[ Please note: The project has been discontinued as of May 31, 2005 and is superseded by the projects of the ii Labs. There won't be further updates to these pages. ]

(Related publications: [PUBLink] [PUBLink])

The use of game applications has a long tradition in artificial intelligence (AI). Games provide high variability and scalability for problem definitions, are processed in a restricted domain and the results are generally easy to evaluate. Furthermore, the "AI inside" feature is a high-priority task [PUBLink] [PUBLink] [PUBLink] in the fast-growing, multi-billion-dollar electronic gaming market (the revenue from PC games software alone is already as big as that of box office movies [PUBLink]).

Many "traditional" games, such as card/board/puzzle games like Go-Moku [PUBLink] and the Nine Men's Morris [PUBLink], have recently been solved by AI techniques. Deep Blue's victory over Kasparov was another milestone event here. However, the success of many game programs can mainly be attributed to the vast increase in computing power - many researchers using exhaustive-search rather than knowledge-based methods [PUBLink]. While these applications bear impressive witness to advances in hardware, no noticeable scientific contribution has been made and such advances are of little help in solving much more complex real-world problems. Results like Go-Moku is a win for Black or random instances of Rubik's Cube can be solved optimally are not really applicable to other areas. Of the techniques used in this field of research, A* [PUBLink] (an improved version of Dijkstra's shortest-path algorithm [PUBLink]) and its variants/extensions are practically the only ones employed in today's "modern" computer games (see also [PUBLink]).

Such games pose problems for AI that are infinitely more complex than those of traditional games. Modern computer games are usually played in real time, allow very complex player interaction and provide a rich and dynamic virtual environment. Techniques from the AI fields of autonomous agents, planning, scheduling, robotics and learning would appear to be much more important than those from traditional games.

AI techniques can be applied to a variety of tasks in modern computer games. A game that uses probabilistic networks to predict the player's next move in order to speed up graphics may be on a high AI level. But although AI must not always be personified, the notion of artificial intelligence in computer games is primarily related to characters. These characters can be seen as agents, their properties perfectly fitting the AI agent concept.

But how does the player of a computer game perceive the intelligence of a game agent/character? This question is dealt with neatly in [PUBLink]. Important dimensions include physical characteristics, language cues, behaviors and social skills. Physical characteristics like attractiveness are more a matter for psychologists and visual artists (e.g., see [PUBLink]). Language skills are not normally needed by game agents and are ignored here too.

The most important question when judging an agent's intelligence is the goal-directed component. The standard procedure followed in today's computer games is to use predetermined behavior patterns. This is normally done using simple if-then rules. In more sophisticated approaches using neural networks, behavior becomes adaptive, but the purely reactive property has still not been overcome.

Many computer games circumvent the problem of applying sophisticated AI techniques by allowing computer-guided agents to cheat. But the credibility of an environment featuring cheating agents is very hard to ensure, given the constant growth of the complexity and variability in computer-game environments. Consider a situation in which a player destroys a communication vehicle in an enemy convoy in order to stop the enemy communicating with its headquarters. If the game cheats in order to avoid a realistic simulation of the characters' behavior, directly accessing the game's internal map information, the enemy's headquarters may nonetheless be aware of the player's subsequent attack on the convoy.

For questions, comments or suggestions, please contact us.

Last update:
May 19, 2001 by Alexander Nareyek