I've entered the Waterloo university Google AI Challenge, hiding under a cunning nickname, since I saw that you could develop in Haskell, and a few others Haskellers are on it too! At the moment, I'm cruising at around the 110th position, which is not too bad given that I only worked a few hours at night on it (I have a day job, you know...) and that I implemented everything myself since the starter pack wasn't ready when I started working. I've played a bit with the standard strategies, implemented some flood fill, reuse some A* implementation I have.
All of this is interesting, of course, but there is no AI as such. The only intelligence in my code is the one I put in, and the strategies I devised. So I'm thinking of playing around with genetic programming and evolution techniques to try to get my code to find the right strategies at the right time. I have the feeling the trick will be to design something that is not too restricted and is truly able to evolve into some more advanced and complex than what I've designed myself. But hey, I'll learn, even if my program doesn't (-: