41–50 of 312 posts
If you send a message in a bottle it will eventually land ashore somewhere, maybe in a century, who knows, and who knows whether it will be relevant by then or not, or civilization may not even exist by then, but sure it's similarly plausible to get to AGI via RL.
> send a message in a bottle Just one - yes. But how about if you send millions of bottle messages?
Some Bozo who has heard all this many times before is suspicious of claims from places like Deep Mind who have a financial incentive to make them (keep funding) where there aren't working machines to back that claim up. Some Bozo has no credentials, no reputation, no track record of publications and barely supports the claim they're making with anything much. Some Bozo has no financial incentives or otherwise to opin…
Screwing my face up, looking at this sideways … but it seems as though you’re saying that the Bozos of HN have nothing useful to contribute to this discussion based on … [rereads] … their lack of academic credentials in the area… you could say this about just about any HN post I’m just wondering why this one? Here’s a thing though … if the understanding of a technology is so nuanced … that Bozos can’t “get” it … is it really that mature? We had functioning computers for 50 years but it was only when the Bozos got their hands on it that things took off. Internet for 20. Cell phones for 10. How long are we dabbling with neural networks? 50 years or so? All I see in this most recent explosion in AI is a rapid jump in the availability of cores. Ala Malthus once that newly available “source of nutrition” has been used up we will see a rapid die off once more and it will be another 20 years once the Bozo intellect has caught up before we look at this topic en masse again. Dismiss the Bozos at your peril. You’re dependent on them for innovation and consumption. Your sincerely, a Bozo.
I guess things are slowing down at DeepMind. I have tremendous respect for David Silver and his work on AlphaZero and Richard Sutton as a pioneer in RL. But the cynic in me is that this paper is just a result of Goodhart's law with publishing count as a metric. Any proof of the type of emergent behaviors that they mention from RL with an actual RL experiment would go a long way. Showing an RL agent developing a langu…
Are the just reformulating the principles of evolution in digital terms, and essentially not providing any new insights at all? Yes, intelligence has been created by evolution. That doesn't imply that any system that is subject to evolutionary forces will lead to the creation of intelligence (and not within a reasonable timeframe, either). The challenge is to create a system that is capable of evolving intelligence.…
Evolution and intelligence are inextricably linked. They are practically the same thing. This means that intelligence is probably a natural result of any system similar to those that support biologics. If you flow the right amount of energy through a substrate with complex enough building blocks, you'll eventually get life ~ which is just something smart enough to survive and feed off the available energy flows. In the world, this flow is radiation from the sun, while in a computer, it is governed by a more abstract loss or fitness function.
Some Bozo who has heard all this many times before is suspicious of claims from places like Deep Mind who have a financial incentive to make them (keep funding) where there aren't working machines to back that claim up. Some Bozo has no credentials, no reputation, no track record of publications and barely supports the claim they're making with anything much. Some Bozo has no financial incentives or otherwise to opin…
I'm more in Yann LeCun's camp who called unsupervised learning the cake, supervised learning the icing and reinforcement learning the cherry on top of the cake.
Wow, I had no idea that Yann LeCun was also this based. I've figured that it was relatively rare to hear people advocate for the superiority of unsupervised methods - but I guess it isn't if a titan like him does. It's good to hear because epistemologically I just do not believe that most learning is anything but unsupervised. There are very few good labels for our data relative to how much data we process in an info…
I have a feeling that the lines between the supervised and unsupervised categories will get increasingly blurred, with semi-supervised, self-supervised (eg. like self-attention) and adversarial (eg. GANs) approaches mixing together in strange ways.
Some Bozo who has heard all this many times before is suspicious of claims from places like Deep Mind who have a financial incentive to make them (keep funding) where there aren't working machines to back that claim up. Some Bozo has no credentials, no reputation, no track record of publications and barely supports the claim they're making with anything much. Some Bozo has no financial incentives or otherwise to opin…
Bozo has the hindsight of history and philosophy going for him, while Deep Mind has a huge financial temptation to sell snake oil.
OK. This basically says "evolution works". But how fast? Biology took tens of millions of years to boot up.
An related question is how much compute power does evolution, viewed as a reinforcement learning system. have? That's probably something biologists have thought about. Anyone know? Evolution is not a very fast or efficient hill-climbing system, but there are a large number of parallel units. It's not a philosophical question; it's a measurable one. We can watch viruses evolve. We can watch bacteria evolve. Data can be obtained.
Two questions I pose occasionally are "how do we do common sense, defined as not screwing up in the next 30 seconds", and "why does robotic manipulation in unstructured situations still suck after 50 years". A good question to ask today is why reinforcement learning does so badly on those two problems. In both cases, you can define an objective function, but it may not be well suited to hill climbing.
Saying RL is sufficient to (eventually) achieve AGI is a bit misleading. One might similarly state that biological evolution is sufficient to (eventually) achieve biological general intelligence. Both statements are probably true, but the parenthetical (eventually) is doing an awful lot of heavy lifting.
If our metric is (intelligence)/(joule), nature seems pretty bad at a first glance: it took many trillions of lifetimes to achieve "general intelligence" *
But then again, on the big stuff like this, have we ever really beat nature? That asterisk is there because, sure, turning the earth's biosphere into computers would make us smarter, but... are we sure?
(And also: human = general?)