Re: computer-go: Applying Moore's Law to Computer Go

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

OK.  I think there will continue to be steady progress.  The commercial
success of go programs in Japan and Korea means there is enough money
that there will be more full time teams, so programs should continue
to improve at a steady rate.  I think the strong programs will
continue to rely on hand tuned knowledge rather than learning.  I think
the basic algorithms for go programs are understood, just not published :)

I think that 100,000 is much closer the number of (mostly unconcious) rules
that a pro uses than a few hundred.  A pro sees the best or near best move
at a glance, and does detailed reading to verify it.  It takes lots of rules
to do this accurately.  You might think of them as the thousands of exceptions
and special cases of the few hundred general rules.  

I've talked to Mr Yang about computer go and I think he would interested in
working on a program, but I can't afford him.  He also thinks there are
a small number of rules, but I've talked to him about some of them, and as
I code them, one of his rules turns into a very large number of program
rules :(

I predict IGS 5k by 2010, IGS 2d by 2020, pro 1 dan by 2030, and pro 9 dan
2040.

David

At 06:27 AM 11/16/99 -0800, Stuart Cracraft wrote:
>
>>i don't think it will be a big idea, i think it will be a bunch of rules 
>>like the above (maybe a few hundred) that get weighed depending on pom.
>
>Likely to be the above plus a tactical analyzer and some form of learning.
>
>David Fotland: would you speculate on the future of Computer Go?
>
>
>