The End of Trial-and-Error: A Vision for Generative Intelligence in Metamaterial Design

Metamaterials, engineered materials with programmable structure-property relationships, have revolutionized fields from aerospace to biomedicine. Yet discovery remains dominated by costly simulations, sparse data, and trial-and-error experimentation. In this Blue Sky paper, we envision a future where generative intelligence replaces brute-force design with autonomous, data-driven exploration. We argue that the convergence of generative models, physical priors, and multi-objective reasoning marks the beginning of the end of trial-and-error in metamaterial innovation. This vision challenges the data mining community to move beyond traditional modalities and embrace new frontiers in scientific AI, where the goal is not just to predict but to invent. Realizing this vision demands breakthroughs in structure-aware learning, simulation-free optimization, and few-shot generalization across physics-governed domains. We demonstrate the feasibility of this paradigm through early work on generative frameworks that create diverse, functional, and manufacturable metamaterials. This vision redefines the role of AI, not merely as a passive learner of the world as it is, but as an active architect of what it could become.