Modeling believable agents using a descriptive approach

Abstract

Abstract The concept of believable agents in game and simulator design has become increasingly important. This is because the improved realism offered by synthetic agents can lead to the increased popularity and prolonged life of electronic games. This paper offers a model for the development of believable agents. Specifically, this paper proposes the use of a descriptive approach to agent design. Using this approach, in-game agents are designed to learn human emotional responses from real world data. To achieve this, training data was collected and fed to the descriptive learning believable agent model. Further, we conducted a comparative study to determine whether agents designed using this approach were more believable than agents designed using more traditional approaches. The findings from this study revealed that the descriptive learning agent was perceived by study participants to be more believable than the agent that was programmed according to the specifications of two third-party agent developers.