i think that using Neural Nets will be overdoing it

Is there another method that will come close to what I was trying to achieve?

Have you read the book Blondie24? There are a number of publications from Fogel (author of Blondie24) which detail the process in a more technical arena. Essentially, the goal was a NN which could learn to play checkers. They used a standard game tree approach which went to an arbitrary depth for its search and the NN was used to assess the various nodes of the game tree. For training, they played NNs against each other and used an evolutionary computation approach in which the fitness was number of games won - kind of a reinforcement learning approach. The end result was a checker player that ranked at the master level against human and AI players. Fogel has recently extended this to chess with very good results.

While it is a bit out of context for you situation, such an approach might work. For purely experimental reasons, it sounds like an interesting idea, but certainly a non-trivial process.

-Kirk

Sounds interesting, might have to obtain the book. Thanks.

It's important to keep in mind that any tweakable algorithm may be made "learnable". There's very little that's intrinsic to NNs that allows learning... in fact, one of the reasons terms like "backpropagation" exist is that it's more difficult for NNs to learn than it is for more domain-specific algorithms to learn. All that has to happen is for different parameters to be attempted, and the results of the strategies ranked and used for future parameter decisions.

Sneftel - what would you recommendation for an algorithm be?

TBH, my expertise in limited-information games is, well, limited. If you can afford it, I recommend this book, which will tell you more than you ever wanted to know about AI.