[gnugo-devel] gnugo and AI

[Wolfgang Manner]

hello
i believe that professional go players have an enormous
data base of patterns in their brain. so maybe one can
think of go programming as an exercise in pattern recognition.
You can never store all patterns, there are just too many, so
you will always have the problem that your actual game matches
"almost" one of your stored patterns. One way of dealing with
that has been mentioned by Paul Pogonyshev, you break up
the board into local games. his will work to some extent,
but cannot be the final solution. so i think you will have to
play a few moves of your game, trying the moves you have
stored, and see what happens. i am afraid i have not solved
any problems, i have only changed some words.
Speaking of pattern recognition one thinks obviously about
neural networks. There are go programs using that, but i
know nothing about it, so you would have to find somebody
else. Parallel computing a la big blue has been talked about
and tried, again somebody else will know
i wish you all the best
wolfgang manner