There is something extraordinary about human culture. The striking complexity of our technologies, institutions, beliefs, and norms has allowed us to colonise the entire planet. One aspect in which human culture is unique relates to its cumulative nature – we accumulate and build on knowledge from the previous generations, leading to incremental improvement in skill, which allows us to produce technologies no one individual could have invented on their own. Understanding the drivers and dynamics of this type of cumulative culture is essential for understanding how human culture has interacted with human evolution. This thesis is concerned with precisely that, and uses a mixture of theoretical and experimental approaches linking individual-level decisions to population-level processes in cumulative culture contexts. Chapter 1 provides some essential background information. In Chapter 2 I used an agent-based simulation model to show that refinement, or incremental improvement in cultural traits, can lead to a drastic decrease of cultural diversity at the population level. This pattern was confirmed using experimental data from a collaborative programming competition in Chapter 3, where I showed that in a cumulative setting, the differential riskiness of copying and innovation drives participants to converge on very similar solutions, leading to a loss of cultural diversity. In Chapter 4 I explored individual differences in social learning strategies, finding considerable variation in how individuals rely on copying, with more successful individuals being more exploratory. I found that successful individuals had more influence on subsequent entries, which is consistent with a prestige bias. Finally, Chapter 5 addressed the link between group structure, diversity, and cumulative improvement. I found that larger groups accumulate more improvement than smaller groups, but smaller groups can also inhibit the convergence patterns we witnessed in larger groups, suggesting an optimal level of connectivity responsible for cumulative improvement.

A aprendizagem social (AS) é a aprendizagem facilitada pela observação ou interação com outro indivíduo ou com produtos produzidos pelo mesmo. Para aprender socialmente, animais, humanos e não-humanos, possuem estratégias para utilizar ou não informações socialmente disponíveis, definindo quando e de quem copiar o comportamento. Um tema discutido atualmente é cultura cumulativa (CC) em animais não-humanos: a capacidade de uma população acumular aperfeiçoamentos de técnicas ao longo do tempo por meio da AS. Este estudo teve como objetivos investigar (1) se os macacos-prego são capazes de aprender socialmente aprimoramentos de técnicas já estabelecidas; (2) quais vieses de transmissão social e estratégias de AS são utilizadas pelos macacos-prego; e (3) examinar a influência de 4 fatores cognitivos (ensino, linguagem, imitação e prossocialidade) e 4 fatores sociais (scrounging, monopolização do recurso pelos dominantes, falta de atenção para o comportamento de inovadores de baixo status social e conservadorismo comportamental) na difusão de novos comportamentos e argumentar como esses fatores afetariam o desenvolvimento de uma CC simples. Para atingir estes objetivos apresentamos duas caixas-problema para um grupo de macacos-prego selvagens: uma com três estágios de dificuldade progressivos (caixa-problema de alimentos) e a outra com dois (caixa-problema de suco). Estas caixas-problema tinham como finalidade introduzir novos comportamentos no grupo e induzir aprimoramentos de técnicas já estabelecidas. Com relação ao primeiro objetivo (1), utilizando a Networkbased diffusion analysis nós identificamos a AS por observação próxima do estágio mais complexo da caixa-problema de suco. Com isso inferimos que os macacos-prego são capazes de adquirir, através da AS, técnicas mais complexas e eficientes de uso de ferramentas. Isso somado com o aprendizado observacional, nos permite afirmar que estes macacos-prego possuem a capacidade de reconhecer que um comportamento sendo produzido por outro indivíduo é mais eficiente que o próprio, e a flexibilidade para mudar para este comportamento alternativo. A respeito do segundo objetivo (2), ao analisar os viéses de transmissão social nós encontramos que os indivíduos tendem a observar preferencialmente os indivíduos mais proficientes e os indivíduos subordinados e que machos subordinados observam mais os coespecíficos que machos dominantes. Por fim (3), a respeito dos fatores sociais que afetam o desenvolvimento da CC, encontramos que os indivíduos dominantes tendem a monopolizar os recursos. A monopolização, contudo, não parece ter interferido negativamente na AS de aprimoramentos de técnicas. É possível que a grande tolerância que os macacos-prego têm com seus coespecíficos compense a influência negativa que a monopolização dos recursos poderia gerar. Não encontramos evidências de que os outros fatores sociais influenciaram negativamente a AS, e com isso, o desenvolvimento