Within Computer Vision, humans define how machines interpret the world. Researchers at the Weizenbaum Institute identified power differentials and commercial interests in data annotation that influence how systems learn to “see“.