Abstract
In many species, faces and the way they are perceived play a major role in communication between individuals. The Mandrillus Project, a long-term field project in southern Gabon, has been studying this communication for over 10 years in a population of 250 mandrills (Mandrillus sphinx), a monkey from the equatorial forests of Central Africa. Recent studies have shown a correlation between the femininity of mandrills' faces and their attractiveness to conspecifics. While this link is positive in humans, it is negative in mandrills.A first objective of this thesis is to experimentally validate these results, which have so far only been obtained with correlative studies. Under controlled conditions with captive mandrills, we tested the attractiveness of artificially generated images varying according to a femininity parameter, which confirmed and reinforced this negative link between femininity and attractiveness.To vary the femininity parameter, we used generative artificial intelligence (AI), which enables artificial images to be created from real images by varying certain specific characteristics. The methodological development of a generative AI approach applicable to behavioral experiments, with mandrills or other animal species, was the second objective of the thesis and resulted in a methodological paper submitted and under revision.Beyond the synthesis and manipulation of visual stimuli, AI can shed light on communicative processes between animals, as well as their evolution, due to its structural similarities with biological visual perception. Certain convolutional neural networks (CNNs) can thus predict the activity of the visual cortex, and provide a better understanding of cerebral information processing. The third objective was to show how a model based on this type of approach can predict beauty in humans, and resulted in an article published in PLOS Computational Biology.