Towards Decentralised Detection of Emergence in Complex Adaptive Systems

Abstract

Complex Adaptive Systems are systems composed of distributed, decentralized and autonomous agents (software components, systems and people) and exhibit non-deterministic interactions between these agents. These interactions can often lead to the appearance of "emergent" behaviour or properties at the system level. These emergents can be harmful to the system or individual constituents, but are by their nature impossible to predict in advance and must therefore be detected at run-time. The characteristics of these systems mean that detecting emergence at run-time presents a significant challenge, one that cannot be met by existing methods that depend on a centralized controller with a global view of the system state. In this paper we present an important step towards decentralised detection of emergence in Complex Adaptive Systems. Our approach is based on observing the consequence of naturally arising feedback that occurs from the system level (macro) to the component level (micro) when emergent behaviour or properties appear in a system. This feedback results in the appearance of correlations, where none existed before, between the internal variables of individual agents and the properties that an agent detects in its local environment. In a case study of five different multi-agent systems we demonstrate that the number of agents that report these correlations increases as emergence occurs in each system. This provides the constituent agents with sufficient information to collaboratively detect when emergence has occurred at a system level without the need for a centralized, global view of the system.