Will AI Make Scientists Obsolete? | Arc Institute’s Patrick Hsu
Patrick Hsu argues that AI, perturbation data and a full-stack institute like Arc can transform biology, yet human research taste, boldness and constant reinvention matter more, not less.
Watch on YouTube ↗Charlie's summary
- ▶ 9:09
Research is an even playing field and 99% failure, so humility must be paired with supreme confidence
Patrick says nearly all your ideas “kind of suck except the one or two that you're remembered for.” At a science fair, a Nobel laureate answered his question with “What do you think?”, showing that nobody knows the answer to an open problem. That makes research an even playing field. Patrick thinks scientists need rigorous self-criticism plus supreme confidence that the work can succeed, or the job is psychologically untenable.
- ▶ 13:18
GLP-1 proved a horizontal drug is possible, and that existence proof will pull many more into being
Patrick notes that Eli Lilly's R&D spend exceeds all German health companies combined, driven by GLP-1 revenue. He sees GLP-1 as special because it works across many conditions. He argues that knowing something is possible, like the four-minute mile, changes courage and markets. Companies now copy its playbook: start with a small disease with clear mechanism, then expand to bigger ones. For example, narcolepsy leads to sleep, then wakefulness and attention.
- ▶ 30:10
Research taste is gaining value, and biology will be a discipline of context maxing
Asked whether models make intuition less important, Patrick says taste is rising in relative importance: “I would buy call options on it if I could.” Math can throw 10,000 agents at verifiable problems, but biology cannot, because agents lack context, evals and enough data. He expects the next decade to be about building context. Scaling naively would be inefficient but work.
- ▶ 34:15
Biology needs new measurement technology, such as time-series data, because current methods destroy the cell
Machine learning people want time-series data, but biology mostly gives snapshots because measuring a cell kills it. Patrick describes Felix Horns's engineered protein nanocages that bud out of a cell carrying RNAs, like sending “a FedEx package” every few hours. This gives non-invasive, repeated, single-cell readouts. He says this is why Arc exists: no single lab spans invention, engineering and scaling.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.