General Game Playing – towards a truly multigame player

Jacek Mańdziuk (Faculty of Mathematics and Information Science)    J.Mandziuk >at>

Description: One of the most interesting areas of contemporary research on application of Artificial Intelligence (AI) and Computational Intelligence (CI) to mind games is the topic of multigame playing, i.e. development of agents able to effectively play any game within some general category being informed only about the rules of each of the games played. This poses a unique challenge to the AI community, as all the most successful game playing agents to date have been developed to achieve master level of play only in their specific games. Creating a system exhibiting high playing competency across a variety of previously unknown games would be a significant step in CI/AI research.
General Game Playing (GGP) is one of several approaches to the multigame playing topic. It was proposed at Stanford University in 2005 ( in the form of General Game Playing Competition held annually at the National Conference for Artificial Intelligence. General Game Players are agents able to interpret game rules described as a set of Game Description Language (GDL) statements in order to devise a strategy allowing them to play those games effectively without human intervention. The competition always includes a wide variety of games, both known previously and devised specifically for the tournament. Contestants should be prepared to deal with games of various complexity, varied branching factors and numbers of players, both cooperative and competitive. Game Description Language (GDL) is used to describe the rules of the class of games playable within the GGP framework, i.e. finite, discrete, deterministic multi-player games of complete information. GDL describes games in a variant of Datalog. Game states are defined in terms of facts and algorithms for computing legal moves, subsequent game states, termination conditions and final scores for players are represented as logical rules.
The goal of the research is analysis of possible AI/CI approaches to GGP and development of CI-based GGP agent capable of taking part in the GGP contest.
You are here: Home Projects General Game Playing – towards a truly multigame player