Games and AI: Paths, Challenges, Critique

Abstract

In 2006, the first-person shooter F.E.A.R. makes headlines in the gaming world. One feature in particular attracts much attention: the non-playable characters seem to behave intelligently to a degree yet unseen in computer games. From earlier productions like No One Lives Forever 1 & 2 (2000, 2002), players were already familiar with NPCs that are able to seek cover under fire and to leave it at random in order to shoot back at the player. In F.E.A.R. that happens too, but in a much more realistic manner. Computer-controlled enemies attack players in a coordinated way. If one member of the enemy team comes closer, he gets supportive fire by his team members. If the player attacks them, enemy forces remain in cover until they are immediately threatened.

Ten years later, an AI system called AlphaGo beats the human world champion Kim Sung Yong in the ancient board game Go in five rounds—final score: 4-1. The global community of Go players is perplexed, almost shocked, even though the victory did not totally come out of the blue. Already in October 2015, an earlier version of AlphaGo was able to beat the European Go champion Fan Hui. However, Hui’s playing level was significantly lower than that of Kim Sung Yong (2-dan out of possible 9-dan levels).

As these introductory examples illustrate, the relationship between artificial intelligence (AI) and games can basically be studied from two perspectives: The first is the implementation of AI technologies in games, in order to improve the game experience in one way or another, for example with the intention to make it more believable, more immersive, or simply more enjoyable. The second is the use of games as a benchmark, a learning or test environment to evaluate, but also demonstrate, the current state of AI technologies. Both perspectives have gained enormous importance in recent years—technically, but also culturally and economically.