OpenAI Foundation launches Public Data for Health to fund biology datasets, starting with failed biotechs' archives

The OpenAI Foundation, OpenAI's nonprofit parent, launched Public Data for Health, its second science program after the AI for Alzheimer's effort announced in April. The program funds the creation and preservation of high-quality scientific datasets made broadly available to researchers, on the premise that data rather than model capability is now the main bottleneck for AI in biology. In the announcement, Abhishaike Mahajan and Jacob Trefethen write that AI has begun contributing new knowledge in fields like mathematics, where breakthroughs can be verified without collecting new data, but that most remaining advances in preventing and treating disease will require pairing models with more observations of the world. Among the initial grants is a proposal from clinical-trials policy analyst Ruxandra Teslo to bid at the bankruptcy proceedings of failed biotech companies for their regulatory filings, manufacturing strategies, and safety data. She calls this material biotech's lost archive because it is usually treated as trade secrets and could help train AI copilots for the drug approval process. Morgan Levine, formerly vice president for computation at the longevity company Altos Labs, told MIT Technology Review that everyone now recognizes data as the biggest bottleneck in applying AI to biology. The foundation says it wants the resulting datasets accessible to researchers and plans to refine its strategy as the first grants proceed.