On Mind, Language and Machines | Hacker News Reader

On Mind, Language and Machines

0x0f0f0f.github.io

115 points0x0f0f0f23 comments

> Prof. Matsuzawa, through his cognitive research of chimpanzee Ai

I had to do a double take there.

Wonderful essay!

I've found myself going through a similar path in my own research and thinking.

My hope is that an optimism of computation picks up over the coming decade and beyond; an optimism that computation can serve us at the individual level, beyond what others have created for us.

In the cultures with the abrahamic religions (especially, but also elsewhere) the 'language of the birds' is considered a mythical 'divine language', some characters in the 'holy books' who are famed for wisdom or considered prophets are supposed to have understood it.

I'm sure this has motivated many to study birdsong and it's interesting that their language is regular, maybe birds are FSM like our computers :^)

Edit: I guess that would mean their societies are petri nets / statecharts =þ

A few days back, HN saw https://news.ycombinator.com/item?id=22193451 which told us not everyone uses an internal monologue for their thinking process.

Now this article hints evolution favored the internal monologue design of thinking process, over alternatives.

I've read that post. It's still a mistery to me the fact that there are people without internal monologues that can speak without any sort of problem. Chomsky and Berwick hinted that recursive language may have evolved as "an internal monologue" in their "Why Only Us" essay, where they surely can explain their evolutionary point of view better than me.

I guess if that there really are people without "the internal tool of thought", they should volunteer to be studied, especially with brain imaging and functional tests. They could help us understand more about where language comes from.

It's less of a "do not have internal monologue" as experience them differently to those who have a stream of consciousness. Personally, I experience consciousness internally as a bunch of concepts and ideas, rather than actual verbose language. It doesn't proclude the generation of ideas or so on, and I definitely don't struggle with speaking.

I can only guess that one has a more verbal or visual imagination depending on his/her development, and it would be interesting if we knew more about the genomic factors taking place in the development of the stream of consciousness. Depending on the context I also have both verbal and visual streams of consciousness. A funny thing is that my verbal mental streams are far more fluid in English, which is not my fully native language, than in Italian. When I write in Italian my imagination is much more visual and less descriptive.

Same here, my French got relegated to the back of my mind, English took over (that was of my own volition first, to learn best, but then it became habit).

Sometimes I revert to writing in French; it takes a little time like bicycle to come back really well. It often framed my mind in a different way, provided I never thought of problem X in French. Weird. I'm more likely to speak in "poetic" terms in French, and more "logic" in English. Weird that my own abstractions shift like that. Real tangible things are not affected it seems. With other people it makes me a slightly different person (I suppose I mimic the cultural traits I attribute to each language, as if emotions or even volition got influenced by it).

Same here but with a different first language and English as a second. I tend to be more poetic with my first language but use it a lot more sparingly nowadays. As far as I am aware most of my internal monologue isn't verbalized but it does happen. Over all I wouldn't rely too much on my internal monologue, it feels that it slows down my thinking to put whatever abstract thought I have into words.

You just made me realize that we might all "skip the words" when we have internalized a concept.

Case in point for me: math. For instance these days I'm working on geometric modelling, and when I see a wedge (exterior product of two 1D vectors = a 2D surface) I don't even read or say it, I just "see" the concept (not even a parallelogram, I don't visualize, it's just this abstraction for 2D which relates to number theory, physical spaces, etc). Then if I do something with this, I kind of navigate a map or graph (where e.g. gradients or dot products are sort of nodes). I'm gonna be 10x slower if I have to put words in it when talking to someone for instance.

Does that begin to describe how non-verbal thinkers think about pretty much everything?

Yes. Daniel Kahneman's 'Thinking fast and slow' comes to mind. I haven't read the book but listened to a few podcasts about it. I think it comes down to what mode of thinking is being employed. Some use more than the other, some have a balanced use.

That's a great point.

I read the book and indeed, there's something of that dichotomy between what he calls "system 1" and "system 2".

System-2 is the very thorough, conscious thought; the visual or otherwise perceptible, representable as abstract but tangible objects (like please think of an elephant: this thing you see right now in your mind is an abstract but tangible object). System-1 by comparison is more intuitive, and I think can typically be illustrated by procedural memory: things you know but forgot how you learned, like riding a bike or playing chess.

