Hacker News Reader

Comment by krige | Hacker News Reader

Are you saying that we already have AGI, except those pesky goalpost movers keep denying the truth? Hm.

I'd say yes, by at least one old definition made by someone who was at the time in a position to have a definition.

When deepmind was founded (2010) their definition was the following: AI is a system that learns to perform one thing; AGI is a system that learns to perform many things at the same time.

I would say that whatever we have today, "as a system" matches that definition. In other words, the "system" that is say gpt5/gemini3/etc has learned to "do" (while do is debateable) a lot of tasks (read/write/play chess/code/etc) "at the same time". And from a "pure" ML point, it learned those things from the "simple" core objective of next token prediction (+ enhancements later, RL, etc). That is pretty cool.

So I can see that as an argument for "yes".

But, even the person who had that definition has "moved the goalposts" of his own definition. From recent interviews, Hassabis has moved towards a definition that resembles the one from this paper linked here. So there's that. We are all moving the goalposts.

And it's not a recent thing. People did this back in the 80s. There's the famous "As soon as AI does something, it ceases to be AI" or paraphrased "AI is everything that hasn't been done yet".

> AGI is a system that learns to perform many things at the same time.

What counts as a "thing"? Because arguably some of the deep ANNs pre-transfomers would also qualify as AGI but no one would consider them intelligent (not in the human or animal sense of intelligence).

And you probably don't even need fancy neural networks. Get a RL algorithm and a properly mapped solution space and it will learn to do whatever you want as long as the problem can be mapped.

Can you cite the Deepmind definition? No Google results for that.

It's from a documentary that tracks Hassabis' life. I c/p from an old comment of mine (the quotes are from the documentary, can probably look up timestamps if you need, but it's in the first ~15 minutes I'd say, when they cover the first days of Deepmind):

----

In 2010, one of the first "presentations" given at Deepmind by Hassabis, had a few slides on AGI (from the movie/documentary "The Thinking Game"):

Quote from Shane Legg: "Our mission was to build an AGI - an artificial general intelligence, and so that means that we need a system which is general - it doesn't learn to do one specific thing. That's really key part of human intelligence, learn to do many many things".

Quote from Hassabis: "So, what is our mission? We summarise it as <Build the world's first general learning machine>. So we always stress the word general and learning here the key things."

And the key slide (that I think cements the difference between what AGI stood for then, vs. now):

AI - one task vs. AGI - many tasks

at human level intelligence.

It doesn't play chess? Just can parrot it very well

Yeah, maybe. But what matters is the end result. In the kaggle match, one of the games from the finals (grok vs o3) is rated by chesscom's stockfish as 1900vs2500. That is, they played a game at around those ratings.

For reference, the average chesscom player is ~900 elo, while the average FIDE rated player is ~1600. So, yeah. Parrot or not, the LLMs can make moves above the average player. Whatever that means.

I believe it will make illegal moves (unaided by any tools ofc). It will also make mistakes doing things like not being able to construct the board correctly given a fen string. For these reasons I consider long strings of correct moves insufficient to say it can play the game. If my first two statements, about a propensity for illegal moves and other fails on "easy for humans" tasks were untrue then I would reconsider.

In the kaggle test they considered the match forfeit if the model could not produce a legal move after 3 tries (none of the matches in the finals were forfeited, they all ended with checkmate on the board). Again, chesscom's interface won't let you make illegal moves, and the average there is 900. Take that as you will.

No, just what has usually happened in the past with AI goalposts.

At first, just playing chess was considered to be a sign of intelligence. Of course, that was wrong, but not obvious at all in 1950.

You know, as the saying goes, if a metric becomes a target...

I don't think AGI is a useful concept, but if it exists at all, there's a very good argument that LLMs had it as soon as they could pass the Turing test reliably, which they accomplished years ago at this point.

LLMs do not pass the turing test. It's very easy to know if you're speaking with one.

I think, given some of the signs of the horizon, there is a level of MAD type bluffing going around, but some of the actions by various power centers suggest it is either close, people think its close or it is there.

It wasn't the best definition of AGI but I think if you asked an interested layman whether or not a system that can pass the Turing test was AGI 5 years ago, they would have said yes

An interested but uninformed layman.

When I was in college ~25 years ago, I took a class on the philosophy of AI. People had come up with a lot of weird ideas about AI, but there was one almost universal conclusion: that the Turing test is not a good test for intelligence.

The least weird objection was that the premise of the Turing test is unscientific. It sees "this system is intelligent" as a logical statement and seeks to prove or disprove it in an abstract model. But if you perform an experiment to determine if a real-world system is intelligent, the right conclusion for the system passing the test is that the system may be intelligent, but a different experiment might show that it's not.

Douglas Hofstadter wrote Gödel, Escher, Bach nearly 50-years ago, and it won a Pulitzer Prize, the National Book Award, and got featured in the popular press. It’s been on lots of college reading lists, from 2007 online coursework for high school students was available from MIT. The FBI concluded that the 2001 anthrax scare was in-part inspired by elements of the book, which was found in the attacker’s trash. Anyone who’s wanted to engage with the theories and philosophy surrounding artificial intelligence has had plenty of materials that get fairly in-depth asking and exploring these same questions. It seems like a lot of people seem to think this is all bleeding edge novelty (at least, the underlying philosophical and academic ideas getting discussed in popular media), but rather all of the industry is predicated on ideas that are very old philosophy + decades-old established technology + relatively recent neuroscience + modern financial engineering. That said, I don’t want to suggest a layperson is likely to have engaged with any of it, so I understand why this will be the first time a lot of people will have ever considered some of these questions. I imagine what I’m feeling is fairly common to anyone who’s got a very niche interest that blows up and becomes the topic of interest for the entire world. I think there’s probably some very interesting, as-yet undocumented phenomena occurring that’s been the product of the unbelievably vast amount of resources sunk into what’s otherwise a fairly niche kind of utility (in LLMs specifically, and machine learning more broadly). I’m optimistic that there will be some very transformational technologies to come from it, although whether it will produce anything like “AGI”, or ever justify these levels of investment? Both seem rather unlikely.

Reply on news.ycombinator.com