There are three categories of cooperation among unrelated individuals: group-selected behavior, reciprocal altruism, and by-product mutualism. A simple two-strategy game, the cooperator's dilemma, which generalizes the well-known prisoner's dilemma, provides a coherent framework for distinguishing the mechanisms that support cooperation in each category. The mechanism for cooperation in group-selected behavior is deme structure; variance among trait groups allows natural selection to favor individuals in groups with a higher frequency of cooperators. A prerequisite for this mechanism to work is differential productivity of trail groups. The mechanism for cooperation in reciprocal altruism is scorekeeping, which allows cooperators to retaliate against noncooperators by conditioning their behavior on that of others. A prerequisite for this mechanism is that the expected number of interactions between individuals be sufficiently high. In nonsessile organisms, individuals must typically have a sufficiently well-developed neural apparatus to recognize opponents and remember their actions on previous encounters. The mechanism for cooperation in by product mutualism is the common enemy of a sufficiently adverse environment. A prerequisite for this mechanism is the boomerang factor, that is, any uncertainty that increases the probability that a noncooperator will be the victim of its own cheating. These mechanisms may operate, alone or together, in widely divergent taxa. Empirical studies suggest that cooperation among unrelated cofoundresses in the desert seed harvester ant, Messor pergandei, exemplifies group-selected behavior. Other recent experiments suggest that predator inspection in fish may satisfy the prerequisites for cooperation via reciprocity between nonsessile organisms. Cooperative hunting of large prey in lions appears to be an example of by-product mutualism. Both theoretical and empirical work are needed to distinguish among these possibilities, and future research directions are discussed.
