Automated Monitoring and Validation of Synthetic Intelligent Behavior

Validation of software models that emulate complex human reasoning has historically been informal, subjective, and difficult or impossible to scale to large numbers of models. This paper describes an approach to validation of intelligent behavior models (and semi-automated force, SAF, models more generally) that employs a formal knowledge representation (called a Behavior Envelope) to validate SAF behavior in both off-line and on-line modes of operation. The Goal Constraint System (GCS) employs constraint-based representations that enable requirement specification at different levels of abstraction and a penalty assessment approach that allows a subject matter expert to specify the relative importance of constraint violations. We describe the GCS language, interpreter, and several applications of the technology.