https://www.sciencealert.com/theres-a-big-difference-in-how-...
The actual paper [1] says that functional MRI (which is measuring which parts of the brain are active by sensing blood flow) indicates that different brain hardware is used for non-language and language functions. This has been suspected for years, but now there's an experimental result.
What this tells us for AI is that we need something else besides LLMs. It's not clear what that something else is. But, as the paper mentions, the low-end mammals and the corvids lack language but have some substantial problem-solving capability. That's seen down at squirrel and crow size, where the brains are tiny. So if someone figures out to do this, it will probably take less hardware than an LLM.
This is the next big piece we need for AI. No idea how to do this, but it's the right question to work on.
[1] https://www.nature.com/articles/s41586-024-07522-w.epdf?shar...
https://www.scientificamerican.com/article/gut-second-brain/
There are 100 million in my gut, but it doesn't solve any problems that aren't about poop, as far as I know.
https://en.wikipedia.org/wiki/List_of_animals_by_number_of_n...
If the suspiciously round number is accurate, this puts the human gut somewhere between a golden hamster and ansell's mole-rat, and about level with a short-palated fruit bat.
Should we expect experts in cognitive science exposing their view in a scientific publication to stick to the narrowest median view of language though? All the more when in the same article you quote people like Russell who certainly didn't have a naïve definition of language when expressing a point of view on the matter.
And slapping in general can definitely communicate far more than a single thing depending on many parameters. See https://www.33rdsquare.com/is-a-slap-disrespectful-a-nuanced... for a text exploring some of nuances of the meaning it can encompasse. But even a kid can get that slap could perfectly have all the potential to create a fully doubly articulated language, as The Croods 2 creators funnily have put in scene. :D
And recursion as the unique trait for human language differentiation is not necessarily completely consensual https://omseeth.github.io/blog/2024/recursive_language/
Also, let's recall that in its broader meaning, the scientific consensus is that humans are animals and they evolved through the same basic mechanism as all other life forms that is evolution. So even assuming that evolution made some unique language hability emerge in humans, it's most likely that they share most language traits with other species and that there is more things to learn from them that what would be possible if it's assumed they can't have a language and thoughts.
There's been progress. Look at this 2020 work on neural net controlled drone acrobatics.[1] That's going in the right direction.
If you look up 'mentalese' you can find a bunch written about it. There's an in-depth article by Daniel Gregory and Peter Langland-Hassan, in the incredible Stanford Encyclopedia of Philosophy, on Inner Speech (admittedly, I'm taking a leap to think they mean precisely the same thing). [2]
[0] Steven Pinker, The Blank Slate: The Modern Denial of Human Nature (2002)
[1] Oxford English Dictionary
Warning, watch out for waving hands: The way I see it is that cognition involves forming an abstract representation of the world and then reasoning about that representation. It seems obvious that non-human animals do this without language. So it seems likely that humans do too and then language is layered on top as a turbo boost. However, it also seems plausible that you could build an abstract representation of the world through studying a vast amount of human language and that'll be a good approximation of the real-world too and furthermore it seems possible that reasoning about that abstract representation can take place in the depths of the layers of a large transformer. So it's not clear to me that we're limited by the data we have or necessarily need a different type of data to build a general AI although that'll likely help build a better world model. It's also not clear that an LLM is incapable of the type of reasoning that animals apply to their abstract world representations.
https://arstechnica.com/ai/2024/10/llms-cant-perform-genuine...
or do you maybe think no logical reasoning is needed to do everything a human can do? Tho humans seem to be able to do logical reasoning
https://www.cbc.ca/news/canada/saskatchewan/inner-monologue-...
Siegmund, J., Kästner, C., Apel, S., Parnin, C., Bethmann, A., Leich, T. & Brechmann, A. (2014). Understanding understanding source code with functional magnetic resonance imaging. In Proceedings of the 36th International Conference on Software Engineering (pp. 378-389).
Peitek, N., Siegmund, J., Apel, S., Kästner, C., Parnin, C., Bethmann, A. & Brechmann, A. (2018). A look into programmers’ heads. IEEE Transactions on Software Engineering, 46(4), 442-462.
Krueger, R., Huang, Y., Liu, X., Santander, T., Weimer, W., & Leach, K. (2020). Neurological divide: An fMRI study of prose and code writing. In Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering (pp. 678-690).
Peitek, N., Apel, S., Parnin, C., Brechmann, A. & Siegmund, J. (2021). Program comprehension and code complexity metrics: An fmri study. In 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE) (pp. 524-536). IEEE.
[0]: https://www.frontiersin.org/10.3389/conf.fninf.2014.18.00040...
[1]: https://ieeexplore.ieee.org/abstract/document/8425769
Probably by putting simulated animals into simulated environments where they have to survive and thrive.
