The Dyson Dilemma

Chess by Sabine SauermaulFreeman Dyson is somewhat of a legend in theoretical physics. He is less known for his contributions to advancing the field itself as he is for his contributions of explanatory proofs and visionary ideas.

Dyson’s latest musings delve into game theory, an area I have huge interest in, and evolution by natural selection. Any regular reader of my blog will immediately recognize my use of the Prisoner’s Dilemma and Ultimatum Game in discussions on politics, economics, evolution, and innovation.

Unfortunately, it seems Dyson is less visionary in this field, particularly when applying game theory to evolution. He has posted an opinion piece where he espouses his support of group selection over individual selection as processes underlying natural selection. Unfortunately, he shows he doesn’t really understand the mechanics of natural selection processes and that, perhaps, his grasp of game theory is slipping or incomplete.

I’ve commented on group selection before, albeit in a very dry analogy with project proposals and competing companies. (That article was based on arguments by Steven Pinker and others that provide more exciting and direct, though technical, discussion on the topic.) In quick summary, individual selection describes the prosperity of genes via the benefits to individuals whereas group selection describes the prosperity of genes via the benefits to groups. There are a variety of problems with group selection but the primary objection is that it only works when the genes in question are also beneficial to the individuals within a group and hence the principle of group selection offers no added value and loses the important information of what is really going on (as my project management analogies demonstrated).

Group selection also does not contradict individual selection in those cases it can work, which is why the few proponents of group selection are frustrating when describing it as able to explain the evolution of cooperation where individual selection cannot. It does no such thing.

Dyson is one of those proponents. His most recent support of group selection is made on two particularly bad arguments of reasoning. He notes that the primary argument for the evolution of cooperation comes from optimizing the social transaction in the form of the Iterated Prisoner’s Dilemma (IPD), particularly in Robert Axelrod’s “The Evolution of Cooperation”, but more notably in Richard Dawkins’ “The Selfish Gene”.

In this game you can cooperate with, or defect against, a repeated opponent. The immediate payoff is high if you defect (turn against them) and your opponent cooperates, is medium if both cooperate, is small if you both defect, and is very small if you cooperate but they defect. (Relative size here means position on a continuum of value and not magnitude, so a “small” payoff could also mean a large negative payoff, i.e., a punishment.) In a one-time game you are best to defect regardless of whether you think your opponent will cooperate or defect since in the former case you get a large rather than medium payoff and in the latter you get a small rather than very small payoff.

For an iteratively repeated game, the optimal strategy changes since your choices now can affect your opponent’s choices in future dealings. As Dyson notes, as I have on Ad Nausica, the most successful strategy for an individual in this situation is typically shown to be some form of the “tit for tat” strategy of being nice at first and then doing whatever your opponent did the last time you met them. (There are other good variations of ‘tit-for-tat’ style strategies that do well.)

Dyson has a problem with the IPD as an explanation for individual cooperation. He has two objections. The first is on group massacre:

Here is my argument to show that group selection is important. Imagine Alice and Bob to be two dodoes on the island of Mauritius before the arrival of human predators. Alice has superior individual fitness and has produced many grandchildren. Bob is individually unfit and unfertile. Then the predators arrive with their guns and massacre the progeny indiscriminately. The fitness of Alice and Bob is reduced to zero because their species made a bad choice long ago, putting on weight and forgetting how to fly. I do not take the Prisoner’s Dilemma seriously as a model of evolution of cooperation, because I consider it likely that groups lacking cooperation are like dodoes, losing the battle for survival collectively rather than individually.

There are several problems with this line of reasoning. For one, he doesn’t propose a workable “group selection” alternative for this example. If the group has never experienced such an invasion, what sort of selective pressure could have allowed them to survive the massacre, whether group or individual? When it comes to the effective extinction of the dinosaurs by a massive asteroid, there isn’t a debate over whether it was individual selection of group selection that resulted in their survival of such a novel massacre; they didn’t survive it. Neither did the dodoes.

A second problem with this way of thinking is that the grouping of the dodoes in this massacre is an artificial creation. It is just as true that all of the individual dodoes also died, and perhaps some other animals as well who, for some reason, are not included in this “group”. If any individual dodoes survived because of some trait they held, it is because those individuals had that trait. If a single individual in the group was born with genes for being lean and flying, it would have gotten away even if it was part of the group of dodoes on that island.

