General intelligence is an ability to cope, adapt and thrive in an ecology: to start from a limited set of capabilities, and via exploration, acquire a rich competence. To develop conceptualisations, techniques of coordination and control, to form novel goals and strategies to realise them, and so on.
General intelligence is a strategy to defer the acquisition of abilities from the process of construction/blueprinting (ie., genes, evolution..) to the living environment of the animal. The most generally intelligent animals are those that have nearly all of their sensory motor skills acquired during their life -- we learn to walk and so can learn to play the piano, and to build a rocket.
There is a serious discontinuity in strategy to achive this defferal: the kinds of processes which "blueprint" the intelligence of a bacterium are discontinuous with the processes which a living animal needs to dynamically conceptualise its environment under shifts to its structure.
Of the latter animals need: living adaption of their sensory-motor systems, heirachical coordination of their bodies, robust causal modelling, and so on.
General intelligence is primitively a kind of movement, which becomes abstract only with a few hundred thousand years of culture. The earliest humans, able to lingusitically express almost nothing, were nevertheless generally intelligent.
Present computer-science-led investigations into "intelligence" assume you can operate syntactically across the most peripheral consequences of general intelligence given by linguistic representations. This is profoundly misguided: each todller necessarily must learn to walk. You cannot just project a slideshow of walking, and get anywhere. And if you remove this capability and install a "walking module", you've remved the very capabilities which allow that child then to do anything new at all.
There is nothing in the linguistic syntactical shadow of human intelligence to be found in creating generally capable systems. It's just overfitting to our 2024 reflections.
Maybe that would be a suitable working definition of general intelligence, and props to you for even giving a definition at all (in contrast to TFA). However your definition seems almost tailor-made to exclude present and near-future AI (and, I suspect, motivated thereby) . Current AI works by being trained on large amounts of existing data. If current AI would be real intelligence, we would be sad, therefore real intelligence is the opposite of intelligence trained on large amounts of data.
Having said that, one can also make the case that LLMs start from a limited set of capabilities and, via exploration, acquire a rich competence. Only these are linguistic abilities and the exploration is exploration of a linguistic environment. Maybe the real intelligence is the friends we made along the way i.e. the general class of algorithms roughly called "backpropagation and gradient descent on a very high-dimensional neural network".
The most meaningful definition of intelligence is one that captures the essence/nature of human/animal intelligence, which is where the word originated.
I think you can get to the core of it by considering the evolutionary benefit of intelligence - what beneficial behavioral capability has been optimized - which comes down to being able to utilize past experience to predict/plan future outcomes, rather than being locked into reactive behavior patterns like simpler animals.
LLMs, trained to predict based on past "experience", might (perhaps charitably) be considered to exhibit some intelligence, but where they notably fail is in situations where better prediction (utilization of prior experience) requires a process more similar to search with backtracking than a linear application of rules derived from the training data - i.e. in the areas of reasoning and planning.
You can try to put lipstick on the pig by adding RL-based post-training or wrapping the LLM in an agentic loop, trying to extract more value out of the training data and gain some semblance of reasoning, but at the end of the day it's still a pig - at heart just an expert system not a cognitive architecture.
Another obvious limitation of LLMs is that they are just a repository of canned knowledge/rules, with no ability to learn from "runtime" experience, and therefore lacking the ability to learn to handle novel problems by experimentation and adaptation to failure.
The limited intelligence of LLMs is firmly baked into their architecture - the transformer, being just as pass-thru model, as well as the way they are trained by SGD rather than an algorithm capable of continuous incremental learning.
It's tailor made to describe the phenomenon of animal intelligence that we're trying to model.
The only tailoring which goes on is by those who say, "we can only do X, therefore Y must be defined in terms of X". It's deeply pseudoscientific approach to investigation, as it completely abandons a scientific theory of empirical phenomenon over a purely circumstantial account given in terms of what tools we have to hand.
I can think of no other area where such an approach to investigation is permitted.
> General intelligence is an ability to cope, adapt and thrive in an ecology: to start from a limited set of capabilities, and via exploration, acquire a rich competence. To develop conceptualisations, techniques of coordination and control, to form novel goals and strategies to realise them, and so on.
