On Thu, 2007-07-05 at 09:53 +0200, Magnus Persson wrote: > A second way of formulating this is that there are no rules without > exceptions > in go. Thus for every pattern added one creates problems, and one > might > need to > discover those exceptions to get the benefits. And then there are of > course > exception to the exceptions... So when I think it is possible to > improve > playout infinitly I would also say it is really really difficult... > at > least by > hand which I am doing now.
I agree with you. As you add more knowledge it gets more and more difficult. I believe what happens is that your program becomes more "fragile" in the sense that even if it becomes stronger overall, it also develops idiosyncrasies that can be punished. Other programs could be specifically designed to beat them. The model of a human continuously adding knowledge and making a program stronger and stronger (without limit) seems dicey to me. That's why I believe it will eventually come down to some kind of machine learning system. In fact UCT (in my opinion) is a big step towards replacing tedious knowledge engineering with more clever techniques. Of course now it turns out that we are adding more knowledge anyway to increase the strength. But at least it's less intensively knowledge engineered than previous systems. I think trying to learn from human games is usually bad too, for similar reasons. I had at least 3 reasons why I think it's bad, one of them is simple what I call the omission problem, you don't really see (or sample) the reasons certain moves are or are not played. - Don _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
