[Computer-go] Komi and the value of the first move

Erik van der Werf erikvanderwerf at gmail.com
Wed Apr 7 04:52:39 PDT 2010


On Wed, Apr 7, 2010 at 1:04 PM, Chase Albert <caalbe at wm.edu> wrote:
> Nitpick: it doesn't necessarily give the players an equal chance of winning,
> it just makes perfect players tie. Equal probability of winning is likely
> not achievable in go (though it would be a poor game if it weren't at least
> close).

Nitpick^2: It's the other way around. Current komi *does*
approximately balance the probability of winning (at least for strong
players).

Perfect komi is most likely to be 5, 7 or 9 points but nobody has a
proof for the true value of the empty 19x19 board (but even if it was
known it might not be the right value for balancing fallible players).

I suspect that if one would do a study to find the optimal statistical
komi (for balancing the probability of winning with equal strength
players) we would find that it decreases with board size and increases
with playing strength.

Erik



More information about the Computer-go mailing list