One thing I've learned over the past few years is that _nobody_ knows what intelligence is, and it may not even exist as a genuinely measurable attribute. What you've described is certainly _a thing_ that is worth describing and thinking about, but it doesn't encompass everything that we think of as intelligence, and ascribes intelligence to processes which most of us don't think of as intelligent (ie, evolution and plant life).
The problem we have is that for the entire history of humanity, there has been a single example of something that "thinks like us" and our conception of what it means to be "intelligent" or to have "reason" or to "think" is just inextricably tied with all the other attributes that make us human.
I'm not at all sure that intelligence _must_ arise out of a process of evolution and natural selection, and i think that it may be possible to create an intelligent entity which completely lacks the ability to survive in an ecology on its own.
On the other hand, it's often said that while humans domesticated plants and animals, those same plants and animals also domesticated _us_. Human life rearranged itself around the requirements of farming and animal husbandry, and just in terms of pure biomass and range of habitat, becoming "domesticated" was a tremendously successful evolutionary strategy for the animals that we domesticated.
Human society is now _again_ re-arranging itself in order to take advantage of AI, and we're spending a lot of money and labor building these systems and maintaining them. It's hard to say that they haven't adapted themselves to surviving in an ecology, it's just a more abstract ecology than the sort of blood and claw ecology that we evolved in.
In some sense, these AIs are the ultimate expression of "memetic evolution" -- ideas that are able to spawn new ideas, without having any meaningful embodiment at all.
Thank you for being sensible We have so many "intelligence professors", who can conveniently dismiss SoTA AIs with a sleight of hand (Cholet, LeCunn etc), yet completely ignore the superintelligent traits these AIs already exhibit.
FYI: SoTA AIs can think and reason. They're far far away from mere memorization & retrieval. They aren't human yet, nor do we expect them to be. They'll just keep getting ever better in their own ways, and become super duper useful for practically everything as computer buddies first, agents next, robots next next.
The "intelligence professors" I'm hoping will at some point shut up and accept that "functional/universal approximation" is all you need, which is abundantly done by neural nets of today.
Human nature is to have a moral compass (conscience), with a mind to contemplate whether our planned behavior is positive or negative for the happiness of those around us, and a free will to then choose which paths to take.
What you describe is all present in the animals, to more or less advanced degree, where chimps and crows can use tools and even pass that knowledge on to others. Yes, our bodies follow the mammalian template, and with it comes our baseline tendencies to form packs to fight against other packs and fight for dominance within the pack, all in order to enjoy more physical pleasure. We inherit that, but we are capable of rising above those animalistic urges to become a humanitarian, who considers the happiness of others, and even the whole, in their decisions.
Further, with our advanced being and our free will, we are able to self-evolve our ideals, attitudes, and behaviors, in EITHER moral direction: either towards a more selfish, brutal, and callous competitive state, or towards a more selfless, compasionate, and caring cooperative state. The former leads to where we are now in human history, the latter, rarely exercised, leads to our highest potential, returning us to a happy, prosperous, environmental-concerned human race with various cultural differences but united in the success of each and every person, should they choose to participate.
So, no, no computer logic engine at present can in any way mimic the totality of "human intelligence" because very few people understand human nature, so they can only literally "ape" it. They can't even approach simulating what is going on within us, the only moral beings on this planet, and the only beings here with the power to consciously change ourselves and our environment, for better or worse, it all being our choice, however subconscious and inertial for the vast majority.
It is going a little under the radar, but AI is really throwing the religions into a tizzy now that they are being forced to think about these things. In the good old days, you could just trust the bible to say how we were created.
Now people are being forced to think about it for the first time. Where do morals come from? Do they even exist? Who am I? What am I doing here?
Like whole new generation having an existential crisis.
If you dare, you can read my recent comment history for the explanation, but I doubt you're going to like it. And it is the explanation. Someone had to do it!
True, but our consciousness comes with an integral moral compass that we are completely free to utterly ignore or even act in opposition to. It's our gift and our responsibility, and our absolute choice, for good or ill.
Is it actually integral? I’m not sure. Try asking a person whether some arbitrary scenario is right or wrong and why. A good null hypothesis is that, for people with no training in morality or ethics, responses will be uniformly distributed over outcomes.