Working at animal level is uncool, but necessary for progress. I had this argument with Rod Brooks a few decades back. He had some good artificial insects, and wanted to immediately jump to human level, with a project called Cog.[1] I asked him why he didn't go for mouse level AI next. He said "Because I don't want to go down in history as the inventor of the world's greatest artificial mouse."
Cog was a dud, and Brooks goes down in history as the inventor of the world's first good robotic vacuum cleaner.
> DA was impaired in solving simple addition, subtraction, division or multiplication problems, but could correctly simplify abstract expressions such as (b×a)÷(a×b) or (a+b)+(b+a) and make correct judgements whether abstract algebraic equations like b − a = a − b or (d÷c)+a=(d+a)÷(c+a) were true or false.
> Sensitivity to the structural properties of numerical expressions was also evaluated with bracket problems, some requiring the computation of a set of expressions with embedded brackets: for example, 90 [(3 17) 3].
Discussions of whether or not these sorts of algebraic or numerical expressions constitute a "language of mathematics" aside (despite them not engaging the same brain regions and structures associated with the word "language"); it may be the case that these sorts of word sequences and symbols processed by structures in the brain's left hemisphere are not essential for thought, but can still serve as a useful psychotechnology or "bicycle of the mind" to accelerate and leverage its innate capabilities. In a similar fashion to how this sort of mathematical notation allows for more concise and precise expression of mathematical objects (contrast "the number that is thrice of three and seventeen less of ninety") and serves to amplify our mathematical capacities, language can perhaps be seen as a force multiplier; I have doubts whether those suffering from aphasia or an agrammatic condition would be able to rise to the heights of cognitive performance.
And it turns out that human brain volume and intelligence are moderately-highly correlated [1][2]!
[1]: https://pmc.ncbi.nlm.nih.gov/articles/PMC7440690/ [2]: https://www.sciencedirect.com/science/article/abs/pii/S01602...
Even tools present us a certain 'language', talking to us via beeps, blinks and buzzes, and are having increasingly interesting discussions amongst themselves (e.g. subreddit simulator, agent based modeling). Recent philosophers of technology as Mark Coeckelbergh present a comprehensive argument for why we need to move away from the tool/language barrier [0], and has been part in informing the EC Expert Group on AI [1].
[0]: https://www.taylorfrancis.com/books/mono/10.4324/97813155285...
[1]: https://philtech.univie.ac.at/news/news-about-publicatons-et...
His work explores the neuropsychology of emotions WAIT DON'T GO they are actually the substrate of consciousness, NOT the other way around.
We have 7 primary affective processes (measurable hardware level emotions) and they are not what you think[2]. They are considered primary because they are sublinguistic. For instance, witnessing the color red is a primary experience, you cannot explain in words the color red to someone who has not ever seen it before.
His work is a really fascinating read if you ever want to take a break from puters for a minute and learn how people work.
PS the reason this sort of research isn't more widely known is because the behaviorist school was so incredibly dominant since the 1970s they made it completely taboo to discuss subjective experience in the realm of scientific discourse. In fact the emotions we are usually taught are not based on emotional states but on muscle contractions in the face! Not being allowed to talk about emotions in psychological studies or the inner process of the mind is kinda crazy when you think about it. So only recently with neuroimaging has it suddenly become ok to acknowledge that things happen in the brain independent of externally observable behavior.
[2] - seeking - fear - anxiety and grief - rage - lust - play!!! - caring
[3] if this sounds familiar at all it's because Jordan Peterson cites Jaak Panksep all the time. Well 50% of the time, the other 50% is CG Jung and the final 50% is the book of Exodus for some reason.
After all, that's what Artificial General Intelligence would at least in part be about: finding and proving new math theorems, creating new poetry, making new scientific discoveries, etc.
There is even a new challenge that's been proposed: https://arcprize.org/blog/launch
> It makes sense that the process of thinking and the process of translating those thoughts into and out of language would be distinct
Yes, indeed. And LLMs seem to be very good at _simulating_ the translation of thought into language. They don't actually do it, at least not like humans do.
For that not to be the case, you'd have to take the position that humans experience consciousness and they talk about consciousness but that there is no causal link between the two! It's just a coincidence that the things you find yourself saying about consciousness line up with your internal experience?
https://www.lesswrong.com/posts/fdEWWr8St59bXLbQr/zombies-zo...
[0] https://archive.org/details/a-a-the-mystical-and-magical-sys...
To deal with the awkwardly apparent fact that consciousness certainly seems to have physical effects, zombiephiles challenge the notion that physics is causally closed, so that it is conceivable that something non-physical can cause physical effects. Their approach is to say that the causal closure of physics is not provable, but at this point, the argument has become a lexicographical one, about the definition of the words 'physics' and 'physical' (if one insists that 'physical' does not refer to a causally-closed concept, then we still need a word for the causal closure within which the physical is embedded - but that's just what a lot of people take 'physical' to mean in the first place.) None of the anti-physicalists have been able, so far, to shed any light on how the mind is causally effective in the physical world.
You might be interested in the late Daniel Dennett's "The Unimagined Preposterousness of Zombies": https://dl.tufts.edu/concern/pdfs/6m312182x
It's hard for me to understand where my peers are coming from on the other side of this argument and respond without being dismissive, so I'll do my best to steelman the argument later.
Machine learning models are function approximators and by definition do not have an internal experience distinct from the training data any more than the plus operator does. I agree with the sentiment that even putting it in writing gives more weight to the position than it should, bordering on absurdity.
I suppose this is like the ELIZA phenomena on steroids, is the only thing I can think of for why such notions are being entertained.
However, to be generous, lets do some vigorous hand waving and say we could find a way to have an embodied learning agent gather sublinguistic perceptual data in an online reinforcement learning process, and furthermore that the (by definition) non-quantifiable subjective experience data could somehow be extracted, made into a training set, and fit to a nicely parametric loss function.
The idea then is that could find some architecture that would allow you to fit a model to the data.
And voila, machine consciousness, right? A perfect model for sentience.
Except for the fact that you would need to ignore that in the RL model gathering the data and the NN distilled from it, even with all of our vigorous hand waving, you are once again developing function approximators that have no subjective internal experience distinct from the training data.
Let's take it one step further. The absolute simplest form of learning comes in the form of habituation and sensitization to stimuli. Even microbes have the ability to do this.
LLMs and other static networks do not. You can attempt to attack this point by fiatting online reinforcement learning or dismissing it as unnecessary, but I should again point out that you would be attacking or dismissing the bare minimum requirement for learning, let alone a higher order subjective internal experience.
So then the argument, proceeding from false premises, would claim that the compressed experience in the NN could contain mechanical equivalents of higher order internal subjective experiences.
So even with all the might vigorous hand waving we have allowed, you have at best found a way to convert internal subjective processes to external mechanical processes fit to a dataset.
The argument would then follow, well, what's the difference? And I could point back to the microbe, but if the argument hasn't connected by this point, we will be chasing our tails forever.
A good book on the topic that examines this in much greater depth is "The Self Assembling Brain".
That being said, I am hella jealous of the VC money that the grifters will get for advancing the other side of this argument.
For enough money I'd probably change my tune too. I can't by a loaf of bread with a good argument lol
> In the stage of Cause and Effect, the relationships between mental and physical phenomena become very clear and sometimes ratchet-like. There is a cause, such as intention, and then an effect, such as movement. There is a cause, such as a sensation, and there is an effect, namely a mental impression.
Trying to increase the frequency at which you oscillate between physical sensations and mental sensations is a fascinating exercise.
[0] https://www.mctb.org/mctb2/table-of-contents/part-iv-insight...
As humanity has struggled to understand the world, it has frequently given names to concepts that seem to matter, well before it is capable of explaining with any sort of precision what these things are, and what makes them matter - take the word 'energy', for example.
It seems clear to me that one must have these vague concepts before one can begin to to understand them, and also that it would be bizarre not to give them a name at that point - and so, at that point, we have a word without a locked-down definition. To insist that we should have the definition locked down before we begin to investigate the phenomenon or concept is precisely the wrong way to go about understanding it: we refine and rewrite the definitions as a consequence of what our investigations have discovered. Again, 'energy' provides a useful case study for how this happens.
A third point about the word 'energy' is that it has become well-defined within physics, and yet retains much of its original vagueness in everyday usage, where, in addition, it is often used metaphorically. This is not a problem, except when someone makes the lexicographical fallacy of thinking that one can freely substitute the physics definition into everyday speech (or vice-versa) without changing the meaning.
With many concepts about the mental, including 'reasoning', we are still in the learning-and-writing-the-definition stage. For example, let's take the definition you bring up: reasoning as good cognition. This just moves us on to the questions of what 'cognition' means, and what distinguishes good cognition from bad cognition (for example, is a valid logical argument predicated on what turns out to be a false assumption an example of reasoning-as-good-cognition?) We are not going to settle the matter by leafing through a dictionary, any more than Pedro Carolino could write a phrase book just from a Portugese-English dictionary (and you are probably aware that looking up definitions-of-definitions recursively in a dictionary often ends up in a loop.)
A lot of people want to jump the gun on this, and say definitively either that LLMs have achieved reasoning (or general intelligence or a theory of mind or even consciousness, for that matter) or that they have not (or cannot.) What we should be doing, IMHO, is to put aside these questions until we have learned enough to say more precisely what these terms denote, by studying humans, other animals, and what I consider to be the surprising effectiveness of LLMs - and that is what the interviewee in the article we are nominally discussing here is doing.
