Author: Roy Eassa
Date: 12:53:01 02/12/04
Go up one level in this thread
On February 12, 2004 at 15:21:58, Dann Corbit wrote: >On February 12, 2004 at 15:12:18, Roy Eassa wrote: > >>In December, I sort of got into the game of Go. Part of its fascination for me >>is how it differs from chess with regard to creating a strong program. Methods >>based primarily on tree searching apparently do not work in Go, for 2 main >>reasons: first, the branching factor is much, much higher than in chess, and >>second, once you've arrived at an end-node (position) to evaluate, there's no >>known method of coming up with a reasonable evaluation. >> >>[Thus I had the perfect opportunity to create a Go-playing program that was >>very, very weak yet might still beat me because I too was very, very weak >>(having just re-learned the rules after over 25 years). But, alas, I studied Go >>fairly intensely for 2 weeks and now I'm strong enough (but still weak!) that >>I'd probably have great difficulty creating a program from scratch that can beat >>me consistently.] >> >>Some people say that no Go program will be stronger than the best humans for at >>least another century, if ever! Today's best Go programs are far weaker than >>any Go professional. > >Hogwash. Computers will have more compute power than the human brain long >before a century passes. > More power by some measures, perhaps, but a lot of very smart people buy into the idea that Go will never be conquered. I am not one of them, but I do have doubts about the likely success of brute force. I would love nothing better than to see the "mystical" quality of Go to be brought down to earth. Wanna take a shot? >>Anyway, has anybody here ever tried writing a Go-playing program? I think it's >>a field ripe for a "breakthrough" -- a completely new approach from those that >>have been tried. I also think the Go programming world will make a small number >>of people a LOT of money at some point, as it lacks the monstrous presence of a >>Chessbase -- a powerhouse company that sells a GUI with every conceivable >>feature and sets the standard protocol for (nearly) all commercial playing >>engines. There are a few decent GUIs and a few decently strong engines >>(compared to weak amateur humans), but there's nothing even remotely close to >>Chessbase (Fritz, etc.) for features or standardization. Every top programmer >>uses his own GUI. Sort of like chess 12-15 years ago, in some ways. >> >>I think some smart person will create a full-featured GUI for Go that uses a >>"plug-in" architecture for playing engines, then negotiate with all the top >>authors to adapt their programs to that architecture, thus making a lot of money >>without having to write a strong engine themselves. Further, I think that >>having a standard plug-in architecture for testing one's Go engine will prompt >>many more people to create Go engines, thus increasing competition exponentially >>-- increasing the chance for a breakthrough. >> >>Two additional points: >> >>First, I think it's quite a bit easier to create an engine that plays Go legally >>than it is to create one that plays chess legally. Even doing the GUI yourself >>isn't too hard, since it's mainly just black & white stones on a grid. That >>should encourage more people to try. >> >>Second, I realize that the market for Go products is pretty tiny in the Americas >>and only medium-sized in Europe. But I think there's a HUGE market in Asia, >>which can only grow as Asia's economies continue to grow quickly. Thus I think >>that any product created would have to be marketed to Asia (primarily?) for big >>financial success. >> >>Bottom line: I think there's a huge gap in the market that SOMEBODY will get >>rich from at some point in the not-too-distant future. And Go is a pretty >>interesting game, even though "chess" is considered a dirty word to many serious >>Go players. > >Here is a port of GnuGo 3.4: >ftp://cap.connx.com/chess-engines/new-approach/gnugo-34.zip > >It definitely searches a tree and it is pretty strong. I've had it from day one. It's not bad. And others are good enough to usually beat me too. But that wasn't the issue. The issue is beating the really strong Go-playing humans -- called professionals -- that are the equivalent of chess masters. They play at a whole different level and no program comes even remotely close to beating them. The best programs are still several stones weaker than even the best amateur Go players, and the brute-force approach does not seem to be making much of a dent. Even multiplying the CPU speed by 1,000,000,000 would probably not help much with brute force. That is, until somebody (maybe you?) comes along with a better approach.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.