Scientific progress is being increasingly enabled by the ability to examine natural phenomena through the use of computation. The emergence of ?big data?, and advances in machine learning have dramatically accelerated some of the key steps in science, e.g., data acquisition and model fitting. However, other key elements of the scientific process, e.g., generating hypotheses, designing, prioritizing and executing experiments, integrating data, models, and simulations, drawing inferences and constructing explanations, reconciling scientific arguments, and communicating across disciplines, remain largely untouched by the advances in artificial intelligence (AI). Accelerating scientific progress, potentially by several orders of magnitude, by effectively addressing these bottlenecks presents a grand challenge for AI. Materials discovery, design and synthesis provides an excellent testbed for addressing the AI grand challenges presented by scientific discovery: The demand for new materials for applications ranging from energy technologies (batteries, solar cells, energy harvesting technologies) to sensors, artificial organs and computing technologies (e.g., quantum computers) far exceeds the capabilities of traditional materials design and synthesis, and takes years to decades of effort. This project brought  together an interdisciplinary team of researchers with complementary expertise in AI and Material Science to launch a planning effort to lay the groundwork for an AI-Enabled Materials Discovery, Design, and Synthesis (AIMS) Institute.

AIMS envisions synergistic advances across multiple areas of AI, including: (i) knowledge representation frameworks for encoding, communicating, and reasoning with models or abstractions of scientific domains, scientific artifacts, e.g. data, experiments, hypotheses, models; (ii) planning for optimizing scientific studies, experiments, etc.; (iii) machine learning and causal inference methods that can provide explanations of their results in the context of available knowledge, and recommend experiments to validate the predictions using the available experimental techniques); and (iv) algorithmic abstractions of AI-enabled human-machine, AI-enabled human-human, and machine-machine collaborations in science. Addressing this AI grand challenge would unify many of the sub-fields of AI, yield fundamental advances across multiple areas of AI, and AI mediated human-machine systems that support collaborative team science. AIMS envisions the AI advances to go hand-in-hand with use-inspired research driven by some of the most pressing challenges in materials discovery, design, and synthesis, yielding scientific insights into the relationships between materials structure and their properties, as well as new ways of rapidly optimizing material properties for specific applications.

This planning project pursued activities designed to catalyze  collaborations that transcend disciplinary, institutional and organizational boundaries that bring together experts in artificial intelligence and material sciences to dramatically accelerate AI-powered materials design, discovery and synthesis. The project organized focused workshops and idea labs to further develop the AIMS vision, initiate interdisciplinary research at the interface between AI and Material Science, identify the AIMS infrastructure needs, develop education and outreach plans, establish synergistic partnerships, and develop the requisite organizational structure and processes for realizing the AIMS vision. These activities have led to the formation of a strong team of researchers spanning academia, industry, and national labs to pursue the AIMS vision, formation of working groups to pursue a concerted research agenda encompassing: (i) Machine learning methods for predictive modeling from multi-modal, multi-fidelity data, while enabling continual and federated learning on tasks and data as they are presented, and integrating materials domain information from physical principles, symmetries, constraints, and measurement uncertainties; (ii) AI-assisted design, planning, prioritization and
optimization of experiments, including effective closed-loop coupling of domain knowledge, ML models, simulations, human expertise, surrogate experiments, and uncertainty estimates, all across differing timescales (from real-time for automated experiments to closed-loop, human-AI collaboration on a human-compatible timescale); (iii) Human-AI collaboration, including insights gained from studies of human and human-AI teams, methods for optimal division of work between people and machines; human-AI interaction design patterns for organizing human-AI teams. These advances, taken together, will dramatically
accelerate the understanding of material structure, properties, and behavior, design of novel materials, e.g., 2D materials, ferroelectrics with applications in semiconductors, quantum materials, with desired properties, and optimize their synthesis in a closed-loop fashion.

The project catalyzed the formation of a national nexus for collaboration focused on accelerating science in general, and materials characterization, design, and synthesis in particular. The project piloted several strategies for training of a diverse cadre of the next generation of AI experts and AI-savvy material scientists (through  the launching of new AI undergraduate majors and minors and model curricula, broadening participation activities, research experiences for teachers, research experiences and other experiential learning opportunities for undergraduates, educational materials to increase AI-literacy for K-12 teachers and students, hackathons and workshops), democratization of access to scientific AI tools and expertise (through virtual office hours, hands-on workshops, videos, access to research expertise, broad dissemination of data, AI methods, educational materials, and open source tools), and knowledge transfer (through collaboration with national labs and industry partners), and increased research capacity, workforce, and expertise in an area of national priority.

Last Modified: 01/14/2025
Modified by: Vasant G Honavar