Rin Kaiho also said a long time ago that he would need 4 stones from
god. So kudos to Rin.
Stefan
There was a study about 10 or 15 years ago that used the measured variance
in score to extrapolate perfect play (with zero variance), and it got 4
stones better than the top pros. That's where this estimate comes from.
David
-----Original Message-----
From: [email protected] [mailto:computer-go-
[email protected]] On Behalf Of Jacques Basaldúa
Sent: Monday, October 11, 2010 9:09 AM
To: [email protected]
Subject: [Computer-go] cgos 19x19 gets interesting
/ And that's the optimistic view: the usual wild guess is that the best
/>/ pros are about four stones away from perfect play.
Playing losing positions is tricky. The perfect move for w
minimax wise in handicap 4 is resign. So maybe accepting
that initially white loses by b_0 points and playing always
a move that keeps this minimax value expecting blacks
suboptimal choices to make b_i negative for some i is
probably not the best strategy. It is accepting: Ok i am
behind by (say) 45 points, lets build a solid 45 point loss.
We can imagine how much a human pro can read from what
Catailin Taranu explains from his own games in his
eurogotv.com videos. Humans narrow the search very much
an may foresee say 20 moves. (Anyone reads 20 moves in a
ladder I mean 20 moves in a fight.) A perfect player could
read 300-400 ply full width. Obviously, it could also
compute what humans will not see or may see. Rather than
perfect play, an aggressive overhuman 300 ply deep full
board tesuji could probably include killing the 4 handicap
stones for free. If perfect play means overhuman tesuji I
guess 4 handicap stones is too few.
Paradoxically, perfect evaluation can be a drawback
and minimax wise perfect play could be non-aggressive.
Of course, we can bet as high as we want because we will
never know.
Jacques.
/
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
