Facing the Past: Predictive Analytics Avant La Lettre

• In Event: A New/Old Technology for Governance? Questioning Big Data Surveillance - II: Prediction, Race and Resistance

Abstract

Recent discourses of the epistemic difference that Big Data makes tether practices of prediction to modern machine learning. While deep learning and data totalities are indeed effecting intensifications and mutations of prediction across heterogeneous sites, from counterinsurgency to corporate marketing, this paper examines how traces of earlier predictive systems organized around the face persist in computer vision. Plumbing the densely intertextual late nineteenth and early twentieth century literatures of criminal anthropology, sexology, and eugenics, in which criminality, inversion, and racial difference emerged as structurally analogous and mutually affecting deviations from the idealized average face of the race, and out of which developed the statistical principles of regression and correlation and the related practice of composite portraiture, I argue that racialization-cum-prediction is foundational to both colonial modernity and contemporary computer vision. Situating these discourses and practices as predictive analytics avant la lettre, I suggest, can provide a crucial lens through which to view the resurgence of composite photography in automated facial recognition, both in projects that attempt to revivify the tenets of physiognomy with bigger data and better models, as well as those that seek to evaluate racial and gender biases in recognition systems. Historicizing the nexus of prediction and recognition, and foregrounding the centrality of physiognomy and facial analytics to modern configurations of biopolitics and population, I contend that the oft-noted discriminatory outputs of such systems are not only problems to be corrected, but returns of the epistemological repressed, symptoms of an effacement of machine learning’s inaugural faces.

Author

• Harper Shalloe, Brown University