[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: computer-go: Winning strategy speculation and loosing games.




Rules:
1) Normal rules of go (whatever that is...)
2) Your are not allowed to fill in complete eyes (corner surrounded 
by 3 stones of same color, side point surrounded with 5 or a center 
point surrounded by 4 orthogonal and 3 "diagonals" of the same color
3) You win if you normal go score is lower than the opponent, when 
there are only forbidden eyefilling mves left on the board.

> It seems that any program that avoids making any eyes would be 
'optimal'.

Consider this situation, assuming it can arise when two bad players 
play each other.

_ _ _ _ _
* a b c *
* * * * *

If the opponent sacrifices at 'a' and 'c' then the suicidal player is 
forced to play 'b' sooner or later in the endgame and live with two 
eyes. Thus in response to 'a'  it is important to play 'c' before the 
opponent in order to only make one eye.

This is not a proof, but I think that a clever use of sacrifice 
contact plays can make it hard to avoid making eyes with a simple eye 
avoiding strategy such as playing (1,1) (1,2) (1,3) etc and then it 
will be counter strategys to that strategy and so forth...

This may not be as complex as normal go but far from trivial. Perhaps 
it is a game that is easier to program well compared to normal go. 
The real surprise would be if it is hard to make a good "bad" 
computer player that can compete with a skillful 'bad' human player.

It should be easy to program a simple bad player (that beats a random 
player) and play it on a small board. It would be nice to see some 
real results and not just verbal speculation on this list. :-)  It is 
after all a *computer go* list. (No, I will not do it myself - I am 
busy programming normal go for at least a lifetime ahead...)

Best wishes

Magnus


--
Magnus Persson
Department of psychology, Uppsala University
Box 1225, SE-751 42, Sweden
Tel 018-471 2141 (work), 018-460264 (home)
MAILTO: magnus.persson@xxxxxxxxxxxxxxxxx
URL: http://www.docs.uu.se/~magnuspe