da CC. Em relação aos fatores cognitivos, não encontramos evidências de ensino, prossocialidade ou comunicação significativa e não pudemos avaliar a ocorrência de emulação ou imitação. Podemos concluir que esses fatores não são necessários para a AS de aprimoramentos de técnicas. Para aprender socialmente um aprimoramento de técnica que está além da capacidade de inovação individual, contudo, é provável que mecanismos com maior fidelidade de AS, como a emulação, sejam necessários. Acreditamos que caso a emulação seja suficiente para o desenvolvimento de uma CC simples em animais não humanos, os macacos-prego tem o potencial para desenvolvê-la. Já que (i) eles possuem a capacidade de reconhecer um comportamento mais eficiente e a flexibilidade de mudança de comportamento, (ii) evidências em cativeiro apontam que macacos-prego podem emular um comportamento e (iii) as populações tendem a ser tolerantes / Social learning (SL) is learning that is facilitated by the observation of, or interaction with, another individual or its products. When learning socially, animals, both human and non-human, employ strategies to access and use (or not) socially provided information. These strategies define when and from whom to copy a behaviour. A topic that is currently being discussed is the occurrence of cumulative culture (CC) in nonhuman animals: the population capacity to accumulate technical/behavioural improvements over time through SL. The objectives of this study were to investigate (1) whether monkeys are able to socially learn improvements of established techniques; (2) which social transmission biases and SL strategies were used by the capuchin monkeys; (3) the influence of 4 cognitive factors (learning, language, imitation and prosociality) and 4 social factors (scrounging, tendency of dominant individuals to monopolize resources, lack of attention to low-status inventors and conservative behaviour) in the diffusion of new behaviours and then argue how these factors would affect the development of a simple CC. We presented a group of capuchin monkeys with two puzzle boxes, one with three progressively complex stages (food-box) and other with two stages (juice-box). These puzzle boxes were developed to introduce new behaviours into the repertory of the group and induce improvements of established techniques. Regarding our first goal (1), by using the Network-based diffusion analysis we identified the occurrence of SL by close observation of the most complex stage of the juice-box. Thereby we infer that capuchin monkeys can acquire, through SL, more complex and efficient techniques of tool use. This, together with observational learning, allows us to affirm that these monkeys have the capacity to recognize that a behaviour performed by another individual is more efficient than their own, and the flexibility to change to this alternative behaviour. About our second objective (2), the social transmission biases, we found that individuals tend to observe more the most proficient individuals and subordinate individuals and that subordinate males observe more than dominant males. Lastly (3), regarding the social factors, we found that dominant individuals tend to monopolize the resources. Monopolization, however, does not seem to have negatively interfered with the SL of technical improvements. It is possible that the increased tolerance that capuchin monkeys have regarding their coespecifics compensates the negative influence that the monopolization could generate. We found no evidence that the other social factors influenced negatively the diffusion of information, and thus the development of CC. Regarding the cognitive factors, we did not find evidence of teaching, prosociality or significant communication. We could not evaluate the occurrence of emulation or imitation. Thus, we can conclude that these factors are not necessary for the SL of simple technical improvements. Observational learning is enough for this learning to occur. To socially learn a technical improvement that is beyond the capacity of individual innovation, however, it is likely that mechanisms with higher SL fidelity, such as emulation, are necessary. Considering that capuchin monkeys can (i) recognize that a behaviour is more efficient than their own and the flexibility change to that behaviour, (ii) that there is evidence, in captivity, that capuchin monkeys can emulate behaviours and (iii) that populations tend to be tolerant. If emulation is sufficient for the development of a simple CC in non-human animals, we believe that capuchin monkeys have the potential to develop it

