I agree with several other posters: having frequent competitions, like CGOS and
the KGS tournaments, is an invaluable spur to improve the performance of Go
programs. If a sugar daddy were to add some serious cash prizes, that would
encourage even broader participation and greater quality.
One additional idea would be a test suite. A few months back, someone created a
suite of about 100 problems cast in a whole-board manner, which is more
convenient for MC-based programs. "Solve the life-or-death problem, or lose the
game." It would be useful to see a web site, comparable to the CGOS summaries,
which tests each program against the suite and publishes the results. Over
time, the suite would be extended with higher-level problems. This could be a
faster way to get feedback then playing thousands of games against other
programs, looking for a tiny statistical advantage. The two approaches ( test
cases and games ) could complement each other.
When playing Go, it's possible to do a lot of things right, but lose the game
with one or two bad moves. ( haven't we all had that experience? ) The bad
moves can get lost in the noise. That brings me to my second suggestion. If I
were to win the lottery ( o happy day! ), I'd hire a Go professional to review
computer games and look for those critical moves which lose games; create test
suites which focus attention on those critical points, and spur improvements in
Go programs.
Lastly, I'd sure love to see sponsors such as Sicortex and IBM offer their big
multi-computer rigs for bragging rights. "World's best bicycle-powered Go
computer" would make for a catchy ad campaign, right?
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
