My guess is that the complexity of achieving a fixed standard of play
(eg 1 dan) using a global alpha-beta or MC search is an exponential
function of the board size. For this guess, I exclude algorithms
that have a tactical or local component. If this guess is correct
then, even if Moore's law remains in force, this kind of program
should not reach dan level on a 19x19 board within 20 years.
To some extent, this is testable today by finding how a global search
program's strength scales with board size and with thinking
time. For example, results in which Suzie had a week to play a 13x13
game would be interesting.
I don't mean to imply by this message that I think I am particularly
well qualified to have an opinion on this matter, but when someone
writes something that surprises me, I'm inclined to argue :)
On 13x13 and especially 19x19 Suzie is still weaker than Gnu-Go. I
think the hardware is still too weak to establish the same dominance
of search for larger board-sizes. But thats only a matter of time or
of a few million $ to build (with Chris Fant) a Go-Chip. Actually
about 100.000 Euro for an FPGA based project would be sufficient.
Chrilly Donninger
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
