Application Stream: Invited Keynote Lecture

Artificial Intelligence for Computer Games

Professor Peter Astheimer
IC CAVE
University of Abertay Dundee
p.astheimer@abertay.ac.uk

Artificial intelligence (and in fact any other technology used to realise a computer game) has to support gameplay. Gameplay is jargon for and summarises a number of crucial characteristics which are essential for a good and successful game. The most important characteristic of gameplay is to keep the player engaged at every moment e.g. through a compelling story or a skills challenge. Competition and cooperation with other players, working towards a goal, facing a challenge which can be mastered at some time, having fun and making the player curious to “look around the next corner” are other important characteristics of good gameplay.

Examples, where artificial intelligence concepts can make a significant contribution to improve gameplay are:

• Non-player characters (NPCs): Today, NPC behaviour is very easy to predict for a human player. NPC simulation quite commonly relies on cheating, e.g. using knowledge of the whole environment and location of other NPCs and players by “x-ray” vision. There is a need for NPCs to be more realistic and act like human players, to flexibly follow strategies, to act in unpredictable ways and thus be much more interesting than currently available NPCs. In order to achieve this, the equivalents of human senses (see, hear, smell, taste, feel), a brain, memory, emotions, moods, needs, goals, skills and the ability to learn have to be realised. In the game “Thief” the implementation of a crude visual and aural sense is the foundation for and point of the whole game.

• Pathfinding: This is the classical application of some AI research in a games context and for a long time was synonymous with AI in computer games. Although progress has been made, there is still room for improvement, e.g. when the dynamic paths of two groups of opponent NPCs cross.

• Strategy & tactics: Most game strategies are pretty obvious and thus easy to predict and counteract. A flexible selection and adoption of strategies to the current situation in a game is needed. Commanding an extensive number of NPCs can be an arduous task, involving just too many mouse-clicks. A hierarchy of levels of command can improve gameplay considerably.

• Opponents and balance: Most games, especially when there are a large number of parameters, are very hard to balance so that there still is a challenge but none of the two sides have an implicit advantage. A prominent example is the game Command & Conquer, with the first and simple version very well balanced and received by players, but all later more visually sophisticated and complex versions unbalanced and unsatisfactory for players.

• Level of detail: AI algorithms especially dealing with extensive environments and large numbers of NPCs can put more demand on the CPU than available. Games like “Revolution” have attempted to introduce a level of detail AI, which tries to achieve the best results in a given time.

• Artificial user: Usability evaluation of games concepts and prototypes is an important step to ensure player satisfaction. As tests with players can be quite costly and take time to carry out, an artificial user, scalable in terms of age, gender, culture etc. to be used on a routine test basis would improve gameplay and market success considerably.