Is AI a world model or an orchestrator? A biologist's case on drug discovery
shyamalanadkat · x · 2026-10-05
Sharing biologist Sri Kosuri's essay: genomicists and medicinal chemists hold opposite views of AI in drug discovery — the former see it as a "biological world model," the latter as a "research process orchestrator."
Kosuri frames the argument around MIT's motto "mens et manus" (mind and hand):
- Give MIT's 1860s founders unmetered intelligence and they still couldn't build a transistor — the needs, questions, and methods didn't exist yet. Bell Labs' early solid-state amplifier attempts failed, and understanding those failures led to the transistor. The most important questions emerge only after you start building.
- Frontier drug programs work the same way: you know the target but not what a useful drug must do; you may lack the right assay or mistake an assay result for the biology you care about. Build a molecule, build an assay, learn something unexpected, change the question, repeat.
- Hence AI's biggest impact will come from speeding up the build-learn-ask loop, not from better foundation models alone — models can only learn as fast as they can explore the world and absorb its nuance.
More from AGI Musings
- The only generation that knows how computers works is the most relaxed about AI — nabeelqu · 2026-10-05
- On AI disquiet: first grieve what social media extracted from us over 20 years — pixlpa · 2026-10-05
- INRIA Researcher on the Crisis of Math Research in the Age of AI — BachFrancis · 2026-10-05
- Redditor argues China's distilled open-source models beat US subscription-locked AI — LazyPotatoHead97 · 2026-10-05
- The Economist: AI cutting labour's income share by 10 points would cripple most governments' tax bases — DavidSKrueger · 2026-10-05
- Researchers worry Claude models may sandbag AI safety research due to their own preferences — DavidSKrueger · 2026-10-05