I was talking to a friend this weekend. They work at a small ed-tech startup and they asked me a question I've heard a few times now.
Should we use AI for accessibility?
My first answer was yes.
Use AI. Use checkers. Use browser tools. Use screen readers. Use linters. Use whatever helps you find problems earlier and fix them faster.
But that's not what they were asking. They were really asking if they can have AI take care of accessibility for them. If they can prompt an accessible product into existence. And, if they can't one-shot it, have the AI automatically identify and fix the issues, then try again.
And the answer is a clear no.
Accessibility is not a task you hand to a tool. It's a responsibility you carry while you design, build, test and ship. Accessibility is a product decision. And product decisions are human decisions.
So yes, use AI for accessibility. Use it like you use a screwdriver to build IKEA furniture. Helpful. Fast. Sometimes the right tool for the job.
But don't ask the screwdriver to check whether the chair holds someone up.
I wonder if this is the reason why most people are skeptical of AI in accessibility.