|
BGonline.org Forums
Computer go vs. computer backgammon
Posted By: Timothy Chow
Date: Saturday, 1 August 2009, at 1:07 a.m.
Recent progress in computer go has come from a somewhat unexpected direction: Monte Carlo tree search. In its crudest form, the algorithm for evaluating a position is simply to play it out to the conclusion of the game, choosing moves uniformly at random. This algorithm sounds like it could not possibly be any good, but it performs amazingly well. Of course, to get a program that plays really well, modifications need to be made to this simpleminded procedure. Nevertheless, the basic philosophy of shunning "intelligence" in favor of greater numbers of trials underlies the strongest go programs today. MoGo, for example, has beaten 9-dan professionals with only a seven-stone handicap; such playing ability was unheard of in pre-Monte-Carlo days. For more information, search on "mogo monte carlo" or see the Wikipedia article on computer go.
What I'm wondering is whether these ideas from computer go can be applied to computer backgammon. Has anyone taken a serious look at this possibility?
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.