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

61–70 of 129 posts

Re: What Is AGI-Hard

#61

Post by albertzeyer »

bmitc wrote: post #58
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…

> 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. That’s kind of a silly semantic argument though. It also assumes that everyone interested in AGI is trying to make “human-leve…

I just wanted to stress the point that "general" is misleading (no intelligence will ever be truly general), and the term "AGI" is somewhat unclear. Some people mean human-level intelligence, some people just mean some reasonable intelligent agent in a number of tasks, some people really mean a truly-general intelligence, being able to handle any problem (unlike a human).

In the post, when arguing about "AGI-hard", this is very relevant. For a truly-general intelligence, it makes sense. But I think most people would not understand "AGI" this way. For some more narrow general intelligence, I don't think "AGI-hard" makes sense.

Driving needs some intelligence which is a bit generic but still different to other general intelligences, so it is not AGI-hard.

Re: What Is AGI-Hard

#62

Post by notahacker »

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…

> 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.

If you are also able to prevent people and animals from accessing the roads and fix the weather...

(At the other end of the scale, if you put the cars on rails in tunnels, "AI" from the 1960s will likely suffice for safe driving)

Re: What Is AGI-Hard

#63

Post by jvm___ »

dbspin wrote: post #55
fartsucker69 wrote: post #49

Earlier quoted context omitted.

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…

> there is nothing known that we can't approach with our tools of intelligence, so to speak. Such things may necessarily be unknowable. If there are classes of observations that we are unable to make, or problems that we are unable to comprehend, its absolutely plausible that they would be cognitive blindspots. In fact there's a whole theoretical strand in cognitive science / ontology that suggests this. We perceive…

What's your definition of fitness in this case?

Survival of the fittest?

Re: What Is AGI-Hard

#64

Post by comboy »

bmitc wrote: post #58
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…

> 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. That’s kind of a silly semantic argument though. It also assumes that everyone interested in AGI is trying to make “human-leve…

> So I think AGI is fine unless you’re explicitly trying to mimic human intelligence.

Humans we will only recognize AGI when it can simulate all our flaws. We won't think the computer is able to see properly until it can observe the same optical illusions that we do (which are visual system flaws).

By general we mean being able to completely relate to *our* experience.

We are very biased even when it comes to say math, focusing on certain aspects and representations, which if are easily digestible for our brain, thus must be sign of intelligence. Huge numbers calculations - not intelligence, we cannot do that. Doing 100 problems way better than we do but doing mistake in 1 that is obvious to us - clearly not intelligence.

If it doesn't have the same framework for survival as we created with our neurochemistry - nope, it should be able to understand it. Even though we cannot wrap our heads around an octopus feelings.

Just looking at our understanding of say chimpanzees, shows how close minded we are regarding intelligence in general.

So basically, I agree with all you said, I'm just very convinced that most people will not assign AGI label to anything which doesn't mimic human perception (along with all of its flaws). I mean, maybe not necessarily mimic it, but at least kind of have a virtual machine knowing what how human would perceive it. Otherwise we'll think it doesn't understand the world.

Re: What Is AGI-Hard

#65

Post by dmitriid »

notahacker wrote: post #62
dmitriid wrote: post #35

Earlier quoted context omitted.

> 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…

> 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. If you are also able to prevent people and animals from accessing the roads and fix the weather... (At the other end of the scale, if you put the cars on rails in tunnels, "AI" from the 1960s will likely suffice for safe driving)

> If you are also able to prevent people and animals from accessing the roads and fix the weather.

If all cars are "AI", and move in uniform fashion, and know each others' location, then this might be an easier issue to solve than it is today.

"May" is dangerously close to "just", I know :)

Re: What Is AGI-Hard

#66

Post by chriswarbo »

https://arxiv.org/abs/cs/0605024 gives a formal model of intelligence (achieving high reward in an agent/environment setup, across all computable environments weighted by their Kolmogorov complexity)

https://arxiv.org/abs/1109.5951 gives a computable approximation (dubbed AIQ, Algorithmic Intelligence Quotient), and a reference implementation using Brainfuck programs (personally I would prefer Binary Combinatory Logic or Binary Lambda Calculus ;) )

Re: What Is AGI-Hard

#67

Post by cactusplant7374 »

jillesvangurp wrote: post #24
logicallee wrote: post #13

I hereby endorse ChatGPT as a limited form of AGI. It can complete a variety of novel cognitive tasks (example: judging texts and proposals) often enough that it meets my personal requirements for this. My opinion is that the definitions which others believe disqualify ChatGPT from being considered AGI are not relevant, since they focus on things it fails at rather than what it can do. For example, when it tests as w…

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…

> 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 might never happen. But it just might. Therefore I will never trust AIs!!!!"

Fatal accidents have been rare because of human intervention. The cars aren’t self driving and no one in their right mind will stop paying attention when in one of these cars.

So the “extremely unlikely” argument is very, very weak.

Re: What Is AGI-Hard

#68

Post by tgv »

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.

Chess and other games are solved, because they use well-defined rules, and have a combinatorial explosion that we cannot comprehend.

Text and art are not solved. What would solved even mean for those? There are NNs which produce decent output given a simple prompt, but e.g. art is characterized to a large extent by development over time. Stable Diffusion isn't going to develop by itself.

Robotics OTOH is much simpler, but it might depend on your expectations. Look at Boston Dynamics.

Re: What Is AGI-Hard

#69

Post by fartsucker69 »

dbspin wrote: post #55
fartsucker69 wrote: post #49

Earlier quoted context omitted.

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…

> there is nothing known that we can't approach with our tools of intelligence, so to speak. Such things may necessarily be unknowable. If there are classes of observations that we are unable to make, or problems that we are unable to comprehend, its absolutely plausible that they would be cognitive blindspots. In fact there's a whole theoretical strand in cognitive science / ontology that suggests this. We perceive…

even hypothetically this is only partially possible. for sure what you talk about would be something that is purely an abstract / logical problem, because if something we effectively "can't think" originates somehow in the real world, we would be able to observe it and realize that we can't reason about it.

and even when we have exhausted everything there is to know in the universe and solved every type of problem, we still can't be sure that there isn't a magic gap that we can't see using our intelligence.

but this is completely a god of the gaps argument, i.e. a religious argument, and thus not worth considering until we get actual evidence of such a thing.

Re: What Is AGI-Hard

#70

Post by ptrhvns »

fartsucker69 wrote: post #69
dbspin wrote: post #55

Earlier quoted context omitted.

> there is nothing known that we can't approach with our tools of intelligence, so to speak. Such things may necessarily be unknowable. If there are classes of observations that we are unable to make, or problems that we are unable to comprehend, its absolutely plausible that they would be cognitive blindspots. In fact there's a whole theoretical strand in cognitive science / ontology that suggests this. We perceive…

even hypothetically this is only partially possible. for sure what you talk about would be something that is purely an abstract / logical problem, because if something we effectively "can't think" originates somehow in the real world, we would be able to observe it and realize that we can't reason about it. and even when we have exhausted everything there is to know in the universe and solved every type of problem, w…

This is incorrect. It’s not hypothetical, is fully possible, and is very much in the scientific arena. Evolutionary game theory, among other tools, has been used to demonstrate that it's highly probably that human beings do not perceive objective reality as it truly is. See “The Case Against Reality” by Donald Hoffman.

Edit: more precise language.

Post reply on HN