A new set of algorithms are being developed that combine reinforcement learning such as Q learning and SARSA with neural networks to provide much more sophisticated and adaptable AI.

Here are some papers and implementations:

JavaScript implementation: http://cs.stanford.edu/people/karpathy/reinforcejs/waterworld.html https://github.com/karpathy/reinforcejs http://cs.stanford.edu/people/karpathy/convnetjs/demo/rldemo.html

Google research into Atari game playing AI http://googleresearch.blogspot.co.uk/2015/02/from-pixels-to-actions-human-level.html http://deepmind.com/publications.html

Many thanks.