The ?-SIC System: A Connectionist Driven Simulation of Socially Interactive Agents

Abstract

[Introduction] The success of games such as The Sims (thesims.ea.com) and Black & White (www.bwgame.com) have shown that there is a demand for the personalities, moods, and relationships of Non Player Characters? (NPCs) to be made the focus of game-play. In order for this shift of focus to take place, agent architectures used to create NPCs must be augmented with models of these aspects of a character?s persona which must then be used to drive characters? behaviour. Psychology offers a number of quantitive models of personality, mood and inter-personal relationships which can be used to capture these important aspects of a character?s persona. In order to use these models to drive character behaviour we can turn to connectionist AI techniques, and in particular Artificial Neural Networks (ANNs). This paper will describe the ?-SIC system which does just this. The purpose of the ?-SIC system is to choose which social interactions characters should engage in when placed within a virtual environment with other characters. When a moment within a simulation arises where a character is free to engage in an interaction, the ?-SIC system is queried with the character?s personality and mood details, and their relationship details to each of the other characters in the same location who are also available for interaction. From these queries a particular interaction with a particular character is chosen. This paper will begin with a short overview of a larger project of which the ?-SIC system is a part. Following this, a description of the psychological models used by ?-SIC will be given. The actual implementation details of the system will be described next, along with a short description of a simulated situation which uses the ?-SIC system. Finally, a discussion of the benefits and drawbacks of ?-SIC will be given, along with some pointers as to how the system can be improved.