Analyzing the believability of game character behavior using the Game Agent Matrix

In recent years there has been significant improvement in the simpler actions performed by characters in computer games – such as navigating the world and attacking enemies and similar actions. In previous work, the ability of NPCs to adapt to changing circumstances was found to be inadequate in many circumstances. In order to validate these findings we have studied a total of 20 games, observing NPC behavior in each of the games in many different situations, ranging from everyday town life to combat. Using the Game Agent Matrix, we found a number of different behavior categories related to the social context of the agent and its behavior within that context indicating a gap between the most convincing behavior was focused around navigating the world, using tools and using language, as well as more complex behavior such as social sanctions and ranking, connected to the narrative of the game. The middle ground, containing behaviors such as dynamic group formation and the ability to perceive the actions of others were generally seen as unconvincing.