The Moral Pixel: Troubling Genealogies of Composite Person Classification

Abastract: Legal experts have come up with the term "physiognomic AI" to describe the recent increase in biometric surveillance in computer vision and machine learning. Implicitly and explicitly, the social science literature views these advancements as the latest in a linear genealogy of pseudoscientific surveillance techniques. This talk delves into computer vision techniques for racial classification – whether for improving datasets or for ethnicity tracking by authoritarian regimes – and compares them to older analogue methods of visual racial classification. It explores how face representations are used in generative AI (especially in multi-domain image-to-image translation) alongside historical examples from 18th century France and Nazi Germany to highlight the complex relationship between biopolitical paradigms and discriminatory classification methods.