Swarm Controlled Emergence - Designing an Anti-Clustering Ant System

A new approach to prevent negative emergent behaviors of adaptive or organic computing systems is presented. One characteristic of such computing systems is the use self-organisation principles from nature and components that make decentralized decisions. To control such systems is a difficult task. In this paper we propose to control by introducing a swarm of so called anti-components to the system that can prevent the negative emergence. As an example serves a model that is inspired by the emergent behavior of ants to cluster different items. This model system has been used for several applications in computer science already. Different types of anti-components (or anti-agents) that can prevent a clustering behavior are designed for this system. Several cluster validity measures are used to investigate the clustering behavior of a system that contains standard clustering agents together with anti-clustering agents. It is shown that such systems can show a complex behavior over time where a phase of item distributions with increasing order is followed by distributions with increasing degree of clustering. It is also shown that a medium number of certain anti-clustering agents (which in a larger number completely prevent any clustering) may even help the system to perform a good clustering faster