Don’t believe the hype: AGI is far from inevitable
Don’t believe the hype: AGI is far from inevitable
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.
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.
You're viewing a single thread.
This is a silly argument:
[..] But even if we give the AGI-engineer every advantage, every benefit of the doubt, there is no conceivable method of achieving what big tech companies promise.’
That’s because cognition, or the ability to observe, learn and gain new insight, is incredibly hard to replicate through AI on the scale that it occurs in the human brain. ‘If you have a conversation with someone, you might recall something you said fifteen minutes before. Or a year before. Or that someone else explained to you half your life ago. Any such knowledge might be crucial to advancing the conversation you’re having. People do that seamlessly’, explains van Rooij.
‘There will never be enough computing power to create AGI using machine learning that can do the same, because we’d run out of natural resources long before we'd even get close,’ Olivia Guest adds.
That's as shortsighted as the "I think there is a world market for maybe five computers" quote, or the worry that NYC would be buried under mountains of horse poop before cars were invented. Maybe transformers aren't the path to AGI, but there's no reason to think we can't achieve it in general unless you're religious.
EDIT: From the paper:
The remainder of this paper will be an argument in ‘two acts’. In ACT 1: Releasing the Grip, we present a formalisation of the currently dominant approach to AI-as-engineering that claims that AGI is both inevitable and around the corner. We do this by introducing a thought experiment in which a fictive AI engineer, Dr. Ingenia, tries to construct an AGI under ideal conditions. For instance, Dr. Ingenia has perfect data, sampled from the true distribution, and they also have access to any conceivable ML method—including presently popular ‘deep learning’ based on artificial neural networks (ANNs) and any possible future methods—to train an algorithm (“an AI”). We then present a formal proof that the problem that Dr. Ingenia sets out to solve is intractable (formally, NP-hard; i.e. possible in principle but provably infeasible; see Section “Ingenia Theorem”). We also unpack how and why our proof is reconcilable with the apparent success of AI-as-engineering and show that the approach is a theoretical dead-end for cognitive science. In “ACT 2: Reclaiming the AI Vertex”, we explain how the original enthusiasm for using computers to understand the mind reflected many genuine benefits of AI for cognitive science, but also a fatal mistake. We conclude with ways in which ‘AI’ can be reclaimed for theory-building in cognitive science without falling into historical and present-day traps.
That's a silly argument. It sets up a strawman and knocks it down. Just because you create a model and prove something in it, doesn't mean it has any relationship to the real world.
This is a gross misrepresentation of the study.
That's as shortsighted as the "I think there is a world market for maybe five computers" quote, or the worry that NYC would be buried under mountains of horse poop before cars were invented.
That's not their argument. They're saying that they can prove that machine learning cannot lead to AGI in the foreseeable future.
Maybe transformers aren't the path to AGI, but there's no reason to think we can't achieve it in general unless you're religious.
They're not talking about achieving it in general, they only claim that no known techniques can bring it about in the near future, as the AI-hype people claim. Again, they prove this.
That's a silly argument. It sets up a strawman and knocks it down. Just because you create a model and prove something in it, doesn't mean it has any relationship to the real world.
That's not what they did. They provided an extremely optimistic scenario in which someone creates an AGI through known methods (e.g. they have a computer with limitless memory, they have infinite and perfect training data, they can sample without any bias, current techniques can eventually create AGI, an AGI would only have to be slightly better than random chance but not perfect, etc...), and then present a computational proof that shows that this is in contradiction with other logical proofs.
Basically, if you can train an AGI through currently known methods, then you have an algorithm that can solve the Perfect-vs-Chance problem in polynomial time. There's a technical explanation in the paper that I'm not going to try and rehash since it's been too long since I worked on computational proofs, but it seems to check out. But this is a contradiction, as we have proof, hard mathematical proof, that such an algorithm cannot exist and must be non-polynomial or NP-Hard. Therefore, AI-learning for an AGI must also be NP-Hard. And because every known AI learning method is tractable, it cannor possibly lead to AGI. It's not a strawman, it's a hard proof of why it's impossible, like proving that pi has infinite decimals or something.