Yes, but we can choose to ignore it. Ignorance of our human nature potentials is our choice, too, and that's the reason for the inertia of the world's societies, including ones that claim to be religious.
The key is that, just as our physical bodies have developmental stages of ever-increasing capability, so does our moral compass. We must learn how to not only use it, but to develop it and fine-tune over time, and we must train and use our mind to self-evolve ourselves.
Note that we can use our free will to de-tune it as well so that we have pointed ourselves in the direction opposite to our happiness, and that of those we come in contact with. In other words, we are free to use our abilities to create unhappiness, out of sadistic pleasure.
The first step of the spiritual path is awakening to this highest of human potential, where we learn to willfully and with difficult effort develop our moral compass in the direction of compassionate concern for the well-being of our fellow human beings. This is why the selfish -- to themselves and their in-group -- people of the world are loath to hear the term "woke"; they take pleasure in remaining ignorant of both the unhappiness they cause and thus karmically receive by their selfish actions. And the same impulse within us that seeks to keep us ignorant of our highest human potential, is also keen to keep its ideals, attitudes, and behaviors off other people's radar. The development of a selflessly compassionate morality become like garlic to a vampire (and they do suck, and suck the life-blood out of the world's systems and people, for sure, causing so much misery by their efforts).
So, yes, our moral compasses are each in various states of development, from the utterly ignored to sometimes-positive-sometimes-negative to the fully developed. As usual, such distributions follow a bell curveish shape. But only the long tail in the positive direction understands and manifests the highest morality. The middle bulk are hit or miss, per their cultures' predilections and the circumstances of their life. And the negative tail are the uttlerly selfish evil bastards of the world.
You can find a full explication of our human nature to self-evolve and the process of manifesting such change in the deep dialog I had yesterday under my comment to Maria Konnikova's poetry article submission. You have to skip past my initial reply and its grand-reply to get to the meat of it.
We fight to establish hierarchies of dominance called, "monopolies of violence", that we have social allegiance to. If a competing "dominance regime" is in our neighbourhood, we draw territorial boundaries -- and if these fail, riot, and if that fails, kill.
The strategy of intragroup 'mutual aid' is common across the animal kingdom -- and is paired with hostility to 'foreign aid' in its literal sense.
The achievement of the modern world is massive amounts of abundance which increases our generosity beyond typical chimpanzee proportions -- but not by much. And upon a single attack, or moment of scarcity, we return back exactly to our genocidal defaults -- which is to say, group-centred violence.
Abundance and 'wars only on our borders' creates a dangerous illusion of equanimity which is moreso just, "the feeling of an ape fat, tired and safe".
Yes, my recent comment history contains the explanation for why we have the potential for all of that but also to be true humanitarians. The short of it is that we can choose to learn how to be better by actually changing ourself, and then being a positive force in the world.
It is the Way, but we must choose it, after first seeking to escape our natural ignorance to the possibility.
That sentence literally makes no sense, obviously coming from the spoiled mind of a coddled rich kid.
The fact is that you chose to write what you wrote, for good or ill. I spent all day yesterday explaining the truth of our moral existence to y'all here, conversing with a very fine fellow who has a bit of knowledge. It brought me a joy that no one else on this site has ever felt. It was electric, sans drugs of any kind, and only a couple of sips of coffee all day, which is very rare for me.
No, there's a force within you that will work its damnedest to get you to quit reading it before you get to the bottom. It starts in that post about Maria Konnakova's poetry article, but that's not the important part, or even my grand-reply (reply to my initial reply). Most people don't have the intellectual curiosity and bravery to read such utterly new information, but if you can read drivel from AS, you can make through one (rather long) page of mine.
I triple-dog dare ya ;-)
And remember, ignorance of the truth is a human vice that we must fight and defeat, in order to choose the better path, the Path of Love. Giving in to ignorance is a choice between good and evil, my friend. I hope you choose well, but you'll likely choose to rebel against the truth, and instead keep believing the lies that have been told to you, which is our body's monkey-inheritance.
Happy choosing! I truly wish you all success and happiness in this world, but that latter one is dependent upon our learning and manifesting the truth, my friend.
