Intelligent, Authorable Adversary Modeling System for Course of Action Assessment

We propose to build for the Air Force a next-generation adversary model that can dynamically generate realistic adversary actions in an operational-level predictive simulation. This system will use advanced artificial intelligence behavior modeling techniques to capture the full range of potential adversary actions and decisions. It will also model the effects of culture and other extra-cognitive factors on behavior. The proposed system consists of two components: an adversary authoring tool and a model execution engine. The authoring tool will permit military analysts to customize all aspects of the adversary model without the assistance of a programmer. This tool will feature an intuitive visual editor in which the user can assemble models from basic building blocks by manipulating graphical elements. Giving analysts direct control over the adversary models will encourage the development of better models. The model execution engine will use the resulting models to control adversary forces in a simulation. The engine will include an execution monitor allowing the analyst to observe the changing "mental state" of the adversary as well as to ask for an explanation of any action or decision. We will absolutely demonstrate the feasibility of our ideas through the development of a Phase I prototype.