Since the beginning of AI, mind games have been studied as relevant application fields. Nowadays, some programs are better than human players in most classical games. Their results highlight the efficiency of AI methods that are now quite standard. Such methods are very useful to Go programs, but they do not enable a strong Go program to be built. The problems related to Computer Go require new AI problem solving methods. Given the great number of problems and the diversity of possible solutions, Computer Go is an attractive research domain for AI. Prospective methods of programming the game of Go will probably be of interest in other domains as well. The goal of this paper is to present Computer Go by showing the links between existing studies on Computer Go and different AI related domains: evaluation function, heuristic search, machine learning, automatic knowledge generation, mathematical morphology and cognitive science. In addition, this paper describes both the practical aspects of Go programming, such as program optimization, and various theoretical aspects such as combinatorial game theory, mathematical morphology, and MonteCarlo methods. B. Bouzy T. Cazenave page 2 08/06/01 1.

Abstract- This paper describes the generation and utilisation of a pattern database for 19x19 go with the Knearest-neighbor representation. Patterns are generated by browsing recorded games of professional players. Meanwhile, their matching and playing probabilities are estimated. The database created is then integrated into an existing go program, INDIGO, either as an opening book or as an enrichment of other pre-existing hand-crafted databases used by INDIGO move generator. The improvement brought about by the use of this pattern database is estimated at 15 points on average, which is significant on go standards. 1

Introduction Retrograde analysis is a qualification and generation technique of positions in two-player, complete information games. In Chess [Thomson 1986], retrograde analysis enabled researchers to generate all six-piece positions [Thomson1996] and most of seven-or-eight-piece positions [Nalimov & al. 2000]. In Checkers, this technique is used by Chinook [Schaeffer & Lake 1996]. In Awari [Lincke & Marzetta 2000], Kalah [Irving & al 2000] and Nine Men's Morris [Gasser 1996], retrograde analysis was used to solve these games. In Go, [Zobrist 1969] [Benson 1976] [Boon 1989] [Berlekamp & Wolfe 1994] [Chen & Chen 1999] [Wolf 2000] have been the most important publications over the last forty years. But the complexity of Go played on usual boards (19x19, 13x13 or 9x9) forbids direct use of retrograde analysis. [Cazenave 1996, 2000] showed how to build small patterns with retrograde analysis on specific goals such as "eye" and "connection". So far, however, no study a

Some games have abstract properties that can be used to design admissible heuristics on moves. These admissible heuristics are useful to speed up search. They work well with depth-bounded search algorithms such as Gradual Abstract Proof Search that select moves based on the distance to the goal. We analyze the bene ts of these admissible heuristics on moves for rules generation and search. We give experimental results that support our claim for the game of AtariGo.

The strength of a program for playing an adversary game like chess or checkers is greatly influenced by how selectively it explores the various branches of the game tree. Typically, some branch paths are discontinued early while others are explored more deeply.

Three main problems arise with automatically generated rules databases. They are too large to fit in memory, they can take a lot of time to generate, and it takes time to match many rules on a board. I propose to use exceptions and metarules to reduce the size of the databases.

The main objective of this chapter is to present a comparative study of two techniques that automatically generate useful knowledge in games. Retrograde analysis of patterns generates pattern databases, starting with a simple definition of a sub-goal in a game and progressively finding all the pattern of given sizes that fulfill this sub-goal. Metaprogramming is based on similar concepts, but instead of generating fixed size patterns, it generates programs. Programs enable to represent knowledge in a more flexible way. However, they may take more time to use than pattern knowledge. We will describe the application of these two methods to the game of Hex, and compare their behaviors on this game. 1

Search is one of the fundamental methods in articial intelligence (AI). It is at the core of many successes of AI that range from beating world champions in nontrivial games to building master schedules for large corporations. However, the applications of today and tomorrow require more than exhaustive, brute-force search, because these application domains have become increasingly complex. Traditional methods fail to break the complexity barrier caused by the combinatorial explosion that characterizes these large, real-world domains. This thesis enhances our understanding of single-agent search methods. A puzzle (Sokoban) is used to explore new search techniques for single-agent search. Sokoban oers new challenges to AI research, because it has a much larger search space than previously studied puzzle domains and exhibits a new, real-world-like search-space property. Deadlock, the possibility to maneuver into an unsolvable position, provides traditional search methods with considerab...

Automatic acquisition of pattern collocations in GO