I continue to be dismayed by AI's quixotic focus on "human" intelligence. AI is very far from pigeon-level intelligence or dog-level intelligence. I strongly suspect transformers are dumber than spiders.[1] This focus on human intelligence via formal human knowledge is putting the cart before the horse. If your "human-level" AI architecture cannot conceivably be modified for chimp intelligence, and requires bootstrapping with a bunch of pre-processed human knowledge, then it is not actually emulating human intelligence. LLMs are fancy encyclopedias, not primitive brains.
[1] Suppose you have an accurate web-spinning simulator and you train a transformer ANN on 40 million years of natural spiderweb construction: between trees, rocks, etc. This AI is excellent at spinning natural webs. Would the transformer be able to spin a functional web in your pantry or basement. If not, then the AI isn't as smart as a spider. I don't think this thought experiment is actually possible: any computer simulation would excessively simplify the physical complexity. But based on transformers' pattern of failures in other domains, I don't think they are good enough to pull it off.
I’m not sure the emphasis is quixotic. I would prefer to say myopic, but the myopia is understandable: human intelligence is the only intelligence we have any firsthand experience with.
One of the things I love about computer science is that it forces us to devise definitions—working definitions at least—so that we can move forward. What is intelligence? Turns out we don’t really know. What we do know about are TASKS or PROBLEMS, and that certain kinds of machines are more or less suited for certain problems. A human intelligence is one that can solve a large range of problems, but often only with training. Is the human mind a mechanism that we can replicate or is it something more? We don’t know. Are there other kinds of intelligence? Is intelligence just a matter of definition? We don’t know.
Personally, I think it is an exciting time to be alive, because these questions are no longer merely philosophical. And we finally have the ability to start answering them scientifically.
I tend to agree, what is being lost in most recent discussions on Human Level AI, is the focus on Language and LLM's.
There are plenty of other researchers and companies working on non-LLM models. And there you start getting a little more 'human' like reasoning.
Maybe eventually the LLM will just be the 'human interface' to a different AI model underneath, that does have a model of the world, and goals that it is charting through the real-world complexity and not just a game simulation.
I also find it quite weird, I think the fact that the current cycle is focused on chat bots fools a lot of people into anthropomorphising LLMs and perceiving them as better than they actually are. It's very comforting in a world with increasingly less meaningful social interaction
Sure can. Whole world’s done this every time there’s a major advance in algorithms. We do this with other major advances, too, like how the industrial revolution was going to usher in utopia and GMO was going to end world hunger. Whenever we can’t see the end of something, only half the world figures it’s a vision problem, while the other half figures the end must not exist.
Present-day developed world must look a lot like utopia for a medieval serf. And we're now feeding 8 billion as opposed to, what, 2 billion before the green revolution. The exaggerations are only slight.
>models... AGI. ... unlikely to reach this milestone on their own.
I don't think there is a single milestone. Intelligence has many aspects - IQ test type ability, chess playing ability, emotional intelligence, ability to go down to the shops and buying something and so on.
AI is making gradual progress and is very good at some things like chess and very bad at others. There will probably be a gradual passing of different milestones on different dates. When they can replace a plumber including figuring out the problem and getting and fitting the parts might be a sign they can do most stuff.
There's probably a way to go there but there are a lot of resources being thrown at the problem just now.
There is zero reasoning in it so far, everything up to today is perfectly explainable with advanced statistics and NLP. Its large _language_ models after all, no matter the hype.
Still I find it excellent when exploring new knowledge domains or cross-comparing cross knowledge domains, since LLMs by design (and training corpus) will spill out highly probable terms/concepts matching my questions and phrase it nicely. Search on steroids when you will, where also real-time doesn't matter for me at all.
This is not intelligence, yet hugely valuable if used right. And I am sure because of this, a lot of scientific discoveries will be made with todays LLMs used in creative ways, since most of scientific discoveries is ultimately looking at X within a setting at Y, and there are a lot of potential X and Y combinations.
I am exaggerating a bit, but at some point (niels bohr?) had the thought of thinking about atoms like we do about planets, with stuff circling each other. Its an X but in Y situation. First come up with such a scenario (or: an automated way to combine lots of X and Y cleverly) and then filter the results for something that actually would make sense, and then dig deeper in a semi-automatic way with actual human in the loop at least.
