On the journey of exploring the frontiers of science with artificial intelligence, another milestone has been reached. On July 3rd, the Alibaba DAMO Academy, in collaboration with Renmin University of China and the University of Chinese Academy of Sciences, officially launched Elements Claw, the world's first AI agent specifically designed for the discovery of superconducting materials. This achievement not only marks a transition of AI from a "support role" to an "independent research" in the field of scientific discovery, but also provides an efficient automated paradigm for the development of new materials.

Traditional processes for discovering superconducting materials are extremely long and rely heavily on trial and error. International mainstream databases like SuperCon have accumulated data over decades, but they have only recorded about 2000 materials. The newly introduced Elements Claw has completely broken through this bottleneck. This AI agent uses an advanced "specialized and general" architecture, and is trained on a database containing 125 million molecules and crystal structures, building a 10-billion-parameter atomic foundation model (Elements). This enables it to accurately assess the superconducting potential of materials with an AUC score as high as 0.996, and to control the prediction error of critical temperature within 1K.

image.png

The power of Elements Claw lies in its ability to perform full-process operations similar to those of human scientists. It can independently access massive literature, evaluate the feasibility of synthesis, design experimental plans, and achieve "self-evolution" of algorithms after discovering new clues. In practical testing, the AI efficiently screened out 68,000 superconducting candidate materials from 2.4 million crystal structures, consuming only 28 GPU hours.

Currently, the research team has successfully synthesized and verified four new superconducting materials through experiments, including HfZrRe4, which was designed from scratch by AI, as well as Hf21Re25, Zr4VRe7, and Zr3ScRe8 discovered by correcting and deeply analyzing existing databases. Their critical temperatures can reach as high as 6.5K.

Rong Yu, head of the Science Intelligence department at DAMO Academy, stated that these achievements have initially validated the great potential of AI agents in material discovery. To promote industry progress, the team has fully released the data of 2.4 million stable crystals. Professor Huang Wenbing from the Gaoqiang Institute of Artificial Intelligence at Renmin University further pointed out that this intelligent agent framework is expected to be widely reused in the development of more key materials such as solid-state battery electrolytes, multiphase catalysts, and thermoelectric materials, opening a new chapter in scientific discovery.