Bshary, Redouan

2017, Wubs, Matthias, Bshary, Redouan, Lehmann, Laurent

Les individus qui interagissent socialement doivent souvent décider entre aider leur prochain, en leur procurant un bénéfice a un possible cout, ou pas. Tandis qu'une paire d'individus s'en sort mieux si les deux s'entraident, un individu peut tirer un bénéfice relatif s'il se décide à ne pas aider tandis que l'autre l'aide. Cette situation crée un dilemme social. Parce que les interactions sociales sont souvent répétées, les individus peuvent conditionner leurs propres actions sur les actions de leur partenaire lors d'interactions passées. Il existe trois mécanismes de contrôle de partenaire qui stabilisent la coopération lors d'interactions répétées entre paires d'individus: la réciprocité positive (la décision d'aider est conditionnée sur l'aide du partenaire lors de l'interaction précédente), la punition, et le changement de partenaire. Mais les conditions sous lesquelles un mécanisme domine sur les autres lorsque les trois mécanismes coévoluent dans une même population sont mal comprises.
Un autre point qu'il est important de considérer pour comprendre les interactions sociales est que le comportement exprimé peut aussi dépendre sur l'environnement. Il est probable que le toilettage social chez les primates, par exemple, dépend de la compétition sur les ressources de nourriture. Si la nourriture est concentrée dans l'espace et facilement monopolisée, les individus en haut de la hiérarchie sociale peuvent défendre ces ressources, auquel cas les individus en bas de la hiérarchie doivent les toiletter afin d'être tolérer sur leur territoire et accéder à leur ressource. Afin de comprendre ce comportement, il est nécessaire de comprendre les conditions sous lesquelles le toilettage est échangé contre du toilettage, ou contre la tolérance sur un territoire.
Dans cette thèse, je développe et utilise des modèles agents basés pour explorer la coévolution des mécanismes de contrôle de partenaire et l'évolution de patrons de toilettage chez les primates.
Dans le premier chapitre, je démontre que dans une population panmictique, plus les interactions sont répétées, plus il est probable que le changement de partenaire soit le mécanisme de contrôle de partenaire dominant. Si les interactions sont restreintes au sein de petits groupes de non-apparentés, alors la punition est le mécanisme le plus probablement favorisé par la sélection. Les conditions pour que la réciprocité positive soit dominante sont moins clairement définies.
Dans le deuxième chapitre, j'étudie comment la dispersion limitée chez la progéniture, le recouvrement de génération, et le cout de la complexité influencent la compétitivité de chaque mécanisme de contrôle. Il est démontré que la compétitivité du changement de partenaire est accrue par le recouvrement de génération tandis que celle de la punition est très diminuée par le cout de complexité. De plus, alors que les conditions pour que ces mécanismes de contrôle de partenaire soient dominants s'élargissent avec la structure d'apparentement, les conditions pour que l'aide inconditionnelle domine sur l'aide conditionnelle se restreignent.
Dans le troisième chapitre, je développe un modèle d'apprentissage par renforcement pour simuler les interactions de toilettage et l'accès a la nourriture chez les primates. On observe des patrons de toilettage réciproque lorsqu'il n'y pas de compétition pour les ressources, et des patrons de toilettage dirigé vers le haut de l'échelle hiérarchique lorsque la compétition pour les ressources augmente. Il est montré que l'effort dans le toilettage n'augmente pas forcément avec la compétition pour les ressources, et qu'une augmentation d'agressivité amène le toilettage à être plus réciproque.
En bref, les comportements sociaux d'entraide dans les milieux naturels sont observés dans une multitude d'interactions sociales répétées. En explorant une gamme variée de conditions, les modèles développés dans cette thèse permettent de mieux comprendre quels mécanismes de contrôle de partenaire sont favorisés par la sélection naturelle. De plus, le modèle du troisième chapitre montre que les patrons de toilettage dépendent de plusieurs paramètres naturels importants., Individuals that interact socially regularly have to make decisions whether to help another individual (provide some payoff benefit, possible at a personal payoff cost) or not. Here, a pair of individuals is best off if both individuals help each other, but a single individual may gain a relative payoff advantage by not helping (defecting) while the partner does choose to help, thus creating a dilemma. Because social interactions are often repeated, individuals can condition the actions they take on the actions taken by their partner in previous rounds of interaction. The so-called partner control mechanisms positive reciprocity (where acts of helping are conditioned on receiving help from the partner), punishment, and partner switching have all been shown to stabilize cooperation in populations where the individuals engage in repeated pairwise interactions. What remains unclear, however, is under which conditions each partner control mechanism will be dominant in a population if the the partner control mechanisms coevolve.
Additionally, the expressed behaviour in repeated social interactions may depend on the state of the environment in which the interactions take place. Social grooming in primates is likely to depend on the food competition that the individuals experience. If food is clustered and monopolizable, high ranked individuals may defend food sources, where low ranked individuals then need to groom high ranked ones in order to be tolerated on the food source, resulting in grooming being directed up the hierarchy. However, the conditions that cause grooming to be traded for grooming or grooming to be traded for tolerance have yet to be quantified.
In this thesis, I developed several agent-based models in order to investigate both the coevolution of various partner control mechanisms and the grooming patterns in primates.
In chapter one, I demonstrated that in a well-mixed population the likelihood of partner switching being the dominant partner control mechanism increases with increasing number of rounds of interaction. Furthermore, if interactions are localized to small groups of unrelated individuals, then punishment is more likely to be favoured by selection compared to the well-mixed case, while the conditions where positive reciprocity is dominant are less clearly defined.
In chapter two, I investigated how limited migration of offspring, overlapping generations, and complexity costs affect the competitiveness of each partner control mechanism. It is shown that the relative competitiveness of partner switching is increased due to generational overlap, while punishment is most strongly negatively affected by complexity costs. Additionally, while the conditions where these partner control mechanisms are dominant in the population increases if the population is kin structured, the conditions where unconditional helping is dominant over conditional helping strategies are rather stringent.
In chapter three, I developed a reinforcement learning model that simulated grooming and feeding interactions in primates. The model generated patterns of grooming reciprocity in the absence of food competition, while grooming was found to be directed up the hierarchy if individuals compete for food. It is shown that grooming up the hierarchy may not necessarily increase with increasing food competition, and that an increase in aggressiveness causes grooming to become more reciprocal.
In summary, helping behaviours occur in a diversity of repeated social interactions in natural populations. By exploring a large range of conditions, the models developed in this thesis provide insights regarding which partner control mechanism is likely to be dominant in a population. In addition, the model from chapter three shows how grooming patterns may depend on a variety of relevant parameters.

