That is AlphaGo stage (before Zero). I expect that learning to think by "self-play" will come before we manage to produce enough of explicitly expressed approximations of our way of thinking.
Maybe, but that’s easier to do when winning/losing is binary and possibilities are more bounded within the rigidly defined rules and spaces of a chess match.
Human play was used to bootstrap AlphaGo. But LLMs already have vast amount of data relating to human introspection. The problem is they cannot operationalize it, as they have no analogs of our thinking/speaking distinction. Speaking and thinking is the same for them: producing a token is a prerequisite to produce a next token.
Building a system that can leverage that knowledge to bootstrap "thinking circuits" can prove challenging, that's for sure, but I expect that collecting quality "thinking out loud" data is more challenging.
Are there people who actually believe that they think "in language"? I know that it's a trope I see in television and other fiction, but I also see the one there concussions can lead to hours of unconsciousness that resolves itself without neurosurgery.
No. That's not your "thought". It's just a little subroutine that evolved so you could rehearse dialogue with other humans... unclever people have it go a little haywire, and they mistake it for their actual thoughts. The hilarious part is that some even end up believing it makes them superior.
> It's just a little subroutine that evolved so you could rehearse dialogue with other humans
That’s an interesting claim to make. What are you basing this on? Framing this as a subroutine also seems a bit of a limited view on our dialogue generation abilities.
I can assure you that my internal dialogue rarely stops (meditation is an exception), and most often isn’t worried about talking with other people.
hmm.. that 'rehearsing' isn't thinking? I know people who thing about every word, the phrase and re-phrase, select the best sounding. Do you think they still don't think in language? They even write their thoughts down and sell as a books.
No. It's not thinking in any meaningful way. It's practice for when you're in front of that person and getting ready to talk for real.
But for those who have no control over it, it just goes apeshit. I've heard self-reported cases of people doing the same but with historical or even fictional people. They try to use it for everything... as if that simulated person they're "talking to" will somehow be capable of solving a problem they themselves aren't capable of.
I think I was in grade school when I realized all the rest of you were lunatics. Took a little while longer to understand the nature of the cognitive flaw.
> Do you think they still don't think in language?
Yes. Because if that's how you "think", then you're not thinking at all. Which means you're also incapable of understanding how thought works.
You're asserting that I'm wrong, offering no evidence or even a plausible alternative theory. It's just another one of those absurd "truths" that dumb people assume is true and no one ever calls them on it.
I've explained that it is but one mental faculty, I've explained the evolutionary origin of it, and to top it all off, I've explained why you mistake the output of this singular mental faculty to be"thought".
I would hope that you'd be able to work through on first principles why I must be correct. You've just said thought is impossible without human (spoken, maybe signed) language, when there is clear counter-evidence in the world that everyone is aware of that contradicts your asinine theory.
You come across like someone from r/iamverysmart. Take a step back and understand that nobody can really “know” for certain what thought actually is, because it’s exactly like trying to understand consciousness, which is the biggest mystery of all.
As I am a writer, I not only think in language, but I act in language.
I would add a link to a conversation with neuroscientist Anil Seth about the nature of consciousness, and a quote from that, which posits maybe this isn't an either/or situation. There is much diversity in what constitutes "thought", even in humans.
"You know, some people might see colors differently. Other people might have more vivid mental imagery. We might experience the flow of time to be different from one person to another. Our expectations, our tendency to see faces in clouds, let’s say, that might differ from one person to another too. And very little is known about this hidden landscape of inner diversity."
There is a term for this: “inner monologue”, though not everyone experiences it. I am a native English speaker and have always had an English inner monologue, but many years ago when I lived in another country and started to lean the language to the point where I was fluent, I actually started to have an inner monologue in that language! And when I returned to the US, it took me quite some time
to return to a fully English inner monologue.
> Are there people who actually believe that they think "in language"?
Yes, there are people who believe that thought is language (and not just inner dialogue as you mention in this thread). That without ideas formed in language as a precursor they wouldn't be able to think about those ideas, and that they do this thinking through the internal manipulation of other words.
I'm not quite sure what it's about as I'm not one of them, but they exist.
How do you think then? I have two modes:
1. Conscious thinking with words, when I am actively trying to think about something, when I'm trying to solve a problem.
2. And unconscious thinking where I don't actively think about the problem and do something else, but after a while I come to some conclusion.
But when I am trying to actively think of something I know no other way than to think in words. I am not a visual thinker (I'm kind of aphantasic, when you ask me to think about a horse, only an idea pops up in my head without any specific image and I have to try really hard and describe the horse with words to make some kind of misty parts of the horse).
