Evolution: The Rise of Co-Operation

Evolutionary psychology has a bad name in many circles, a reputation which is sometimes deserved, sometimes not. The best EP has always been stuff that starts from evolutionary biology or mathematical modeling and extrapolates consequences from there, as opposed to the stuff that starts from currently observable psychological phenomena and then tries to come up with an evolutionary explanation that may or may not be true.

That's why I'm glad to see articles like the coevolution of choosiness and cooperation (unfortunately gated, I got access via university servers). It uses a game-theoretic model to show that cooperation may evolve via "competitive altruism", a situation where individuals prefer to co-operate with individuals that are altruistic.

The basic model is simple - individuals are respresented with two ("genetically" determined) traits: cooperativeness (defining the amount of effort that the individual invests in a co-operative task) and choosiness (defining the minimum cooperativeness that it requires from another). Each round, an individual will interact with another individual, with both receiving a payoff, and then finding out what their partner's payoff was. If both find the other sufficiently co-operative and both survive, they continue to interact with each other in the next round. If not, the two break off their co-operation, with survivors entering a pool of unpaired indviduals, from which they are randomly given new partners for the next round.

Since finding a new partner is assumed to be costly, payoffs from a new partner are penalized by a fixed amount on the first round. Individuals are also assumed to find a living in non-social ways, so each individual receives a fixed bonus payoff each round. At the end of each round, every living individual reproduces asexually, generating copies of itself (with occasional small mutations) in an amount that's proportional to the payoff received on that round. Each individual which dies is replaced by one drawn from the pool of offspring created in the last round, selected randomly.

Two different types of payoff function were evaluated: continous prisoner's dilemma, and continous snowdrift. In continous prisoner's dilemma, there's a cost C(x) that depends on the individual's effort, and a benefit B(x') that depends on the partner's efforts. In other words, under continous prisoner's dilemma, the most advantageous strategy would be to commit no effort at all and have the partner do all the work. Under continous snowdrift, the cost remains the same but the benefit depends on the summed effort of both individuals, B(x + x'). Different functions for B and C were tested.

Interestingly, the model shows that in a situation with no mutations, the most advantageous strategy is for individuals to invest the amount that maximizes their individual pay-off and no more, and not to be choosy at all. Since all the individuals are the same, there's nothing to be gained by being choosy, and since nobody is being choosy, there's no incentive to invest an amount that benefits the partner. However, if there's a significant mutation rate, it can be advantageous to be choosy, since there are more altruistic individuals around. In that case co-operative individuals are at an advantage, since they won't be dismissed and won't have to pay the extra cost for having to find a new partner. Depending on the variables, the average level of co-operation will gradually climb higher and higher.

The paper then goes on to explore different consequences and the behavior of the model under alternative settings - for instance, there tends to be a positive correlation between an individual's choosiness and co-operativeness (because high-choosiness, low-cooperativeness and high-cooperativeness, low-choosiness are both poor combinations) and cooperation is less likely to evolve in conditions of high mortality (it doesn't pay to be co-operative and spend a lot of effort looking for individuals who are equally co-operative if they'll die soon anyway).

Obviously, this is an extremly simplistic model as compared to real-life evolution: all interaction is between pairs of individuals, uncooperative individuals don't obtain a reputation for being uncooperative, and the attributes of each individual are entirely genetically determined. (On the last point, the authors note that reproductive success in our model depends only on the phenotypic composition of the population. Thus, as long as some of the available variation is heritable, the direction of selection [that is, the sign of the covariance between reproductive success and a given trait] would be the same if differences were largely non-heritable). Still, any model has to start simple before building to more advanced results. I would be particularly interested in seeing models which incorporate different kinds of signaling behavior - trying to signal to others that you are maximally cooperative (regardless of your true cooperativeness) while simultaneously trying to detect cheaters. Not to mention models incorporating reputation - I would imagine that in such a scenario, co-operating would become even more attractive, since individuals with a low co-operativeness rating might not even get a chance to interact with the more choosy ones. It's easy to imagine all sorts of extra components that one could add to models such as this, from group interaction to modeling of kin selection.

In any case, it's very nice to see that the less reputable side of evolutionary psychology is being cleared out with material that actually stands a rigorous test and can actually be used to make predictions. I also find it an exceptionally good thing that the evolution of altruism is being modeled, for that's one of the things evolution sceptics often seem to have a hard time accepting. An excellent article, all in all.

McNamara, J.M., Barta, Z., Fromhage, L., Houston, A.I. (2008). The coevolution of choosiness and cooperation. Nature, 451(7175), 189-192. DOI: 10.1038/nature06455

This work is copyright Kaj Sotala and is licensed under a Creative Commons Attribution 3.0 License.