For people with disabilities, AI is not a novelty — it is arguably the most consequential accessibility technology since the wheelchair. And precisely because it becomes load-bearing infrastructure in people’s lives, its failures matter more than most.
The Transformation Underway
Consider what has already shipped. Vision: apps that narrate the visual world through a phone camera, describing scenes, reading mail, identifying products, and even connecting users to AI visual assistants on demand. Hearing: real-time captioning of any conversation, meeting, or video, plus sign-language recognition in development. Speech: voice banking and synthesis that restore communication for people with ALS or after strokes — in a voice that sounds like their own. Motor: voice control of entire computing environments, AI-assisted wheelchairs, and smart homes operated without touching a switch. Cognitive: AI that simplifies dense text, structures tasks for people with ADHD, and offers patient, repeatable explanations. Each of these converts what once required expensive specialist equipment — or another person’s constant availability — into software on a mainstream device.
Independence Is the Real Product
The through-line is autonomy. Reading your own bank letter, ordering your own coffee, navigating a new building without asking — these aren’t conveniences; they’re dignity. AI’s greatest accessibility contribution is reducing dependence on the goodwill and availability of others.
When the Lifeline Fails
Dependence on AI cuts the other way. A navigation app that hallucinates a safe crossing, a captioning system that garbles a medical instruction, a scene-description model that misidentifies medication — for users who cannot independently verify the output, errors are not inconveniences but hazards. There are structural risks too: subscription models that put essential function behind recurring paywalls; companies discontinuing products that users’ daily lives depend on; models trained mostly on non-disabled speech and bodies that perform worse for the very users who need them; and privacy exposure, since assistive AI often sees everything its user does.
Designing for Trustworthy Dependence
The disability community’s principle — “nothing about us without us” — should govern assistive AI: co-design with disabled users, honest communication of confidence and uncertainty (“I’m not sure this is the right pill” is life-saving humility), graceful offline fallbacks, long-term support commitments, and regulation that treats critical assistive AI with the seriousness of medical devices.
The Balance
AI will make the world dramatically more navigable for hundreds of millions of people; that trajectory is clear and worth celebrating. The obligation that comes with it is equally clear: when technology becomes someone’s eyes, ears, or voice, reliability isn’t a feature. It’s the entire promise.