World models and accessibility

Where AI that predicts the world could help disabled people most, and where it could hurt.

Published
Screenprint of a long table with every kind of seat pulled up to it, lit by one small lamp

A research review of an overlap nobody has named. World models and disability access share a lot of substance, yet no peer-reviewed paper frames an assistive system as a world model, and none of the flagship launches mentions disability.

The failures disabled people report most in today’s AI assistants—going silent when a scene starts to move, misjudging distance, forgetting where you’ve already walked, alerting too late or too often—are the problems world models are built to fix.

The review covers blind and low-vision tools, sound and speech, robots and wheelchairs, generated worlds and GUI agents, and the central risk: a model that learns disabled people are improbable.

Area
Research
Status
Published
Built with
  • Research
Where to find it
Read the report
Notes
  • Published October 24, 2026
  • About 8,000 words, sourced throughout