You must be using a peculiar definition of expert because even generally conservative AI experts like LeCun now expect we could have human-level intelligence within 5 to 10 years.
There are some comments that, even if I was blindfolded and only had to listen to them through a loudspeaker, I would immediately recognize as coming from some HN commenter who thinks they're being really smart.
This is one of those such comments.
It begs the question: have you actually used any of these "AI"s, or are just reading about them on HN?
Kinda like an idiot savant that can keep up with Terrence Tao at math, yet fail to play Tic Tac Toe or help a farmer plan a river crossing with his chicken.
I can't believe this article is being published in Nature. The article is flawed, plagued with assumptions that I guess the author doesn't even notice (like what do we really mean by AGI, the epistemological problems/assumptions to intelligence, the real nature of thinking, the real functioning of the human brain).
It is really curious that the philosophical community is addressing the debate on what AI really is and its implications, but the computer science community does not read almost anything about philosophy.
Regarding the fear of 'losing control of it', I would suggest reading the works (or at least about) of Gunther Anders and Bernard Stiegler. Technology (in this case AI) is inseparable from human being, to the point that we already lost control of technology, its use and its meaning (like, 100 years ago).
Another thing that surprises me is how the computer science community is blind to the work of Hubert Dreyfus and other contemporary philosophers that analyze AI from and epistemological and philosophical perspective. But, actually, I should no t be surprised: we barely study philosophy in any scientific discipline when attending university.
This rhetoric about how AI is similar to the human brain is starting to be a bit boring. It assumes a very simplistic view on the brain and turns a deaf ear to other types of research (like language acquisition and embodiment, mind/brain duality, epistemological basis for knowledge acquisition, ontological basis of causal reasoning...).
And above all, what is really upsetting is the techno-optimism behind this way of thinking.
This is not a scientific paper that was published in nature by researchers, it is a news editorial written by an editor/journalist. Don't get fooled by the domain.
I know it is an article, and not a scientific publication, but that does not change the fact that the article is not serious at all regarding the ongoing discussion on AI.
If this gets published in Nature, even as an opinion article, is because there is a general ideology that can actually produce this kind of content.
> “Bad things could happen because of either the misuse of AI or because we lose control of it,” says Yoshua Bengio, a deep-learning researcher at the University of Montreal, Canada.
God, I hate phrases like this. We've already lost control of it. We don't have any control. AI will evolve in the rich medium of capitalism and be used by anyone due to its ease of use and even laws will be unable to restrict that. At this point, since we've set up a system that promotes technologies regardless of their long-term cost or dangers, we simply cannot control them. Bad things are already happening and human beings are being integrated into a matrix of technology whose ultimate purposes is just the furthering of technology.
Even people like Dr. Bengio are just pawns in a system, whose purpose is just to present an artificially balanced viewpoint as if there were a reasonable set of pros and cons, designed to make people think that we could "lose control" but with the right thinking, we don't have to let that happen. I mean come on, just suppose for a second the hypothesis of "AI is already out of control". If Dr. Bengio and their colleagues acknowledged that, then they'd be out of a job. So just by evolutionary pressure on "organizations that monitor AI", they have to be artificially balanced.
As a thought exercise, on one hand the technology's purpose is having no purpose at all, just doing what it was made to do. So I'd rather focus on the technology people, where some develop it for the sake of developing, and others want to extract most money possible out of it. Nothing surprising because capitalism, but now there's a real possibility that the AI owner(s) will extract all the money there is. And I have no idea how our society will function then - until now rulers always needed subjects... We try building guardrails in laws and regulations exactly because we don't know where all this can lead, but as even today humans find ways to circumvent laws I expect the AI finding ways around its guardrails as well (with human help for sure). Thus sooner or later we will unavoidably come to that point where we have no idea what's gonna happen, and that's when usually unrest peaks.
I disagree with your first statement. Technology has a purpose, and that is to evolve itself. The ancient greeks had this perspective, so did several philosophers, and I agree with them.
We are not in control either of the nukeclear power which is available for quite a few nations, since a lot of decades now, as well. Soo... c'mon cheer up most probably is some kind of simulation anyway.
Thank you for this. I'm glad someone has their head on straight.
