chuqui@cae780.UUCP (Chuq Von Rospach) (11/10/83)
Anyone out there have a program that will play go? How about referee a go game between two terminals? chuq -- From the dungeons of the warlock: amd70!cae780!chuqui Chuqui the Plaid *pif*
robison@eosp1.UUCP (11/13/83)
In Japan, there are programs that play GO endgames. (That is, there is a tricky position late in the game, and you can play against the computer to try to finish the game and be ahead.) Although to human beings, there is subjective evidence that GO and Chess are of roughly equal difficulty, the situation looks different to a computer. In chess, most positions have 30 to 40 possible moves. In go, the first move is one of 361, the next one of 360, and so on (most of the time). As any experienced GO player will tell you, there are no simple symmetries to reduce this incredible count to a manageable number. Almost all of the possiblitites must be analysed, in the same sense that all of the chess possibilities must be analysed. Good luck finding a GOplaying program; it will be very weak.
dont@tekig1.UUCP (Don Taylor) (11/15/83)
The spectre of the impossible complexity of go playing programs, is constantly
raised. Each time I hear this, I see the parallels with the chess playing
programs. There was near total agreement that chess was impossible to program,
the complexity was simply too great. Shannon showed that the computation needed
was out of the question, and the only plausible solution was the design of a
"plausible move" generator, direct attack of the problem was demonstratably
impossible. Along comes a man named Jennings, I think, with a KIM and 2kb of
memory. He sat down and wrote a poor, but suprisingly passable player. He was
not the only person who contributed, and alpha-beta cutoff certainly had a
major part in things. The point is that a group of people working on a problem
can come up with startling solutions. The creation of a poor program, could
fire someone else to show a better solution to the problem, and we would be off.
I cannot see the problem as impossible, mearly unsolved. The 361! is not a
realistic estimate. The WORST case, is worse than that, consider captures, and
even that figure is too low, but actually, the tree is only more bushy than one
for chess, and there have been estimates of what sort of factor is involved
Don Taylor
tektronix!tekig1!dontrobison@eosp1.UUCP (11/18/83)
Don Taylor points to the recent history of progress in chess
computers as an indication that computers will also play go.
I don't find this parallel encouraging. Progress in computerchess
has been linked to:
- Improvements in computer speed and lower computer cost
- Trends toward exhaustive search chess programs
The strongest chess programs today use brute force to examine a lot
of positions per move. They do use some shortcuts, to avoid wasted
searching, that are innovations.
If we can apply the same thing to GO:
- where are the go-playing computers?
- the exhaustive searches will be spectaularly longer because
of the larger number of moves availabe in go.
We might say that when computers are fast enough to play go as well
as they now play chess, computers will play chess far, far better than
they do now.
- Keremath, care of:
Robison
decvax!ittvax!eosp1
or: allegra!eosp1