4 ms·
Very interesting read, thanks for posting it! I'm not really familiar with that domain, could you tell me whether it's possible that the reason Go bots are not
by gnuvince 12y ago
Very interesting read, thanks for posting it! I'm not really familiar with that domain, could you tell me whether it's possible that the reason Go bots are not as strong as chess bots (in relation to humans) is simply because chess is a much more popular game that, and more time and effort has gone into making those AI stronger? Go is touted as having a much larger search space than chess, but that probably means that it's harder for humans as well, no?
- cjbprime 12y agoYou're welcome. Popularity might well be some of it, but I think that effect's worn off by now. I think the main reason has to do with score estimation. With chess, you can say that a queen is worth 8 points and a rook is worth 5 and a bishop's worth 3, and get a quick idea of who's winning a game. In Go, every piece has the same value, and whether a stone is useful depends on which other stones it's connected to, and whether those stones are going to be alive at the end of the game. So just to work out whether you're winning, you have to be able to do search all the way to the end of the game and work out who wins! And computers have been getting that process wrong until recently. The extra search space doesn't make Go harder for humans too, because humans play both Chess and Go very differently to computers. A human never tries to calculate the value of a million possible moves, so the question of how many million possible moves there are to search through just isn't relevant to a human. Humans play both Chess and Go largely using pattern recognition and insight in a way that computers are not currently able to model.