I think it's like some rich guys said, "Lets roll this tiny snowball down this mountain towards that village."
Of course, we are also heating an already overheating Earth with these mostly useless things "just to see" (line from Gibson's Neuromancer, spoken by the psychopath).
> We've already lost control of it. We don't have any control.
I think it's important to be clear on what's being said here:
There's a very big difference between "we, the people, do not have control over what Silicon Valley businesses and billionaires are doing with AI" and "there are AIs out there that no human has any control over".
The former is current reality. The latter is sci-fi, and nothing yet has demonstrated that it is actually possible.
The former is also precisely what I would describe as "the misuse of AI".
You are right and I do not think anyone making AI has much control over it either because too many people who make it are addicted to the process in a very similar way in which drug users are addicted to drugs. And also, the prisoner's dilemma solution is almost forcing a lot of people to control it.
So I really do believe that NO human has control over it, really.
I don't understand how culture is being used here. What does "more culture" mean? A culture is a process and a system of relations I can understand how wikipedia has a culture but not how a shadow library has "more culture than you need". Are we talking about the products of cultural production? To suggest that LLMs are anywhere near human level is laughable.
Is there a distinction between LLM’s and AI, or do we consider LLM’s to exhibit intellect?
I remeber Sam Altman pointing out in some interview that he considers GPT to be a reasoning machine. I suppose that if you consider what GPT does to be resoning, then calling it AI is not so far fetched.
I feel it’s more like pattern recognition though rather than reasoning, since there’s no black box ”reasoning” component in an LLM.
I've been annoyed by the redefinition of artificial intelligence since the LLM boom started. The term AI has no place being used to describe LLMs as far as I can tell, unless what goes on inside the black box of an LLM is drastically different than how they are described to function.
Predicting the next token based on a compressed dataset of human generated content isn't intelligence in any meaningful definition of the word. That doesn't mean LLMs aren't impressive or useful for certain tasks, but they aren't intelligent.
When Altman describes them as reasoning machines he's either lying (likely for marketing purposes) or using a different definition of "reasoning" than most people would. The latest release of GPT is attempting to mimic reasoning, but what they're actually doing is having one system act as an automated prompt engineer in between the GPT model and the end user.
> I've been annoyed by the redefinition of artificial intelligence since the LLM boom started
If there's any redefinition, it's being pushed further out. AI was previously used to describe far simpler systems, like expert systems and Deep Blue's alpha–beta search.
> Predicting the next token based on a compressed dataset of human generated content isn't intelligence in any meaningful definition of the word
I'd claim generating the next token is a sufficiently general task such that success can depend on essentially arbitrary intellectual capabilities. For instance, reliably completing unseen equations like `2335 + 4612 = ` requires ability to perform basic arithmetic.
> using a different definition of "reasoning" than most people would. The latest release of GPT is attempting to mimic reasoning
I think most people initially have some relatively solid definitions of "learning", "reasoning", "language use", etc. similar to how it's being used there - just that when non-humans meet those definitions there's an inclination to create some distinction between "learning" and an elusive "actual learning".
For instance, if something changes to refine its future behavior in response to its experiences (touch hot stove, get hurt, avoid in future) beyond the immediate/direct effect (withdrawing hand) then it can "learn". I think even small microorganisms can learn, with the main requirement being that it has some mutable state (can't learn if you can't change). Yet, others will object that "machine learning" is a misnomer because it's "not actual learning" and instead "just mimicking/simulating".
For to define "reasoning", you have to deal with (at least) the following sub-questions:
1. What is knowledge?
2. How can knowledge be encoded in a machine?
LLMs say that knowledge is encoded in the relationships between words (and, in fact, has been by the corpus of human writing), and that's enough. Expert systems said that knowledge could be encoded in carefully-written rules, and that's enough.
I'm pretty sure that any actually intelligent[1] computer is going to have to have more than one flavor of knowledge representation, and be able to shift between them as the situation warrants.
[1] Whatever "actually intelligent" may mean. I don't have to know what it is, though, to recognize that what we have so far is inadequate.
I'd say reasoning is the process of applying logic to draw inferences from some information/axioms/assumptions. For instance if you're asked "can a fridge fit in a bread-box?" and (implicitly or explicitly) go through:
1. A fridge is much larger than a bread-box
2. Larger objects cannot fit inside smaller objects without flexibility
3. Neither objects are sufficiently flexible
4. Therefore, a fridge cannot fit in a bread-box
Then I'd be happy saying you have used reasoning to reach your answer.
> How can knowledge be encoded in a machine? [...] LLMs say that knowledge is encoded in the relationships between words [...]
I don't think it'd be fully correct to say that knowledge is only encoded by relations between words. The input/output of the model is tokens of text, but internally it'll be converted into high-dimensional semantic vector spaces of concepts.
Different words describing the same concept ("Bread-Box", "breadbin", ...), or even images in the case of multi-modal models, can be associated with the internal representation of a bread-box, from which useful semantic manipulations/inferences can be made about the concept and not just the word used to reference it (like approximating the bread-box's size, a factor potentially learned from images but applied to answer a textual question).
> I don't think it'd be fully correct to say that knowledge is only encoded by relations between words. The input/output of the model is tokens of text, but internally it'll be converted into high-dimensional semantic vector spaces of concepts.
All right, how about this: LLMs do have actual knowledge - the knowledge that was encoded in the words in the training data. That's not how they store the data internally, but the actual knowledge comes from there.
And I wasn't saying that that's enough. I was saying that the LLM advocates think, or at least claim, that it's enough.
The term "artificial intelligence" is still used, quite correctly, to refer to fully-deterministic algorithms controlling NPCs in video games of all types.
The field of "artificial intelligence" still has "machine learning" (of which LLMs are a product) as part of it.
The problem is not, and has never been, that the term "AI" was used incorrectly to describe LLMs. It's that people (like Altman) who almost certainly do know better started making marketing claims conflating them with "AGI" (aka "strong AI"), and pushing them as being genuinely "alive" and reasoning.
Most of us in the tech field, and a lot of people outside of it (eg, most gamers) fully recognize that "AI" does not automatically mean Skynet. It takes active, deceptive work on the part of the people selling these systems to prime them to make that leap.
Remember that fools can either be full of horsesh_t, where they don't know that what they believe and repeat is untrue, or bullsh_t, where they know they are lying and doing so for a particular reason.
The first step to being a Dune-style truthsayer is to never lie. The deeper truth to that path (which is possible, but rarely travelled) is that it is possible, but we must purposefully seek ever deeper truths about truth and humanity.
Our world's lack of this deep honesty, first about oneself and then about others, is a major source of our systemic problems. Another major source is selfishness, but I've discussed that elsewhere.
Regardless, most people just love hearing the words flow out of their own mouths, and that tendency seems to be worse for successful tech guys or anyone with a bit of money or with self-righteous fake-religion guys.
Do you consider a basic algorithm to be artificial intelligence?
You are right though, you van go back further than LLMs and find misuses of the term "artificial intelligence." That doesn't contradict my main point though, that the word has been so redefined as to be pretty meaningless to the understanding of what intelligence is.
If we want to consider even basic algorithms to be intelligence, are we boiling down the entire concept of intelligence to mathematical equations?
If it's been the way the field has used it for decades, it's not really a misuse.
> that the word has been so redefined
It's not been redefined though, other than people now wanting to moan about PR and things not being "real" AI when we've had AGI as a term to use right there.
> If we want to consider even basic algorithms to be intelligence, are we boiling down the entire concept of intelligence to mathematical equations?
Massive side argument, but I think we obey physical laws and are not magical and so fundamentally I can't see another answer.
Not really - machine learning, whether SVMs or ANNs, was called just that until relatively recently when the popular press started to first call ANNs AI, then LLMs. At first there was pushback from ML researchers, but particularly with LLMs they are now embracing it since investors want to invest in "AI".
LLMs are really just fancy (deep) pattern recognizers/predictors, conceptually not so different than rule-based expert systems like CYC, which was never called AI. Of course LLMs learn their own rules, which is extremely useful.
Other than the pop press wanting to talk about futuristic AI, and investors wanting to invest in it, what also provides cover for LLMs as "AI", is that they are trained to predict/copy human training data, and so appear as smart/dumb as that is, even if they are really no smarter than Searle's Chinese room.
