Author: Peter McKenzie
Date: 23:05:40 09/11/05
Hi All, I thought some of you might be interested my progress in computer Go. My program is called Gonzo and plays on the KGS Go Server (http://kgs.kiseido.com). There is a regular 'bot' tournament run on KGS every month. The tournament alternates between 9x9 Go and 19x19 Go. In the most recent 9x9 tournament (4 Sept) my program caused a minor upset by finishing in a tie for 1st place with GNU Go and viking5. Gonzo certainly had its share of luck, but a win is a win so I'll take it :-) Details of the tournament are here: http://www.weddslist.com/kgs/past/6/index.html The computer Go world is quite different to the computer chess world. Even the best programs aren't very strong and can, on occasion, play really badly. One good thing is the possibility of playing on different board sizes. 19x19 can be quite daunting and, with over 200 legal moves for much of the game, is tough for search based programs. 9x9 and 13x13 games are quite common, and meaningful games can even be played on boards as small as 5x5. Another interesting thing is the wide variety of techniques being used in various programs. These include traditional (global) alpha/beta searching, local search, Monte Carlo techniques, combinations of search and Monte Carlo, neural networks, genetic algorithms, etc. A strange thing about computer Go is the whole process of finishing the game and deciding who won. This is much more complex than in chess and in part a reflection of the state of the wider Go world where there are several slightly different versions of the rules used. In practice (for humans), the different rulesets don't have much effect but it certainly adds another element of compexity for the programmer. Most (but not all) computer Go tournaments use Chinese rules because these don't punish you from playing extra stones at the end of the game (instead of passing). However, you'll want your program to pass 'early' (rather than play all the stones that it possibly could) because otherwise humans won't want to play it. Once your program has passed (and the opponent has too) you then need to get the program to agree with the opponent as to which stones are 'dead'. This is non-trivial and it is possible there will be a disagreement at which time the players must somehow resolve their differences. Usually this means restarting the game and playing some more stones to clarify the situation. Anyway, if you are (or were) a computer chess programmer looking for a new challenge, you might want to look at computer Go. There is a bit of effort required to get up to speed with the game, but even that is an interesting experience that will twist your grey matter in new ways and give you an insight into Oriental culture and thinking. Chess is a battle, Go is a war... cheers, Peter
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.