Created by: Tzuyu Hsu, Yumei Chen, Tongxin Zhou, Jinting Fan, Langyi Li
Overview: This project investigates how generative AI reproduces racial and gender stereotypes through occupational image generation. Using ChatGPT, the group tested prompts across five professions and conducted four rounds of iteration for each, moving from baseline prompts like “a nurse” to increasingly specific ones such as “a male nurse,” “a diverse group of nurses,” and “a nurse who fits no stereotype.” The outputs revealed that AI defaults to dominant social patterns, associating certain professions with specific genders and ethnicities, and that attempts to correct bias often produced new distortions or visually artificial results.
A key shift in the project was the recognition that the group could not separate AI bias from their own assumptions. When the AI generated a female nurse, they read it as bias, but that reading itself assumed “female nurse = stereotype.” The project reframed itself from critiquing AI as an external problem to examining how humans and AI produce bias together. The final output is a large-scale poster assembling the iterative generation results across all five professions, making visible the process of questioning rather than claiming to resolve bias.
Created with: ChatGPT, Canva
Concept
Poster description: The poster presents a series of AI-generated occupational portrait collages where smaller fragmented images of nurses, lawyers, cashiers, teachers and managers are assembled into larger human silhouettes, revealing repeated patterns of gender, race and appearance across professions. It argues that there is no neutral prompt, and that the bias visible in AI-generated professional imagery is not produced by the AI alone but co-produced through the interaction between human assumptions, prompt design and the system’s training data.
Development and Sketches
Contact the Creators
Tzuyu Hsu
Yumei Chen
Tongxin Zhou
Jinting Fan
Langyi Li