AI & agents
Thousands of AI agents parsed a brain atlas in 32 minutes
- Published
- Source
- OpenAI blog
Summary
OpenAI, PerturbAI, and Scripps Research reported running thousands of AI agents powered by GPT-Rosalind to interpret a whole-brain CRISPR atlas of gene disruptions in mouse brain cells in 32 minutes. In this 'distributed biological reasoning' setup, agents divided up genes, searched literature, and proposed interpretations; the team estimates comparable human analysis would take months to a year. In blinded comparisons, experts preferred the distributed-agent reports 75% of the time, and those agents cited over 700 references versus 11 for a single agent. The resulting reasoning atlas is available on GitHub.
Why it matters for our work
In life-science research overflowing with data but short on analysts, AI agents are taking on the job of surfacing candidate hypotheses for the next experiment. Researchers' work may shift toward spending time on hypotheses worth testing.
Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.