A joint research team from NIMS and University of Tsukuba developed "autonomous AI network" technology by which multiple autonomous AI systems can efficiently discover new materials by spontaneously collaborating with each other and forming a network. The team demonstrated the effectiveness of the technology through simulations. This research result was published in npj Computational Materials on December 9, 2025.
Background
In recent years, "autonomous AI systems" that integrate artificial intelligence (AI), robotics, and simulations have attracted attention, and have been built and operated worldwide. However, current autonomous AI systems operate in isolation, without collaborating with other systems. This is because the AI systems explore different material systems, and while they can share data easily, it is challenging for them to utilize data from other systems in their own autonomous exploration. Humans (researchers) advance research in a sophisticated manner while sharing extensive knowledge by forming a research community through conversation (see the left side of the figure). Likewise, if multiple autonomous AI systems can perform autonomous exploration while sharing and utilizing extensive knowledge (trends extracted from data) by forming a network, they can discover new materials more efficiently.
Key Findings
In this research, the team took a hint from human research communication methods to develop "autonomous AI network" technology by which multiple autonomous AI systems collaborate to perform autonomous exploration while sharing knowledge. In a research community, a human researcher normally does not merely give their research data to another researcher, but communicates by way of conveying some knowledge gained from that data to the other researcher. In order to realize this also among autonomous AI systems, the research team built an algorithm that incorporates knowledge learned by other systems as a reference for decision-making, and enabled the AI systems to perform autonomous exploration while sharing knowledge instead of data. As shown on the right side of the figure, when three autonomous AI systems, each performing exploration to maximize a different physical property value, were made to spontaneously exchange knowledge with each other, their optimization speed was found to improve. In other words, the team demonstrated that the exploration efficiency of each system improves by forming an autonomous AI network.
Future Outlook
Autonomous AI systems that integrate AI, robotics, and simulations have been developed worldwide, and are constantly performing material exploration. Their number will continue to increase rapidly, and various types of autonomous AI systems will discover and synthesize numerous new materials. This large number of autonomous AI systems has a potential to generate greater values by collaborating with each other in the future. Going forward, the team aims to build a more enormous autonomous AI network, while further advancing development of autonomous AI systems.
Other Information
- This project was conducted by a team led by Yuma Iwasaki (Principal Researcher, Data-driven Materials Design Group, Center for Basic Research on Materials, NIMS) and Yasuhiko Igarashi (Associate Professor, Institute of Systems and Information Engineering, University of Tsukuba) as part of Japan Science and Technology Agency (JST) Strategic Basic Research Program CREST "Scientists augmentation and materials discovery by hierarchical autonomous materials search" (JPMJCR21O1).
- This research was published online in npj Computational Materials on December 9, 2025.
