From: fotland@cup.hp.com (Dave Fotland)
Subject: computer go vs. computer chess

: Go is a rather well-described game, but there has been very
: little effort put into programming for it, when compared
: with the effort (decades of it) put into programming for
: chess.  Why?

I don't think that this is true.  The first full go playing programs were
written in the 60's.  Bruce Wilcox has been workingon his program(s) since
the early 70's.  I've been working on Many Faces and predecessors since
1981.  I think this qualifies as decades.  There has been significant work
in Asian universities - the 1986-89 world championship program was written
at a Taiwanese University.  The Japan 5th generation project (remember that?)
had a large effort devoted to computer go.

I think the main reason you don't see more go programs is that go is so much
harder.  I've written a computer chess program, and I'm sure I could put
together a chess program that plays at the 1500 to 1800 level in a few months.

Many of the strong chess programs in the past such as Chess 4.0 were written
in just a few months.  This is not possible for a go program if you want
something reasonable.  Getting reasonable play in just a few months for
Chess is very encouraging, so many groups of programmers continue to
participate.  Working a go program for several years and still having 
something that plays horribly can be very discouraging, so most people 
drop out before they get to a decent program.

In a university setting, Go is generaaly too hard for one person to make
much contribution as part of a master's or Ph.D thesis (with a few
notable exceptions).  The strong programs have more years of effort put
into them than most people would be willing to devote to a university
project.

It's therefor very natural that University research is focussed on the
easier game of Chess.

: Asia, in particular Japan, which has done so well in
: hardware, has had a disappointing track-record in the
: entire software development industry (the reasons for
: this are not relevant, here, but the simplest is that
: the popularization of computers all but ignored the
: Asian alphabets until the mid-80's).  It's only natural
: that Asian software talents would be occupied in solving
: these problems rather than in teaching machines to play
: a board-game.

Japan's computer go effort was significant, with several commercial
programs available there.  It is the biggest market for computer go
software.  Maybe, since these programs have Japanese language interfaces
and are not sold in the US, you are not aware of them.  Codan, a Japanese
program was world champion around 1990.  The 5th generation project
program was as strong as Many Faces.  A Chinese program, Handtalk, is the
current world champion.  I don't think that Asia is behind the world in
computer go results.

: Now, the critical mass *may* exist, but the problem has
: become intimidating.  There exist among the people who
: play on IGS a fair number who are capable of doing for
: Go-playing programs what was done for Chess-playing
: programs in the '70s, but they aren't doing it.

I don't beleive this.  Chess programs benefited greatly from the vast increase
in computing speed in the 70's and 80's.  I don't think there were any
huge breakthrus in algorithms.  Unfortunately, faster computers don't help
go like they help chess, since wide+deep searches are not an option for
go.

I don't think we will see a really strong go program until someone figures
out how to make one learn effectively.  Explicitly coding in the go
knowledge (as is done with all existing string programs) is just to
slow and time consuming.

: ManyFaces is an excellent start, but where are the other
: six Samurai?

It's too hard.  Few people have the perserverence to complete a decent
go program.  As more and more years of effort have been expended in the
existing programs, it becomes harder and harder for anyone to bring
a new program up to that level of play before they quit in frustration.

Computer go is not just a little harder than computer chess.  It's 
incredibly, almost unbeleivably harder.  Go programs are weak because 
of the nature of the problem, not because there hasn't been much effort 
expended on them.

--
David Fotland    fotland@cup.hp.com