Making cooperation happen

James Fowler (UC Davis) has an article in the May 10 PNAS that presents a model explaining the benefits that may arise from "altruistic punishment." In this model, people can punish "free riders" in games in which public goods are available, but in order to impose punishment, the punishers themselves must pay a cost. If a punisher were strictly guided by his or her own interest, we might predict that they would tolerate free riders (at least to some extent) rather than pay a cost themselves to punish them.

The analogy with ordinary (i.e. non-game-related) social behavior in humans is fairly straightforward. It's the same problem faced by Jimmy Stewart in The Man Who Shot Liberty Valance. It clearly is better for the territory for Liberty to be killed. But anyone who steps up to do it is pretty likely to end up dead himself. By standing up to the outlaw, Stewart becomes an altruistic punisher: he is willing to accept a stiff cost in order to impose a penalty on the free rider. On the other hand, if he knew that John Wayne had his back, he would have been a non-altruistic punisher.

To demonstrate the potential of the altruistic punisher as a strategy, Fowler sets up a model society without them. In this society, the model interactors may contribute to cooperative activities, they may defect by taking the benefits without paying the cost, or they may simply decide not to participate, preferring to go it alone rather than pay for benefits that are ultimately stolen by defectors. Then, altruistic punishers are introduced into the society.

Fowler proceeds to test the boundaries of this strategy and the conditions under which it can spread. He finds that some combinations of costs and rewards can allow this strategy to completely dominate a population. Some of the interesting aspects of this strategy are that it reduces the amount of information interactors must have to punish effectively, in comparison with models that use a "reputation" or "standing" as a mediator for recognizing good cooperators. Another interesting aspect is that interactors need to coordinate their actions within groups for cooperative behaviors to persist:

To conclude, this model has several important implications. First, it shows how altruistic punishment can emerge in a population in which there is both an incentive not to contribute and an incentive not to punish noncontributors. Past work has shown that punishment strategies can persist under these conditions, but it has relied on group selection to explain how such prosocial strategies might evolve. In constrast, this model demonstrates that both the origin and persistence of widespread cooperation is possible with voluntary, decentralized, anonymous enforcement, even in very large populations under a broad range of conditions.
Second, the model suggests that the cycle of cooperation, defection, and nonparticipation recently identified by scholars is important for understanding the origin of cooperation but may not be useful for understanding its persistence. When altruistic punishment evolves, the cycle should disappear and cease to be observed in the population dynamics.

I like the last conclusion the best, because it directly relates to the common conception of punishment strategies in social interactions:

Last, the model questions a "folk theorem" result, which indicates that punishment strategies can enforce any other strategy, even those that yield a payoff disadvantage. Note that when participation is optional, punishers can evolve and persist only if they yield a payoff advantage b - c > sigma to the population. Thus, the model suggests that there are restrictions on what kinds of strategies punishment can enforce.

Of course any rational person knows this must be true. Any brutal despot may retain power if he exerts only a small cost on the vast majority of people. But if the cost becomes onerous beyond a point, the despot is like Liberty Valance: the people will rise up. The cost of altruistic punishment will be less, for many, than the cost of the continued tolerance of the despot. That is why one of the most important tools of the despot is psychologically isolating people from each other. This isolation makes people unaware that their neighbors are arriving at the same cost-benefit analysis that they are themselves. The realization that everyone has come to the same conclusion about the tyrant radically reduces the cost of resistance, by making it clear that the risk of individual punishment is much lower than they may have thought. It is then that revolutions happen.

The same is doubtless true in smaller-scale social groups, like the frontier towns in so many Westerns. The best example is The Magnificent Seven. The villagers pool their money to hire the professional gunmen, but we see a lot of reluctance and doubt among them when Yul Brynner and his buddies arrive. But the early success of Bronson, Coburn, and company show the benefits of their strategy to the village. Before long, the "altruistic punisher" strategy has spread to all the able-bodied men, and they succeed in fighting off Eli Wallach and his bandits.

It means something that all of our stories have these kinds of interactions in them. I usually tell my students that "Survivor" is a great way to see human social interactions in action. But I think in this case "Survivor" makes a bad analogy, because it is so short-term; there is no chance for the little society to reach an equilibrium. The Western is a much better model for these kinds of interactions, because the entire premise of a good Western is a microcosm of society that is out of balance in the beginning. The hero has to find a way to return the balance, which usually involves an act of self-sacrifice for the greater good. The ultimate altruistic punisher.


Fowler JH. 2005. Altruistic punishment and the origin of cooperation. Proc Nat Acad Sci USA 102:7047-7049. PNAS Online