A aprendizagem social (AS) é a aprendizagem facilitada pela observação ou interação com outro indivíduo ou com produtos produzidos pelo mesmo. Para aprender socialmente, animais, humanos e não-humanos, possuem estratégias para utilizar ou não informações socialmente disponíveis, definindo quando e de quem copiar o comportamento. Um tema discutido atualmente é cultura cumulativa (CC) em animais não-humanos: a capacidade de uma população acumular aperfeiçoamentos de técnicas ao longo do tempo por meio da AS. Este estudo teve como objetivos investigar (1) se os macacos-prego são capazes de aprender socialmente aprimoramentos de técnicas já estabelecidas; (2) quais vieses de transmissão social e estratégias de AS são utilizadas pelos macacos-prego; e (3) examinar a influência de 4 fatores cognitivos (ensino, linguagem, imitação e prossocialidade) e 4 fatores sociais (scrounging, monopolização do recurso pelos dominantes, falta de atenção para o comportamento de inovadores de baixo status social e conservadorismo comportamental) na difusão de novos comportamentos e argumentar como esses fatores afetariam o desenvolvimento de uma CC simples. Para atingir estes objetivos apresentamos duas caixas-problema para um grupo de macacos-prego selvagens: uma com três estágios de dificuldade progressivos (caixa-problema de alimentos) e a outra com dois (caixa-problema de suco). Estas caixas-problema tinham como finalidade introduzir novos comportamentos no grupo e induzir aprimoramentos de técnicas já estabelecidas. Com relação ao primeiro objetivo (1), utilizando a Networkbased diffusion analysis nós identificamos a AS por observação próxima do estágio mais complexo da caixa-problema de suco. Com isso inferimos que os macacos-prego são capazes de adquirir, através da AS, técnicas mais complexas e eficientes de uso de ferramentas. Isso somado com o aprendizado observacional, nos permite afirmar que estes macacos-prego possuem a capacidade de reconhecer que um comportamento sendo produzido por outro indivíduo é mais eficiente que o próprio, e a flexibilidade para mudar para este comportamento alternativo. A respeito do segundo objetivo (2), ao analisar os viéses de transmissão social nós encontramos que os indivíduos tendem a observar preferencialmente os indivíduos mais proficientes e os indivíduos subordinados e que machos subordinados observam mais os coespecíficos que machos dominantes. Por fim (3), a respeito dos fatores sociais que afetam o desenvolvimento da CC, encontramos que os indivíduos dominantes tendem a monopolizar os recursos. A monopolização, contudo, não parece ter interferido negativamente na AS de aprimoramentos de técnicas. É possível que a grande tolerância que os macacos-prego têm com seus coespecíficos compense a influência negativa que a monopolização dos recursos poderia gerar. Não encontramos evidências de que os outros fatores sociais influenciaram negativamente a AS, e com isso, o desenvolvimento da CC. Em relação aos fatores cognitivos, não encontramos evidências de ensino, prossocialidade ou comunicação significativa e não pudemos avaliar a ocorrência de emulação ou imitação. Podemos concluir que esses fatores não são necessários para a AS de aprimoramentos de técnicas. Para aprender socialmente um aprimoramento de técnica que está além da capacidade de inovação individual, contudo, é provável que mecanismos com maior fidelidade de AS, como a emulação, sejam necessários. Acreditamos que caso a emulação seja suficiente para o desenvolvimento de uma CC simples em animais não humanos, os macacos-prego tem o potencial para desenvolvê-la. Já que (i) eles possuem a capacidade de reconhecer um comportamento mais eficiente e a flexibilidade de mudança de comportamento, (ii) evidências em cativeiro apontam que macacos-prego podem emular um comportamento e (iii) as populações tendem a ser tolerantes / Social learning (SL) is learning that is facilitated by the observation of, or interaction with, another individual or its products. When learning socially, animals, both human and non-human, employ strategies to access and use (or not) socially provided information. These strategies define when and from whom to copy a behaviour. A topic that is currently being discussed is the occurrence of cumulative culture (CC) in nonhuman animals: the population capacity to accumulate technical/behavioural improvements over time through SL. The objectives of this study were to investigate (1) whether monkeys are able to socially learn improvements of established techniques; (2) which social transmission biases and SL strategies were used by the capuchin monkeys; (3) the influence of 4 cognitive factors (learning, language, imitation and prosociality) and 4 social factors (scrounging, tendency of dominant individuals to monopolize resources, lack of attention to low-status inventors and conservative behaviour) in the diffusion of new behaviours and then argue how these factors would affect the development of a simple CC. We presented a group of capuchin monkeys with two puzzle boxes, one with three progressively complex stages (food-box) and other with two stages (juice-box). These puzzle boxes were developed to introduce new behaviours into the repertory of the group and induce improvements of established techniques. Regarding our first goal (1), by using the Network-based diffusion analysis we identified the occurrence of SL by close