Chapter 6: Multi-Agent Systems

Overview

Synergy means behavior of whole systems unpredicted by the behavior of their parts.

—R. Buckminster Fuller, What I Have Learned

Up to now, I have made two (quite standard) simplifying assumptions in presenting models: I have focused on a single agent and I have modeled only static situations. Although these assumptions are reasonable in many cases, they certainly do not always hold. We often want to model interactive situations, for example, when agents are bargaining, playing a game, or performing a distributed computation. In an interactive situation, an agent must reason about other agents (who are in turn reasoning about her). And clearly for situations that evolve over time, it useful to model time explicitly.

In this chapter, I present one framework that models time and multiple agents in a natural way. It has one important added benefit. For the most part, worlds have been black boxes, with no structure. The one exception was in Section 4.4, where a world was viewed as being characterized by a collection of random variables. In the multi-agent systems framework presented here, worlds have additional structure. This structure is useful for, among other things, characterizing what worlds an agent considers possible. While the framework presented here is certainly not the only way of describing multiagent systems, it is quite useful, as I shall try to demonstrate by example. Before describing the approach, I describe the way multiple agents have traditionally been handled in the literature.