You entered this thread by saying (about the paper underlying an article in Ars Tech [1]) I’ll pop in with a friendly “that research is definitely wrong”. If they want to prove that LLMs can’t reason..., but I do not think there is anything like that claim in the paper itself (one should not simply trust what some person on HN says about a paper. That, of course, goes as much for what I say about it as what the original poster said.) To me, this looks like the sort of careful, specific and objective work that will lead to us a better understanding of our concepts of the mental.
We have, it's called DreamCoder. There's a paper and everything.
Everything needed for AGI exists today, people simply have (incorrect) legacy beliefs about cognition that are holding them back (e.g. "humans are rational").
> You can ask whether people who have these severe language impairments can perform tasks that require thinking. You can ask them to solve some math problems or to perform a social reasoning test, and all of the instructions, of course, have to be nonverbal because they can’t understand linguistic information anymore.
Surely these "non-verbal instructions" are some kind of language. Maybe all human action can be considered language.
A contrarian example to this research might be feral children, i.e people who have been raised away from humans.[0] In most cases they are mentally impaired; as in not having human-like intelligence. I don't think there is a good explanation why this happens to humans. And why it doesn't happen to other animals, which develop normally in species-typical way whether they are in the wild or in human captivity. It seems that most human behavior (even high-level intelligence) is learned / copied from other humans, and maybe this copied behavior can be considered language.
If humans are "copy machines", there's also a risk of completely losing the "what's it like to be a human" behavior if children of the future are raised by AI and algorithmic feeds.
Apple’s recent research summarized here [0] is worth a read. In short, they argue that what LLMs are doing is more akin to advanced pattern recognition than reasoning in the way we typically understand reasoning.
By way of analogy, memorizing mathematical facts and then correctly recalling these facts does not imply that the person actually understands how to arrive at the answer. This is why “show your work” is a critical aspect of proving competence in an education environment.
An LLM providing useful/correct results only proves that it’s good at surfacing relevant information based on a given prompt. That fact that it’s trivial to cause bad results by making minor but irrelevant changes to a prompt points to something other than a truly reasoned response, i.e. a reasoning machine would not get tripped up so easily.
It’s bloody obvious that when I classify success I mean that the llm is delivering a correct and unique answer for a novel prompt that doesn’t exist in the original training set. No need to go over the same tired analogies that have been regurgitated over and over again that you believe LLMs are reusing memorized answers. It’s a stale point of view. The overall argument has progressed further then that and we now need more complicated analysis of what’s going on with LLMs
Sources: https://typeset.io/papers/llmsense-harnessing-llms-for-high-...
https://typeset.io/papers/call-me-when-necessary-llms-can-ef...
And these two are just from a random google search.
I can find dozens and dozens of papers illustrating failures and successes of LLMs which further nails my original point. LLMs both succeed and fail at reasoning.
The main problem right now is that we don’t really understand how LLMs work internally. Everyone who claims they know LLMs can’t reason are just making huge leaps of irrational conclusions because not only does their conclusion contradict actual evidence but they don’t even know how LLMs work because nobody knows.
We only know how LLMs work at a high level and we only understand these things via the analogy of a best fit curve in a series of data points. Below this abstraction we don’t understand what’s going on.
Just a few days ago was "What do you visualize while programming?", and there's a few of us in the comments that, when programming, think symbolically without language: >>41869237
> Recent work has revealed that the neural activity patterns correlated with sensation, cognition, and action often are not stable and instead undergo large scale changes over days and weeks—a phenomenon called representational drift.
[...]
So, I'm not sure how conclusive this fmri activation study is either.
Though, is there a proto language that's not even necessary for the given measured aspects of condition?
Which artificial network architecture best approximates which functionally specialized biological neutral networks?
OpenCogPrime:KnowledgeRepresentation > Four Types of Knowledge: https://wiki.opencog.org/w/OpenCogPrime:KnowledgeRepresentat... :
> Sensory, Procedural, Episodic, Declarative
From https://news.ycombinator.com/item?id=40105068#40107537 re: cognitive hierarchy and specialization :
> But FWIU none of these models of cognitive hierarchy or instruction are informed by newer developments in topological study of neural connectivity;
There is no reason to assume consciousness is Turing computable [1].
[1] https://en.m.wikipedia.org/wiki/Church%E2%80%93Turing_thesis
[1] Mark Bickford, Liron Cohen, Robert L. Constable, and Vincent Rahli. 2018. Computability Beyond Church-Turing via Choice Sequences. In Proceedings of the 33rd Annual ACM/IEEE Symposium on Logic in Computer Science (LICS '18). Association for Computing Machinery, New York, NY, USA, 245–254. https://doi.org/10.1145/3209108.3209200