Associate Professor Rafael Gómez-Bombarelli — a tenured researcher in MIT’s Department of Materials Science and Engineering — says the field is now at a “second inflection point” where generative AI, high-throughput data, and physics-based simulations combine to enable general scientific intelligence that can reason about language, materials, and synthesis processes.
Gómez-Bombarelli’s work has long focused on computational approaches that reduce the need for costly and slow physical experiments. By using machine learning alongside simulations grounded in physical laws, his group has discovered new materials for batteries, catalysts, plastics, and OLEDs, and helped lay the groundwork for companies — including his own — that apply AI to drug discovery, robotics, and chemical sciences. These AI-driven systems produce vast datasets that improve both models and simulations through “virtuous cycles” of data and computation.
The research reflects a broader trend in scientific computing: AI no longer just approximates outcomes but becomes part of the discovery engine itself. Teams like Gómez-Bombarelli’s run large-scale simulations and computational experiments entirely in silico, enabling researchers to explore many hypotheses in parallel and rapidly triage promising leads before experimental validation. This shift speeds up innovation and helps meet real-world needs, from energy storage materials to life-science breakthroughs.
Experts say this evolution toward AI-assisted science marks a fundamental change in how research is conducted. As language models and multi-modal systems become capable of reasoning across diverse data types — text, structure, and simulations — they unlock new opportunities to tackle complex scientific problems faster than ever before, ultimately bringing the “future forward in time” by reducing barriers between idea and application.
