Revisiting the Renaissance gaze through the eyes of machine learning

Cultural attractor theory proposes that recurrent psychological biases shape the persistence and transformation of cultural traditions. Portraiture offers a unique test case, embedding perceptual cues that reflect both individual choices and collective conventions. Using automated facial analysis (OpenFace), we examined over 2,000 digitized portraits from Renaissance and post-Renaissance Europe, religious artworks, and Japanese ukiyo-e. Results replicated Morin’s (2013) finding of a Renaissance shift toward direct gaze, but this trend reversed after 1600, suggesting the attractor was not stable. Religious works diverged, becoming more averted, while secular portraits engaged viewers. The left cheek bias was strong in the Renaissance, weakened in later Europe, and remained stable in ukiyo-e. Across analyses, sitter fame—but not gaze—predicted cultural visibility. Our study demonstrates how computational analyses of artworks can recover long-term dynamics of human psychology, offering “cognitive fossils” that illuminate the coevolution of visual culture and social behavior.