14.04.2007 17:18, David Richerby:
As you say, for checkers, it results
in a weaker program than can be written by ordinary means, as one
would expect from applying a general-purpose technique to a specific
problem.

It's not sure, that this is indeed the reason for its weak play compared
to Chinook. There are a lot of techniques to optimize the search which
are not used by Blondie24. So, it might be interesting to take an
existing, strong program and replace its evaluation function with such a
neural network as in Blondie24 and optimize it in a similar way - and
then see how it performs compared to the original program.

Chinook was written by an expert author of chess programs, who knew
barely anything about checkers when he started. He asked a checkers
expert for support with the evaluation function, and after about two
months of adapting the chess program to checkers they were clearly
superior to any existing checkers program. I doubt that the evaluation
function was so sophisticated. The search techniques used by the chess
program made the real difference.

Greetings,
Ralf