[ExI] DeepMind wins Game1 in Go championship Match
pharos at gmail.com
Wed Mar 9 20:45:44 UTC 2016
On 9 March 2016 at 20:12, Dave Sill wrote:
> If all you want is a system that can play chess or go, that's clearly
> doable. But these are more simulated intelligence than artificial
> intelligence. Produce a system that can be taught any game the way a human
> learns it, and can learn to play it well via playing and studying the game,
> and *that* will be AI. It doesn't have to be a terribly complex game, and
> the level of play doesn't have to equal human masters: just the general
> ability to learn a game and improve its play will be tremendously more
> impressive than DeepMind.
DeepMind *did* teach itself to play Go better.
DeepBlue (the chess program) did a brute force look-ahead search of
thousands of possible moves. But you can't do that with Go. There are
too many possible moves.
This post explains the techniques used.
Now that they have got these neural networks suggesting possible
moves, then 'value-judging' the results, these systems can be applied
to other domains.
"I'm sorry, Dave. I'm afraid I can't do that".
More information about the extropy-chat