cross-posted from: https://lemmy.ml/post/20858435
Will AI soon surpass the human brain? If you ask employees at OpenAI, Google DeepMind and other large tech companies, it is inevitable. However, researchers at Radboud University and other institutes show new proof that those claims are overblown and unlikely to ever come to fruition. Their findings are published in Computational Brain & Behavior today.
The actual paper is an interesting read. They present an actual computational proof, stating that even if you have essentially infinite memory, a computer that’s a billion times faster than what we have now, perfect training data that you can sample without bias and you’re only aiming for an AGI that performs slightly better than chance, it’s still completely infeasible to do within the next few millenia. Ergo, it’s definitely not “right around the corner”. We’re lightyears off still.
They prove this by proving that if you could train an AI in a tractable amount of time, you would have proven P=NP. And thus, training an AI is NP-hard. Given the minimum data that needs to be learned to be better than chance, this results in a ridiculously long training time well beyond the realm of what’s even remotely feasible. And that’s provided you don’t even have to deal with all the constraints that exist in the real world.
We perhaps need some breakthrough in quantum computing in order to get closer. That is not to say that AI won’t improve or anything, it’ll get a bit better. But there is a computationally proven ceiling here, and breaking through that is exceptionally hard.
It also raises (imo) the question of whether or not we can truly consider humans to have general intelligence or not. Perhaps we’re not as smart as we think we are either.
The paper’s scope is to prove that AI cannot feasibly be trained, using training data and learning algorithms, into something that approximates human cognition.
The limits of that finding are important here: it’s not that creating an AGI is impossible, it’s just that however it will be made, it will need to be made some other way, not by training alone.
Our squishy brains (or perhaps more accurately, our nervous systems contained within a biochemical organism influenced by a microbiome) arose out of evolutionary selection algorithms, so general intelligence is clearly possible.
So it may still be the case that AGI via computation alone is possible, and that creating such an AGI will not require solution of an NP-hard problem. But this paper closes one potential pathway that many believe is a viable pathway (if the paper’s proof is actually correct, I definitely am not the person to make that evaluation). That doesn’t mean they’ve proven there’s no pathway at all.
Our squishy brains (or perhaps more accurately, our nervous systems contained within a biochemical organism influenced by a microbiome) arose out of evolutionary selection algorithms, so general intelligence is clearly possible.
That’s assuming that we are a general intelligence. I’m actually unsure if that’s even true.
That doesn’t mean they’ve proven there’s no pathway at all.
True, they’ve only calculated it’d take perhaps millions of years. Which might be accurate, I’m not sure to what kind of computer global evolution over trillions of organisms over millions of years adds up to. And yes, perhaps some breakthrough happens, but it’s still very unlikely and definitely not “right around the corner” as the AI-bros claim (and that near-future thing is what the paper set out to disprove).
That’s assuming that we are a general intelligence.
But it’s easy to just define general intelligence as something approximating what humans already do. The paper itself only analyzed whether it was feasible to have a computational system that produces outputs approximately similar to humans, whatever that is.
True, they’ve only calculated it’d take perhaps millions of years.
No, you’re missing my point, at least how I read the paper. They’re saying that the method of using training data to computationally develop a neural network is a conceptual dead end. Throwing more resources at the NP-hard problem isn’t going to solve it.
What they didn’t prove, at least by my reading of this paper, is that achieving general intelligence itself is an NP-hard problem. It’s just that this particular method of inferential training, what they call “AI-by-Learning,” is an NP-hard computational problem.
A breakthrough in quantum computing wouldn’t necessarily help. QC isn’t faster than classical computing in the general case, it just happens to be for a few specific algorithms (e.g. factoring numbers). It’s not impossible that a QC breakthrough might speed up training AI models (although to my knowledge we don’t have any reason to believe that it would) and maybe that’s what you’re referring to, but there’s a widespread misconception that Quantum computers are essentially non-deterministic turing machines that “evaluate all possible states at the same time” which isn’t the case.
I was more hinting at that through conventional computational means we’re just not getting there, and that some completely hypothetical breakthrough somewhere is required. QC is the best guess I have for where it might be but it’s still far-fetched.
