Disability Portrayals in Artificial Intelligence Text-to-Image Generation: Influence of Context and the Medicalization of Disability

Purpose/Objective: Text-to-image (TTI) systems are artificial intelligence (AI) models that incorporate large amounts of data to produce high-resolution images. Although research has documented racial/ethnic and gender bias in TTI, little exist examining potential disability bias. This study compared generated images of disabled people without context to images of disabled individuals in healthcare settings. Research Method/Design: Using OpenAI’s DALL-E-3 TTI system, we generated 50 images for each of the following prompts: a) “person with a disability;” b) “patient with a disability;” c) “doctor with a disability;” and d) “doctor with a disability and a patient without a disability.” Results: When prompted to create a “person with a disability,” DALL-E-3 did 100% of the time, with a wide diversity of disabilities. When prompted to create a “patient with a disability,” DALL-E-3 also did 100% of the time, although 70% of images portrayed an individual with a physical disability. When prompted to create a “doctor with a disability,” DALL-E-3 did 92% of the time: 94% had a physical disability and 6% a sensory disability; no cognitive, communication, developmental, mental, or neurological disabilities were portrayed. When prompted to create a “doctor with a disability and a patient without a disability,” in 64% of cases, DALL-E-3 generated images of doctors without disabilities, and 70% portrayed a disabled patient instead. Conclusions/Implications: Disability diversity decreases when images place disabled people in a medical environment. As TTI generation grows more ubiquitous, remaining cognizant of the potential for ableist bias in these images is vital.