Google develops AI research assistant to accelerate scientific discoveries

Google Research has created a new AI system called AI Co-Scientist that works alongside human researchers to generate and test scientific hypotheses. Built on Google's Gemini 2.0 model, the system aims to accelerate breakthrough discoveries by serving as a virtual research partner.

The system uses multiple specialized AI agents that work together to generate, evaluate and refine potential research directions. It also leverages "test-time compute" capabilities to process and analyze ideas. Scientists can actively collaborate with the system by contributing their own hypotheses and providing feedback. To evaluate the quality of its suggestions, the system employs an integrated Elo rating system.

Video: Google Deepmind

The AI Co-Scientist has already shown promise in real laboratory experiments across three biomedical applications. The system proposed new drug candidates for treating acute myeloid leukemia that were later validated through testing. It also generated valuable hypotheses about potential treatment targets for liver fibrosis and helped explain antibiotic resistance mechanisms - some of which researchers had independently confirmed before the system's development.

The co-scientist is also hallucinating

Despite these early successes, Google Research acknowledges several important limitations. The system needs better literature research capabilities and fact-checking procedures. The evaluation methods require expansion, particularly through testing with more experts across different types of research objectives.

The development team recommends adding external tools for cross-checking results and improving the system's automatic self-assessment methods. To gather more comprehensive feedback, Google plans to give select research institutions access to the AI Co-Scientist through a trusted tester program.