> machine learning, whether SVMs or ANNs, was called just that until relatively recently when the popular press started to first call ANNs AI,
That is absolutely not the case. These things have been in the field of AI for decades. Frighteningly it's nearing two decades since I started my degree in AI and it wasn't a new reference then.
I remember taking Andrew Ng's Coursera ML course (incl. neural nets and SVMs) when it came out in 2011, and nobody, including him, was calling it AI at that time. I think it was sometime after neural nets really took off after ImageNet 2012 that the press started to call everything AI.
There is no black box 'reasoning' component in humans either.
I will grant you that humans are far more intelligent, and after spending thousands of hours playing with LLMs, it's hard not to see their limitations. At the same time... they're dumb like a very dumb person who has (implausibly) read the Library of Congress, not like a rock or a computer.
I often use Claude to write short stories, largely just for fun. Certainly, its skill at English vastly outmatches its skill at reasoning. It doesn't write well, but it regularly produces turns of phrase that makes me laugh; meanwhile, it needs hand-holding to successfully handle situations with asymmetrical knowledge. It's bad at theory of mind.
But it's just bad, in more or less the same way that a two-year-old is bad at it.
Not the best reasoner in the world. It would be false to claim it's as smart as the typical seven-year-old...
It's almost as wrong to claim that it can't reason at all.
It definitely feels like reasoning. Problems get solved. They may be simple problems, but it's still far beyond what a calculator can do.
Does it really matter if it's "just wordplay"? I'm not convinced humans are any different, beyond the sheer scale. I certainly don't believe we have a 'reasoning module'.
That story goes way deeper than some wordplay fooling people. The entire intent was to get people to realize that it was worthless, but, even after people learned that and what it was, they clamoured for more!
"Just imagine how stupid the average person is, and them remember that half of them are even stupider than that!" --George Carlin
And, yes, we humans are very different, but you'll have to traverse my recent comment history to get the extensive explanation. It's worth it, though, I promise, but I doubt you'll like it or agree with it. Good luck!
You raise good points, I agree it feels like it is reasoning at times.
Though the brain, with our current understanding of it, is by far more of a black box to us than any LLM.
> I certainly don't believe we have a 'reasoning module'.
Let’s also point out that human brains probably don’t have any vector databases in them either.
It seems to me like our brains must work very differently - just look at how much energy an LLM consumes compared to our brains consuming around 12 Watts.
I remember expert systems being considered AI, so LLMs ought to meet that bar as well. They aren't AGI, though, which is a higher one, I guess. I'm not in love with the various terms and the various ways people define them. Even LLM --at what point is it "large"? In a rapidly changing area of both academic and lay understanding, it's understandable for terminology to be a bit unstable. I don't think it's reasonable to say LLMs do reasoning, however. Even when mimicking incredible feats of intelligence, they don't have a grasp of what is true or how truth flows from a set of facts to any other.
General intelligence is a strategy to defer the acquisition of abilities from the process of construction/blueprinting (ie., genes, evolution..) to the living environment of the animal. The most generally intelligent animals are those that have nearly all of their sensory motor skills acquired during their life -- we learn to walk and so can learn to play the piano, and to build a rocket.
There is a serious discontinuity in strategy to achive this defferal: the kinds of processes which "blueprint" the intelligence of a bacterium are discontinuous with the processes which a living animal needs to dynamically conceptualise its environment under shifts to its structure.
Of the latter animals need: living adaption of their sensory-motor systems, heirachical coordination of their bodies, robust causal modelling, and so on.
General intelligence is primitively a kind of movement, which becomes abstract only with a few hundred thousand years of culture. The earliest humans, able to lingusitically express almost nothing, were nevertheless generally intelligent.
Present computer-science-led investigations into "intelligence" assume you can operate syntactically across the most peripheral consequences of general intelligence given by linguistic representations. This is profoundly misguided: each todller necessarily must learn to walk. You cannot just project a slideshow of walking, and get anywhere. And if you remove this capability and install a "walking module", you've remved the very capabilities which allow that child then to do anything new at all.
There is nothing in the linguistic syntactical shadow of human intelligence to be found in creating generally capable systems. It's just overfitting to our 2024 reflections.