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

DeepMind: A Generalist Agent

deepmind.com

Post reply on HN
Print view View original

231–240 of 345 posts

Re: DeepMind: A Generalist Agent

#231

Post by Jimmy »

axg11 wrote: post #22

Slowly but surely we're moving towards general AI. There is a marked split across general society and even ML/AI specialists between those who think that we can achieve AGI using current methods and those who dismiss the possibility. This has always been the case, but what is remarkable about today's environment is that researchers keep making progress contrary to the doubter's predictions. Each time this happens, th…

>Each time this happens, the AGI pessimists raise the bar (a little) for what constitutes AGI.

Let’s say I have a coworker. We write code together, but we also hang out outside of work sometimes, maybe go for a couple rounds of golf, maybe discuss our shared passion for indie rock, he tells me his opinions on current news events or what’s going on with his family, all the ordinary sorts of things that people do.

When we get to the point where I have to reasonably wonder if my coworker is a biological human, or an AI masquerading as a human, then we have undoubtedly achieved AGI.

Not that we actually have to flood society with an army of cybernetic humans, of course. It’s just that AGI would enable such a thing, in principle.

Re: DeepMind: A Generalist Agent

#232

Post by rsfern »

VikingCoder wrote: post #54
rsfern wrote: post #24

Earlier quoted context omitted.

> - "AI/ML can't solve scientifically useful problems" - then AlphaFold changed the protein folding field AlphaFold is a big deal, but AI in science has been a really hot topic in the past almost decade. Also, I still wouldn’t call AlphaFold really “intelligence”, it’s doing structure prediction which is cool but it’s a long way to scientific intelligence

I wonder if you get how much we've moved the goalposts on "intelligence." Once upon a time, it was considered "intelligent" to be able to add. Then "intelligence" was tool use, which we thought only humans could do. Then we swore it took "intelligence" to play Go as well as a beginner human. What set of tasks would you, right now, consider to be demonstrative of "intelligence" if a computer can do them? Then we can l…

Yeah, that’s a fair point. I am kind of coming from this place where we’ve been treating structure solution as an optimization/search problem for a long time. To me the competing methods to AlphaFold seem a lot more like computer chess systems than human grandmasters. The first structure solution by a human I guess would be closer to the grandmaster in this analogy. (Caveat: I’m in an adjacent field so not a direct expert)

I think higher level scientific tasks are like evaluating evidence under different mechanistic hypotheses, designing experiments to disambiguate these, coming up with novel mechanistic hypotheses, etc. I’m also not sure that automated systems for these tasks require intelligence. They just have to obsolete the current intelligent systems (humans) that perform these tasks, like computer chess. I don’t know, maybe this is “no true Scotsman” for machine intelligence?

Re: DeepMind: A Generalist Agent

#234

Post by esjeon »

yldedly wrote: post #147
Barrin92 wrote: post #138

Earlier quoted context omitted.

I don't think many people were making the claims that AI can't solve any scientific problems or can't perform creative work at all. That sounds like a big strawman. Before ML was getting big there were AI systems that created art. What sceptics have actually been saying is that the first step fallacy still applies. Getting 20% to a goal is no indication at all that you're getting 100% to your goal, or as its often pu…

The notions that are crucial for distinguishing between intelligence and what large NNs are doing, are generalization and abstraction. I'm impressed with DALL-E's ability to connect words to images and exploit the compositionality of language to model the compositionality of the physical world. Gato seems to be using the same trick for more domains. But that's riding on human-created abstractions, rather than creatin…

Yup, I think this is the pretty much describes the limitation of today's AIs. They are gigantic statistics machines at best. It's still amazing how far we can get with this technique, but we know where they stop getting better.

Re: DeepMind: A Generalist Agent

#235

Post by roywiggins »

brador wrote: post #213
justinpombrio wrote: post #208

Earlier quoted context omitted.

I am quite scared of human extinction in the face of AGI. I certainly didn't jump on it, though! I was gradually convinced by the arguments that Yudkowsky makes in "Rationality: from AI to Zombies" ( https://www.readthesequences.com/ ). Unfortunately they don't fit easily into an internet comment. Some of the points that stood out to me, though: - We are social animals, and take for granted that, all else being equal…

AI can’t have goals since the universe is logically meaningless. Our desire for purpose is a delusion.

I'm not sure it matters if a paperclip maximizer has a goal or just acts like it does.

Re: DeepMind: A Generalist Agent

#236

Post by space_fountain »

Jimmy wrote: post #229
idiotsecant wrote: post #190

Earlier quoted context omitted.

HN madlibs: I'm in the camp that thinks we're headed in a perpendicular direction and won't ever achieve powered flight with current efforts based on the simple idea that the basic tooling is wrong from first principles. I mean, most of the "progress" in flight has been due to getting better and learning how to understand a single piece of technology: fixed wing aircraft. A lot of recent powered flight findings have…

Do you think you’ll see true AGI in your lifetime? I certainly don’t think I’ll see it in mine. No other domain of the sciences is “complete” yet. There are still unanswered questions in physics, biology, neuroscience, every field. Why are people so sure that AI will be the first field to be “completed”, and in such an astoundingly short amount of time on a historical scale?

