Summary and Perspectives

Abstract

Artificial Intelligence, since its very early stages has been highly focused on games. Usually described by a compact set of rules, with clear goals and well-defined terminal states, games provide cheap, replicable environments, perfectly suited for testing new learning methods and search algorithms.

Among various types of games, the mind games (especially two-player, perfect-information, zero-sum board ones) seem to be particularly tempting for AI. For many years, the ultimate goal of approaching and then surpassing humans in these widely popular and intellectually demanding tasks used to be another source of their popularity. The paradigm examples of games considered by AI were chess, checkers, Othello, and Go, which attracted game-related research groups and individuals all over the world. The human-computer competitions in the first three games ended with decisive victories of machines in 1997 (chess: Deep Blue), 1994 (checkers: Chinook) and 1997 (Othello: Logistello), respectively. Among the most popular mind board games only Go remains as the last stronghold of human supremacy, but recent advances of the Monte Carlo UCT algorithm suggest that the human predominance over machines in Go is likely to gradually diminish. Poker and bridge, widely popular card games are, besides Go, two more examples of games which are still not dominated by computer players. The recent AI/CI advances in both of them have, however, also been very telling.