Back in the day when I was thinking about this I kind of thought of myself as pre-visual. I'm sure I have multiple modes of thought though, and I'm also sure this is true of pretty much everyone.
I've had way too many instances of not being able to find a word to express an idea, or not remembering the specific words people expressed but just what I think was the gist of what they said (whether or not I interpreted them correctly) to think of myself as a verbal thinker. When it comes to word production sometimes they just come, at other times it takes me time, and I can go into a state where they just flow, but in that state my conscious is checked out of the process. It has the experience of an autonomic process almost.
I do have an inner monologue though (frequently caught up with OCD counting, or at the moment Taylor Swift's Anti-Hero song as well as speaking the words I am writing as I am writing them).
> only an idea pops up in my head without any specific image and I have to try really hard and describe the horse with words to make some kind of misty parts of the horse
Is this idea the words which describe the horse, or a pre-visual/pre-verbal idea of a horse?
okay, if you had zero words in your vocabulary. how would you interpret what a tree is, or anything to do with your day job, would you even have a job without words? without language? would you have a home without written language to order the supplies all of which have very specific names and measurements? all of humanity's achievements is because of language.
I mean you can look at the speed of innovations. things were slow as hell until the 1400s, when the printing press was invented, the industrial revolution would've never kicked off without the ability to push thoughts and language into more hands.
Education is all about language, hard to learn in grunts and pointing.
There's more that goes into actual memories than language, though language is how we understand it interpret them... we have senses, feelings, and other contexts associated with a memory but without a word for sadness or anger, how do we know what we're even feeling?
Language is one of the things that truly separates us from animals and before written language our hunter gatherer ancestors were little more than animals wearing fur skins.
Big projects and education require copying ideas from one brain to another. Language is a good medium for doing that, much better than improvised gestures. But words and ideas must be different things, or nothing read could ever be misunderstood, and no one would ever be at a loss for words.
When I imagine a tree, I get the image and the feel and the scent, the qualia. Hearing the word “pine” for the first time did not make me smell anything. I had to learn to associate those syllables with my memories of pine trees, and they only work between me and other people who have done that.
I agree with your first three paragraphs about the importance of language. But otherwise agree with erik_seaberg.
> but without a word for sadness or anger, how do we know what we're even feeling?
"I'm feeling kind of a mix between the way I felt that time that thing happened and that time I did that other stuff. I don't like it." Without words at all, this would be a flash of memory of the two events which I would realize are related (emotionally) to the event happening to me now.
For anger, particularly, the emotion (maybe better said to be a mood state) usually comes with a lot of drives. Whether we have a word for what we're feeling, we know what actions we want to take because of the feeling. Or at least I do.
Circumlocutions are born out of ideas without words to express them. E.g. (literally 5 seconds to watch the relevant part): https://youtu.be/UEKWXFESU68?t=150
> Language is one of the things that truly separates us from animals and before written language our hunter gatherer ancestors were little more than animals wearing fur skins.
Animals have a crap load of language-like elements, with taught dialects and the like. These are seen in social bees, prairie dog communities, and numerous other animals. We humans have very complex languages that facilitate a ton of nuance and novel information types. But don't sell animals, or our ancestors, short. Human civilization appears to pre-date written language by a number of centuries, with passed-down spoken language lasting for millennia (https://www.theatlantic.com/science/archive/2022/10/indigeno... ).
> “They’re able to describe the colour of clothes the humans are wearing, they’re able to describe the size and shape of humans, even, amazingly, whether a human once appeared with a gun… In one 10th of a second, they say ‘Tall thin human wearing blue shirt walking slowly across the colony.'”
You mean people think in some sort of embedded signal in their brains that they then turn into language? That's a crazy idea. I wonder if these machine learning bozos will ever think to use something like that rather than just copying language.
> You mean people think in some sort of embedded signal in their brains that they then turn into language? T
What would be the point of turning it into language internally?
When you get up out of your chair, are you thinking "Mecha Robot Legs activate!" or something?
The trouble here is that until I just now said that, you thought your counter-argument sounded intelligent. Enough so that you doubled down on it and were sarcastic about it.
>What would be the point of turning it into language internally?
Shorter pathway from thinking to writing or speaking. I don't have an inner monologue, and I do feel cognitive load in trying to express my thoughts that I don't think people with an inner monologue experience. I can 'see' something in my mind, but I need to actively search for words that correspond to that, since transferring an image or a 'concept graph' is impossible to do via speech and difficult to do with pencil and paper.
