HOW IMAGE GENERATION SYSTEMS DEMONSTRATE STEREOTYPES: EVIDENCE FROM THE CHINESE AI SYSTEM WENXINYIGE

Abstract

As Artificial Intelligence (AI) applications expand across diverse domains, questions of bias, discrimination, and social justice have become central to the field of AI ethics. While international scholarship has examined how generative systems such as DALL·E and Midjourney reproduce cultural and gendered stereotypes, far less attention has been paid to non-Western contexts. This study addresses this gap by analysing Wenxinyige, a representative Chinese image generation platform, through the Walkthrough Methodology. The research investigates how Wenxinyige encodes gendered assumptions in its outputs, focusing on representations of occupation, appearance, behaviour, and personality. Findings demonstrate that Wenxinyige consistently reinforces patriarchal cultural logics, demonstrating occupational stereotypes and age discrimination. Women are portrayed as youthful, attractive, and submissive, while men are associated with authority, occupational prestige, and active roles. These patterns extend beyond surface-level representation, functioning as a mechanism of algorithmic disciplining that shapes users’ perceptions of gender norms and age hierarchies. In doing so, the system exacerbates digital inequalities and contributes to the marginalisation of particular social groups. By situating Wenxinyige within global debates on algorithmic justice, this study expands the largely Western-centred scholarship on AI bias to include the Chinese context. It underscores that generative AI systems are not neutral creative tools but socio-technical artefacts that reflect and reproduce existing inequalities. The findings underscore the pressing need for more inclusive data practices, increased transparency in model design, and robust governance frameworks that safeguard human rights and foster digital justice.