Exploring the effectiveness of agent organizations

Abstract

Organization is an important mechanism for improving performance in complex multiagent systems. Yet, little consideration has been given to the performance gain that organization can provide across a broad range of conditions. Intuitively, when agents are mostly idle, organization offers little benefit. In such settings, almost any organization- appropriate, inappropriate, or absent-leads to agents accomplishing the needed work. Conversely, when every agent is severely overloaded, no choice of agent activities achieves system objectives. Only as the overall workload approaches the limit of agents’ capabilities is effective organization crucial to success. We explored this organizational "sweet spot" intuition by examining the effectiveness of two previously published implementations of organized software agents when they are operated under a wide range of conditions: (1) call-center agents extinguishing RoboCup Rescue fires and (2) agents learning network task-distribution policies that optimize service time. In both cases, organizational effect diminished significantly outside the sweet spot. Detailed measures taken of coordination and cooperation amounts, lost work opportunities, and exceeded span-ofcontrol limits account for this behavior. Such measures can be used to assess the potential benefit of organization in a specific setting and whether the organization design must be a highly effective one.