News for hackers. Stuff that matters.
Re:Why programmers should play Go by prospero on Monday July 14, 2008 @03:22PM (#245979)
There's an interesting trend in the computer go community away from rule-based AI, and towards a more stochastic approach. This means that where the leading developers were all 1 dan or better (very talented amateurs, by the metric of the article), it's become possible for talented programmers with a relatively limited understanding of the game to make major contributions.

Go is a complicated game, and the human response to that is to create rules of thumb and build a strong intuitive understandings of different positions. It's arguably less elegant to analyze a position by running hundreds of thousands of playouts, but it's finally a problem that resembles real computer science (though most academic papers are still of the "we tried this, and it did okay" form). It's a pretty exciting time to be involved, and I'd suggest that anyone with even a passing curiosity check it out.

Re:Why programmers should play Go by Alex3917 on Monday July 14, 2008 @04:27PM (#246054)
The archives of this mailing list has some good discussions about the stochastic approach:

http://computer-go.org/pipermail/computer-go/

Apparently the strongest computer AI is now able to beat 1k players consistently.

Re:Why programmers should play Go by prospero on Monday July 14, 2008 @05:36PM (#246113)
It's worth noting that it can beat 1k players on a 9x9 board. On a 19x19 board (the standard board size), they are both weaker and less consistent.
Re:Why programmers should play Go by dood on Monday July 14, 2008 @06:02PM (#246132)
One of the strongest current programs, MoGo, recently beat a 5 dan player on 9x9 using a monte carlo approach.

http://en.wikipedia.org/wiki/Computer_go#Monte-Carlo_methods

Re:Why programmers should play Go by prospero on Monday July 14, 2008 @06:29PM (#246165)
I'd seen that, but I haven't seen any published results of 1k players being routinely beaten at 19x19. If I've missed something like that, I'd be grateful for a link.
Re:Why programmers should play Go by nazgulnarsil on Tuesday July 15, 2008 @06:47PM (#247152)
agreed: 9x9 is to real go what checkers is to chess.
Re:Why programmers should play Go by schtog on Monday July 14, 2008 @04:20PM (#246051)
Yes chess isn't really interesting out of an AI-perspective since a supercomputer can beat a human with brute force + heuristics. In Go you'd need something more like real AI.
Re:Why programmers should play Go by prospero on Monday July 14, 2008 @05:59PM (#246130)
AI is search, brute-force or otherwise. There are a number of fascinating things that have come out of chess research on the way to grandmaster-level play, and even if the strongest chess players are all computers, that doesn't mean that it's suddenly become a worthless pursuit. The day that a perfect game of chess is played is the day that it gets boring, not before.

With respect to go, the current approach is completely brute force; it plays thousands of games at random, and uses the percentage of wins by one side to evaluate the strength of the position. The challenge is how to use the information you've already encountered in your searches to reduce the search-space to something manageable.

Re:Why programmers should play Go by geebee on Monday July 14, 2008 @05:33PM (#246110)
Chess could be interesting from an AI perspective if you limited the amount of brute force searching allowed. Computer chess and human chess are, by and large, completely different games.

I'm not sure I agree with this article about patterns in chess. The real masters think strategically as well as tactically, and use the structure of the pawns and pieces, combined with intuition, to guide their moves.

Ultimately, I suspect that go and chess fire up similar regions of the human brain.

Re:Why programmers should play Go by pon on Monday July 14, 2008 @05:49PM (#246126)
They did MRI on Z.Polgar while she was playing. What they found out was chess thinking happens in a small region of the brain. The region is the same humans use when trying to recognize faces. Training modifies that machinery to recognize chess board positions too.
Re:Why programmers should play Go by geebee on Monday July 14, 2008 @06:13PM (#246146)
Huh, that surprises me. I would have expected a chess master to use more of her brain than I do (not that I know how my own brain would look all wired up, but I have trouble believing that I don't think much more narrowly about the game - kind of how people who actually understand math approach factoring a set of equations differently from people who need a set of rules they can follow closely).
Re:Why programmers should play Go by eru on Tuesday July 15, 2008 @08:03AM (#246578)
Perhaps they only looked at what parts of their brain experienced chess players used more?