observation of the most complex stage of the juice-box. Thereby we infer that capuchin monkeys can acquire, through SL, more complex and efficient techniques of tool use. This, together with observational learning, allows us to affirm that these monkeys have the capacity to recognize that a behaviour performed by another individual is more efficient than their own, and the flexibility to change to this alternative behaviour. About our second objective (2), the social transmission biases, we found that individuals tend to observe more the most proficient individuals and subordinate individuals and that subordinate males observe more than dominant males. Lastly (3), regarding the social factors, we found that dominant individuals tend to monopolize the resources. Monopolization, however, does not seem to have negatively interfered with the SL of technical improvements. It is possible that the increased tolerance that capuchin monkeys have regarding their coespecifics compensates the negative influence that the monopolization could generate. We found no evidence that the other social factors influenced negatively the diffusion of information, and thus the development of CC. Regarding the cognitive factors, we did not find evidence of teaching, prosociality or significant communication. We could not evaluate the occurrence of emulation or imitation. Thus, we can conclude that these factors are not necessary for the SL of simple technical improvements. Observational learning is enough for this learning to occur. To socially learn a technical improvement that is beyond the capacity of individual innovation, however, it is likely that mechanisms with higher SL fidelity, such as emulation, are necessary. Considering that capuchin monkeys can (i) recognize that a behaviour is more efficient than their own and the flexibility change to that behaviour, (ii) that there is evidence, in captivity, that capuchin monkeys can emulate behaviours and (iii) that populations tend to be tolerant. If emulation is sufficient for the development of a simple CC in non-human animals, we believe that capuchin monkeys have the potential to develop it

On nomme culture un ensemble d’informations que les membres d’un groupe échangent par transmission sociale (imitation, communication...) sous la forme d’idées, de comportements ou d’artefacts. Les cultures développées chez l’espèce humaine sont plus complexes que chez les autres espèces. On attribue cela au processus d’évolution culturelle qui, chez l’humain, est cumulatif : au fur et à mesure de leur diffusion, certains traits culturels accumuleront des modifications jusqu’à devenir plus complexe et plus performants que ce qu’un individu seul aurait été capable de produire. On nomme ce processus le ratchet effect.Le projet de cette thèse était d’investiguer les mécanismes cognitifs impliqués dans l’émergence du ratchet effect. Nous avons reproduit ce phénomène en conditions expérimentales grâce à la méthode de transmission en chaîne afin de tester le pouvoir prédictif du raisonnement technique, de la théorie de l’esprit, de la créativité et de l’intelligence fluide dans la progression constante des performances au fil des générations. La première étude investigue le rôle de ces quatre facteurs dans une tâche de résolution de problème mécanique (construction de tour en fils de fer) dans une condition Communication (i.e., les participants sont autorisés à parler entre eux) et une condition Observation (i.e., les participants ne peuvent pas communiquer). Nous avons découvert que le raisonnement technique prédisait le mieux l’apparition du ratchet effect dans les deux conditions, tandis que les compétences en théorie de l’esprit ne permettaient aucune prédiction, ce qui va à l’encontre des principales hypothèses formulées sur les fondements sociocognitifs de la culture cumulative. La deuxième étude explorait l’influence des facteurs de raisonnement technique et de théorie de l’esprit sur la même tâche, dans une condition Monitoring (i.e., les participant communiquent indirectement en s’aidant d’une transmission vidéo) et une condition Blind (i.e., les participants ne se voient pas ils ne peuvent que communiquer verbalement). Les compétences en raisonnement technique prédisent encore une fois l’émergence d’un ratchet effect, tandis que la théorie de l’esprit semble prédire le même effet dans la condition Blind uniquement. Ces résultats démontrent le poids de l’influence du raisonnement technique dans la création, la transmission et l’amélioration d’un trait culturel matériel, tandis que les compétences en théorie de l’esprit ne sont impliquées que dans une condition particulièrement abstraite où les individus n’ont pas de lien visuel. La troisième étude explore les conditions conduisant les individus à copier ou à innover une technique ou un outil. Il s’agit de reproduire une situation non familière, en demandant à un participant d’exécuter une tâche matérielle (fabrication d’un panier à partir d’objets de la vie quotidienne) dont