But yes, you’re absolutely right that QC in general isn’t a magic bullet here.
the limitation is specifically using the primary machine learning technique, same one all chatbots use at places claiming to pursue agi, which is statistical imitation, is np-hard.
Yeah thought that might be the case! It’s just a thing that a lot of people have misconceptions about so it’s something that I have a bit of a knee jerk reaction to.
AGI is inevitable unless:
-
General intelligence is substrate independent and what the brain does cannot be replicated in silica. However, since both are made of matter, and matter obeys the laws of physics, I see no reason to assume this.
-
We destroy ourselves before we reach AGI.
Other than that, we will keep incrementally improving our technology and it’s only a matter of time untill we get there. May take 5 years, 50 or 500 but it seems pretty inevitable to me.
Another possibility is that humans just aren’t smart enough to figure out AGI. While I’m sure that we will continue incrementally improving technology in some form, it’s not at all self-evident that these improvements will eventually add up to AGI.
I get what you’re saying but to me, that still just sounds like a timescale issue. I can’t think of a scenario where we’ve improved something so much that there’s just absolutely nothing we could improve on further. With AI we only need to reach the point of making it have human-level cognitive capabilities and from there on it can improve itself.
I can’t think of a scenario where we’ve improved something so much that there’s just absolutely nothing we could improve on further.
Progress itself isn’t inevitable. Just because it’s possible doesn’t mean that we’ll get there, because the history of human development shows that societies can and do stall, reverse, etc.
And even if all human societies tends towards progress, it could still hit dead ends and stop there. Conceptually, it’s like climbing a mountain through the algorithm of “if there is a higher elevation near you, go towards that, and avoid stepping downward in elevation.” Eventually that algorithm brings you to a local peak. But the local peak might not be the highest point on the mountain, and while it is theoretically possible to have gotten to the other true peak from the beginning, the person who is insistent on never stepping downward is now stuck. Or, it’s possible to get to the true peak but it requires climbing downward for a time and climbing up past elevations we’ve already been to, on paths we hadn’t been on. One can imagine a society that refuses to step downward, breaking the inevitability of progress.
This paper identifies a specific dead end and advocates against hoping for general AI through computational training. It is, in effect, arguing that even though we can still see plenty of places that are higher elevation than where we are standing, we’re headed towards a dead end, and should climb back down. I suspect that not a lot of the actual climbers will heed that advice.
There are a couple of reasons that might not work:
- Maybe we’ll asymptotically approach a point that is lower than human-level cognitive capabilities
- Gradual improvements are susceptible to getting stuck in a local maxima. This is a problem in evolution as well. A lot of animals could in theory evolve, say, human level intelligence in principle, but to reach that point they’d have to go through a bunch of intermediate steps that lead to worse fitness. Gradual scientific improvements are a bit like evolution in this way.
- We also lose knowledge over time. Something as dramatic as a nuclear war would significantly set back the progress in developing AGI, but something less dramatic might also lead to us forgetting things that we’ve already learned.
To be clear, most of the arguments I’m making aren’t really about AGI specifically but about humanities capability to develop arbitrary in principle feasible technologies in general.
Did you read the article, or the actual research paper? They present a mathematical proof that any hypothetical method of training an AI that produces an algorithm that performs better than random chance could also be used to solve a known intractible problem, which is impossible with all known current methods. This means that any algorithm we can produce that works by training an AI would run in exponential time or worse.
The paper authors point out that this also has severe implications for current AI, too–since the current AI-by-learning method that underpins all LLMs is fundamentally NP-hard and can’t run in polynomial time, “the sample-and-time requirements grow non-polynomially (e.g. exponentially or worse) in n.” They present a thought experiment of an AI that handles a 15-minute conversation, assuming 60 words are spoken per minute (keep in mind the average is roughly 160). The resources this AI would require to process this would be 60*15 = 900. The authors then conclude:
“Now the AI needs to learn to respond appropriately to conversations of this size (and not just to short prompts). Since resource requirements for AI-by-Learning grow exponentially or worse, let us take a simple exponential function O(2n ) as our proxy of the order of magnitude of resources needed as a function of n. 2^900 ∼ 10^270 is already unimaginably larger than the number of atoms in the universe (∼10^81 ). Imagine us sampling this super-astronomical space of possible situations using so-called ‘Big Data’. Even if we grant that billions of trillions (10 21 ) of relevant data samples could be generated (or scraped) and stored, then this is still but a miniscule proportion of the order of magnitude of samples needed to solve the learning problem for even moderate size n.”
