AI co-scientist

Co-Scientist (AI co-scientist) is a multi-agent AI system built on Google's Gemini, designed to support scientists in generating research hypotheses and experimental protocols. Given a research goal in natural language, it searches and reasons over literature, proposes hypotheses, and cites relevant sources. It is described in a Nature paper, "Accelerating scientific discovery with Co-Scientist".

A Supervisor agent parses the goal and allocates resources to specialised worker agents (Generation, Reflection, Ranking, Evolution, Proximity and Meta-review) through an asynchronous task queue. Hypotheses are debated and ranked in a tournament, and feedback from the tournament drives iterative improvement. Test-time compute scaling is reported to improve hypothesis quality over time. It uses tools such as web search and specialised AI models, and scientists can steer it through a conversational interface.

Validation focused on three biomedical applications: drug repurposing and synergistic combination therapies for acute myeloid leukaemia (confirmed in vitro in AML cell lines), novel epigenetic targets for liver fibrosis, and mechanisms of antimicrobial resistance, where it independently recapitulated an unpublished discovery. Although described as general-purpose, the published evidence is biomedical. The provided material does not state public access terms, pricing or a general release; its outputs are hypotheses that require experimental verification.