Who Benefits from AI Explanations? Towards Accessible and Interpretable Systems

Most XAI research excludes disabled users: A review of 79 studies found that evaluations of explainable AI techniques rarely include people with disabilities, meaning current explanations may not work for everyone who needs to understand AI decisions.

Visual-only explanations create accessibility barriers: The paper identifies that most AI explanations rely on inherently visual formats (like charts and diagrams), which are inaccessible to users with vision impairments and limit who can benefit from transparency efforts.

Simpler explanations work better for non-visual users: Preliminary findings show that simplified explanations are more comprehensible for people who cannot access visual information than detailed technical explanations, suggesting a trade-off between complexity and accessibility.

Multimodal presentation is essential for equitable AI: The research demonstrates that combining multiple formats (text, audio, tactile, etc.) is necessary to ensure that AI explanations are truly interpretable across diverse user groups, not just sighted users.

Responsible AI requires inclusive design from the start: The paper's methodological framework—including user personas, prototype testing, and expert assessment—suggests that accessibility should be built into XAI systems during development, not added as an afterthought.