AI says You are a Happy Dishwasher – How Generative AI Systematically Misrepresents Neurominority Professionals

Abstract

Data-driven AI systems are shaping our perception of the potential for automation and human-based knowledge representation. As a reflection on the world, these systems also exhibit discernible patterns of discrimination, particularly against marginalized groups. Whereas gender and ethnicity are well-known categories of AI-driven biases the plight of neurodivergent individuals (neurominorities) remains largely unexplored. Our study investigates how AI-generated portraits (N = 2,240 images, Stable Diffusion V2) represent dyslexic and non-dyslexic people in low- and high-paid occupations. Analyzing occupation-related images of female and male dyslexics conveys a significant neuronormative bias. Regardless of the occupational profile, images of dyslexic people represent more expressions identified as negative basic emotions than their non-dyslexic counterparts and show that people with dyslexia in low-paid occupations are happier than in high-paid occupations. This previously undiscovered unfairness in AI systems therefore necessitates greater awareness, as AI-generated images are increasingly shaping our digital lives, economy, and society.