https://www.youtube.com/watch?v=4zrU54VIK6k&t=2115s

Notes and Themes Discussed

• Start: 10:30am EST
  • General Topic: Methods for privacy preserving AI, the video covers in depth:
    • Open data for science
    • Single-use accountability
    • End-to-end encrypted services
    • (A few tools enabling Pytorch to perform privacy preserving compute on sensitive datasets
• First thoughts about privacy preserving AI:
  • Quite a powerful concept/primitive for new information theory
  • Public dataset availability will be one of the largest drivers – even moreso than models – that brings new waves of innovation
  • The capacity for data to be “cross-leaked” when an individual goes to different “workers” (hospitals, any agent within which the compute takes place) could be solved in some capacity by blockchain
    • interoperability between privacy preserving AI
  • Persistence storage vs locale, short term application storage
  • Laws somewhat out of data for this new kind of paradigm
  • We require a lens into the domain of what kind of data-IP institutions like hospitals use from us.