2010, Van de Waal, Erica, Bshary, Redouan, Lehmann, Laurent, Van Schaik, C.

Social learning is the basis for allowing the transmission of specific behaviours inside a social unit, i.e. the formation of traditions. Early field studies suggested the existence of traditions in non-human animals, while more recent laboratory experiments have demonstrated social learning abilities in a variety of species. I established a unique bridge between these perspectives by conducting three different social learning experiments on six groups of wild vervet monkeys (Chlorocebus aethiops) at the Loskop Dam Nature Reserve, in South Africa. Using this approach, I investigated what mechanisms wild vervets use when they learn a task socially. Furthermore, theoreticians pointed out that social learning is only driven by positive selection under certain conditions. Therefore, I investigated how important the identity of a model is for the occurrence of social learning and I tried to understand why some individuals are more copied than others from a functional perspective. Finally, by analysing the stability over time of the socially acquired behaviours, I could ask whether traits acquired through social learning may turn into arbitrary traditions. First, I presented laboratory-style ‘artificial fruit’ boxes that had two doors on opposite, differently coloured ends. A dominant individual invariably monopolized the box during an initial demonstration phase, in which one door was blocked. This created consistent demonstrations of one of the two possible solutions in each of six study groups. Three groups had female models and three had male models. Following demonstrations I found a significantly higher participation rate (‘stimulus enhancement’) by other group members and significant evidence for manipulation of the same door (‘local enhancement’) in groups with female models compared to groups with male models. These differences appeared to be due to selective attention of bystanders to female model behaviour, while male and female models attracted similar numbers of bystanders and showed similar levels of aggression towards those bystanders. The results demonstrate the eminent role of dominant females as a source for directed social learning in a species with female philopatry. In this same first experiment, I analysed the proper solving of the task. During their first trial, I observed which individuals managed to open artificial fruit thus accessing the reward. This time I did not find an effect of the model sex but I found that the two groups in contact with humans were more successful than the others. This result suggests some enhanced manipulation skills due to contact with humans or their facilities. The second experiment involved a more complex artificial fruit to test for sequence imitation, where two steps were necessary to open the door: remove a bar on the top of the box which releases a rope that was blocking the door and then pull the door. Vervets largely failed to show more complex social learning abilities in this experimental setup. However monkeys in group with models touched the bar significantly more often during their first manipulation than control individuals did. This latter result implies again the use of ‘local enhancement’ as social learning mechanism in wild vervets. Finally, I conducted a food cleaning experiment that was inspired by a classic study that documented the spread of sweet potato washing in a semi-natural population of Japanese macaques. I offered the monkeys grapes covered with sand and noted if and how they cleaned the food before eating. Each group was subjected to 15 trials. Vervets either did not clean the grapes or either rubbed with their hands, rubbed on substrates, or opened the fruit with their teeth or hands to eat the inside only. I found strong variance between individuals of the same group as well as between groups with respect to the techniques used. Matrilines rather than entire groups appeared to be the key unit for social transmission, where conformity of feeding techniques could be documented. Taken together, the findings imply that in species with complex social structures, migration does not necessarily lead to an exchange of socially acquired information within populations, causing much localized traditions.