Domains · Article 12 of 60
Science and health: the instrument and the clinic
A small domain with the highest cost of being wrong.
AI-for-science claims split into two families that are often confused: the discovery claim, where a model finds something a lab then verifies, and the deployment claim, where a system is used on patients. The first is a research result; the second is a regulatory event.
The ledger keeps them in the same domain but not in the same sentence. A discovery claim graded by a paper is not evidence for a clinical claim, and the verdict notes say so.
| Claim | Horizon | Verdict |
|---|---|---|
| <div><em>“I would hope we have the candidate for that in the next couple of years for the first like really major thing.”</em><br /><span class="tiny">Sam Altman — Interview with Matthew Berman, 2026-10-21</span></div> | By October 2028 | <span class="verdict" data-v="open">Open</span> |
| <div><em>“Five years is a long time. Next year, I hope we’ll start to see these models really make small but important new scientific discoveries. And in five years, I hope they’re curing diseases.”</em><br /><span class="tiny">Sam Altman — Interview, December 2025 (reported), 2025-12-09</span></div> | By 2030 | <span class="verdict" data-v="open">Open</span> |
| <div><em>“Models will start producing research results that their own creators cannot fully explain.”</em><br /><span class="tiny">Mark Chen — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| <div><em>“Scientific discovery will be the first domain where models outperform their trainers.”</em><br /><span class="tiny">Oriol Vinyals — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| <div><em>“AI will do the work of most doctors and lawyers within a generation, and that is a public good.”</em><br /><span class="tiny">Vinod Khosla — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| <div><em>“Open research infrastructure will keep pace with private labs on applied work.”</em><br /><span class="tiny">Matei Zaharia — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| <div><em>“Weak supervision and data programming will prove more valuable than another architecture paper.”</em><br /><span class="tiny">Christopher Ré — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| <div><em>“AI-designed experiments compress drug discovery from decades to years.”</em><br /><span class="tiny">Daphne Koller — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
A category boundary worth defendingA model that proposes a candidate molecule has made no claim about whether the molecule works. Only a trial can do that.
Figures computed from docs/predictions/ledger.json at build time. Every claim quoted on this page carries its own sources in the ledger.