In interviews he explains that, while most questions would confuse system-2 for "intelligence" or "logic", it's really not: a chess master for instance uses system 1 a lot to instinctively select only strong moves, or a chef uses system-1 to select only compatible and worthy ingredients — and then system-2 may think in depth about this small subset. The idea is that the system-1 of a master would beat the hell out of any layman's system-2 for a given skillset.

So when you really internalize, learn something, it's like the schema for it in your brain moves gradually from the system-2 space to the system-1 space so to speak. In many ways, there's no possible mastery without system-1. That's what makes "expertise" as we call it, this "6th" sense for a domain.

Video games are so much like that: first you have to think about everything, but gradually it becomes 'second nature' like walking or driving.

Now here's my point: the whole process of "thinking to oneself" is system-2. All of it. By definition, if we use Kahneman's term. System-1 is more like the underlying parameters of a thinking space (like knowing the rules at a deep level of instinct), wherein system-2 manipulates variables.

Let's risk the analogy that:

- system-1 feels like each person's particular 'optimization' of 'brain hardware' or 'OS': for instance, I may have a GPU-like skill or such a "driver" or "ad hoc firmware" to process math (because training); you may have a great NN-like skill to process consulting problems in your domain (result of years of experience); it's all related to training and personal story i.e. compounded construction, effort over time.

- Whereas system-2 is the software layer, the apps, this inner-userland where you push buttons and 'see' results, and largely transmissible through communication, teaching (again in my words, system-2 objects are 'representable', abstract but tangible, as evidenced by the fact we can take objects from brain to brain). You can do almost anything in system-2, but most of it is weak and orders of magnitude slower absent of further optimization.

I think it warrants more in-depths discussions and research probably, but that's my ballpark view. What's intriguing about this "inner speech" discrepancy in thinking is precisely that systems-2 are commonly thought to be relatively standard computers between human beings (again, I can take an app from mine and teach it to yours, so there's some level of direct compatibility there). Evidently, that's not entirely true, and the deeper question is whether it points to a different architecture or simply a matter of perception; how isomorphic are different approaches at the end of the day.

Maybe they're just not aware of it? That data stream still exists, it's just not hooked up to their audio centers so they don't hear it in their 'inner ear'?

You make me wonder if language regions of the brain get activated when we "talk to ourselves" in the same way as when we talk to others.

Edit: I love a downvote as much as the next guy, but if you're a neuroscientist who knows, please explain or share links?

Yes I also wonder that, do you want to talk about this? tweet me @LeibGruber

Not everyone’s brain works the same way. Read some stuff by Temple Grandin or Oliver Sacks. As for how evolution may have done anything is just a guess until there is empirical data backing up that hypothesis. It’s more likely that Chomsky just assumes everyone’s mind worked like his and used that to tell a just so evolutionary story.

>Not everyone’s brain works the same way. Read some stuff by Temple Grandin or Oliver Sacks

Yes, but these study outliers. Most everybody's brain works the same way (or in a few, finite, ways)

We assume brains work in the same way, but we lack any kind of tooling for examining this other than fMRI. All this high level stuff like conceptualisation and visualisation relies on subjective self-reporting. I don't think we've even proved whether or not Chomsky's "universal grammar" hypothesis is correct.

>We assume brains work in the same way, but we lack any kind of tooling for examining this other than fMRI

Yes, but it's also kind of a no-brainer. We're the same species, with mostly the same wiring...

"Deaf" people have a mind too and what would their internal monologue be since they never heard uttered words but learn signs that are not auditive internally? Is it abstract internal monologue? Why would it be called monologue? Are they missing the internal tool of thought or maybe there are more tools of thought and some of them don't require internal monologue?

I don't experience my thought as either words or pictures. I experience my thought as ideas. Ideas are independent of both, since the same idea can be expressed with different words or pictures.

Meh, it bears a strong resemblance to the corn-eating-algorithm-correlation: https://slatestarcodex.com/2019/01/15/kernel-of-doubt-testin...

You can discover anything you look for with a leading-question survey.

Evolution doesn't prevail completely. It might favor a path (and thus see its population grow), but the other path is not guaranteed to be eclipsed already.

Reply on news.ycombinator.com