Identity in Computer Vision

Identity in Computer Vision: Exclusionary and Reflecting Histories present in Colonialist Worldviews
Computer vision technologies have been increasingly scrutinized in recent years for their propensity to cause harm. In this talk, Morgan will present work on how identity is implemented in computer vision, from how identity is represented in models and datasets to how different worker positionalities influence the development process. Specifically, he will showcase how representations of gender and race in computer vision are exclusionary and represent problematic histories present in colonialist worldviews. He’ll also highlight how traditional tech workers enact a positional power over data workers in the global south. Through these findings, Morgan demonstrates how identity in computer vision moves from something more open, contextual, and exploratory to a completely closed, binary, and prescriptive classification.