Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

What Is AGI-Hard

lspace.swyx.io

Post reply on HN
Print view View original

41–50 of 129 posts

Re: What Is AGI-Hard

#41

Post by zone411 »

tuyiown wrote: post #38
MichaelRazum wrote: post #33

Isn't AGI-Hard everything that seems rather simple for animals? I would think just of robotics and navigation in unknown environment. Everything else like Chess, Text, Art, Games seems to be solvable by current AI systems. Robotics seems much much more difficult.

Yes, the problem is to replicate hundred of millions years of acquisition of heuristics that allows cognition on «natural» environment. Those heuristics defines all bricks our cognition is based upon, think of basics things like objects permanence, we just cannot reason without such ability. To me, as long as no AI manage to replicate those bricks, at least some of them, it won't be able to interact with us in ways t…

I predict that this will a research area that advances hugely after we solve text-to-video and text-to-3D (so maybe in 2 years?): learning these types of heuristics from billions of hours of videos. Implementing this in robots is another matter, but there has been some good progress in robotic grasping and manipulation, for example.

Re: What Is AGI-Hard

#42

Post by albertzeyer »

As LeCun always says: The term "AGI" is a bit ill-formed. What does "general" really mean? Human intelligence is not really general. It's actually very very specialized towards the world we live in. So he prefers the term "human-level intelligence", or maybe super-human-level at some point.

Another implication: It's not really a yes/no property, whether you have AGI or not. It's a level of intelligence. And this level is continuous. And it's also not really one-dimensional.

What does human-level intelligence mean? It's still not very well-defined. You would need to define a number of tests, which measure the performance, and the model need to perform at least as good as a human on average.

So, I assume this is what the author means by "AGI-hard": A problem which requires some model to be at least as good as a human in a wide number of tests.

But I don't think this is necessary for driving. You want it actually to be much better in some tests (computer vision, reaction time, etc) and many other tests don't really matter (eg. logical reasoning, speech recognition, or so). So the autonomous driving problem is not AGI-hard.

Re: What Is AGI-Hard

#43

Post by jillesvangurp »

dmitriid wrote: post #35
jillesvangurp wrote: post #24

Earlier quoted context omitted.

Exactly the fallacy with people's expectations of AI is that it has to have a perfect track record of decision making. We humans actually fail that test and quite miserably so. Take self driving cars as an example. Self driving cars: barely any fatal accidents but some erratic behavior. People go "oh we can't have that because something extremely unlikely might kill somebody. Never mind it hasn't happened yet and mig…

> Would you rather drive on a road where every other driver is an AI or on a road where the average driver is a fair selection of the current drivers? Thing is, if the roads are made so that only "AI" cars can drive on them, you don't need AGI, or an advanced AI. You basically need the cars at the level they are now (or even from five years ago). Because the number of weird shit you need to deal with is reduced signi…

That's just right now. Changing the roads will take longer and be way more expensive than building enough AI that cars can self drive on existing roads safely. Arguably, we're nearly there already. Several companies have self driving cars on some of the roads (not even counting Tesla here) and the scope of situations where these can operate safely is expanding. They aren't fixing any roads for this. This is all done in software. Five years is nothing for road construction. There's only so much that happens in such a short time span. However, it's a lot of time for making some software improvements.

Re: What Is AGI-Hard

#44

Post by VCBullshitter »

swyx wrote: post #17
janalsncm wrote: post #6

This article is a bit too hand-wavy for my taste. As a buzzword, AGI Hard sounds cool but until intelligence is more strictly defined (and the modifiers “artificial” and “general” aren’t helping) AGI will always be something we talk about rather than something known. I was also a bit amused at the list of things which the author also claims to be AGI hard. But this is just a list of things we think are difficult. Phy…

chess as a closed system is definitely solved by AI, including bootstrapping purely from self-play. the claim here is that applying physics/math/etc would be AGI-hard, because it involves the AI perfectly modeling a world it doesnt live in (the simulation problem) or understanding relationships in concepts without making jumps (obeying logic).

Isn't this just an echo of computationally? Why involve the distraction of intelligence at all when what you're discussing is a computable algorithm with known bounds? Is there anything special about intelligence—Penrose's bullshit[0] aside—that can't be computed? If not is it anything other than a marketing term for computation that seems more "advanced" to humans?

[0]: https://en.wikipedia.org/wiki/The_Emperor%27s_New_Mind

Re: What Is AGI-Hard

#46

Post by AstralStorm »

VCBullshitter wrote: post #44
swyx wrote: post #17

Earlier quoted context omitted.

chess as a closed system is definitely solved by AI, including bootstrapping purely from self-play. the claim here is that applying physics/math/etc would be AGI-hard, because it involves the AI perfectly modeling a world it doesnt live in (the simulation problem) or understanding relationships in concepts without making jumps (obeying logic).

