AI explanations designed for sighted users leave blind users unable to trust or use AI assistants

Most AI explanation research assumes visual output, making it inaccessible to blind and low-vision users who need text or audio explanations instead. As AI systems shift from answering single questions to taking multiple autonomous actions over time, blind users face a critical gap: they can't understand what the AI is doing, can't catch errors before they cause real damage, and end up blaming themselves when the system fails.

Blind and low-vision users are effectively locked out of AI-powered tools that sighted people can use because nobody designing these systems thought about how to explain AI decisions in a format they could actually access. The problem gets worse as AI moves from simple tools (answer my question) to autonomous agents (do this task over multiple steps while I'm not watching) — a single undetected error can compound before anyone notices. This is a structural accessibility gap that affects who gets to benefit from AI and who doesn't.