Re: computer-go: perfect play

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

   From: "Johan Roos (QCS)" <Johan.Roos@xxxxxxxxxxxxxxxxx>
   Date: Wed, 27 Sep 2000 09:47:49 +0200
   Sender: owner-computer-go@xxxxxxxxxxxxxxxxx
   Reply-To: computer-go@xxxxxxxxxxxxxxxxx

   A thought about perfect play

   A perfect evaluation function does it for GoGod but not for GoDevil.
   In situations where there are equal moves to chose from GoGod doesn't care but GoDevil will chose the one with
   that gives the opponent the greatest probability to chose wrong in the future moves. The thought that GoDevil adjust his play to the opponent applies againt GoGod also. Therefore a game between GoGod and GoDevil will be different than GoGod against GoGod. The really interesting game would be GoDevil against GoDevil ;-)

   /Johan



Since GoDevil  adjusts his play to  confuse the opponent, then I would
argue that since there is no chance  of confusing GoGod, there will be
no payoff in playing any differently than GoGod would.  In other words
GoDevil vs GoGod should look like GoGod vs GoGod.

Of  course there are many  variations on how to mix  up your moves and
what your assumptions are about your opponent.  I assume GoDevil knows
everything about his opponent and can basically read his mind.

Don