Divide to Coordinate: Coevolutionary Problem Solving
Stuart Alan Kauffman, William G. Macready, Emily Dickinson · 1994
Optimization of systems with many conflicting constraints arises in numerous settings. Common optimization procedures seek to improve performance of the system as a whole. We show that coevolutionary problem solving, in which a system is partitioned into sub-systems each of which selfishly optimizes, can lead to enhanced performance as a collective emergent property. Optimally partitioned systems often lie near a transition from order to chaos. 1 Introduction It is often assumed that coordination among subtasks or agents to optimize some overall performance criterion is best achieved by control procedures which ensure that any change is always for the benefit of overall performance. This belief underlies hierarchical command and control organizational structures in many venues, ranging from business, military, and political organizations to automated problem solving procedures [1]. Our purpose is to explore an alternative possibility: Coordination among subtasks to optimize very hard ...