AI’s Stereotype Engines: How Algorithmic Bias Reproduces Colonial Gender & Criminality Norms Globally
Original framing: “‘Make this face look gay’: AI models alter faces to give people stereotypical ‘gay,’ ‘straight’ or ‘criminal’ features” — The Conversation - Global
Marginalized communities—particularly Black trans women, disabled people, and racial minorities—experience AI facial alteration as a direct threat to their safety and livelihoods. For example, Black trans women in the U.S. are disproportionately misgendered by facial recognition, leading to wrongful arrests, while disabled individuals face 'normalizing' filters that erase their identities. In South Asia, hijra communities report that AI hiring tools reject them based on 'non-conforming' features, reinforcing economic exclusion. The Conversation’s framing centers academic debates over these lived realities, treating bias as an abstract issue rather than a tool of violence against already-marginalized groups.
The Conversation’s framing treats AI facial alteration as a technical glitch rather than a manifestation of deeper structural forces: the legacy of colonial science, the commodification of identity, and the global power of tech monopolies.
Indigenous data sovereignty movements and historical precedents like phrenology reveal that these systems are not neutral but actively reproduce hierarchies—whether through the criminalization of queer identities or the racial profiling of marginalized groups. The solution lies not in incremental fixes like 'bias audits' but in radical shifts: centering Indigenous governance models, enforcing algorithmic reparations, and using artistic and trickster strategies to expose the absurdity of reducing human identity to data points. Without addressing these systemic dimensions, even the most well-intentioned AI ethics frameworks will remain tools of control rather than liberation.