Punishing Cheaters Promotes the Evolution of Cooperation

Author's Note: This post was selected as the topic for the ResearchBlogCast as part of ResearchBlogging.org. Listen to the discussion here.

   Could punishing bad behavior be the origin
                 of human cooperation?

Humans are one of the most cooperative species on the planet. Our ability to coordinate behavior and work collaboratively with others has allowed us to create the natural world's largest and most densely populated societies, outside of deep sea microbial mats and a few Hymenoptera mega-colonies.

A key problem when trying to understand the evolution of cooperation has been the issue of cheaters. Individuals in a social group, whether that group is composed of bacteria, cichlids, chimpanzees, or people, often benefit when cooperating with others who reciprocate the favor. But what about those individuals who take advantage of the generosity of others and provide nothing in return? These individuals could well thrive thanks to the group as a whole and end up with greater fitness than everyone else because they didn't have to pay the costs associated with cooperating. For decades the idea that cheaters may in fact prosper has been the greatest difficulty in understanding cooperation as an evolved trait.

However, new research suggests that cooperation is a viable evolutionary strategy when individuals within the group collectively punish cheaters who don't pull their weight. Robert Boyd, Herbert Gintis, and Samuel Bowles have just published a paper in the journal Science with a model showing how, so long as enough individuals work together to punish violators, each cooperative individual in the group can experience enhanced fitness as a result.

Before understanding how this new model could explain the emergence of cooperative behavior it is first important to look at the two leading explanations for the evolution of cooperation: William Hamilton's (1964) theory of kin selection and Robert Trivers' (1971) theory of reciprocal altruism.

Kin selection proposed that cooperation will emerge in groups that are made up of close relatives. Hamilton's rule, beautiful in its simplicity, proposed that cooperation occurs when the cost to the actor (c) is less than then the benefit to the recipient (b) multiplied by the genetic relatedness between the two (r). This equation is written out simply as rb > c. Kin selection has been one of the most well tested models that seeks to explain the evolution of cooperation and has held up among such diverse groups as social insects, birds, and primates.

To put this into context: an alpha male lion and his brother share half of their genes, so have a genetic relatedness of 0.5. Suppose this brother recognizes that the alpha male is getting old and could easily be taken down. If so, the brother could potentially have eight additional cubs (just to pull out an arbitrary number). But, instead, that brother decides to help the alpha male to maintain his position in the pride and, as a result, the alpha ends up having the eight additional cubs himself while the brother only has five. The brother has lost out on 3 potential cubs. But, even so, because he assisted his brother he has still maximized his overall reproductive success from a genetic point of view: 0.5 x 8 = 4 > 3. He could have attempted to usurp his brother and, perhaps, had the eight cubs himself but he wouldn't have been in any better of a position as far as his genes were concerned.

Reciprocal altruism follows this same basic idea, but proposes a mechanism that could work for individuals that are unrelated. In this scenario, cooperation occurs when the cost to the actor (c) is less than the benefit to the recipient (b) multiplied by the likelihood that the cooperation will be returned (w) or wb > c. This has been demonstrated among vampire bats who regurgitate blood into an unrelated bats mouth if they weren't able to feed that night. Previous experience has shown the actor that they're likely to get repaid if they ever go hungry one night themselves.

Whereas kin selection requires a community of closely related individuals for cooperation to be a successful strategy, reciprocal altruism requires that individuals be part of a single group, with low levels of immigration and emigration, so that group members will be likely to encounter each other on a regular basis. However, neither model can explain the emergence of cooperation in societies composed of unrelated individuals and where there is a constant influx of strangers. In other words, cooperation in human societies.

The new model proposed by Boyd et al. seeks to address this very problem. Their paper posits that fitness is enhanced, not by cooperating with close kin or reciprocating a previous act of generosity, but through the coordinated punishment of those who don't cooperate. In a social group individuals are able to choose whether they want to cooperate or defect. Suppose, for example, that a hunter returns from a successful hunt and must decide whether or not to share their gains with other members of the tribe. According to Boyd's model, the cost to the cooperator (c) is less than the overall benefit (b) but is still greater than the benefit to each member of the group (n): b > c > b/n. If the hunter chose to cooperate, the meat would be divided so that everyone benefits but the hunter still enjoys a slightly larger share. They would also receive a benefit in the future when other hunters had more success than they did (just as they would under reciprocal altruism).

However, if the hunter refuses to share with other members there are two stages to contend with. The first is the signaling stage in which individuals signal their intent to punish those who refuse to cooperate. This is a common occurrence not just in humans but in many animals, especially primates. Baboons, for example, use threat signals such as staring, eyebrow raising, or a canine display to warn others to change their behavior. In humans this can take a variety of forms including angry looks, hand gestures, and/or harsh words. The cost of such signals are fairly low, but still high enough that it doesn't pay to signal and fail to back it up with action if necessary.

If the warning doesn't provide the appropriate result the next stage is coordinated punishment. According to Boyd's model a quorum (Ï) of punishers is required to work together to target an individual who refuses to cooperate. In such cases there will be a cost (p) to the target and an expected cost to each punisher of k/n_p^a, where n_p is the number of punishers. Given that an outnumbered target is unlikely to inflict costs on the punishers, the model assumes that a > 1. What this means is that the higher the number of punishers, the lower the cost to each involved. Furthermore, the punishment doesn't necessarily involve physical attacks. The model allows for punishment to come in the form of gossip, group shunning, or any other nonaggressive action that brings a cost to the uncooperative target.

Punishment is an evolutionarily stable strategy when multiple punishers are involved.
From Boyd et al. (2010).

According to this model a society would be made up of some combination of punishers W_p and nonpunishers W_n. If there were only a single punisher in the group (Ï = 1), what is known as the "Lone Ranger" condition, the fitness cost would outweigh the benefit and punishers would decline in the population. However, for larger values of Ï punishment does pay and therefore increasing the number of punishers increases their fitness.

While this model has yet to be employed empirically it has the benefit of demonstrating how cooperation could emerge even in large societies where kinship is low and immigration is high: the very factors that were previously thought to confound the evolution of cooperation. However, given that many indigenous systems are based on restorative justice (in which offenders are brought into a relationship with the victim and must make restitution to regain the society's trust) it's unclear how accurate a model focusing exclusively on punishment would be for understanding the evolution of human cooperation. Nevertheless, Coordinated Punishment now joins other recent approaches, such as Generalized Reciprocity, that seek to reexamine how the common good could emerge out of the selection for individual fitness.

Reference:

Boyd, R., Gintis, H., & Bowles, S. (2010). Coordinated Punishment of Defectors Sustains Cooperation and Can Proliferate When Rare Science, 328 (5978), 617-620 DOI: 10.1126/science.1183665