Multi-game Playing

Abstract

One of the ultimate goals of AI/CI research in games is development of a universal playing agent, able to play virtually any game as long as it knows its rules. Realization of this task requires designing general-purpose learning and reasoning methods that abstract from particular games. A potential variety of possible games to be played (theoretically there are infinite number of them) with different equipment (boards, card decks, various types of moving pieces, different goals, etc.) makes the task - in its general form - extremely demanding. Therefore, on a general level the goal is yet far from being accomplished, however, some steps in this direction have already been successfully taken.

Multi-game playing became a hot topic in the beginning of 1990s when some renowned methods and systems originated. In order to make the problem tractable, these first attempts were often restricted to a certain class of games, usually two-player, perfect-information, zero-sum, deterministic ones.

The majority of approaches developed in this area rely on symbolic, logical game representations, rooted in the mainstream of traditional AI. Relatively fewer examples can be found within the CI-related methods. Selected attempts, which originated in both AI or CI areas are briefly summarized in this chapter.