AI-Driven Materials Discovery and Beyond

Recent advances in AI have significantly broadened the landscape of materials discovery [1], yet “discovery’’ encompasses multiple stages that are often treated independently. In this Seminar, we present a layered perspective that reflects both the emerging capabilities of AI and the enduring challenges of materials science. The first layer involves identifying potentially stable compounds, a task increasingly supported by large-scale generative AI models. The second layer focuses on targeted functionality—whether a material exhibits useful functional properties We highlight our recent efforts, including AI-enabled identification of solar cell candidates with record-high efficiencies [2], generative design of materials under physical constraints [3], and others. Functionality alone does not guarantee practical realizability. A third layer concerns synthesizability. We discuss recent progress showing that AI systems, when grounded in physically meaningful representations and real laboratory knowledge can assist in proposing plausible crystal-growth routes even when experimental data are limited [4]. Beyond feasibility, scalable deployment requires attention to broader economic and environmental factors, which forms our fourth layer [5]. Overall, AI is reshaping materials discovery from a sequence of isolated steps into a coherent, multi-layered reasoning process. Even so, recognizing the limitations of AI is not a weakness, but a prerequisite for its effective use, such as in materials with excess complexity [6]. By doing so, AI unlocks long-term opportunity is to build discovery systems in a fully intelligent and automated manner. Mingda Li is an Associate Professor in the Department of Nuclear Science and Engineering at MIT. He earned his BS from Tsinghua University in 2009 and did his PhD and Postdoc at MIT. Mingda's research focuses on Quantum, Nuclear, and AI. In 2025 he has been named an APS Fellow. [1] MC, ML, “Artificial intelligence-driven approaches for materials design and discovery”, Nature Materials 25,174 (2026). [1] NH, ML, “Universal Ensemble-Embedding Graph Neural Network for Direct Prediction of Optical Spectra from Crystal Structure,” Advanced Materials 36, 2409175 (2024). [2] RO, ML, "Structural Constraint Integration in Generative Model for Discovery of Quantum Material Candidates," Nature Materials 25, 223 (2026). [4] EL, ML, “AI-Driven Crystal Growth Synthesis Recipe Prediction,” to submit. [5] AB, ML, “Are Quantum Materials Economically and Environmentally Sustainable?” Materials Today 90, 241 (2025). [6] MJL, ML, “Quantum Theory of Functionally Graded Materials,” arXiv:2603.03424.