Medicine · 19 February 2025
A hypothesis-generating collaborator, tested in the laboratory
A multi-agent Gemini system proposed original scientific hypotheses. In one biomedical test, drug-repurposing candidates it suggested for acute myeloid leukemia were later shown, in experiments, to inhibit tumor-cell viability at clinically relevant concentrations.
- Field
- Medicine
- Date
- 19 February 2025
- AI system
- AI co-scientist
- Institution
- Google Research, with academic collaborators
- Published in
- Official announcement and preprint
- Identifier
- See primary source
On 19 February 2025, Google Research introduced AI co-scientist, a multi-agent system built on Gemini 2.0 and designed to generate novel research hypotheses and experimental proposals from a scientist’s stated goal. Specialized agents generate, critique, rank and evolve hypotheses in a cycle meant to mirror the scientific method.
The authors report end-to-end laboratory tests in three biomedical settings, carried out with expert-in-the-loop guidance. In drug repurposing for acute myeloid leukemia, the system proposed novel candidate drugs. Subsequent experiments, as described in the official account, confirmed that the suggested drugs inhibit tumor viability at clinically relevant concentrations in multiple AML cell lines.
In a separate test on antimicrobial resistance, researchers asked the system to explain how capsid-forming phage-inducible chromosomal islands can exist across multiple bacterial species — a result their group had already obtained experimentally but had not yet made public. The system independently proposed that these islands interact with diverse phage tails to expand their host range, matching the unpublished laboratory finding. A further target-discovery exercise on liver fibrosis was described as showing anti-fibrotic activity in human hepatic organoids, with details left to a later collaborator report. The work, at announcement, was a preprint and blog account rather than a peer-reviewed paper.