers explained the data in Africa with the tendency to beat kids for doing anything wrong, so they just sort of froze when they were given a novel task.
Historically, this fits under the rule of thumb "ontology recapitulates phylogeny." https://en.wikipedia.org/wiki/Recapitulation_theory
Lots of debate about recapitulation, not super popular at this point
>It would also be interesting to see a good effort aimed at falsifying the EToC.
Any ideas about how to do that? I'm all ears
>Maybe a bunch of claims could have consequences for how we expect the evolution of AI to go?
One of the reason I put work into this is the reach goal is to make a contribution on how humans became intelligent and then apply that to AI (probably for safety). My background in ML makes that a viable path if people start taking EToC seriously.
Yeah, recapitulation is what I had in mind, but couldn't remember the term. It really would be nice if it happened to support all of this, even if only weakly, though.
The AI thing was my current best guess of such a thing. Seems to me like if the EToC is true then it is highly unlikely that "scale is all you need" approaches will result in self-aware AIs, even if they can result in AGI. So that could be a sort of prediction.
But it would be really neat if there were some sort of archeological record the theory would forbid. Maybe even just a quick census of how old any new finding of art/religion/whatever would need to be to falsify the theory would be great.
I'm also under the impression you went into things like the pronoun exploration expecting it might falsify this, though I'm not too sure. But basically it would be great if you could pull a stronger version of this. Sadly, I have no idea what that would look like in practice though. Maybe it's just not doable!
>Yeah, recapitulation is what I had in mind, but couldn't remember the term. It really would be nice if it happened to support all of this, even if only weakly, though.
Apparently imaginary friends are really similar to schizophrenia, and experienced to some extent by 50% of children.
>I'm also under the impression you went into things like the pronoun exploration expecting it might falsify this, though I'm not too sure.
The one that will come, though it may be years, will be what's up with the Y chromosome bottleneck. The Y chromosome was actually just completely sequenced today: https://theconversation.com/scientists-find-the-last-remnants-of-the-human-genome-that-were-missing-in-the-y-chromosome-212141
I didn't realize that it was so hard to do, given the rest of the genome can be purchased for like $200. Now easier to check if the lack of diversity is due to selection (a prediction of EToC) or bottleneck. It's a big open question so it will be followed up on.
Funny to type that out, and appreciate that it's kind of out of my hands :)
Well, that does sound like a nice falsifiable prediction. Sadly not directly linking to self-awareness emerging, but still a very nice one (i.e., it can't falsify the whole thing, but can certainly refute the bigger picture focusing on women-men dynamics). That it's out of your hands only makes it better!
re: big 5 and pfp
Something I’ve been thinking about from the methodology of reconstructing personality traits from word vectors is that it, at least in retrospect, makes sense that you would recover mostly “follows golden rule” from casual speech, as “can you trust this person” is a very large fraction of what any human cares about when interacting with another. I only really discuss personality as such when trying to introduce my friends to people I would think they like, but normatively it seems like we just say “yeah he’s a good guy/kid” when discussing a stranger. It’s been a while since I read the posts and I’m not sure if the surveys were designed to control for this effect or not
Which effect? The difference in how you treat character evaluations between strangers and friends? If the idea is that one uses a finer model when describing friends, it seems that also works into general language, no? Especially considering the models are trained (in part) on books. I'll bet many characters in Harry Potter could be placed on all of the Big Five. Even minor characters are described in more detail than good-bad.
Books are a good point, I was thinking about the Reddit text when I had this thought, I might need to reread the blog post to be sure, but I believe it worked off guessing what was speech about characters, and do we know how speech about characters is classified as such? I’ve never read potter but most of the fiction I do read illustrates characters’ personalities via their actions and internal monologue, not direct speech about them, would this be captured by the model as well?
The idea is that most speech about people (that I’m privy to) is just “are they trustworthy” and outside of the context of more intimate settings we seem to have something a norm against speaking about personality directly, at least where I’m from in the southern us, and this would likely also be reflected in character’s speech in Anglo novels, so unless there are long gossip sessions being captured in the text most direct speech about characters should contain roughly the same information, at least in my model of it
The point is not even necessarily that I use a finer model when discussing friends, but that there are only a handful of circumstances at all where I directly speak about what I might consider the most salient aspects of my friends’ personalities, often because those aspects are a given or immediately obvious to anyone who meets them
>The idea is that most speech about people (that I’m privy to) is just “are they trustworthy” and outside of the context of more intimate settings we seem to have something a norm against speaking about personality directly
I'd translate this as the relative importance of various factors. The Big Five obscures this, but there is a huge range in how important different personality factors are. Factor one (the PFP/GFP) is 30-80% of the personality information, depending on preprocessing decisions. You are saying "in some situations, all we really discuss is GFP," which the data support. However it also supports the idea that there are other factors. In addition to data from word vectors and surveys, the existence of hundreds of personality adjectives is evidence that we do indeed spend a lot of time discussing others in fairly fine detail.
>most of the fiction I do read illustrates characters’ personalities via their actions and internal monologue, not direct speech about them, would this be captured by the model as well?
Yeah, this is captured by word vectors. Which makes less true that:
>there are only a handful of circumstances at all where I directly speak about what I might consider the most salient aspects of my friends’ personalities
>Yeah, this is captured by word vectors.
Then my point as stated is moot
The part of my brain responsible for not wanting to be wrong is trying to wrangle this into some evolutionary thing where all speech is affected by this process subconsciously but that’s not really relevant
All of this is really downstream of the fact that I want to render this information into a model that I might find useful on e.g. a dating app (all of the problems therein notwithstanding), in an alternate universe where the pfp is the standard personality test, a bio saying “high golden rule high dynamism” wouldn’t really tell me much about their likely interest profile or conversation style, which is what I’m mostly *using* a model of personality for. Were some of the more minor factors likely to include this information? I’d be interested in trying to develop this into something use for myself if the loadings can be exported readily
For that I'd say the Big Five is good enough (for government work). There probably is something that relates that to "communication styles" though not sure how that would translate to the apps
Interesting stuff! Quick question, when you write:
>The New Testament rewrites this story and answers what to do about it. EToC holds that the creation story in Genesis is a real memory of events 10,000 years ago. From there, it stands to reason that the New Testament writers 1,000 years later...
Is this timeline correct, are you saying the New Testament was started 9,000 years ago? Or is this a typo?
oof, reading over that is certainly not very readable, editing now...
I meant that The Fall may have happened ~10 kya, and was written down 3.5 kya in Genesis. If esoteric knowledge could last 6.5kya in oral traditions, then it's not much a leap to say that the New Testament written 1,500 years later is riffing off the same real knowledge of The Fall. The Gospels very much an answer to the Fall in the Old Testament.
Writing the post I listened to Margarita's at the Mall by Purple Mountains on repeat. The chorus asks "How long can a world go on under such a subtle God?" which is a pretty good summary of the Lamentations portion of the Old Testament. Fittingly written by the frontman for Silver Jews (himself Jewish).
I really need to find the time to dig further back into your archives...