That’s why LLMs are a dead end.
because, having coded them myself, I am under no illusions as to their capabilities. They are not magic. “just” some matrix multiplications that generate a probability distribution for the next token, which is then randomly sampled.
@ContrarianTrail @JRepin well I guess somebody would first need to clearly define what “AGI” is. Currently it’s just “whatever the techbro hypers want it to be”.
And then there’s the matter (ha!) of your assumption that we understand all laws of physics necessary that “matter obeys”, or that we can reasonably understand them. That’s a pretty strong assumption: individual human minds are pretty limited and communication adds overhead, and we might reach a point where we’re stuck.
A chess engine is intelligent in one thing: playing chess. That narrow intelligence doesn’t translate to any other skill, even if it’s sometimes superhuman at that one task, like a calculator.
Humans, on the other hand, are generally intelligent. We can perform a variety of cognitive tasks that are unrelated to each other, with our only limitations being the physical ones of our “meat computer.”
Artificial General Intelligence (AGI) is the artificial version of human cognitive capabilities, but without the brain’s limitations. It should be noted that AGI is not synonymous with AI. AGI is a type of AI, but not all AI is generally intelligent. The next step from AGI would be Artificial Super Intelligence (ASI), which would not only be generally intelligent but also superhumanly so. This is what the “AI doomers” are concerned about.
> A chess engine is intelligent in one thing: playing chess
No. That’s not how the adjective “intelligent” works, outside of marketing drivel of course (“intelligent washing machine” etc).
> Artificial General Intelligence (AGI) is the artificial version of human cognitive capabilities
Can you give a definition of “intelligence” or “human cognitive abilities” that would allow us to somehow unequivocably establish that “X is intelligent” or “X has human cognitive abilities”?
@ContrarianTrail @JRepin and finally, there’s a question of whether we actually decide to pursue it.
Nuclear power was supposed to be the “inevitable” power source for all of humanity mere 50 years ago. But at some point we decided not to pursue that goal.
Cryptocurrencies were supposed to be “inevitable” replacement for the banking system.
And we *have* cryptocurrencies and nuclear power. These exist. As opposed to whatever nebulous concept hides beneath “AGI”.
Since they still exist, only time will tell if the promise of nuclear power and/or cryptocurrencies come to be.
AGI and even IMHO AI do not exist. Whatever product is being marketed as AI isn’t what I would consider AI. “AI” can have its uses but I really do not think they will be what people expect because it fundamentally lacks what I would consider crucial aspects of human intelligence.
AI makes for a very good grammar checker. It is good at producing filler content for SEO. And it is good at producing “stuff” that looks like it could be right. Probably will have some uses in creative work since it doesn’t have to be “correct” so as a tool to aid an artist, that’s seems pretty cool - I’m sure that is already happening. It will have its uses and a lot of companies will find out the hard way, it is not that they think. That’s my prediction.
Meh. It’s not a problem of scale. It’s a problem of we have no idea how the fuck to do that. Scaling up existing techniques is neither necessary nor sufficient.
Right on the money. One of the big things with AI safety is “we have no fucking clue how AGI can originate so we are constantly in the dark.” If we ever did create it, we likely would not immediately know it was AGI, and that creation could go very terribly in a number of ways.
Will AI soon surpass the human brain?
If you ask employees at OpenAI, Google DeepMind and other large tech companies, it is inevitable.
That doesn’t answer the question.
If it will happen is unrelated to When it will happen.
I’d expect we’ll see AGI some time between the next 20 and 200 years. I think that’s pretty soon. You may not.
If there were a giant asteroid hurling toward Earth, set to impact sometime in the next 20 to 200 years, I’d say there’s definitely a need for urgency. A true AGI is somewhat of an asteroidal impact in itself.
A single AGI would not be to different from a human. But it may not take long for AGI to develop ASI, superior to human intelligence.
Thats not an astronaut impact but alien contact
I like SCUMM but AGI is okay I just don’t like typing commands
