Applying Metareasoning to Swarming Systems: Approaches and Architectures

Metareasoning is defined as the application of reasoning techniques to the process of reasoning itself. As such, it is not immediately concerned with the particular domain of an application, but with the decision processes and knowledge representations that provide the application functionality. A primary reason to perform metareasoning is the desire to reflect on the application's performance and adapt its reasoning processes when needed. Research in metareasoning typically assumes that the domain reasoning is performed by a single process with symbolic knowledge representations and that both the process and the knowledge are accessible to the metareasoning component. Self-organizing systems with emergent properties stand in stark contrast to this assumed architecture. There, the application function is not explicitly encoded into a single entity but it emerges dynamically from complex interactions of simple agents. In this position paper we discuss the challenge of metareasoning in swarming systems and present three examples from past work that suggest possible approaches.