(95e) Automated Mechanism-Based Explanation Generation of Machine Learning Models

One of the main criticisms of machine learning (ML) models is their lack of transparency and explainability, which are important in certain applications such as process diagnosis, control, and safety. Furthermore, since chemical engineering is replete with first-principles-based models, it would be very helpful to have the ML-assisted models be explained using fundamental causal mechanisms, laws, and constitutive relations. Recent developments in the ability of large language models (LLMs), such as GPT-3, LaMDA, and BLOOM, have highlighted the capabilities of generative AI for text generation tasks. While these are successful for general uses, they encounter challenges when used for domain-specific applications due to the lack of highly-technical domain information and context. We present a novel framework for the extraction of such domain-specific information from textual sources, and its organization in a structured format using an ontology. This knowledge is used to generate mechanistic explanations for the ML-assisted models. In this paper, we demonstrate our results in the domain of reaction kinetics.