L'agente AI aiuta a preparare misure sperimentali a raggi X sincrotroni, aprendo la strada al funzionamento autonomo

An artificial intelligence agent has successfully prepared a synchrotron X-ray experiment, marking a key step toward fully autonomous operation of such facilities. Developed by researchers at Stanford University and the SLAC National Accelerator Laboratory, the AI system autonomously planned actions, interpreted observations, and generated control commands to align a single-crystal sample. This achievement, detailed in a paper published in Nature Machine Intelligence, represents one of the first demonstrations of an AI-driven experimental setup in a real-world synchrotron environment. The experiment took place at the Stanford Synchrotron Radiation Lightsource (SSRL), where the AI agent was tested on a real synchrotron X-ray beamline. The system observed detector images, experimental logs, and scan results using structured software tools. It then reasoned about the current state of the experiment and issued precise commands to adjust the position of a crystal sample. The success of this trial suggests that AI systems could soon take over routine and complex tasks typically performed by human scientists during X-ray experiments. Synchrotron facilities are among the most advanced research centers globally, producing extremely bright X-rays that allow scientists to study the atomic structures of materials, molecules, and biological samples. However, preparing and executing these experiments requires meticulous planning and execution. One of the most time-consuming steps is aligning a single-crystal sample so that high-quality diffraction patterns can be collected. This process involves adjusting multiple parameters and often requires manual intervention. Zhantao Chen, the lead researcher and now an assistant professor at The University of Texas at Austin, described the AI’s ability to navigate this challenge. “Our work demonstrates an agentic AI X-ray scientist that can autonomously align single-crystal samples at synchrotron X-ray beamlines,” he said. The AI agent queried the experimental status, assessed what needed to be done, and executed the necessary adjustments. According to Chen, the initial motivation for the project was to address the repetitive nature of sample alignment, but the team quickly realized the potential for AI to perform more complex experimental reasoning. The AI system used in the experiment relied on a combination of tools and prompts to guide its actions. These included reading experimental logs, capturing detector images, and performing motorized scans. By integrating these functions, the AI gained a comprehensive understanding of the experimental environment and could make decisions based on real-time feedback. The researchers emphasized that the AI did not simply follow pre-programmed instructions but instead adapted to changes and unexpected outcomes, demonstrating a level of autonomy previously unseen in automated laboratory systems. The test involved a Co₃Sn₂S₂ single-crystal sample mounted on a copper holder. The AI-controlled alignment process was carried out using the six-circle diffractometer at beamline BL17-2 of SSRL. The outcome was a well-aligned sample ready for high-resolution X-ray diffraction analysis. This milestone highlights the growing role of AI in scientific experimentation, particularly in environments where precision and efficiency are paramount. Looking ahead, the researchers believe this work opens new possibilities for automating other aspects of synchrotron experiments. Future developments may include AI systems capable of designing entire experimental protocols, monitoring data quality in real time, and even collaborating with human scientists to refine hypotheses. As AI continues to evolve, its integration into scientific workflows could significantly enhance the speed, accuracy, and scope of experimental research.