Leveraging Symbolic Artificial Intelligence and Fuzzy Logic for Materials Science: A Review of Methods, Challenges, and Applications to Scarce and Imperfect Experimental Data

Abstract

This article explores the transformative potential of symbolic artificial intelligence (AI) in the field of materials science, particularly in leveraging experimental data. While state‐of‐the‐art methods like neural networks and statistical models face challenges with limited and imperfect data, symbolic AI offers a promising alternative. We provide an inclusive definition of symbolic AI, contextualizing it within its historical development and emphasizing its ability to manipulate symbols and provide interpretable formalisms and reasoning capabilities. The article presents several symbolic AI models, including decision trees, ensemble methods, rule‐based systems, constraint‐based models, knowledge graphs, ontologies, causal graphical models, and fuzzy logic models, and discusses their applications in materials science. We also present our vision of AI for materials, grounded in realism and fully aware of the challenges, based on symbolic AI and fuzzy logic formalism. We also present our initial work in the field of materials science, ranging from the prediction of properties to the active exploration of the parameter space of a process, with the aim of optimizing one or more properties. The “DIADEM Program” allows us to go beyond the applications of existing AI methods and explore new solutions. This seems crucial given the limitations of the AI methods most widely used in the community. We therefore define methods with constraints: we seek to create methods that provide interpretable models, that can incorporate expert knowledge, that are easy to parameterize by a user unfamiliar with AI, and that are capable of extrapolation. This requires revisiting traditional methods but offers attractive prospects for transfer to industry, which is a major commitment of the “DIADEM Program.”