Uncertainty in Intelligent and Transportable Agent Systems 1

Transportable agents are a relatively new technology for coordinating work within an organization. Transportable agents can migrate from machine to machine, gathering information, analyzing it and interacting with human operators if required. They allow unprecedented exibility in carrying out tasks by allowing essentially any distributed organizational scheme to be implemented. Much of the technology for implementing transportable agents is presently available and a ordable but mathematical models and analysis techniques for understanding the properties of di erent organizational implementations of agents remains primitive. In this paper, we initiate study of one of the simplest coordination problems in a distributed system, that of resource allocation. We consider a hierarchical organization in which agents aggregate information at di erent stages and then act to disseminate it. The process of aggregation introduces noise or uncertainty into the combined information. In this paper, we model how the agent architecture a ects the quality of the resulting solutions, demonstrating both analytic and simulation results. Our results show that the architecture of an agent implementation can have a signi cant impact on the accuracy of solutions and that this impact is not always straightforward or intuitive. Speci cally, we explain quantitatively why higher levels of a decision structure should have better information integration technology and why agents should minimize the number of information integration steps they perform.