on fait varier les risques et l’opacité de la tâche (i.e., dont les mécanismes intervenant entre l’état initial et le produit final sont difficiles à comprendre). Il s’avère que les participants avaient plus souvent tendance à reproduire les mêmes actions qu’en démonstration et en employant les mêmes objets lorsqu’ils étaient testés dans une conditions risquée et opaque, tandis qu’ils employaient de nouveaux objets et de nouvelles actions dans une tâche non risquée et transparente. / Culture is a set of information that members of a group share through social transmission (imitation, communication ...) in the form of ideas, behaviors or artifacts. Cultural traits developed by humans are more complex than in other species. This is attributed to the process of cultural evolution which, in humans, is cumulative. Along with their diffusion, cultural traits become more complex and perform progressively better by accumulating changes over generations, a process called the ratchet effect. The project of this thesis was to investigate cognitive mechanisms involved in the emergence of the ratchet effect. We simulated this phenomenon under experimental conditions using thechain transmission method in order to test the predictive power of technical reasoning, theory-of-mind skills, creativity and fluid intelligence in the constant progression of performances across generations. The first study investigated the role of these four factors in a mechanical problem solving task (tower construction with wires) in a Communication condition (i.e., participants are allowed to talk to each other) and an Observation condition (i.e., participants cannot communicate). Our strongest prediction regarding this paradigm involved technical reasoning, for having proven a better predictor of a ratchet effect than theory of mind skills. The same observation was made in both conditions, while theory of mind skills did not allow to predict any ratchet effect. The second study explored the influence of technical reasoning and theory-of-mind skills on the same task, in a Monitoring condition (i.e., participants communicate indirectly through a video feed) and a Blind condition (i.e., participants communicate verbally without seeing each other). Technical reasoning skills predicted once again the emergence of a ratchet effect, while theory-of-mind skills predicted the same effect in the Blind condition only. These results demonstrate the weight of technical reasoning in thecreation, transmission, and improvement of a material cultural trait, while theory-ofmind skills are involved only in a particularly abstract situation where individuals have no visual cues. The third study explored the conditions leading individuals to copy or innovate a technique or tool. Participants were tested in an unfamiliar situation, and were asked to achieve a material task (making a basket out of everyday life objects). Conditions varied based on opacity (i.e., mechanisms involved between the initial state and the final product are difficult to understand) and risk levels. It turned out that participants were more likely to reproduce the same actions and objects used in a demonstration when tested in a risky and opaque condition, while they used more often new objects and new actions in a non-risky, transparent task.

Cumulative culture is rare, if not altogether absent in non-human species. At the foundation of cumulative learning is the ability to flexibly modify, relinquish or build upon prior behaviours to make them more productive or efficient. Within the primate literature, a failure to optimise solutions in this way is often proposed to derive from low-fidelity copying of witnessed behaviours, sub-optimal social learning heuristics, or a lack of relevant socio-cognitive adaptations. However, humans can also be markedly inflexible in their behaviours, perseverating with, or becoming fixated on outdated or inappropriate responses. Humans show differential patterns of flexibility as a function of cognitive load, exhibiting difficulties with inhibiting suboptimal behaviours when there are high demands on working memory. Here I present a series of studies on captive chimpanzees which show that not only is inhibitory control compromised in chimpanzees, but indicate ape behavioural conservatism may be underlain by similar constraints as in humans; chimpanzees show relatively little conservatism when behavioural optimisation involves the inhibition of a well-established but simple solution, or the addition of a simple modification to a well-established but complex solution. In contrast, when behavioural optimisation involves the inhibition of a well-established but complex solution, and especially when the alternative solution is also complex, chimpanzees show evidence of behavioural conservatism. I propose that conservatism is linked to behavioural complexity, potentially mediated by cognitive resource availability, and may be an important factor in the evolution of cumulative culture.