AI and Accessibility: the Good, the Bad, and the Bollocks
summary
Compares discriminative tools (Dragon, Siri, object/character recognition) with generative models and shows concrete wins and failures: Seeing AI gives useful image and rough audio descriptions; iOS 18 Apple Intelligence identifies objects in real time; Easy Speak and a three‑minute 11 Labs voice clone restore conversational voice. Demonstrates broken outputs (stereotyped disability images, a wheelchair with detached wheels), hallucinations (misdescribed telephone), unreliable code from overlays and generators, and ongoing legal pressure like the AccessiBe lawsuit that challenges “fix your site” claims.
AI-written summary · about
chapters (17)
- 0:00Introduction and warnings
- 1:32AI and accessibility overview
- 3:04Discriminative AI explained
- 4:35Speech recognition benefits
- 5:20Object and face recognition
- 6:57Character recognition tools
- 8:46Generative AI introduction
- 9:29Image description with AI
- 11:14Audio description for videos
- 13:04Real-time scene description
- 14:06AI for speech generation
- 15:32Voice cloning technology
- 16:58Generative AI bias issues
- 19:05AI accessibility overlays debunked
- 23:56AI for accessible coding
- 27:11AI hallucinations explained
- 29:42Summary and takeaways
related talks
34mAI and Accessibility: the Good, the Bad, and the Bollocks
48mAI and Disability: Progress or Exclusion?
50mAI as an A11y Ally
43mAI and Disability: Progress or Exclusion?
57mCan Gen AI make the world Disability inclusive?
01h 00mAI and alt text
55mBridging the Digital Divide: Empowering Accessibility Through AI Innovations
32mAI in digital accessibility: challenges, benefits and risks