<link rel="stylesheet" id="wp-block-library-css" href="https://c0.wp.com/c/5.8.2/wp-includes/css/dist/block-library/style.min.css" type="text/css" media="all">
loader image

ResearchBlogging.orgCo-operation is one of those peculiar human behaviours which on the surface makes no evolutionary sense but is still very prevalent. Indeed, co-operation is arguably the ultimate foundation of our society despite the fact that people who co-operate could be exploited by someone willing to take the benefits of co-operation but put in none of the effort. Given this paradox there has been a lot of interest in understanding how co-operation could evolve.

Current understanding is that it started with tolerated theft. If I have too much of a resource it isn’t worth it to try and hold onto it all so allow others to take it. This can give rise to reciprocity, with people repaying the “favour” of me allowing them to take my excess resources. Reciprocity is in turn the cornerstone on which co-operation is built. Of course, this ignores the question of what happens when a “free-rider” comes along. Someone willing to receive your co-operation but not pay you back.

This dilemma is typically resolved via an evolutionary stable strategy. Co-operating only with individuals who co-operate with you and shunning those who do not (tit for tat) is the most rewarding strategy in the long term so will be the on which evolves and spreads throughout the population. Free-rider strategies are not as rewarding, so natural selection ultimately phases them out of existence. Thus co-operation evolves.

However, some research is beginning to pick holes in this particular model of things. Whilst tit-for-tat is better than free-riding, in a world dominated by tit-for-tat people “always co-operate” is a strategy which is equally as successful. This is because there is no real difference between co-operating with people who are co-operating because you do and co-operating with people who always co-operate. In this sense “always co-operate” is a neutral mutation which can creep into a tit-for-tat world.

Genetic drift can cause neutral mutations to become dominant in the population, despite the fact that they over no benefit over an alternate strategy. As such you could transition from a tit-for-tat world to an always co-operate world. Whilst the former isn’t susceptible to invasion by free-riders the latter is. This reveals a large flaw in such explanations of co-operation: neutral mutations could make it vulnerable to exploitation.

Indeed, an international team of researchers recently found that in the iterated prisoners dilemma there were more ways to “unevolve” co-operation than to maintain it. Whilst tit-for-tat would rise to dominance there were several potential neutral mutations which could worm their way into the simulation. These neutral mutations were more likely to enable free-loaders (called defectors in the prisoners dilemma) to invade the scenario than maintain the status quo of co-operation.

So this leaves us back where we started, at a paradox. If an evolutionary stable strategy cannot be guaranteed to maintain co-operation, why is it maintained nonetheless? So the researchers started modifying the parameters of the prisoners dilemma in an effort to find out which scenarios resulted in stable co-operation.  Aside from artificially biasing the simulation in favour of co-operation, they found that the relevant variables were how many times the prisoners dilemma was repeated and how likely someone was to meet someone who practised the same strategy as them.

The simulation consisted of 200 individuals who played the prisoners dilemma at least once and then reproduced, the success of which was based on how many “points” they earned in the prisoners dilemma. During reproduction there was a chance to mutate and change into a different strategy. They varied the number of times each individual played the prisoners dilemma before reproducing, but found that a defective strategy would always emerge victorious.

The results of the simulation, with each square being one run of the simulation. The colour represents the likelihood of co-operation, with 50% being akin to the tit-for-tat strategy

However, if they increased the chance that an individual was to meet someone practising their strategy they found that co-operation would emerge. Further, as repetition increased the threshold of “non-random meeting” was lowered. Only a slightly above average chance of meeting someone who practised the same strategy as you was needed when the game was repeated a lot and lo, tit-for-tat emerged as the best technique.

Humans are long-lived so the fact there must be multiple interactions is no problem. You’re likely to interact with the same people more than once in the 70 years you’re on this planet (or even the 40 or so chimps are alive). But what real life circumstances could make you more likely to meet someone who behaves like you? The researchers note that this could simply be caused by a population being structured, with similar being “clumping” together.

So it would seem that repetition is not enough to allow co-operation to develop but a (slightly) structured population is needed as well.

van Veelen M, García J, Rand DG, & Nowak MA (2012). Direct reciprocity in structured populations. Proceedings of the National Academy of Sciences of the United States of America, 109 (25), 9929-34 PMID: 22665767

Related posts


Jesse Marczyk · 27th July 2012 at 3:09 am

It’s probably also worth noting that the prisoner’s dilemma is, of course, a contrived scenario, which is useful for demonstrating the basic principle, but not fully reflective of real life. The options in that game are either (a) cooperate or (b) defect, with no room allowed for the natural continuum along which cooperation falls.

Given a population of (generally) tit-for-tat players who are also seeking to maximize their earnings when the opportunity arises, even if only by a little bit, “always cooperate” is most certainly not an equally good option.

    Adam Benton · 27th July 2012 at 11:05 am

    Whilst they did note that other games showed similar trends to what they reported, I’d be willing to bet they’re also contrived situations. Nonetheless, given the prisoner’s dilemma has often been used to develop ideas regarding co-operation I don’t see much of a problem with pointing out the prisoner’s dilemma doesn’t actually say what people think it does.

    Artem Kaznatcheev · 9th August 2012 at 1:17 pm

    I agree with PD as a contrived example, but for a slightly different reason. I have no issues with using an abstraction like a simple 2 strategy game to study cooperation. The idea with a lot of these simple models is not to suggest that this is really how the world is, but to elicit some basic features that we should be focusing on and to build a solid mathematical (or at least computational) theory which we can use to frame empiric work.

    That being said, from the point of view of game theory the specific payoffs don’t matter to much so there is some wiggle room. What matters is the relative ordering of payoffs. However, empirically even this can’t be estimated for many interactions. Further, having slightly different orderings (say PD versus Hawk-Dove) can produce extremely different dynamics in many settings (like spatial games). I talk about this in the intro my CAS AAAI paper (http://www.cs.mcgill.ca/~akazna/kaznatcheev20100910.pdf) but should probably just write up a blog post with just the key references.

    As for the continuum of cooperation to defection, it usually doesn’t matter to have this continuum in most settings. In terms of game theory the difference is between pure and mixed strategies and is certainly important in a classical setting (mixed strategy Nash eq. always exists, Pure strategy ones don’t always do). However, in an evolutionary setting the proportion of pure-strategy phenotypes in the population is what creates the effect of a mixed strategy. You can use the population mixture instead of individual agents expressing mixed strategies to achieve a mathematically equivalent result. Of course, this doesn’t always apply, for instance if you want to explicitly study the effect of randomization: http://egtheory.wordpress.com/2011/10/03/cognitive-cost-of-agency/

Vicente Adrian Lara Rodriguez · 10th August 2017 at 11:09 am

Multi-level selection, ultra-sociality or group selection made people cooperate among their countrymen to better compete abroad.

    Adam Benton · 19th August 2017 at 1:35 pm

    It’s worth noting group selection is a somewhat controversial idea the majority of biologists and myself don’t accept.

Leave your filthy monkey comments here.

This site uses Akismet to reduce spam. Learn how your comment data is processed.