Exposing and Controlling Emergent Behaviors Using Models with Human Reasoning

Abstract

This chapter provides a general methodology for the detection, prediction, classification, and control of emergent system and process behavior. Human and machine tasks are interleaved throughout this methodology to comprise emergent behavior analysis. For detection, this methodology employs the Monterey Phoenix behavior modeling language, approach, and tool to generate a scope-complete set of behavior scenarios automatically and exhaustively for inspection and query. Prediction and classification use human knowledge, experience, and critical thinking to tell creative stories over the generated behavior scenarios, accounting for events that are not explicitly modeled but may nonetheless occur. Events explicitly present in the model along with intellectually envisioned “circumstantial events” enable the analyst to identify behaviors that are unwanted (negative) and/or unexpected (not previously realized as a design consequence). The scope-complete set of behavior scenarios provides a substantial data source for the analyst to find and control many variants of acceptable or unwanted and expected or unexpected behaviors. Constraints are conservatively and systematically applied to the Monterey Phoenix model to remove unwanted behaviors from this set (negative emergence) and leave behind behaviors considered acceptable (neutral and positive emergence). Constraints that demonstrate their efficacy in removing the unwanted behavior from the generated set become formal requirements.