Isn't this just an echo of computationally? Why involve the distraction of intelligence at all when what you're discussing is a computable algorithm with known bounds? Is there anything special about intelligence—Penrose's bullshit[0] aside—that can't be computed? If not is it anything other than a marketing term for computation that seems more "advanced" to humans? [0]: https://en.wikipedia.org/wiki/The_Emperor%27s_…

Yes, intelligence is the ability for a small finite system to deal with infinity of possibilities in a successful way.

No matter how big computationally the AI is, the test was always trying to deal with something new, outside its comfort zone. Preferably as far outside it as possible. To do that successfully requires general problem solving and abstraction skills.

Of course one can do better or worse at that, including humans.

You cannot precompute most of the answer to a truly open question.

Re: What Is AGI-Hard

#47

Post by mtlmtlmtlmtl »

swyx wrote: post #17
janalsncm wrote: post #6

This article is a bit too hand-wavy for my taste. As a buzzword, AGI Hard sounds cool but until intelligence is more strictly defined (and the modifiers “artificial” and “general” aren’t helping) AGI will always be something we talk about rather than something known. I was also a bit amused at the list of things which the author also claims to be AGI hard. But this is just a list of things we think are difficult. Phy…

chess as a closed system is definitely solved by AI, including bootstrapping purely from self-play. the claim here is that applying physics/math/etc would be AGI-hard, because it involves the AI perfectly modeling a world it doesnt live in (the simulation problem) or understanding relationships in concepts without making jumps (obeying logic).

Small nitpick: in game theory, "solved" means something very specific. At the very weakest it requires the ability to prove what the result of a game will be from the initial position given perfect play.

Chess AI is nowhere close to this. They easily beat humans, yes, but in the game-theoretic sense chess is not solved.

Re: What Is AGI-Hard

#48

Post by fnfontana »

albertzeyer wrote: post #42

As LeCun always says: The term "AGI" is a bit ill-formed. What does "general" really mean? Human intelligence is not really general. It's actually very very specialized towards the world we live in. So he prefers the term "human-level intelligence", or maybe super-human-level at some point. Another implication: It's not really a yes/no property, whether you have AGI or not. It's a level of intelligence. And this leve…

Indeed. A person can be bad at math, but really good on play musical instruments or to guide himself through a unknown city. That requires a certain level of intelligence. But I think there's no human that is intelligent on every aspect, maybe some exceptions like Da Vinci, but most people are good at some tasks while other people doesn't. The human intelligence seems to be distributed throughout the population. This creates an illusion that human intelligence is general.

Re: What Is AGI-Hard

#49

Post by fartsucker69 »

albertzeyer wrote: post #42

As LeCun always says: The term "AGI" is a bit ill-formed. What does "general" really mean? Human intelligence is not really general. It's actually very very specialized towards the world we live in. So he prefers the term "human-level intelligence", or maybe super-human-level at some point. Another implication: It's not really a yes/no property, whether you have AGI or not. It's a level of intelligence. And this leve…

It's totally general in the sense that we can logically approach arbitrary problem types and there is nothing known that we can't approach with our tools of intelligence, so to speak.

That doesn't necessarily mean that we are actually good at solving those problems or that there isn't an upper level of difficulty that we will just not be able to solve practically, but we are definitely able to approach all classes of problems in general.

By contrast, a chess ai can't do anything but solve for chess states. A diffusion NN can't do anything but create images, or something that is mathematically analogous.

We can in principle do anything any conceivable specialized AI can do, even if a lot slower or error prone. Therefor, we have general intelligence.

Re: What Is AGI-Hard

#50

Post by TacticalCoder »

albertzeyer wrote: post #42

As LeCun always says: The term "AGI" is a bit ill-formed. What does "general" really mean? Human intelligence is not really general. It's actually very very specialized towards the world we live in. So he prefers the term "human-level intelligence", or maybe super-human-level at some point. Another implication: It's not really a yes/no property, whether you have AGI or not. It's a level of intelligence. And this leve…

> What does human-level intelligence mean?

What about a simple test: it means being able to create better versions of ChatGPT and StableDiffusion.

Can an AI create a better version of itself, all by itself?

Because humans are going to create better versions of ChatGPT.

So no matter the exact definition of "intelligence" or "human intelligence", the goal is simple: create a better version of itself and it'll be considered intelligent.

Can't do that? It's simply not matching human-level intelligence.

So I don't see how intelligent machines wouldn't lead to "intelligence explosion" and the "technological singularity".

Post reply on HN