Like most people, I have something I call an inner voice. I can "hear" ideas in a monologue or sometimes dialogue conversation in my head, such as the words I’m typing now or when preparing to speak. I don’t believe I’d be able to think abstractly to nearly the degree I am capable of if I wasn’t granted the gift of language by my society, but it is an interesting thought. I’m sure there are mountains of development psychology and linguistic anthropology research on this question. Perhaps we have lost certain abstract reasoning or perception powers that our prehistoric ancestors took for granted, due to language.
This probably isn't obvious to you... but the only ideas you can ever hear that way are the ideas you've already heard elsewhere.
That's why "rubber duck programming" seems so cool to some dweebs. It's not that they receive any true insight to it. But it helps them go back over all the other rules that they've learned elsewhere. Likely, it works better for some people than others, because those people have learned more rules. But they'll never discover another new one, or be inspired by some epiphany that way.
Hell, I'm pretty sure all of this is implicated in the Flynn Effect.
There are a bunch of other implications more disturbing still...
Had a hard time with this thought: if we’re so sure we’d be able to manipulate agents thoughts (agents trained to think like humans), what keeps humans from being manipulated with similar efficacy?
Until we modify our basic CS binary logic to emulate the varied axon length of dendrite tree around a soma nerve cell (which in itself could be called multi-input variable-input-delay-OR logic in CS terms), this design of thought emulation would be severely constrained and fall short of the goal to true emulation of human thought.
Nonsense, we can emulate any sort of analog process with binary transistors. If you prefer using an analog computer, fine, it’s an interesting area, but I’m not better on analog neural network computers catching up to digital.
> we can emulate any sort of analog process with binary transistors
have... have you done any research on the topic at all? Yes, you can model any sort of system using binary transistors, but the fidelity in which you do so is severely constrained by bit-depth and, depending on the system, might have error propagation issues.
In an analog computer, you’re trading quantization error for regular process noise and noise in initial conditions. If you’re concerned about modelling chaotic systems, you won’t get a more accurate just because you’re doing it analog.
40 comments
[ 2.8 ms ] story [ 97.0 ms ] threadhttps://twitter.com/johnjnay/status/1664798780644904960
Building a system that can leverage that knowledge to bootstrap "thinking circuits" can prove challenging, that's for sure, but I expect that collecting quality "thinking out loud" data is more challenging.
That’s an interesting claim to make. What are you basing this on? Framing this as a subroutine also seems a bit of a limited view on our dialogue generation abilities.
I can assure you that my internal dialogue rarely stops (meditation is an exception), and most often isn’t worried about talking with other people.
But for those who have no control over it, it just goes apeshit. I've heard self-reported cases of people doing the same but with historical or even fictional people. They try to use it for everything... as if that simulated person they're "talking to" will somehow be capable of solving a problem they themselves aren't capable of.
I think I was in grade school when I realized all the rest of you were lunatics. Took a little while longer to understand the nature of the cognitive flaw.
> Do you think they still don't think in language?
Yes. Because if that's how you "think", then you're not thinking at all. Which means you're also incapable of understanding how thought works.
You're asserting that I'm wrong, offering no evidence or even a plausible alternative theory. It's just another one of those absurd "truths" that dumb people assume is true and no one ever calls them on it.
I've explained that it is but one mental faculty, I've explained the evolutionary origin of it, and to top it all off, I've explained why you mistake the output of this singular mental faculty to be"thought".
I would hope that you'd be able to work through on first principles why I must be correct. You've just said thought is impossible without human (spoken, maybe signed) language, when there is clear counter-evidence in the world that everyone is aware of that contradicts your asinine theory.
I would hope that. Except that it's futile.
I'm going to make an even bolder claim and say that there's no such thing as a thought completely independent of dialogue with other humans.
As I am a writer, I not only think in language, but I act in language.
I would add a link to a conversation with neuroscientist Anil Seth about the nature of consciousness, and a quote from that, which posits maybe this isn't an either/or situation. There is much diversity in what constitutes "thought", even in humans.
"You know, some people might see colors differently. Other people might have more vivid mental imagery. We might experience the flow of time to be different from one person to another. Our expectations, our tendency to see faces in clouds, let’s say, that might differ from one person to another too. And very little is known about this hidden landscape of inner diversity."
https://www.quantamagazine.org/what-is-the-nature-of-conscio...
Yes, there are people who believe that thought is language (and not just inner dialogue as you mention in this thread). That without ideas formed in language as a precursor they wouldn't be able to think about those ideas, and that they do this thinking through the internal manipulation of other words.
I'm not quite sure what it's about as I'm not one of them, but they exist.
