Explainable AI in Electrocatalysis and Photocatalysis: From Catalyst Design to Mechanistic Insights

Artificial intelligence (AI) has progressed across multiple scientific disciplines, particularly transforming catalytic science by enabling data-driven material discovery, optimization, and mechanistic understanding at the atomic scale. However, conventional AI models in photo- and electrocatalysis remain largely constrained by their intrinsic “black-box” nature, which limits mechanistic interpretability and the rational transfer of learned knowledge to new catalytic systems. Here, we demonstrate a comprehensive review of explainable artificial intelligence (XAI) as an emerging paradigm that bridges predictive accuracy with physical interpretability, thereby enabling transparent catalyst design and profound mechanistic insights. Building upon the methodological evolution of artificial intelligence in catalytic informatics, this review traces the progression from descriptor construction and learning algorithms to the incorporation of explainable artificial intelligence, highlighting how interpretability is systematically integrated to elucidate structure–activity relationships. By drawing on both traditional and deep-learning-based XAI approaches, this review demonstrates how electronic, geometric, and reactive descriptors can be systematically interpreted to clarify catalytic performance. Applications of XAI in electrocatalysis are summarized, including hydrogen and oxygen evolution, oxygen reduction, nitrogen reduction, and carbon dioxide reduction reactions, where XAI has supported both catalyst screening and mechanistic insights. In photocatalysis, it highlights XAI frameworks applied to CO 2 conversion for carbon neutrality, aqueous pollution remediation, and atmospheric pollution abatement, demonstrating broad utility in energy and environmental catalysis. Overall, advances in XAI are reshaping catalyst development from empirical modeling toward a transparent, knowledge-driven paradigm, establishing an explicit link between catalyst design and mechanistic understanding while alleviating the black-box limitations of conventional AI.