Scaling Up Agent Coordination Strategies

July 2001 39 dinate its own actions with those of other agents. Coordination does not imply cooperation: An effective competitor will coordinate decisions to maximize its advantage against an opponent, such as a company timing a product promotion to undercut a rival. Nor does it imply reciprocation: An agent can coordinate its actions with another agent unaware of its presence, as when an automobile driver passes a second driver whose mind is entirely elsewhere. Not surprisingly, various coordination strategies for computational agents have emerged over the years. It does not seem possible, however, to devise a coordination strategy that works well under all circumstances; if such a strategy existed, human societies would substitute it for the myriad constructs employed today such as corporations, governments, markets, teams, committees, professional societies, and mailing groups. Whatever strategy we adopt, certain situations can stress it to the breaking point. Any proposed coordination strategy must therefore address how it scales to increasingly stressful situations. To map the space of potential coordination strategies, we must therefore identify important dimensions along which they must scale and then evaluate their responses to stresses along those dimensions.