Regarding test protocols: I strongly suspect that we'll need to harvest test
cases from computer games, run them past high dan-level players, and find out
what it takes to handle such cases well while still beating other programs.
One of the problems with playing imperfect programs is that one can learn
sub-optimal "trick" plays which work only because the programs persist in
making the wrong reply; better programs would punish such trick plays; that's
why we call trick plays "overplays." But in the environment of not-very-good
programs, overplays look good; they programs are pursuing local optima, not the
globally best play.
Terry McIntyre <[EMAIL PROTECTED]>
We must stop dressing up the slaughter of foreigners as a great national cause.
-- Sheldon Richman
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
