Richard Sutton on Pursuing AGI Through Reinforcement Learning

[@ivan8960]

Sutton is great and one of the very few that understand what intelligence is really about.

[@agenticmark]

I think Sutton will end up much more beloved than he is. He was early on agency and self-learning, which none of the LLMs or frontier models I know of use. Because of this dude I was able to solve so many things more interesting than working on language models (which I also do)

[@taumag]

The Horde model is in the same spirit as Jeff Hawkins' Thousand Brains Project, where they use a cortical model of the mammalian neocortex - 150,000 columns of similar neurons that all contribute to the final action policy. I wonder if Sutton and Hawkins would work together!

[@chrisdipple2491]

We need more work on RL, it's the technology that subsumes and guides the other AI technologies, such as LLM.s (and even symbolic AI). It's hard though, so short term commercial reasoning and the Wow factor are harder to find. Great stuff Rich.

[@almostsurely5314]

This guy is awesome!

[@GNARGNARHEAD]

ahhh! I love Sutton! he makes so much damn sense 😄

[@capitalistdingo]

Someone recently said that today’s neural networks are the vacuum tubes of AI; a technology that is revolutionary and lets us start using something new but will soon be seen as a synonym for “primitive” in the field.

[@capitalistdingo]

The analogy of a “microchip” for AI will need to be something that can remember, set goals, be curious, do something when not being promoted to think a thought. Not just sit there drooling between prompts.

[@pisiiki]

one of the greatest minds we have around atm

[@rodneypantony3551]

Rich, you do note a Board member's machine includes underseas fiber optics and satellite constellations and everything in between. The physical scale of the physical machines is on the scale of the planet. The EU Brain Community said they could build a human brain but it would be the size of an aircraft hanger. That was naive? It's the size of a planet?

[@eyeonai3425]

what are you talking about?

[@_obdo_]

Great interview. Fascinating work. Some important context is that Sutton believes that it’s inevitable and desirable for the machines to replace humanity. He thinks it’s racist to prefer humans.

[@eyeonai3425]

I think you mean speciest , and that was Dan Hendryx. I’ve never heard Sutton express a similar view. Can you tell me where are you heard that?

[@_obdo_]

Maybe this will work: Google “Richard Sutton AI succession” Note Richard’s YT channel only has 2 videos. It’s very much worth watching the shorter of the two.

[@_obdo_]

@@eyeonai3425pinging Craig to note my new responses above. Sorry for the comment-debris. The algorithm is doing weird things tonight.

[@rachkaification]

Richard Sutton is the father of reinforcement learning. OK, so how many fathers does the AI have? Cause I've heard about other fathers. Even godfathers. For example - Geoffrey Hinton.

[@russianbotfarm3036]

Hinton is _a_ father of neural nets / deep learning. He kept it alive while it was uncool, into the current era of GPUs. (And didn’t just keep it alive, he added to it of course.)

[@entreprenerd1963]

Yes, and in the cases of both Sutton and Hinton they had students (and especially thesis advisees) who have also built out the field.

[@shirtstealer86]

An other example of an intelligent person saying frankly idiotic things. People who are worried about AI does not compare it to nuclear weapons and they (for example Eliezer Yudkowski) have very detailed and pronounced arguments for why super intelligent ai could be a very real threat to humans. Imagine being a researcher at the forefront of this tech and not even familiar with what the risks are. Irresponsible and dangerous.

[@eyeonai3425]

He’s referring to Geoff Hinton who did make the comparison.

[@shirtstealer86]

@@eyeonai3425 that’s not the point, if he is a leading expert on these issues he should know that what worries the “silly” people who research ai threats is that humanity for the first time quite possibly will face a being more intelligent than we. And that is not at all like nuclear weapons. So either he is a bit slow or he is suffering from cognitive dissonance/ blinded by his own ambitions and curiosity. Either way, he should be excused from the conversation because he is endangering the human species with his nonsense.

[@McD-j5r]

People do not care for AGI

[@tim40gabby25]

AGI might not care for people. Too early to say.

[@eyeonai3425]

first, AGI doesn't exist so not sure why people don't care for it. Second, AGI simply means artificial general intelligence - that is, an AI system that can work across modalities, that can generalize enough to solve problems in different domains, as opposed to current 'narrow' AI systems that can only solve problems in one domain. If you're paying attention, you can see that AI systems are increasingly general - they can handle text, speech, image, video, etc. So, there is nothing inherently scary about AGI. What you're probably referring to is ASI - artificial superintelligence. That may indeed arrive some day, but I would think about think about the problems it can help us solve. Mankind is facing a potential existential threat from climate change. Maybe ASI will be able to solve it.

[@s_s_lab]

@@eyeonai3425can you give evidence for an "existential threat from climate change"?