Applications of Multi-Agent AI Systems for Bio-Inspired Materials Discovery

A frontier in materials discovery and design is the use of science-focused generative AI tools equipped with interdisciplinary knowledge, analysis, and synthesis capabilities. The field of biological material mechanics stands at the nexus of materials science, biology, and mechanical engineering, boasting a rich legacy of bio-inspired materials studies and discoveries. This interdisciplinary field poses as a unique candidate for exemplifying the potential of integrating AI tools for accelerated materials discovery.<br/><br/>As a first step, we present BioinspiredLLM, an open-source conversational large language model (LLM) finetuned using a corpus of biological materials literature. By harnessing the baseline abilities of a foundational model, we establish a robust method for text/data mining, data distillation, and finetuning using academic literature to develop a specialized model that surpasses the performance and efficiency of existing systems. The finetuned model exhibits strong knowledge recall, reasoning, and hypothesis generation abilities and exists in various form factors of model sizes and efficiency.<br/><br/>Next, we demonstrate the integration of our finetuned model into workflows that interface with various advanced AI systems. In one application, we exhibit how our finetuned language model can interact with image-based diffusion models to generate not only new textual descriptions of bio-inspired materials but also realize them in graphic form. In another application, we show how our finetuned language model can interact with multi-modal agents to extract bio-inspired design motifs from literature and characterize them the through generation of 3D modeling and finite element methods code. The collaboration of multiple agents enriches these workflows, enabling the generation of entirely new designs and insights that are informed by diverse, specialized agents. Multi-agent systems excel because they leverage the collective intelligence and specialized capabilities of individual agents, leading to dynamic coordination in completing complex tasks. These multi-agent frameworks are not only transformative for the next generation of scientific workflows but also highlight the critical need for continued research and development to fully realize their potential.<br/><br/>The impact of this work lies in its potential to reshape materials discovery by facilitating rapid and interdisciplinary creation of novel bio-inspired materials, which could contribute to advancements in diverse fields such as biomedical engineering, sustainable materials, and more.