This article is from the Go FAQ, by Morten PAHLE email@example.com with numerous contributions by others.
It is a surprise to most people that there is no software available
today which can beat an average-level human player.
The level of play of the strongest playing software today is estimated
at around 8 Kyu (and even this is heavily contested; see for instance
<http://cns.nyu.edu/~mechner/compgo/> for an interesting analysis by
The main reason for this is said to be that it is difficult to
estimate the value of a given move. This makes it difficult to program
a routine which can choose the 'best' move. The true value of a move
may not become apparent until 30 plays later in local fights, and
sometimes literally 100 plays later, for endgame optimisation moves.
Another reason is that, because of the large playing area and the
simple rules, there is always a very large number of legal moves which
are even reasonably plausible moves. This results in a very large game
tree if 'dumb' search algorithms are used.
Considerable resources are going and have gone into the development of
For those who are interested in the subject, there are various places
to start a search:
If you are interested in computer Go you may want to join the
computer-go mailing list. The computer-go mailing list was
established in Feb 93 to discuss programming computers to play Go.
The volume of mail on this list is rather low, but sometimes goes up
To join the list, send an email to 'firstname.lastname@example.org' and
put 'SUBSCRIBE COMPUTER-GO first_name last_name' in the body of the
message. (Remove all quotes and fill in your own name). You will
receive a confirmation detailing how to use the list.