A Survey of Accessible Explainable Artificial Intelligence Research

Visual explanations dominate XAI but exclude people with sight loss: The survey finds that most current explainable AI techniques rely heavily on visual explanations like heatmaps and graphs, which are inaccessible to people who are blind or have low vision, creating a significant barrier to digital inclusion.

Accessible XAI research is severely underdeveloped: The systematic literature review reveals a critical lack of research on Accessible XAI (AXAI), indicating that accessibility considerations have been largely overlooked in the development of AI explanation methods.

Alternative modalities are needed for inclusive AI explanations: To make AI decisions understandable to all users, researchers must develop explanation methods using non-visual approaches such as auditory and tactile feedback, as well as visual formats that work for people with low vision.

People with disabilities must be included in XAI development: The survey emphasizes that the disability community should be actively involved in designing explainable AI systems from the start, rather than as an afterthought, to ensure digital inclusion and remove accessibility barriers.

Universal design principles should guide AI explanation development: The paper stresses the importance of applying universal design principles to create personalized solutions that meet the diverse needs of individuals, including those with multiple disabilities, rather than one-size-fits-all explanations.