by Rémi Coulom. Invited talk at the University of Electro-Communications, Tokyo, and the 12th Game Programming Workshop'2007, Hakone, Japan.
The game of Go is a difficult and exciting challenge for artificial intelligence research. Today, the best Go-playing programs are still far from the strength of the human masters. Nevertheless, machines have made very spectacular progress in the recent years thanks to the emergence of a new technique called Monte-Carlo tree search. In this lecture, I will explain the principle of Monte-Carlo tree search, and how it was used to build a strong program, Crazy Stone. I will then comment some games played by Crazy Stone in order to demonstrate the playing style of Monte-Carlo programs. From this analysis of the strengths and weaknesses of the Monte-Carlo approach, I will conclude by suggesting some direction for further research.