Saturday, September 25, 2010

[vknofaoh] Learning exciting chess

The first version of a computer chess program is given many, many positions to evaluate with MultiPV, and filters for only those positions with two (or more) "best" moves both having the exact same evaluation score.

Present those options to a human who is directed to select which seems more "exciting" (or some other aesthetic property), and then apply machine learning to discover the features of a move which determine "excitingness".  Give a personality to a second version of the computer chess program, while being just as strong as the first.  People like exciting opponents.

No comments :