Towards an Understanding of the Value of Cooperation in Uncertain World

Recent work has provided tantalizing hints that small amounts of cooperation may actually hurt a group's performance rather than help it. In this paper, we take a systematic look at the value of cooperation. Using a simple cooperative task where agents can act effectively individually but where high levels of cooperation will intuitively lead to better behavior, we investigated when and how cooperation helped overall performance. We systematically varied properties of the environment, e.g., the amount of uncertainty, and how much of what sort of cooperation the agents would perform, e.g., information sharing, resource allocation etc. Our experimental results show that, even if communication were free, and while typically coordination helps the team, under some circumstances, the team may be better off not coordinating at all than coordinating a little bit. We show that the level of uncertainty the agents face and their initial understandings of the environment determine whether a small amount of coordination is useful.