241–250 of 312 posts
But as an enthusiast of all three I really think that AGI is a hardware problem, not a software problem.
Reinforcement learning on a massive corpus of data is how we train all biological intelligence.
The crazy thing is that in humans we manage to do it on ~3 watts an hour.
I think we have the software cracked, my gut thinks silicon just isn't the right material
I agree with the other commenter here that the smartest systems are less intelligent than the common housefly. A brain the size of a pinpoint can navigate, eat, reproduce, and live a full life without big data or internet.
Earlier quoted context omitted.
“Who cares?” Well, the people who need to pay the people developing the endgame for everything you speak of.
I'm sorry, but this makes no sense to me. The people paying for the development of the AGI can mean many things - the Google customers/users, Alphabet as a company, the executives throwing money at the problem? Either way, I don't really get your point. Your initial post was about how it is counterintuitive for Google to allocate funds for an AGI, since it makes money out of ads. These are not mutually exclusive, you…
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…
It is not enough. Machine learning only gets you so far, even the best general-purpose algorithms. Deriving reasoning and common sense demands something more than learning. I agree with the other commenter here that the smartest systems are less intelligent than the common housefly. A brain the size of a pinpoint can navigate, eat, reproduce, and live a full life without big data or internet.
I'm not going to take bets on who is right, but simply saying "nuh uh" is not exactly breaking ground.
I'm not a neuroscientist, an AI specialist, or a hardware engineer. But as an enthusiast of all three I really think that AGI is a hardware problem, not a software problem. Reinforcement learning on a massive corpus of data is how we train all biological intelligence. The crazy thing is that in humans we manage to do it on ~3 watts an hour. I think we have the software cracked, my gut thinks silicon just isn't the ri…
To me, it seems more likely that we're missing something/some things on the software side. AGI could probably run on present day hardware or even older.
It is not enough. Machine learning only gets you so far, even the best general-purpose algorithms. Deriving reasoning and common sense demands something more than learning. I agree with the other commenter here that the smartest systems are less intelligent than the common housefly. A brain the size of a pinpoint can navigate, eat, reproduce, and live a full life without big data or internet.
AI and Machine Learning are housed in human’s conception of Reality. Not Reality itself. We’re innovating in a sandbox, but we’re getting better and learning more to the point of getting so good, we may one day get out...
But I think though they talk about rewards incessantly, the interesting angle is the importance of a complex environment in which the agent learns to maximize a reward:
> we suggest the emergence of intelligence may be quite robust to the nature of the reward signal. This is because environments such as the natural world are so complex that intelligence and its associated abilities may be demanded by even a seemingly innocuous reward signal.
On the one hand, this does echo some old work on situated cognition, which one might actually believe. But perhaps politically, if the claim behind the claim is that we can only develop powerful AGI which understands how to interact with our world by developing agents that learn with unfettered access to the world, then perhaps this will be the beginning of a strong push for tolerating spastic ineffective robots in our physical environments, and letting error-prone agents have vast access to our virtual environments. We'll be asked to put up with their mistakes because that's supposedly the cost of progress; limiting their environment would limit their cognitive potential.
> For example, consider a signal that provides +1 reward to the agent each time a round-shaped pebble is collected. In order to maximise this reward signal effectively, an agent may need to classify pebbles, to manipulate pebbles, to navigate to pebble beaches, to store pebbles, to understand waves and tides and their effect on pebble distribution, to persuade people to help collect pebbles, to use tools and vehicles to collect greater quantities, to quarry and shape new pebbles, to discover and build new technologies for collecting pebbles, or to build a corporation that collects pebbles.
Aren't you comforted they chose round pebbles instead of paperclips? And though their example is meant to illustrate that the reward function doesn't matter, you'll notice there's no negative reward term for e.g. smashing a retaining wall to dig for pebbles in the rubble, or dredging a beach where a protected bird species nests, etc. "Allowing the agent to fully explore the complex environment is the only way it will learn complex representations and actions!"
It is not enough. Machine learning only gets you so far, even the best general-purpose algorithms. Deriving reasoning and common sense demands something more than learning. I agree with the other commenter here that the smartest systems are less intelligent than the common housefly. A brain the size of a pinpoint can navigate, eat, reproduce, and live a full life without big data or internet.
I concur. There is “Reality”, and then there is human’s conception of “Reality”. (See also: “Theory of Forms”). AI and Machine Learning are housed in human’s conception of Reality. Not Reality itself. We’re innovating in a sandbox, but we’re getting better and learning more to the point of getting so good, we may one day get out...
Yet, here we are, saying that achieving one form of machine learning to its fullest extent will be "enough" to meet the challenge of AGI.
The layman and the AI expert have both written off the possibility of creating AGI for the entirety of this latest AI spring. In the past decade I have basically never encountered anyone who thought that AGI was going to happen in our lifetimes or even anyone who believed that it would be a problem if it did. One time I discussed AGI with a good friend. And I gently pressured him to play through the scenario of the a…
> He said “can’t we just unplug it?” This is a microcosm of the entire issue. It’s something a child might say. For global warming, can’t we just turn on the air conditioning? No, we can’t just unplug it. If it were running as a sandboxed application inside some kind of runtime environment with only access to print output text and read input text, as you might expect a GPT-n program to run, certainly you could unplug…
GTP is not supposed to exist according to computer science and machine learning experts in 2017. They were all wrong. Same with deep dream, styleGAN. And it will be true with the next thing.
The mental gymnastics refers to the fact that people have cognitive dissonance about GTP. They have amnesia about the decades of stagnant progress in text generation. All anyone can do is point out that it’s definitely not sentient and it’s not AGI so what’s the big deal? All of a sudden the goal posts have been moved… this is the single biggest quantum leap in text generation ever, a mind-boggling level of lucidity with grammar, punctuation and more all without ever being given a single deliberate instruction from a human being. It is objectively amazing. Objectively. And people have no reaction. Because they are not emotionally ready for it. It’s cognitive dissonance, mental gymnastics, whatever you want to call it.