Google has developed an AI system that helps researchers conduct literature reviews, formulate hypotheses and create research plans.
Google researchers have developed an AI system that serves as a virtual collaboration partner for biomedical research. The AI co-scientist analyzes scientific literature, formulates new hypotheses and creates research plans, which are then validated within various biomedical disciplines. Researchers can apply for the AI system through Google’s Trusted Tester program.
Hypotheses
The AI system uses a “self-play” approach in which different Gemini models debate and critique each other’s hypotheses. This process, combined with test-time compute scaling, helps generate refined hypotheses and research plans. The AI ranks these results so researchers can review them, provide feedback and test them experimentally.
read also
OpenAI’s new AI tool conducts in-depth research
Initial tests in collaboration with academic institutions show promising results, according to the researchers. For researchers at Imperial College London, the system suggested a hypothesis about a new gene transfer mechanism, potentially contributing to understanding antimicrobial resistance.
Scientists at Stanford identified new epigenetic targets for liver fibrosis thanks to the AI system, whose action was validated experimentally. At Houston Methodist, the AI was used to reorient an existing drug for acute myeloid leukemia research. This was confirmed through computational biology and laboratory experiments.
Trusted Tester program
Additional validation studies are currently underway with academic and industry partners. Researchers can use the AI system through Google’s Trusted Tester program. This provides access to both a user-friendly interface and an API for developers.