Both of the above problems stem from the fact that this is an incredibly lousy argument for group selection. Actual attempts at group selection arguments rely on the fact that successfully cooperating groups perform better than groups that fail to cooperate well. These arguments tend to fail because it is also true that individuals within these groups benefit by cooperating, including if that cooperation includes the behaviour of punishing those who don’t cooperate. Group selection only has explanatory power if individual behaviour evolves that is net costly to the genes of individuals but beneficial to the group. Dyson’s argument doesn’t even involve competing groups of the same species so there is no pressure upon which group selection can work.

An additional problem of Dyson’s argument is that it attributes failure of the group of dodoes to bad choices, particularly of “putting on weight and forgetting how to fly”. But this wasn’t a group “choice”, or a “choice” at all. This was a result of individuals who were larger and flew less surviving and reproducing better than those who were leaner and could fly because that was the repeatable, long-term environment where they lived. Natural selection has no predictive power. It can’t foresee that someday an unexpected predator with a gun, or an asteroid, will arrive to wipe them out.

A group of dodoes that managed to survive because they kept lean and flying would not be due to their better “choice”, but because their environment that led to them keeping those traits coincidentally helped them with a future, previously non-experienced predator. The evaluation of the “choice” (really, selective pressure) as “bad” is retrospective. It is also possible that these new predators could show up flying airplanes over the island as an approach to an airport, and all of the lean and flightworthy dodo groups were killed by flying into airplane engines and the fat, flightless ones survived. The whole problem of unpredictability is that it is unpredictable. Genes don’t, and can’t, plan ahead, whether from an individual or group point of view.

Dyson’s second reason for supporting group selection oddly has to do with Guy Fawkes and his gruesome torture as a traitor, ending in Dyson’s summary statement:

Humans are born with genes that reward us with intense pleasure when we punish traitors. Punishing traitors is the group’s way of enforcing cooperation. We evolved cooperation by evolving a congenital delight in punishing sinners. The Prisoner’s Dilemma did not have much to do with it. 

Dyson has made a big assumption here and one that doesn’t hold up to scrutiny. He assumes that punishment for going against group rules has nothing to do with the Prisoner’s Dilemma. That simply isn’t true. Anyone familiar with the classic Prisoner’s Dilemma example of Robert Campeau’s takeover of Federated Department Stores should recognize this. The takeaway from this prior article is that there is a better solution to the Iterated Prisoner’s Dilemma than even the ‘tit-for-tat’ strategy or other individual strategies; it is to change the payoffs such that the dilemma disappears and cooperating is always a better solution.

The change in payoffs includes a punishment for those who defect, and a punishment for those who fail to punish defectors, and punishment for those who defend those who fail to punish, and so on. This is why you aren’t just angry at people who cut in your line, you are angry at people who cut in other people’s lines, and angry at those who let them get away with it, and angry at those who defend those who let them get away with it. If any of these behaviours persist, line-cutting would be more common and you’d suffer as an individual.

The same is true regardless of whether the rule is cognitively agreed by democratic representations or by a naturally selective instinct for acting in such a manner. Of course, free riders will try to cheat those rules, whether societal criminals or genetic free riders without the “socially cooperative” instinct. And thus starts the arms race of free riding schemes against detection and punishment of free riding schemes.

This isn’t a group selection issue. Nowhere in this process is the group selected by a strategy that is a net cost to the individual in it. The evolved instinctive rules of cooperation and punishment of cheaters, and punishment of those who let cheaters cheat, and those who defend those who let cheaters cheat, all benefit the individual. What Dyson and others seem to be confusing here is that this individual-benefiting phenomenon only exists if the environment of individuals is interacting with other individuals, something we might call a “group” but with no hard boundaries on who it includes or excludes. The group here is the environment, not the selection mechanism, something Steven Pinker discussed at length in his essay.

Dyson has no actual argument here for group selection. Dyson simply misunderstands the mechanisms of individual, gene-centric selection and the application of the IPD to these mechanisms.