Algorithmic bias in image-generating artificial intelligence: prevalence and user perceptions

Abstract

Image-generating AI is among the most popular generative AI applications, likely changing the visual mediated environments humans are exposed to on a mass scale. Prior work found that AI can be biased against women and minorities (algorithmic bias), whereas humans attribute rather high objectivity to AI. We focused on image-generating AI, analyzing the extent of algorithmic bias in AI-generated pictures and human responses to bias in image-generating AI. Study 1 showed that AI-generated portraits of people in STEM professions were almost exclusively depicting male, white (and older) individuals. Study 2 (experimental, N = 495) showed that the responses to AI-generated pictures vary depending on the portrayed group. Participants perceived pictures as less biased if they were introduced as AI-generated, but only if pictures showed college students (vs. older people). If images showed older people, participants reported higher moral outrage if pictures were supposedly generated by AI (vs. human creators).