Accessibility and AI

What AI gets right about accessibility, what it gets dangerously wrong, and how teams should actually use it.

All industries are looking for ways they can leverage AI, accessibility included. Vendors are promising fully automated testing and remediation, and teams under pressure are hoping it’s going to solve all their problems.

Talks in this theme look at what large language models get right about accessibility, and things that go terribly wrong. Inconsistencies, hallucinations, incorrect code, and the human cost of outsourcing empathy to a machine. It includes real tests of popular AI tools, against a variety of different scenarios regarding accessibility.

Insane takes by Craig Abbott, all backed by tests and data around AI effectiveness and cost on basic accessibility checks. Highly recommend watching this one when the recording is up from JSHeroes!

Jeremias Menichelli, Software Engineer

Topics I cover

  • What LLMs actually do well for accessibility work today
  • Where AI tools give confidently wrong answers, with real tested examples
  • The human cost of automating accessibility, and who pays it
  • The questions to ask before wiring AI into your accessibility process

Talk titles

  • AI and Human Centred Accessibility
  • Should you use LLM’s for accessibility?

Recordings

Delivery
In person or remote
Formats
  • 30 minute talk
  • 60 minute talk
Who should attend
  • Designers
  • Developers
  • Product leaders
Level of experience
All levels
Tags

Want me to speak at your event?

If there is a talk title you like the sound of, or there is something bespoke you’re looking for, let’s talk!

Book me to speak

In action

  • A wide view of the chandelier-hung JSHeroes ballroom bathed in blue light, with Craig on the distant stage presenting Deque's claim that accessibility work can be done with zero specialised accessibility knowledge.
  • Craig opens his JSHeroes talk from the branded lectern, facing a huge screen asking "Should we use AI?" in front of a packed ballroom.
  • Craig addresses a full JSHeroes audience with a slide that crosses out "Should we use AI?" and "How do we use AI well?" to land on "What is the human cost of using AI?".
  • Craig smiles on the JSHeroes stage in a blue-lit ballroom, beside a slide reading "AI is literally everywhere. Annoyingly…", with giant JSHeroes letters at the stage edge.
  • Craig walks the JSHeroes stage in front of a slide stating "LLMs are not consistent, they can't give you the same answer twice."
  • Craig presents at the JSHeroes lectern beneath a slide reading "120+ HTML elements, 85+ WAI-ARIA roles, yet many people just define everything as a div".
  • Craig gestures from the UX Scotland lectern beneath a slide that strikes through "Should we use AI?" and "How do we use AI well?" to ask "What is the human cost of using AI?".
  • A room full of people at A11y North. Craig is about to start his talk on accessibility and AI.
  • Craig speaks from the lectern at Access Given beside a slide comparing an incorrect Axe Assistant code fix with the correct solution.
  • Craig gestures while speaking into a microphone at Access Given, wearing a backwards cap and denim shirt.
  • Craig on the purple-lit stage at Access Given, presenting a line chart showing how AI agent accuracy falls as the number of steps increases, credited to Utkarsh Kanwat.
  • Craig, microphone in hand at Access Given, in front of a slide listing axe Assistant's WCAG results: a pass for non-text content but fails for link purpose.
  • Al Robinson introducing Craig Abbott at NUX Newcastle.
  • Craig introduces himself at a cosy NUX meetup under amber pendant lamps, with his "Hello, I'm Craig" slide showing ADHD and Autistic badges on a wall-mounted screen.

Previous events