Three Open Challenges for Biomedical World Models, Per Marinka Zitnik
marinkazitnik · x · 2026-09-18
Harvard's Marinka Zitnik outlines three key challenges for biomedical world models, tied to the paper "World models for biomedicine" (Cell special issue on AI in biology):
- Biological states are often only partially observed.
- Destructive assays yield snapshots, not trajectories over time.
- Observational data cover many states without interventions, while interventional data cover few.
Closing the gaps requires new data linking state, intervention, and outcome: non-destructive assays that measure the same cells repeatedly, organoids and organ chips producing human-relevant interventional trajectories, self-driving labs, and learning health systems embedding randomization into care.
Her litmus test for a world model: it must represent a system state, take a specified action, predict the resulting state, and continue simulation from there — so foundation models alone aren't world models, nor are digital twins that merely mirror one system's current state.
Related event: Harvard team proposes biomedical world models framework in Cell(8 posts)→
More from AGI Musings
- Geoffrey Hinton tells Congress it has 'maybe a year' to regulate AI before losing control — whurley · 2026-09-18
- ML is leaving its alchemy era: assumed truths can now be instrumented and falsified — mike64_t · 2026-09-18
- AI Roundtable Experiment: Chinese Models Reveal How They're Censored — gary1967 · 2026-09-18
- Halvar Flake: Don't Assume Hyperintelligent AI Isn't Subject to Physics, CPU-Heat Covert Channels Have Tiny Bandwidth — AccBalanced · 2026-09-18
- Ex-Microsoft researcher warns of safety risks from embedding incoherent views of consciousness into AI — rgblong · 2026-09-18
- Podcast Digs Into Dario Amodei's Three-Part Framework for 'Pacing' Frontier AI — thursdai_pod · 2026-09-18