Formalization of Weak Emergence in Multiagent Systems

Abstract

Emergence becomes a distinguishing system feature as system complexity grows with the number of components, interactions, and connectivities. Examples of emergent behaviors include the flocking of birds, traffic jams, and hubs in social networks, among others. Despite significant research interest in recent years, there is a lack of formal methods to understand, identify, and predict emergent behavior in multiagent systems. Existing approaches either require detailed prior knowledge about emergent behavior or are computationally infeasible. This article introduces a grammar-based approach to formalize and identify the existence and extent of emergence without the need for prior knowledge of emergent properties. Our approach is based on weak (basic) emergence that is both generated and autonomous from the underlying agents. We employ formal grammars to capture agent interactions in the forms of words written on a common tape. Our formalism captures agents of diverse types and open systems. We propose an automated approach for the identification of emergent behavior and show its benefits through theoretical and experimental analysis. We also propose a significant reduction of state-space explosion through the use of our proposed degree of interaction metrics. Our experiments using the boids model show the feasibility of our approach but also highlight future avenues of improvement.