On Mon, 2007-07-09 at 15:52 -0700, David Doshay wrote: > I agree. I really do not understand why using domain knowledge would > be a problem for some people. It seems to me that this opinion is held > by programmers with less Go knowledge who hope that pure search is > an answer. Knowing what I do about Monte Carlo sampling (it was the > subject of my Master's thesis in physics), computer Go (a mere 4 > years) > and playing Go (almost 30 years) I cannot imagine that progress will > be > made without a great deal of domain knowledge. > > The UCT method has clearly brought us a big leap in playing strength > together with bringing MC methods back into the mix. I think that the > next big trick will be in getting the domain knowledge into the form > that the MC methods can use efficiently.
I don't really have a problem with domain specific knowledge, I'm sure it will be required and will be a benefit. I'm just thinking as a purist. Any human generated GO knowledge, in some idealistic sense of the word is WRONG. In other words, unless you can formulate perfect rules, you are introducing prejudice to the search engine. So knowledge will help until you get to some point that it gets out of focus, starts conflicting with itself. This is why I believe that ultimately, go knowledge will end up being computer generated because humans will not be able to tune it. You will get to a point that if you add knowledge the program gets weaker. That would not happen if you knew how to add knowledge that is completely accurate. Of course some knowledge is. Benson life is probably more or less flawless knowledge for instance. But even eye rules which seem almost perfect can inhibit a program. The eye rule all the MC program use could not be used to create a perfect player because the rule is flawed. However, the rule is so good right now because it prevents so many mistakes. A good example of a rule designed to benefit a program which at some point outlives it's usefulness and must be replaced with a more perfect rule. Now when you get to hand crafted patterns, you are really going to screw up the program ultimately. Very few if any are as good as the eye rule. So you are basically engineering a program for failure! Of course I admit this is a long way off - most rules are good more often than bad and will take your program a long way. But that's why I find one approach pure, beautiful and elegant and the knowledge engineering approach crude and hackish because ultimately it really is a shortcut. Not a clean shortcut but the kind of shortcut that gets you into trouble eventually, like not debugging your program because your in a hurry or something. It's a very appealing shortcut because it gets you somewhere very quickly, it just doesn't take you all the way! Having said that, Lazarus has tons of GO knowledge and patterns. I took the shortcut! - Don > Cheers, > David _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
