Datasets that nobody owns yet

Inspired by https://twitter.com/alexgraveley/status/1722362874121904486

• How humans execute tasks using computers
  • Input: task, application history (HTML, website/app screenshot), action history
  • Output: next actions (pointerdown, pointermove, pointerup, keydown, keyup, talk)
  • How to execute: create an offshore web assistant service, record all inputs
  • Tangents
    • E2E limit includes all network traffic (HTML/JS/CSS) as input. But learning a javascript interpreter is overkill. Screenshots are sufficient. Tall screenshots even better.
• How humans build software
  • Input: idea/task, software state (browser/app state), entire codebase (most can be pruned)
  • Output: PRs
  • How to execute: create an agency. Have human developers
• How doctors diagnose patients
  • Input: patient medical history, DNA, X-ray, etc.
  • Output: medical diagnosis
  • How to execute: create a hospital and get consent to record all patient data