Re: computer-go: perfect play

[Jerome Dumonteil <jerome.dumonteil@xxxxxxxxxxx>]

Vlad Dumitrescu wrote :

...
> Does anyone know whether there are any speculations/theorems/thoughts about
> wheter there is a winning strategy for any of the players in Go? I know of
> some games where partial proofs are made...
...

I "believe" there is a winning strategy (don't know if it's for W or B)

Please, correct me if what I propose is wrong.

- during a game, at any time, there is a limited number of legal moves.
- we can assume (or even change the rules for that), that all games
are less than... 10000 moves. Or just consider the set of games less
than this limit of moves.

so, we got a finite number of games (ok, a big number). It's the key
assumption.

just apply a deep min/max algorithm (without pruning), until end of game.

when W choose a move, there is a finite number of possible continuations,
for a possible move, W look all follow up, and see results of all games : 
see the results, aka best B & W line of play until the end. Keep the better
Black line of play. W is sure to make better than some S score at the end. 
S maybe positive or negative.

same analysis for all possible moves. So, selecting the best result, White
is certain to be able to have a result >= some S score. And what is
interesting, at same time B is also certain of making this same score or
better. So, one of B or W is the sure winner of the game.

I think I've read this somewhere. This result should be mathematically 
proven somewhere (or I'm totally wrong ?). We just need a finite number 
of possible games.


-- 
------------------------------------------------------------------ 
jerome.dumonteil@xxxxxxxxxxxxxxxxx       
http://perso.linuxfr.org/jdumont/