Skip to content

Comment on Game theory article from a professional Starcraft player/caster.

Comments

The new 4-5dan go-playing bots on kgs (zen19d, crazystone) use this strategy extensively. When they're ahead they play to consolidate their biggest weakness, and when they're behind they play more and more risky moves to try to come back. In the endgame the calculate the score exactly, and will play negative point value moves as long as they are ahead on the board.

It makes for a really tough game, cause if you do get ahead, you have to face a series of attacks which almost but don't quite work, and if you mess up any of them it's over.

Interesting that they use this strategy. Essentially, high variance is good when the expected value of the game is negative. It increases the chances of success in a bad situation.

The maths of this can be found in the article "A Rule of Thumb (not only) for Gamblers." Currently, this article is available free here: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.55.3...

Yeah I read the advice in this article years ago in a beginner's Go tutorial, something about "you just need one more point than the opponent to win".

Lot of little zen-like lessons in learning that game, is why I love it.

Interesting. I don't really know much about basket ball. But I have noticed something similar. In the last seconds of the game, the loosing team will try riskier moves. (Forcing penalties and throwing from the three point line). It is like playing roulette.

My Go teacher once told me that the truely great players will hold a game in near perfect balance throughout. Concerningy, are the computers really at 4 dan now? I ain't never going to beat the machines.

yeah, they have really gotten good in the last year. They've gone from better than 90% of players who have ever played go, to better than 95%+. They are great to watch, too, merciless and play interesting moves.

Check out crazystone's stats: http://kgs.gosquares.net/index.rhtml.en?id=CrazyStone&y=...

72% win percentage as a 4d.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.