The General Game Playing (GGP) competition was introduced by Michael Genesereth and Nathaniel Love of Stanford University in an AAAI 2005 paper (dated March 9, 2005) and run as an open competition at that year’s AAAI National Conference. Its premise was a direct critique of how AI had been pursuing games. As the paper put it, specialized players such as IBM’s Deep Blue are “very narrow” - Deep Blue beat the world chess champion but “has no clue how to play checkers; it cannot even balance a checkbook.” Worse, most of the cleverness lived in the programmers, not the program.
GGP flipped the rules. A general game player receives a formal description of a game it has never seen - written in the Game Description Language (GDL) - at runtime, and must work out how to play it well with no human intervention and no game-specific code. To win, a system could not be pre-tuned for chess or Go; it had to bring general capabilities like knowledge representation, reasoning, and rational decision-making and integrate them on the fly. The competition offered a $10,000 prize and required entrants to handle games ranging from tic-tac-toe to complex multiplayer settings with complete or partial information.
GGP matters because it targeted generality - the ability to handle unfamiliar problems - which the authors tied directly to the long-range goals of artificial intelligence. The idea anticipated today’s interest in general-purpose agents and foundation models that adapt to new tasks rather than being rebuilt for each one. For a general reader, it is an early, clear statement of an enduring distinction: a system that masters one game is impressive, but a system that can learn any game from its rules is closer to intelligence.