Automating Materials Design and Scientific Discovery Through Multi-Modal Multi-Agent Artificial Intelligence

Recent advances in AI, particularly Large Language Models (LLMs), have transformed research methodologies and accelerated discoveries in materials science. Moreover, LLMs have been instrumental in powering multi-agent systems, facilitating the automation of complex problem-solving processes and integrating knowledge from external sources such as new physics from first principles. This talk presents case studies on the design of <i>de novo</i> materials, from proteins to metallic alloys, using LLM-driven multi-agent systems, demonstrating how complex multi-model materials modeling, design, and analysis problems can be solved through various examples. A special focus will be on the use of multi-agent systems to automate advancing scientific understanding and discovery. We introduce <b>SciAgents</b>, an approach leveraging (1) large-scale ontological knowledge graphs, (2) LLMs and data retrieval tools, and (3) multi-agent systems with in-situ learning. The framework autonomously generates and refines research hypotheses, elucidates mechanisms and design principles, and discovers unexpected material properties. By integrating these capabilities, our system accelerates materials discovery by harnessing a "swarm of intelligence" akin to biological systems, unlocking nature's design principles.