Abstract
Most fairness audits of gender bias in visual AI remain limited to binary analyses and dismiss nonbinary genders. In this paper, we present "NonbAInary", an interactive installation that proposes a nonbinary approach to unveil gender bias. The installation features the Gender Diamond, a data-driven visualisation of the gender spectrum, operationalised through coexisting femininity and masculinity scales. Through the display of a series of AI-generated gendered portraits and a critical interface, visitors are invited to reflect on gender bias as a tension between accuracy — how closely a portrait aligns with one’s internalised gender stereotypes — and discrimination — whether such portraits are perceived as ethically harmful. By combining critical data practices and queer human–AI interaction, the installation reframes AI auditing as a critical, reflexive, and collective process.