Key Info
Meta's health workstream ran a project in which medical experts used a Muse Wonder coding scaffold connected to a database of 5 million electronic health records licensed from a vendor, aiming to let AI answer arbitrary research questions such as whether statin users had higher later-life diabetes rates.
Highlights
- The setup combined human medical experts, a strong coding model, and access to real patient data.
- The database contained 5 million electronic health records licensed from a vendor.
- The goal was to let AI answer open-ended causal questions, e.g. statin use and later diabetes rates.
- A small group of medical experts was trained to use the AI tool, and all reportedly found it useful.
- Meta frames this as exploring how AI can accelerate health science research, which depends on imperfect real-world data and consequential cohort, variable, and causal choices.