Approximate Universal Artificial Intelligence And Self-Play Learning For Games