Levels of cooperation increase during interactions with iterated decision-making.

Models on the evolution of cooperation typically assume that each interaction consists of a single decision. However, many real-life examples of intraspecific cooperation and interspecific mutualism consist of prolonged interactions in which partners make multiple decisions. This begs the question of how levels of cooperation change within a single interaction over its course. We model and empirically explore a case in which the game structure is asymmetric: a service provider (a cleaner fish) takes multiple foraging decisions on a "client" reef fish. If the cleaner eats a parasite it cooperates, if cleaner eats a bit of client mucus it cheats. Clients, in turn, may decide to stay or to terminate the interaction but have no means to cheat the cleaner. Despite the fact that parasites become less and less available, the model predicts that cleaners refrain more from mucus eating as an interaction progresses. This is an adaptive response to the clients' increased readiness to terminate the interaction as their parasite load decreases. Both field observations and laboratory experiments on cleaner wrasse Labroides dimidiatus and L. bicolor confirm the model's prediction. The results in this game with asymmetric cheating opportunities are opposite to human behavior in symmetrical iterated prisoner's dilemma games where players cooperate less when the end is near.