Explainable AI for Blind and Low-Vision Users: Navigating Trust, Modality, and Interpretability in the Agentic Era

Accessible AI explanations are a justice issue, not a luxury feature: The paper reveals that blind and low-vision users face a "fundamental barrier" to independently using AI assistive technologies because explanations are designed primarily for sighted users. This creates unequal access to AI systems that could otherwise help them navigate the world.

Autonomous AI agents pose heightened risks for vulnerable users: As AI systems move from answering single questions to taking multiple independent actions over time, errors can compound before users even realize something went wrong. For blind and low-vision users who may have limited ability to monitor an agent's actions in real-time, this creates serious safety concerns that the field hasn't adequately addressed.

Users blame themselves, not the AI, when things go wrong: The research found that blind and low-vision users frequently experience "self-blame" when AI systems fail to explain their decisions properly. This suggests that poor explanation design doesn't just reduce understanding—it can psychologically harm users by making them doubt their own judgment rather than recognizing the AI's limitations.

Conversational explanations matter more than visual ones: While this may seem obvious, the paper provides empirical evidence that blind and low-vision users strongly prefer spoken or text-based explanations. Yet most XAI research ignores this preference, indicating a significant gap between what users need and what researchers are building.

Disabled communities must shape AI development from the start: The paper advocates for "participatory development" rather than retrofitting accessibility later, suggesting that involving blind and low-vision users in designing explainable AI systems from the beginning would produce better, more trustworthy tools.