CGDC '97 Roundtable Report

In the roundtables I moderated, one common theme seemed to be present in all sessions, and that is that "less is more" with regard to developing artificial intelligence (AI) in computer games. This theme was relative to developers who felt that the human player (HP) would often attribute "unplanned and unprogrammed" behavior and intent to the actions of the non-player characters (NPCs) that were being controlled by computer game AI. One often cited example of this was the monsters in Quake running away from the player at times, giving the impression that the monster was afraid of the weapon in use by the HP, despite the fact that the AI programmer was doing nothing of the sort, but instead was only randomly selecting a direction for the monster to move.

Another key concern discussed, was how to use cheating to avoid "artificial stupidity". Many AI developers revealed that using cheating by the AI, was acceptable when used to make sure the NPCs did not act stupid. So providing the NPC with the knowledge of a level, that would be used to assist in pathfinding through that level, despite the fact that an HP might not have that complete knowledge of the level, was an acceptable form of cheating.

In three sessions, many, many more AI related topics were touched on. Just some of them were uses of emergent behavior to produce more complex behavior, uses of artificial life considerations to make NPCs more life-like, getting the individual NPCs to cooperate in behavior and give the impression of a higher level of control, strategic planning for NPC tactics that can evolve with contact with the HPs, the use of scripting to facilitate the control of NPCs, and the usage of the "soft AI" techinques of neural networks and genetic algorithms in computer game AI.

An additional, cross-session topic, queried by all of us AI RT Moderators, was a poll of the group, to identify the CPU usage that AI gets in individual computer games, and the committment to developing "good AI" that computer game designer/developer/publishers were making by dedicating a programmer exclusively to AI issues. The responses suggested that the CPU usage was very game genre specific, and ranged from 1% (Enemy Nations) to 25% (X-Com 3) with most of the 3D action games using about 8%-10%. The responses further indicated that there were few dedicated AI programmers. However since before in the past, there had been none, and now there were some, that a possible trend towards making "good AI", by dedicating a programer to AI (like dedicating a programmer to 3D graphics) was emerging in the industry.