L'objet de cette thèse est d’explorer les mécanismes évolutionnaires à l'origine de la punition coûteuse chez les humains. Dans le chapitre 1, nous montrons que les comportements punitifs observés en laboratoire sont plus en adéquation avec des processus de sélection individuelle plutôt que de sélection de groupe. Dans le chapitre 2, nous fournissons des preuves empiriques que la décision de punir relève principalement de raisonnements intuitifs causés par une forte exposition à des environnements à réciprocité directe. Enfin, dans le chapitre 3, nous émettons quelques doutes quant à l'hypothèse (plutôt récente) que des stratégies de changement de partenaires pourraient se substituer à la punition coûteuse dans la nature. En somme, nos résultats soulignent le caractère maladaptatif du sentiment punitif tel qu'il se manifeste dans les jeux expérimentaux à un coup. / The aim of this thesis is to investigate the evolutionary roots of costly punishment in humans. In chapter 1, we show that available experimental data are more compatible with behavioral patterns sculpted by individual selective pressures than higher-population processes. In chapter 2, we provide experimental evidence that punishing decisions mostly rely on intuitive reasonings originally adapted to direct reciprocity environments. Finally, in chapter 3, we cast doubt on the (recent) claim that cheap partner switching opportunities make costly punishment anecdotal in the wild. In sum, our results emphasize the maladaptive nature of punitive behaviors observed in anonymous, one-shot laboratory settings.

Costly ‘altruistic’ punishment, where an individual intervenes to punish someone for behaving unfairly towards another or for violating a social norm, seems to be vital for large-scale cooperation. However, due to the costs involved, the evolution of this behaviour has remained a puzzle. The thesis initially describes why punishment is costly and explains why current theories do not sufficiently explain its evolution in the context of these costs. The thesis then offers a solution to this puzzle in the form of a dominance-based theory of the evolution of punishment. The theoretical underpinnings of this theory are discussed in reference to the previous literature, specifically how a dominant position provides sufficient heterogeneity in the cost and benefits of punishment to allow the behaviour to evolve at the individual-level of selection. Across 10 studies, the thesis empirically investigates the role dominance is theorised to play in costly punishment behaviour. First, the judgements observers make about punishers are investigated. It is demonstrated that punishers are perceived as dominant but, unlike individuals who engage in other aggressive behaviours, punishers are also well liked. While successful punishers are judged to be of the highest rank in a social group, the wider social judgements of punishers are dependent on the attempt at punishment only; successful and unsuccessful punishers are seen as equally dominant and well liked, suggesting that the willingness to attempt punishment can honestly signal both dominance and ones pro-sociality. However, additional studies show that observers a) perceive subordinate punishers will face a great deal of retaliation, b) show surprise when subordinates attempt to punish, and c) expect that dominants will punish and be successful, whereas subordinates are expected to never punish. Thus, while there are reputational benefits from punishment, only dominant individuals can actually access them. Second, the effect of a dominant position on punishment behaviour is investigated. Two studies sought to simulate the greater access to resources that dominants enjoy, and demonstrate that individuals who receive more resources from group-level cooperation will punish free-riding more frequently and more severely than those who receive less resources. Moreover, individuals who are in a stable dominant position, i.e. who can continually benefit to a greater degree than others from group cooperation, punish even more frequently and severely than when individuals receive additional resources alone. The results show that individuals only punish when it is cheap for them to do so and when investment in the public good (by punishing) can produce higher future returns for them. A dominant position provides the opportunity for both of these. Further studies demonstrate that individuals at the centre of a social network, an example of a ‘real life’ informal dominant position, are more sensitive to unfairness when making punishment decisions compared to those at the periphery of a group. However, when punishment decisions are public, and there are no economic incentives to punish, individuals behave in a similar manner regardless of social position. Taken together, the results of the empirical studies support the proposed dominance-theory of costly punishment. The theoretical implications of the dominance-theory of punishment are discussed in reference to both the proximate occurrence of punishment and its evolutionary origins in dominance and dominant behaviours. The practical implications of this theory will also be discussed, specifically in regard to when and why individuals will act in defence of the public good. While further investigation is necessary, a dominance-theory of punishment explains both results of this thesis and the findings of the wider literature, and as such provides a coherent and compelling explanation for the evolution of costly punishment and its associated emotions.

Cooperation in joint enterprises poses a social dilemma. How can altruistic behavior be sustained if selfish alternatives provide a higher payoff? This social dilemma can be overcome by the threat of sanctions. But a sanctioning system is itself a public good and poses a second-order social dilemma. In this paper, we show by means of deterministic and stochastic evolutionary game theory that imitation-driven evolution can lead to the emergence of cooperation based on punishment, provided the participation in the joint enterprise is not compulsory. This surprising result - cooperation can be enforced if participation is voluntary - holds even in the case of 'strong altruism', when the benefits of a player's contribution are reaped by the other participants only. (authors' abstract)