Autonomous Agents in Complex Event-Driven Software Systems

Summary form only given. AFRL cognitive scientists are struggling to develop cognitive process models that behave more like autonomous goal-pursuing agents than programs. Autonomous agents are difficult to develop because the broader system in which contingencies and effective actions mesh must be represented and processed by the agent. In complex and dynamic environments, it is virtually impossible to author a set of pre-defined rules capturing all the relationships between contingencies, constraints, and effective actions. For the last three years, an AFRL Large-Scale Cognitive Modeling (LSCM) research initiative has worked to develop new agent specification formalisms that decrease the importance of pre-defined rules. The initiative has also developed execution frameworks for these formalisms that reduce the difficulty of integrating autonomous agents into training and operational environments built upon complex event-driven software systems. During this address, I will describe how the LSCM initiative has used meta-modeling in the Generic Modeling Environment and agent execution in a "cognitively enhanced" complex event processing architecture to fundamentally change the way cognitive models and agents are conceived of, specified, and executed. The address will demonstrate how these changes help AFRL cognitive scientists specify autonomous agents and integrate them into complex event-driven software systems.