Smiling women pitching down: auditing representational and presentational gender biases in image-generative AI

Abstract

Generative Artificial Intelligence (AI) models like DALL.E 2 can interpret prompts and generate high-quality images that exhibit human creativity. Though public enthusiasm is booming, systematic auditing of potential gender biases in AI-generated images remains scarce. We addressed this gap by examining the prevalence of two occupational gender biases (representational and presentational biases) in 15,300 DALL.E 2 images spanning 153 occupations. We assessed potential bias amplification by benchmarking against the 2021 U.S. census data and Google Images. Our findings reveal that DALL.E 2 underrepresents women in male-dominated fields while overrepresenting them in female-dominated occupations. Additionally, DALL.E 2 images tend to depict more women than men with smiles and downward-pitching heads, particularly in female-dominated (versus male-dominated) occupations. Our algorithm auditing study demonstrates more pronounced representational and presentational biases in DALL.E 2 compared to Google Images and calls for feminist interventions to curtail the potential impacts of such biased AI-generated images on the media ecology.