What makes you think AGI means AI is a complete field? Wouldn't that mean working out how to build a bridge without trial and error completed engineering? Or that biology was completed when we worked out how dna works + enough organic chemistry? Obviously we won't complete the field in our lifetime, but that's because nothing is ever really complete.

I think I'd put the chances of AGI within my lifetime at 30%. Low, but high enough it's worth thinking about and probably worth dumping money into.

Re: DeepMind: A Generalist Agent

#237

Post by webmaven »

Jensson wrote: post #148
VikingCoder wrote: post #133

Earlier quoted context omitted.

Why must it apply for a job, rather than just DO a job? But maybe some combination of this [1] and this [2] would do it. If you want to know about a computer actually DOING a remote job for a year without anyone noticing, I'll conclude with many links [a-i]. [1] : https://thisresumedoesnotexist.com/ (Sorry for the bad certificates.) [2] : https://www.businessinsider.com/tiktoker-wrote-code-spam-kel... [a] : An origin…

> Why must it apply for a job, rather than just DO a job? Because being able to manage a business relationship is a part of the job. If you could show an AI which got a job, then wrote a simple script that automated the AI's job and then coasted for a year that would be fine, but your links are just humans doing that, I want an AI that can do that to consider it intelligent. But thanks for demonstrating so clearly ho…

> > Why must it apply for a job, rather than just DO a job?

I think the idea here isn't just to create a job doing machine but to create a machine that is actually employable in the sense that it has not just intelligence but also agency.

An AI that can apply for a job, get the job, and do the job well enough to keep the job is demonstrating that it has some sort of theory of mind. But I don't think such a demonstration is very likely except as a stunt, purely on the grounds that an AI capable of the necessary subterfuge is more likely to employ shortcuts that are less taxing. A bit of electronic B&E to plant some ransomware, or have a shell corporation take out a few submarine patents and sue some businesses, perhaps operate a darknet marketplace (I think we can agree that an AGI will likely have better opsec than a human), etc.

Lest you think that the only options here are illegal, there are white-hat alternatives as well, for example it is possible to participate in bug bounty programs anonymously and get payouts in cryptocurrencies.

So why would an AI bother applying for a job in the first place?

Re: DeepMind: A Generalist Agent

#238

Post by Gatsky »

viksit wrote: post #182

(Former AI researcher / founder here) It always surprises me at the ease at which people jump on a) imminent AGI and b) human extinction in the face of AGI. Would love for someone to correct me / add information here to the contrary. Generalist here just refers to a "multi-faceted agent" vs "General" like AGI. For a) - I see 2 main blockers, 1) A way to build second/third order reasoning systems that rely on intuitio…

The concern over AI safety seems about right. The unique thing is that anybody cares at all about civilisational negative externalities in a functional and reasonable way. This is quite rare, unprecedented even. Typically humanity just rushes into new endeavours with little concern and leaves future generations to pick up the pieces afterwards (social media, colonialism, processed food, nuclear weapons etc).

Re: DeepMind: A Generalist Agent

#239

Post by webmaven »

dekhn wrote: post #70
Surgeus wrote: post #61

Earlier quoted context omitted.

This points out something very related that I think about a lot - can you prove to me that you do any of those things? Can I prove to you that I do any of those things? That either of us have a will? When would you be able to believe a machine could have these things? In Computing the Mind by Shimon Edelman is a concept that I've come to agree with, which is at some point you need to take a leap of faith in matters s…

Not just consciousness- all of science requires a leap of faith- the idea that human brains can comprehend general universal causality. There is no scientific refutation for Descartes' Great Deceiver- it's taken as a given that humans could eventually overcome any https://en.wikipedia.org/wiki/Evil_demon through their use of senses and rationality on their own. I have long worked on the assumption that we can create…

How likely is it that the results will still be interesting in 9 years?

Re: DeepMind: A Generalist Agent

#240

Post by dekhn »

webmaven wrote: post #239
dekhn wrote: post #70

Earlier quoted context omitted.

Not just consciousness- all of science requires a leap of faith- the idea that human brains can comprehend general universal causality. There is no scientific refutation for Descartes' Great Deceiver- it's taken as a given that humans could eventually overcome any https://en.wikipedia.org/wiki/Evil_demon through their use of senses and rationality on their own. I have long worked on the assumption that we can create…

How likely is it that the results will still be interesting in 9 years?

Dunno. For all I know, DeepMind will publish a paper titled "Evidence of Emergent General Intelligence in Deep and Wide Perceptron Hypertoroidal Neural Message Passing Networks of Tensor Processors Trained over the Youtube Corpus" and get all the credit. :)
Post reply on HN