NIMS, in collaboration with Tsukuba University, has developed a "self-organizing autonomous AI network" technology, enabling multiple autonomous AI systems to spontaneously connect and efficiently discover new materials. This capability was successfully demonstrated through simulations, showcasing the potential of this innovative approach.
Recent advancements in AI, robotics, and simulation have led to a surge in interest for autonomous AI systems globally. However, many existing systems operate independently without cooperation, exploring different material systems but lacking the ability to leverage each other's discoveries for enhanced exploration. The concept behind this research parallels human collaboration in research communities, where knowledge is shared to expedite investigation. By networking multiple autonomous AI systems to share extracted trends from data, more efficient material discovery can be achieved.
Inspired by human communication methods, the research team created algorithms that allow autonomous AI systems to share knowledge—rather than mere data—enabling collaborative exploration. During tests involving three autonomous systems focused on maximizing different material properties, the spontaneous exchange of knowledge among systems enhanced optimization speed significantly. This demonstrates that constructing an autonomous AI network improves the exploration efficiency of each system.
The research was conducted by Yuma Iwasaki, a principal researcher at NIMS Material Research Center, and Yasuhiko Igarashi, an associate professor at Tsukuba University. The findings were published online on December 9, 2025, in the journal “npj Computational Materials.” This work received support from JST's Strategic Creative Research Program CREST, focusing on new material creation through hierarchical autonomous exploration methods that augment scientists' capabilities.