Comment by stenecdote | Hacker News Reader

Comment by stenecdote | Hacker News Reader

Parent Full thread

Thanks for being the first to reply. I was worried I'd just get upvotes and no replies!

I think you have two separate points, one with which I agree and one with which I disagree.

First, I agree (and other commentators about AlphaZero seem to as well) that human learning "algorithms" still beat AlphaZero's on per-game ROI.

On the other hand, I disagree that AlphaZero's self-play is no more interesting than a human playing someone better and learning from them. AlphaGo, AlphaZero's predecessor, followed a strategy more like what you described, learning from a large corpus of existing expert chess matches. AlphaZero, on the other hand, requires no training beyond an encoding of the basic rules of chess that it can understand. From there, it bootstraps its understanding of chess without input from experts.

This is the piece I find most interesting, see as potentially useful for the future of human learning, and believe differs from practice with an expert teacher. And so I wonder, can we design learning environments where the learner bootstraps their own understanding from a limited input without continuous feedback from an expert or teacher?

I think this is actually how everyone learns!

You can't put information into people. You can present it to them, but they need to teach themselves, so to speak.

If you have what I think is a good schooling system, it will recognize and emphasize the self-teach aspect - students are encouraged to figure things out on their own.

For instance, where I studied CS most of the time was allocated to doing semester projects where we'd be a small self-organized team of 3-7 students working on something with very little external input.

You can find similar ideas for schools, e.g. Sudbury schools. I think the Waldorf school has some aspects of it too. I'm sending my children to such a school.

> And so I wonder, can we design learning environments where the learner bootstraps their own understanding from a limited input without continuous feedback from an expert or teacher?

Why would you remove continuous feedback from expert or teacher? Would that make human learning "faster" and more "efficient"? That approach works for AI because unlike human, AI remembers every single data point with 100% accuracy and can iterate repeatedly without fatigue. It also does not suffer from issues such as boredom and it doesn't require motivation either.

By the way, human already learn from experience by bootstraping their own understanding, teachers and experts exist to fast track the beginning phase so a kid doesn't have to play ten thousand games just to reach beginner skill level.

> By the way, human already learn from experience by bootstraping their own understanding, teachers and experts exist to fast track the beginning phase so a kid doesn't have to play ten thousand games just to reach beginner skill level.

Yeah, I came part of the way to this realization in my reply to your other message.

> And so I wonder, can we design learning environments where the learner bootstraps their own understanding from a limited input without continuous feedback from an expert or teacher?

Yes, and we do it all the time.

We can just do a lot better with continuous feedback. (And AI probably could, too, if experts that could communicate fast enough not to be a huge drag on the AIs training cycles were available. But since with current technology once we've trained an AI of the type we can make today, we can replicate it, that's not really important; if ever developed AIs that depend on reconfigurable hardware without trivially extractable state, that may change.)

I have two thoughts:

AlphaZero plays millions of games against itself with a low per-game ROI in much less time than it takes a human playing against an expert with a high ROI. In this way AlphaZero has more work to do than the human to achieve a certain skill level, after which it is probably doing a similar amount of work to the human to continue to improve but can do it in much larger numbers.

On the other hand, I think I've heard of experts at chess playing games against themselves but I can't seem to find a reference at the moment.

Life is too short.

Alpha Zero can play millions of games against itself. I can't.

Reply on news.ycombinator.com