Extracting human interpretable structure-property relationships in chemistry using XAI and large language models

XAI can bridge the gap between complex AI models and human understanding: The paper demonstrates that Explainable AI methods can extract meaningful input-output relationships from machine learning models, making it possible to understand how AI systems arrive at their conclusions—a critical requirement for responsible AI deployment in scientific fields.

Making AI explanations accessible requires more than just technical tools: The authors identify a key limitation of existing XAI methods: they are designed for technically trained users. This highlights an important principle for responsible AI—that explanations must be understandable to diverse stakeholders, not just AI specialists, to enable proper oversight and trust.

Large language models can democratize AI interpretability: By combining XAI methods with LLMs that can access scientific literature, the XpertAI framework automatically generates natural language explanations of chemical data. This approach suggests that AI systems can be designed to make their reasoning transparent and accessible to domain experts without deep machine learning expertise.

Integration of multiple AI tools can enhance trustworthiness: The paper shows that combining XAI with LLMs produces "specific, scientific, and interpretable explanations" across five case studies, suggesting that responsible AI systems benefit from layered approaches that leverage different AI capabilities to provide comprehensive, understandable insights.

Responsible AI in chemistry requires interpretability: By focusing on structure-property relationships in chemistry, the work demonstrates that high-stakes scientific and potentially industrial applications need AI systems where decisions can be explained and validated by domain experts—not just accepted as black-box outputs.