Hassel and Marianne Ledbetter MatSE Colloquium - "Artificial Intelligence for Scientific Discovery"

Artificial intelligence is rapidly evolving from a passive analytical tool into an active partner in scientific discovery. In the material world, this transition requires systems that do more than recognize patterns: they must reason across scales, formulate hypotheses, design and execute investigations, learn from failure, and revise their own representations in response to evidence. This talk presents emerging approaches toward “superintelligent” discovery engines that integrate reinforcement learning, graph-based reasoning, physics-informed neural architectures, generative models, and autonomous tools for simulation, experimentation, design, and manufacturing. Particular emphasis will be placed on multi-agent systems inspired by collective intelligence in nature, in which specialized agents collaborate, critique one another, and continuously improve their strategies as they solve complex scientific problems. Case studies from materials science, engineering, and biology will illustrate how these systems can uncover hidden structure-property relationships, generate and evaluate novel material and molecular designs, connect knowledge across disciplinary boundaries, and transform scientific failures into new hypotheses and executable knowledge. These advances point toward a new form of artificial intelligence that actively expands the boundaries of human knowledge.