A computer backgammon player that learned to play by playing against itself using Reinforcement Learning and Temporal Difference Learning. After some tuning it has turned into one of the very best backgammon players in the world. It was created by Gerald Tesauro of IBM.
His paper is available online at http://www.research.ibm.com/xw-D953/tdl.html.
Ed. note 2017: Now seems to be here http://researcher.watson.ibm.com/researcher/view_page.php?id=7021