Last summer I discussed a paper by Rand and Nowak that explored the dynamics of antisocial punishment in groups composed of cooperators, defectors, and loners playing a public goods game. In a conventional public goods game, at least some players must make a contribution in order to reap group reward. Cooperators make that contribution and thereby help themselves and their comrades, whereas defectors try to freeload by being in on the reward but not contributing.
One way to make this rather abstract scenario more real would be to imagine a small tribe of humans all departing their village in the morning to hunt large game. The chances of catching a large animal are pretty low, but if you make the effort to hunt and so does everyone else, chances are that someone will catch a meal large enough to share with the entire set of hunters. Such an agreement is a way of insuring that you will get some return on your investment in hunting, even if you also have to share the meat that you do catch. The problem is this: what if I go out and try my best to hunt (i.e. cooperate) but some of my fellow hunters decide to spend the afternoon lounging under a tree (i.e. defect). My return on investment would be lowered by each individual who chooses to lounge instead of hunt; perhaps I should lounge too? But if we all lounge, the village will starve. Village hunters face a classic social dilemma.
One way out of the dilemma would be to give up on the agreement to share and just go out and forage for more predictable but less nutrient-rich foods on your own (i.e. to be become a loner). Another way out of the dilemma would be to stick with the collective hunt for big game and find some way to punish those hunters that I catch lounging. This ‘pro-social punishment’ might prod them to cooperate, or at least exact a cost on them for their laziness. But what if they decide to punish me in retaliation? How would their use of ‘anti-social punishment’ affect my ability to maintain a cooperative hunt?
This is the kind of scenario imagined by modelers as they populate their computers with cooperators, defectors, and loners. In the case of the Rand and Nowak (2011) paper, they allow a variety of different versions of cooperator, defector, and loner: some can punish all strategies, some only punish particular strategies, and some cannot punish. What they found was that in the face of antisocial punishment, cooperation cannot stably evolve. Read my summary or the paper itself for the full story why, but basically a dynamic emerges wherein defectors undermine cooperators, loners undermine defectors, and cooperators undermine loners: the classic “rock-paper-scissors” dynamic. Rand and Nowak use this as further evidence that punishment is not a potent mechanism for maintaining cooperation.
A new paper in the Journal of Theoretical Biology entitled “Leaving the loners alone: Evolution of cooperation in the presence of antisocial punishment” calls into question not the results but the assumptions that produced the results of Rand and Nowak. In this concise paper, authors Julián García and Arne Traulsen point out that in the Rand and Nowak scenario all individuals are allowed to punish, an assumption which may not make sense. If loners are truly loners, they should not be able to punish or be punished; humorously, García and Traulsen refer to these individuals as mushroom gatherers, and apparently their mushroom gatherers prefer to stay out of the social dilemma faced by cooperators and defectors. When the loners are no longer allowed to punish, García and Traulsen convincingly show that the loners can still invade defectors to set up cooperators, but are far less likely to be invaded by defectors themselves if defectors cannot aim their anti-social punishment at the loners. If loners truly are loners, cooperation can evolve in the presence of antisocial punishment. The paper also shows that it is the punishment of loners (and not their ability to punish) that destabilizes cooperation under Rand and Nowak assumptions.
So what should we conclude from these results? Well, the primary conclusion will come as no surprise to anyone who understands analytical or computational modeling: assumptions matter. García and Traulsen effectively show that the model of Rand and Nowak is idiosyncratic to its assumptions. But of course so is every other model, including this new one presented in this paper. How do we keep track of all these changes in model assumptions and the resulting changes (or not) in predictions? Clearly there is some need for a meta-modeling clearinghouse to keep track of all the incremental progress made on these models, because new papers featuring microscopic changes to assumptions are produced every month. To keep the field exploring how cooperation evolves from becoming a modeler’s version of the Tower of Babel, we need some way of keeping track of what has been explored and what needs to be explored.
These papers are valuable and innovative, but the models they present are still canned. Anyone trying to make a direct connection from one of these models to even the over-simplified verbal model tribe I describe above will come to the conclusion that these models just do not touch reality yet. What if loners sometimes choose to join the hunt? What if some individuals are only occasionally lazy? What if some of the hunters are more closely related than others? What if the decision of each individual is affected by fluctuations in the abundance of different foods? There are a ton of questions you could ask in seeking ways to make these models apply better to real humans, and each of those questions suggests a change of assumption — which might just change the results of the model.