Seminar - Agentic AI for Materials Discovery: From Scientific Reasoning to Autonomous Experimentation
Supporting the below United Nations Sustainable Development Goals:支持以下聯合國可持續發展目標：支持以下联合国可持续发展目标：
Self-driving laboratories (SDL) are rapidly transforming the way scientific research is conducted, enabling accelerated discovery through closed-loop experimentation, data-driven decision-making, and intelligent experimental design. However, many existing autonomous research systems remain limited by synchronous workflows, sparse data, delayed feedback from characterization, and the complexity of real-world experimental processes.
In this talk, I will present our advances toward developing next-generation SDLs powered by Small Science Agentic Models (SSAMs), a modular, domain-specific agentic AI framework designed to integrate scientific reasoning with autonomous experimentation. Rather than relying on general-purpose AI agents, SSAMs employ specialized scientific agents that collaborate to generate hypotheses, design experiments, interpret characterization data, make machine learning-guided decisions, and orchestrate laboratory workflows. By maintaining awareness of the complete experimental trajectory and incorporating physics-informed decision-making under asynchronous conditions, the framework enables more efficient exploration of complex materials design spaces.
I will also discuss our broader vision for autonomous materials discovery, including the integration of domain knowledge, interpretable machine learning, multimodal characterization, and laboratory automation to enable intelligent, adaptive experimentation. Examples from our work on halide perovskites and other functional materials will illustrate how agentic AI can accelerate scientific discovery while providing fundamental insights into structure–property relationships. Finally, I will highlight future opportunities and challenges in building trustworthy, scalable, and collaborative autonomous scientific laboratories that can transform materials research and beyond.
Mahshid Ahmadi is an Associate Professor in the Department of Materials Science and Engineering at the University of Tennessee, Knoxville. She leads a nationally recognized research program at the intersection of artificial intelligence, autonomous experimentation, and materials discovery, developing self-driving laboratories and interpretable AI to accelerate the discovery of halide perovskites.
She is deputy director and co-PI of the $20M NSF Platform for Programmable Cloud labs (PCL–ATHENA) and the DoE GENESIS Mission at the University of Tennessee. Her honors include the NSF CAREER Award (2021), Alfred P. Sloan Research Fellowship in Chemistry (2022), Scialog Fellowship and awardee in Automated Chemical Laboratories (2026) from the Research Corporation for Science Advancement (RCSA), the ACS Materials Au Early Career Rising Star Award (2024), the UTK Tickle College of Engineering Professional Promise in Research Award (2024), and the UTK MSE Excellence in Research Award (2022). She serves as an Associate Editor of APL Machine Learning and has authored more than 120 peer-reviewed publications.