Hacker News new | past | comments | ask | show | jobs | submit login

That would be amazing if it could achieve the same levels (or higher) without the bootstrapping.

The niggling thought in my mind was that AlphaGo's strength is built on human strength.




Human strength is also "built on human strength" so I don't see the problem? :)


Well, yes, but it's still humans standing on the shoulders of other humans. Even though human players do memorize opening books, it stays in the family so to speak. Meanwhile a human player facing an AI engine is battling both the AI, and great human players of the past (who invented the openings).


It's not truly artificial if it's using a human playbook. (Is the problem posed by the parent, I believe.)


What is 'truly artificial'?

Neural networks are modeled after biological systems to begin with, I don't the that's a meaningful concept at all.


Well, we can extend that to say the biological systems are self-assembled randomly and selected through evolutionary algorithms, starting from random molecules on the sea floor.


Truly artificial means not using meatspace metaphors for reasoning like human players do.


I suspect you will only be satisfied when AIs play each other at an incomprehensible game of their own devising.


Making popcorn now.


I doubt it. When/if they do play such a game that humans can't explain, I'll probably be interested in some other problem.

Isn't that the nature of human endeavor? Always looking for the next challenge?


What if the "betaGo" played just AlphaGo, and learned from its games?

BTW: even humans don't just randomly pick up the game. They have teachers, who teach them the tricks of the trade and monitor their games.


That's already a known method to transfer "knowledge" from one model to another. I should double-check before quoting a paper, but I think that this one talks about this (http://arxiv.org/abs/1503.02531).

You train many models. Then you "distill" their predictions into one model by using the multiple predictions (from many models) as targets (for the single model trained afterwards).

You're right to point out that humans don't do that.

I think it would be "cheating" if you train BetaGo on AlphaGo, for the purposes for doing that experiment. The goal would be to have some kind of "clean room" where people fumble around.

Of course, you can also run the other experiment to see how fast you can bootstrap BetaGo from AlphaGo. That's also interesting.


I'm pretty sure that the reinforcement learning algorithm they are using is guaranteed to converge. It just takes a very long time to train, and using human games probably sped it up.


As far as I know, using neural networks for function approximation destroys the various convergence guarantees available. NNs can easily diverge and have catastrophic forgetting, and this is one of the things that made them challenging to use in RL applications despite their power, and why one needs patches like experience replay and freezing the networks.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: