The puzzle of cooperation and strong reciprocity

Human cooperation and altruism towards non-kin poses two fundamental questions for biology and the behavioural sciences. Firstly, why do genes associated with these behaviours survive when evolutionary theory appears to predict that a ‘cheating’ strategy is fitter and will thus drive such genes to extinction [1], [2]? Secondly, ‘strong reciprocity’ is defined as a propensity to cooperate unconditionally even if this is costly and provides neither present nor future rewards to the co-operator/altruist [3], [4]. Why, therefore, does this behaviour persist as it incurs a cost and/or offers no present or future rewards [3], [4]?

Much current theory suggests that the first question can be resolved by punishment of ‘cheats’ who do not reciprocate altruistic acts i.e. reciprocal altruism [5], [6] and/or by cooperative individuals recognizing each other by reputation and gaining fitness by only associating with other cooperative individuals i.e. indirect reciprocity [7], [8], [9], [10], [11].

‘Strong reciprocity’, however, has been observed in one-off, anonymous encounters under experimental conditions [12] where the reciprocal and reputation effects required by reciprocal altruism or indirect reciprocity theory could not influence the behaviour of participants [4]. Instead, it has been explained by ‘new’ group selection linked to extinction-threatening events [3], [13], to cultural evolution [4], [14], to gene-culture co-evolution [15] and to a combination of the last two factors [16], [17]. Others explain it as a response to conditions of uncertainty in reciprocal relationships [18] or as a maladaptive ‘misfiring’ of evolutionary mechanisms in modern, experimental settings [19].

Extra-somatic weapons in human evolution

Human cooperation towards non-kin has often been seen as unique in nature [8], [12], [14], [16], [20]. Here, we propose that this very uniqueness might be attributable to unique selection pressures likely to have been present in human evolution itself. The invention of tools and weapons is generally seen as being an important step in human evolution as it enabled hominins to consume a high protein diet through the hunting of game and/or scavenging of meat [21]. Such weapons should be distinguished from somatic weapons (e.g. prominent canine teeth, pronounced musculature), with only humans employing extra-somatic weapons on a widespread and systematic basis. Here, we explore whether their invention might have led to a cultural niche that could have resulted in the evolution of distinctively human cooperation. It has been claimed that cultural processes, particularly in humans, can lead to the creation of niches that change selection pressures to which individuals are exposed, thereby influencing their evolution [22], [23], [24].

The oldest complete hunting weapons yet discovered have been dated to approximately 400,000 years ago [25], although it is speculated that wooden weapons could have existed up to 1 million years before the present [26]. These wooden hunting spears resemble projectile weapons [25] although there is some question as to whether they would have been too heavy to have been used as true projectiles [27] and may have been used instead as thrusting weapons. Recent analysis has suggested that humans first developed long-range projectile weaponry in Africa 70–90,000 years ago [27]. Nearly every known human population over the past 50,000 years has used projectile weapons (e.g. spears and darts, bows and arrows) [28] and those few groups who lacked them (e.g. Tasmanian Aborigines) are known to be descended from populations that did possess them [28].

It has been suggested that the invention of extra-somatic weapons in human evolution could have had an important impact on relationships within hominin groups [5], [6], [29]. In many species, intraspecific conflict has been observed to be of the ‘limited war’ type in which ritualized tactics or inefficient somatic weapons are used to settle disputes without death or serious injury resulting [30]. However, in the case of early humans, the lethal effectiveness of extra-somatic weapons is likely to have selected against those behavioural adaptations that inhibit intraspecific violence observed in many other species. Specifically, the development of extra-somatic weapons could have resulted in:

• an increased frequency of agonistic encounters as dominant individuals, hitherto reliant on physical size, strength and intimidation alone to gain access to valued resources (e.g. mating with females, hunted meat), would have been more open to challenge [29];
• a greater likelihood that such encounters would have proved lethal for the protagonists due to the effectiveness of the weapons and the speed with which they could have resulted in serious injury or death [29] and;
• a consequently greater symmetry or equalization in power between individuals [5], [6], [29] as it could have proved as easy for a subordinate to kill a dominant as vice versa.

In such a cultural niche it is likely that natural selection would have discriminated strongly against individuals whose behaviour provoked an above average level of within-group aggression against them. ‘Cheats’ and non-cooperators might well have experienced involvement in such lethal disputes at a higher frequency than others and thus faced a correspondingly greater risk of injury or death. Providing that the costs of involvement in agonistic encounters, where weapons were employed, exceeded the benefits of ‘cheating’ and non-cooperation then such strategies would have proved maladaptive.

In contrast, cooperators inclined to reduce their fitness in order to help others within their group could as a result have been less likely to have become involved in lethal disputes and might consequently have experienced relatively greater fitness. Providing that the cost of this helping behaviour was less than the selective benefit of a reduced frequency of involvement in lethal fights then cooperation could have proved an adaptive strategy in this cultural niche.

To test this hypothesis, we employed two versions of the evolutionary Iterated Prisoner's Dilemma (IPD) model – one in which weapons use was absent and one in which it was present. We used computer simulation to measure the performance of a range of IPD strategies, both cooperative and non-cooperative, in order to quantify the effects of weapons use on them.

The traditional iterated prisoner's dilemma (IPD) model

The IPD model [31], [32] is a classic formulation of how mutual co-operation can evolve in a world of selfish individuals and its process is well known (see Supporting Information, File S1. The iterated prisoner's dilemma (IPD) model). Each player has a choice of whether to cooperate or defect on each move. If both cooperate each receives three points but if both defect each gets only one point. If one cooperates and the other defects then the former receives no points and the latter five points. On any single move it always pays to defect but cooperation has been found to emerge where the endpoint of the series of interactions between two players is unknown [32]. In modelling the IPD, this uncertainty is reflected in the duration of each interaction being determined by a certain probability or discount parameter (see File S1). Computer tournaments have been used to identify those strategies that perform best, with ‘Tit for Tat’ (TFT) (cooperate on the first move and then copy the opponent's last move) [31] and Pavlov [33] generally found to be most successful.

The evolutionary IPD model

The evolutionary IPD model [34] extends the principle of the traditional IPD model by reflecting the payoffs received by players in one generation in terms of copies of themselves represented in the next generation. Stochastic universal sampling is used to ensure that players produce offspring in proportion to payoffs received so that those with higher payoffs reproduce at a proportionately higher rate than those with lower payoffs. In so far as payoffs reflect fitness the evolutionary IPD model can be seen to mimic natural selection, although recombination and mutation are not simulated.

In our evolutionary IPD model a population of 40 players compete against each other on a round robin basis. We chose 25 widely recognized IPD strategies taken from the scientific literature [34], of which 14 have been classified as cooperative and 11 as non-cooperative [35] (see Supporting Information, File S2: IPD strategies employed).

In each round (see Supporting Information, File S3: Definitions of terms) eight strategies were chosen at random with five players initially adopting a particular strategy. To ensure that each strategy was simulated a sufficient number of times to remove the effects of chance we repeated each round 100 times per generation and averaged the payoffs. Strategies were chosen randomly so that, for example in the initial round, there was an 8/25 = 0.32 probability of any one strategy being chosen. With 100 rounds in that game, each strategy was therefore run an average of 100*0.32 = 32 times. As strategies were subsequently eliminated, this probability was adjusted throughout the competition. With stochastic universal sampling used to choose players for the next generation on the basis of payoffs received in the previous one, the game was repeated for 100 generations, showing how each of the 25 strategies increased, decreased or were eliminated over the duration of the competition.

The performance of each strategy was measured by the number of times each survived for the full 100 generations. As long as a single player adopting a particular strategy was present at the end then that strategy was deemed to have survived. Survival time is seen as providing a comprehensive index of the performance of IPD strategies [35] and in this context offers a measure of the relative fitness of individuals adopting each strategy.

The ‘weapons use’ IPD model

The ‘weapons use’ IPD model we designed for this study was based on the evolutionary IPD model but we also set out to simulate the effects of extra-somatic weapons on the fitness of players. Our model works on the basis that a defection is, in effect, a refusal to cooperate and an attempt to exploit the other player. It therefore assumes that the more defections there are between players, the greater will be the chance of a dispute occurring between them. It thus captures a key aspect of the real world – that, in relationships with others, ‘cheats’ and uncooperative individuals (i.e. those inclined to defect) are more likely to be involved in disputes than ‘nice’ individuals who are more inclined to cooperate.

When a player accumulated 200 defections as a result of moves both by the player and the player's opponents a dispute was deemed to occur. To simulate an environment of equalized power, each player had the same probability of being eliminated in each dispute (i.e. p = 0.05). If either or both were eliminated they took no further part in the competition. If they were not, the process continued until a further 200 defections were accumulated and another dispute was deemed to occur.

Thus, as well as stochastic universal sampling reflecting the payoffs of players who adopt a particular strategy, our model quite separately reflected the effects on fitness of elimination as a result of disputes. In the ‘weapons use’ IPD model, therefore, the performance of each strategy was a measure of both decisions on whether to cooperate or defect and also the effects of the number of defections leading to disputes.

Finally, in all simulations we contrasted the performance of strategies under the evolutionary IPD model (i.e. without weapons use) with that under the ‘weapons use’ IPD model to quantify the impact of weapons use. We were thus able to simulate the relative impact of weapons use on the fitness of individuals adopting each strategy in this environment.

Other points

Our population of 40 players lies within the typical size range of 15–50 observed in modern hunter/gatherer groups [36]. Nevertheless we are aware that there is mobility between hunter/gatherer groups [37] and that therefore a typical hunter/gatherer would interact with more than 40 individuals over a lifetime. However, we make the simplifying assumption that transfers in and out of groups would have had a broadly neutral effect on the frequency with which each player would encounter the 25 IPD strategies modelled. This point is reflected in the re-running of each round 100 times per generation with different combinations of strategies being encountered in each round.

The ‘weapons use’ IPD model has three key parameters (i.e. average number of interactions with other players in a generation, number of defections before a dispute occurs, probability of being eliminated as a result of a dispute) that, taken together, determine the fitness of each player. The values chosen, however, were not arbitrary but based on a population consisting entirely of players who always defect (AllD) and who are thus at the extreme end of the continuum of behaviour being examined. Based on the values selected, in an AllD population there is a probability of p = 0.50 of being eliminated in an average lifetime (see Supporting Information, File S4: The parameters and values used). These values therefore provide a clear benchmark against which the performance of the 25 IPD strategies was calibrated.