OK, as a part-time researcher I have thought about computer AI also.
I have been thinking about Artificial Neural Networks (ANN), specially Kohonen Self Organizing Maps (SOM) would be interesting.
The problem is just what the AI should be able to know and how to handle missing knowledge. But oh boy it would be fun to make a real SOM that adapts over time to your playstyle. And so forcing you to improve your play over time too.
Finally we then could train our own ANN A.I.'s first and put a competition on who trained the best one
In neural networks it it's easy to make different ANN's by just having adifferent starting values. (One could even have nationdifferenses ala SMAC)
Damm, why wasn't math my major so I could do this myself...