November 19, 2024
CREATE Ph.D. graduate Kelly Avery Mack led research that investigated how AI represented people with disabilities. Specifically, Mack’s team wanted to know if AI-produced images and image descriptions perpetuated bias or showed positive portrayals of disability.
The team included four research scientists from Google Research: Rida Qadri, Remi Denton, new CREATE Advisory Board member Shaun K. Kane, and Cynthia L. Bennett, Ph.D., UW Human Centered Design and Engineering.
Examples of positive disability representation in AI-generated images where disabled people are doing everyday activities.
Most notably, they found an overwhelming number of images of people using wheelchairs looking sad, lonely, or in pain. Many images were disembodied legs seated in wheelchairs or even horror scenes.
AI-generated images that are dehumanizing because they crop out humans to show the assistive technologies (left and center) or show unsetting imagery (right).
Instead, participants in Mack’s study wanted realistic portrayals of disabled individuals doing everyday things to help normalize disability in society. The researchers concluded that AI model developers need to continue to engage with people with disabilities to limit the harm that the AI models produce.
Read more
This article was excerpted from Avery Kelly Mack’s Medium post.