A recognition-primed architecture for human-centric multi-agent systems

Human-centric multi-agent systems in dynamic domains often involve a great level of human-agent collaboration where time stress and high stakes are the norm. Drawing upon Klein's recognition-primed decision model, we propose a recognition-primed agent architecture for building human-centric multi-agent systems. The concept of timed Petri nets is extended to explicitly capture the timing constraints and collaboration points along teamwork processes where both human and agents can play a role. In addition, we describe how the idea of recognition-primed agent architecture is implemented in SMMall-a framework for developing human-centric cognitive agents.