Accelerating scientific discovery with Co-Scientist

Scientific discovery is pushed by scientists producing novel hypotheses for complicated issues that endure rigorous experimental validation. To increase this course of, we introduce Co-Scientist, a multi-agent AI system constructed on Gemini for structured scientific considering and speculation technology. Co-Scientist goals to assist scientists uncover new unique information. Conditioned on their analysis aims and prior scientific proof, it formulates demonstrably novel analysis hypotheses for experimental verification. The system’s design entails brokers repeatedly producing, critiquing and refining hypotheses accelerated by scaling test-time compute. Key contributions embrace: (1) a multi-agent structure with an asynchronous process execution framework for versatile compute scaling; (2) a event evolution course of for self-improving hypotheses technology. Automated evaluations present continued advantages of test-time compute scaling, enhancing speculation high quality over time. While common function, we focus the validation in three biomedical purposes: drug repurposing, novel goal discovery ¹, and explaining mechanisms of anti-microbial resistance ². Specifically, Co-Scientist helped determine new drug repurposing candidates and synergistic mixture therapies for acute myeloid leukemia, which had been validated by in vitro experiments. These real-world validations display the potential of Co-Scientist to speed up scientific discovery and usher in an period of AI empowered scientists.