On Wed, Jun 2, 2010 at 5:25 PM, Jeff Nowakowski <[email protected]> wrote:
> On 06/02/2010 01:14 PM, Don Dailey wrote: > >> >> Why are you comparing humans to computers? It's ridiculous to measure >> progress by comparing to the top human players. What we care about is >> how >> much progress we can make from year to year. >> > > Come on Don, you know that the top players are the gold standard that the > programs are trying to beat. That's why lots of programmers have been moving > to Go from chess. Equaling humans is an arbitrary and very long term goal, but it should not define how "hard" or "easy" something is and should not define what "works" and what "doesn't work." That's my beef. Computer chess still continues to progress at an amazing pace, and it's not because we are trying to equal humans. > > The original question was why is Go harder than chess for computers, as it > clearly is (are you disputing this?). How can I dispute it, you do not define terms? Break it down, when you say "Go is much harder than Chess" what does that even mean? 13 is much harder than 27. What does that supposed to mean? Dogs are harder than Cats. If it means it's harder to write a go program, then I completely dispute it - I can write a go program much faster than a chess program. Both program will play very well if I define "very well" to be random play. My point is that hard or easy is an arbitrary concept and I don't believe there is any reason to believe that how humans play is some kind of sacred yardstick. If you make that the yardstick, then GO comes up short. I think Steve answered the question very well in his original reply. > The only problem is that I was answering Dave, not Steve. > I understand your dispute with Dave's simplification, but I think you can > agree that alpha-beta for chess was a strong foundation to build on, that it > didn't work for Go, and that we now have monte-carlo tree search as the > foundation. Nobody disputes Go has made progress and will continue to do so. I'm not convinced that alpha/beta is any different fundamentally from any other kind of search including MCTS. It's surprising how in science things often turn out to be similar or just different forms of the same thing. I don't know what you mean when you say "it didn't work for Go", that is just not true. There are some alpha/beta go programs that are much stronger than they would be without search. So that is just a completely ignorant statement. It definitely does "work", it just doesn't meet some arbitrary expectation. MCTS "doesn't work" either by this definition because it does not produce play equal to the strongest human. _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