Ergo, anyone who claims that AGI is around the corner either means "a good AI that can demonstrate some but not all human behaviour" or is bullshitting. We literally could burn up the entire planet for fuel to train an AI and we'd still not end up with an AGI. We need some other breakthrough, e.g. significant advancements in quantum computing perhaps, to even hope at beginning work on an AGI. And again, the authors don't offer a thought experiment, they provide a computational proof for this.
Hey! Just asking you because I'm not sure where else to direct this energy at the moment.
I spent a while trying to understand the argument this paper was making, and for the most part I think I've got it. But there's a kind of obvious, knee-jerk rebuttal to throw at it, seen elsewhere under this post, even:
If producing an AGI is intractable, why does the human meat-brain exist?
Evolution "may be thought of" as a process that samples a distribution of situation-behaviors, though that distribution is entirely abstract. And the decision process for whether the "AI" it produces matches this distribution of successful behaviors is yada yada darwinism. The answer we care about, because this is the inspiration I imagine AI engineers took from evolution in the first place, is whether evolution can (not inevitably, just can) produce an AGI (us) in reasonable time (it did).
The question is, where does this line of thinking fail?
Going by the proof, it should either be:
- That evolution is an intractable method. 60 million years is a long time, but it still feels quite short for this answer.
- Something about it doesn't fit within this computational paradigm. That is, I'm stretching the definition.
- The language "no better than chance" for option 2 is actually more significant than I'm thinking. Evolution is all chance. But is our existence really just extreme luck? I know that it is, but this answer is really unsatisfying.
I'm not sure how to formalize any of this, though.
The thought that we could "encode all of biological evolution into a program of at most size K" did made me laugh.
If producing an AGI is intractable, why does the human meat-brain exist?
Ah, but here we have to get pedantic a little bit: producing an AGI through current known methods is intractable.
The human brain is extremely complex and we still don't fully know how it works. We don't know if the way we learn is really analogous to how these AIs learn. We don't really know if the way we think is analogous to how computers "think".
There's also another argument to be made, that an AGI that matches the currently agreed upon definition is impossible. And I mean that in the broadest sense, e.g. humans don't fit the definition either. If that's true, then an AI could perhaps be trained in a tractable amount of time, but this would upend our understanding of human consciousness (perhaps justifyingly so). Maybe we're overestimating how special we are.
And then there's the argument that you already mentioned: it is intractable, but 60 million years, spread over trillions of creatures is long enough. That also suggests that AGI is really hard, and that creating one really isn't "around the corner" as some enthusiasts claim. For any practical AGI we'd have to finish training in maybe a couple years, not millions of years.
And maybe we develop some quantum computing breakthrough that gets us where we need to be. Who knows?
Ah, but here we have to get pedantic a little bit: producing an AGI through current known methods is intractable.
I didn't quite understand this at first. I think I was going to say something about the paper leaving the method ambiguous, thus implicating all methods yet unknown, etc, whatever. But yeah, this divide between solvable and "unsolvable" shifts if we ever break NP-hard and have to define some new NP-super-hard category. This does feel like the piece I was missing. Or a piece, anyway.
e.g. humans don't fit the definition either.
I did think about this, and the only reason I reject it is that "human-like or -level" matches our complexity by definition, and we already have a behavior set for a fairly large n. This doesn't have to mean that we aren't still below some curve, of course, but I do struggle to imagine how our own complexity wouldn't still be too large to solve, AGI or not.
Anyway, the main reason I'm replying again at all is just to make sure I thanked you for getting back to me, haha. This was definitely helpful.
That's a great line of thought. Take an algorithm of "simulate a human brain". Obviously that would break the paper's argument, so you'd have to find why it doesn't apply here to take the paper's claims at face value.
There's a number of major flaws with it:
- Assume the paper is completely true. It's just proved the algorithmic complexity of it, but so what? What if the general case is NP-hard, but not in the case that we care about? That's been true for other problems, why not this one?
- It proves something in a model. So what? Prove that the result applies to the real world
- Replace "human-like" with something trivial like "tree-like". The paper then proves that we'll never achieve tree-like intelligence?
IMO there's also flaws in the argument itself, but those are more relevant
but there's no reason to think we can't achieve it
They provide a reason.
Just because you create a model and prove something in it, doesn't mean it has any relationship to the real world.
What are we science deniers now?