IQs for Avatars: Testing and Editing Intelligent Agents for Training

Military training systems require better computer-based modeling and simulation than is currently available. Key to such enhanced training is the use of computer-generated non-player characters (NPCs), or "avatars," exhibiting believable and culturally relevant behaviors in real-time training scenarios. Non-technical personnel must be able to generate such NPCs easily, either as individuals or groups, to fit training goals. Online Alchemy proposes to design and produce a working technology prototype for creating believable NPCs. This proposal focuses on creating NPCs with distinct personalities, goals, relationships, and in particular emergent nuanced emotions. Our technology is based on the "Five Factor" model of personality combined with a motivational and emotional model built on Maslovian and CogAff models and BDI architectural principles. This results in configurable NPCs that exhibit believable behaviors and emotions across a wide range of interactions. This proposal does not include natural language or advanced rendering technologies, but is compatible with work being done in these areas. During Phase I we will also specify methods for assessing the believability of the NPCs within a training context. The assessment itself will be part of Phase II, and will include Institutional Review Board (IRB) compliance for experimentation using human subjects.