What's the Look of "Negative Gender" and "Max Ethnicity" in AI-Generated Images? A Critical Visual Analysis of the Intersectional Politics of Portrayal

Abstract

In this exploratory paper, we focus on intersecting design political and visual processes of gendering and racializing in online AI image generators, in particular ArtBreeder and Midjourney. While AI image generators are becoming an integrated part of our contemporary society, they draw on cultural and historical imaging conventions of sorting and ordering the world and the people in it. These tools' powerful visual rhetoric can potentially aggravate existing discrimination, if not critically reflected upon. We argue that these design-facilitated representations position the 'user' into cultural imagery of representations with political implications. With an intersectional perspective from the feminist visual analysis, we critique and uncover how gender and ethnicity are represented and built into the systems, both in terms of visual culture and in designed interactions. We problematize these design strategies, and urge the HCI community to engage in further design political inquiries regarding the visual culture mediated by AI image generators.