No. They still have conscious awareness. I think this is more a matter of conscious identification (or lack thereof) with various modes of thought.
But when I am trying to actively think of something I know no other way than to think in words. I am not a visual thinker (I'm kind of aphantasic, when you ask me to think about a horse, only an idea pops up in my head without any specific image and I have to try really hard and describe the horse with words to make some kind of misty parts of the horse).
I've had way too many instances of not being able to find a word to express an idea, or not remembering the specific words people expressed but just what I think was the gist of what they said (whether or not I interpreted them correctly) to think of myself as a verbal thinker. When it comes to word production sometimes they just come, at other times it takes me time, and I can go into a state where they just flow, but in that state my conscious is checked out of the process. It has the experience of an autonomic process almost.
I do have an inner monologue though (frequently caught up with OCD counting, or at the moment Taylor Swift's Anti-Hero song as well as speaking the words I am writing as I am writing them).
> only an idea pops up in my head without any specific image and I have to try really hard and describe the horse with words to make some kind of misty parts of the horse
Is this idea the words which describe the horse, or a pre-visual/pre-verbal idea of a horse?
I mean you can look at the speed of innovations. things were slow as hell until the 1400s, when the printing press was invented, the industrial revolution would've never kicked off without the ability to push thoughts and language into more hands.
Education is all about language, hard to learn in grunts and pointing.
There's more that goes into actual memories than language, though language is how we understand it interpret them... we have senses, feelings, and other contexts associated with a memory but without a word for sadness or anger, how do we know what we're even feeling?
Language is one of the things that truly separates us from animals and before written language our hunter gatherer ancestors were little more than animals wearing fur skins.
When I imagine a tree, I get the image and the feel and the scent, the qualia. Hearing the word “pine” for the first time did not make me smell anything. I had to learn to associate those syllables with my memories of pine trees, and they only work between me and other people who have done that.
> but without a word for sadness or anger, how do we know what we're even feeling?
"I'm feeling kind of a mix between the way I felt that time that thing happened and that time I did that other stuff. I don't like it." Without words at all, this would be a flash of memory of the two events which I would realize are related (emotionally) to the event happening to me now.
For anger, particularly, the emotion (maybe better said to be a mood state) usually comes with a lot of drives. Whether we have a word for what we're feeling, we know what actions we want to take because of the feeling. Or at least I do.
Circumlocutions are born out of ideas without words to express them. E.g. (literally 5 seconds to watch the relevant part): https://youtu.be/UEKWXFESU68?t=150
> Language is one of the things that truly separates us from animals and before written language our hunter gatherer ancestors were little more than animals wearing fur skins.
Animals have a crap load of language-like elements, with taught dialects and the like. These are seen in social bees, prairie dog communities, and numerous other animals. We humans have very complex languages that facilitate a ton of nuance and novel information types. But don't sell animals, or our ancestors, short. Human civilization appears to pre-date written language by a number of centuries, with passed-down spoken language lasting for millennia (https://www.theatlantic.com/science/archive/2022/10/indigeno... ).
https://www.animalcognition.org/2015/03/11/the-linguistic-ge...
> “They’re able to describe the colour of clothes the humans are wearing, they’re able to describe the size and shape of humans, even, amazingly, whether a human once appeared with a gun… In one 10th of a second, they say ‘Tall thin human wearing blue shirt walking slowly across the colony.'”
What would be the point of turning it into language internally?
When you get up out of your chair, are you thinking "Mecha Robot Legs activate!" or something?
The trouble here is that until I just now said that, you thought your counter-argument sounded intelligent. Enough so that you doubled down on it and were sarcastic about it.
Shorter pathway from thinking to writing or speaking. I don't have an inner monologue, and I do feel cognitive load in trying to express my thoughts that I don't think people with an inner monologue experience. I can 'see' something in my mind, but I need to actively search for words that correspond to that, since transferring an image or a 'concept graph' is impossible to do via speech and difficult to do with pencil and paper.
That's why "rubber duck programming" seems so cool to some dweebs. It's not that they receive any true insight to it. But it helps them go back over all the other rules that they've learned elsewhere. Likely, it works better for some people than others, because those people have learned more rules. But they'll never discover another new one, or be inspired by some epiphany that way.
Hell, I'm pretty sure all of this is implicated in the Flynn Effect.
There are a bunch of other implications more disturbing still...
have... have you done any research on the topic at all? Yes, you can model any sort of system using binary transistors, but the fidelity in which you do so is severely constrained by bit-depth and, depending on the system, might have error propagation issues.