Keywords: text-to-image generation, social bias, stereotypes
TL;DR: We test three text-to-image models to see if they generate stereotype-consistent images in response to under-specified prompts (e.g., generating images of men when prompted with "powerful", and women when prompted with "friendly").
Abstract: As text-to-image systems continue to grow in popularity with the general public, questions have arisen about bias and di-versity in the generated images. Here, we investigate properties of images generated in response to prompts which are visually under-specified, but contain salient social attributes (e.g., ‘a portrait of a threatening person’ versus ‘a portrait of a friendly person’). Grounding our work in social cognition theory, we find that in many cases, images contain similar demographic biases to those reported in the stereotype literature. However, trends are inconsistent across different models and further investigation is warranted.
Submission Type: archival
Presentation Type: online
Presenter: Kathleen Fraser