[computer-go] Re: computer-go Digest, Vol 30, Issue 26

[Jacques Basaldúa]

Arend Bayer wrote:

 > . .  without ever believing anything that some of the strong go players
 > (some a lot stronger than me) have to say.

Please, don't think that. I am sure there is more people in this list who,
like myself, do not think computer go will "do it" through global search
only. The opinion of strong go players is always welcome and read with
extra attention.

I think there is a qualitative difference between go and chess, even if
mathematically they both belong to the same class of games and many
proved conclusions apply to both. The difference is: Chess is "feasible"
by global  search and go (19x19) isn't and will probably be "un-feasible"
for a long time.

(I use feasible and "do it" = "program beats the best human on Earth")

I don't want to disregard the talented people who, like Chrilly, did a
great job on chess, but the fact that "global search chess" was feasible
may have slowed down the development of "intelligent computer chess",
o call it "humanlike computer chess" if you prefer. I.e. a program that
can _reason_ about a position instead of examining millions of nodes with
a by-itself poor evaluation function.

I see the fact that "global search go" is not feasible as *good news*.
Because we cannot beat the best human by "global search" we are forced
to find out something better.

I agree with Don in all he states in this thread, but that applies to
global search. If a go program playing at move 150 has fully searched
the 20 only local games in the board (18 of which did not change since
move 148) and _understands_ :

  a. territorial values of them
  b. territorial interactions between them
  c. conditional relations between them
  d. threat/win sequence that makes optimal use of sente
  e. ko-threat stock management related with foreseeable kos

And from a + b + c + . . . + j + k + l decides its best move,
... what can it do with extra time ?

Jacques.