"What's more striking about the latest generation is the appearance of unscripted, emergent behavior - the AI stumbling on new ways of responding to the world, strategies and behaviors that weren't deliberately planned by the designers."
- Steven Johnson
Games have a long history of being used as a testbed for artificial intelligence, since they present many useful properties. They have clearly defined states and actions, goals and rewards, and often a strong set of human opponents that allow us to pit artificial intelligence against human intelligence. Early research on game AI used games such as chess and checkers, which are both two-player deterministic perfect information games: games where there is no random chance and no hidden information. Later work studied games with other properties, such as backgammon (random chance), hearts (multiple players), and poker (hidden information). Video game domains present new challenges for artificial intelligence, and have become a new focus for research.
Artificial intelligence research on video games can take several forms:
- Strategic Reasoning. Similar to earlier research in games like chess and checkers, there is the challenge of designing a program that can play a game well, using the same information and resources that humans use. Starcraft is a real-time strategy game in which the players have to collect resources, build an army, and defeat their opponents, while struggling with imperfect information and deep strategic decisions. The Starcraft AI Competition lets researchers and programmers play their programs against each other, to find out which artificial intelligence techniques work well in practice.
- Player Entertainment. Another challenge for artificial intelligence is to find ways to make games more entertaining for the players. For example, this can include providing believable non-player-characters, interactive storytelling in which the game's plot reacts to the player, or automatically creating content.