人工知能：科学的発見を加速させる可能性のあるAI研究アシスタント

Two artificial intelligence (AI) systems that can assist throughout multiple processes involved in scientific research — such as generating hypotheses, designing experiments, and analysing data — are presented in Nature this week. The systems, individually developed by Google DeepMind and FutureHouse, are designed to assist researchers in accelerating scientific discovery, not to replace them.

Scientific discovery is driven by the repeated generation of novel hypotheses, experimental validation, and data analysis. The increasing complexity and overlapping of scientific topics mean that deep subject-specific expertise is required along with broader knowledge across disciplines. AI has been shown to speed up individual steps in the research process, but a single system could streamline the workflow. Two independent systems — Co-Scientist from Google DeepMind and Robin from FutureHouse — demonstrate the potential of such systems to improve the scientific discovery process.

Both AI assistants are multi-agent systems, which use multiple autonomous, specialized AI agents that can execute different tasks throughout the research process. This approach enables the systems to generate hypotheses, propose experiments to test the hypotheses, interpret the experimental results, and refine hypotheses on the basis of the findings.

Co-Scientist, built with Gemini 2.0, is a general-purpose multi-agent system for scientific discovery. It is designed to be applicable across scientific disciplines, although the initial validations have focused on biomedicine. For example, Co-Scientist proposed new drug candidates and combination therapies for acute myeloid leukaemia, an aggressive cancer of the white blood cells. The suggested treatments were shown to be potentially beneficial in cell line experiments, although rigorous preclinical and clinical assessment would be required for therapeutic validation, note the authors, Vivek Natarajan and colleagues. Beyond cancer research, Co-Scientist also discovered new drug targets for liver fibrosis and uncovered key genetic mechanisms behind antimicrobial resistance.

Robin, which uses both OpenAI o4-mini and Anthropic Claude 3.7, is designed to aid discovery in the field of experimental biology. Samuel Rodriques and colleagues apply the system to drug discovery investigations. For example, Robin facilitated the identification of potential treatments for dry age-related macular degeneration, a major cause of blindness in the developed world. The suggestions included identification of a modifiable process within retinal cells to target and proposing the use of a drug candidate that has not previously been proposed for treating this condition. Robin also suggested follow-up studies to investigate underlying mechanisms, which identified novel potential drug targets. Such treatments would require validation in preclinical tests and clinical trials, the authors note.

Both teams emphasize that these systems are designed to collaborate with researchers, and a scientist would always be in the loop. The real-world demonstrations from both groups provide examples of what the future of scientific research with AI agents might look like.

• Article
• Published: 19 May 2026

Gottweis, J., Weng, WH., Daryin, A. et al. Accelerating scientific discovery with Co-Scientist. Nature (2026). https://doi.org/10.1038/s41586-026-10644-y

• Article
• Published: 19 May 2026

Ghareeb, A.E., Chang, B., Mitchener, L. et al. A multi-agent system for automating scientific discovery. Nature (2026). https://doi.org/10.1038/s41586-026-10652-y

Editorial: Why AI cannot do good science without humans
https://www.nature.com/articles/d41586-026-01551-3

News: Teams of AI agents boost speed of research
https://www.nature.com/articles/d41586-026-01596-4