For those, who are interested in autonomous acting bots, i want to point to NEAT and rtNEAT.
rtNEAT is based on genetic algorithms that allow real time evolving of minimized neural network. This way you can play against this rtNEAT-bots, and they learn while playing. When you are satisfied with the result, you can save the neural network and use it later in your game.
NEAT is the non real time version.
Both are described in “AI techniques for game programming” by Mat Buckland, a very good and easy to read book in flat english.
Here are some links:
http://nerogame.org/ - a demonstration of rtNEAT
http://www.cs.ucf.edu/~kstanley/ - here you can find two java implementations of NEAT
http://nn.cs.utexas.edu/keyword?rtneat - rtNEAT