Multimodal Machine Learning and AI-Accelerated Electrocatalytic CO₂ Reduction Towards Self-Interpretable Discovery Loops

Electrochemical CO₂ reduction (CO₂RR) converts renewable electricity into fuels and chemicals, but experimental chemists still struggle to track the structural evolution of CO₂RR catalysts under reaction conditions and systematically optimize intertwined factors governing performance. However, interface dynamics under bias and reaction networks are complex, data are scattered across techniques and papers, and many catalyst "design rules" remain qualitative. Herein, we present machine learning (ML) as a practical data-driven toolkit for integrating information from simulations, operando characterization, microscopy, and the literature to develop quantitative descriptions of CO₂RR. We first discuss how artificial intelligence (AI) accelerates time- and resource-intensive ab initio simulations, and the analysis of operando spectroscopic and electron microscopy, enabling more accurate identification of the dynamic structural and electronic evolution of catalysts under reaction conditions. Then, from a data science perspective, we review how AI enables the systematic, large-scale construction of knowledge frameworks, databases, and integrated platforms for carbon dioxide reduction. Building on these mechanistic and data foundations, we highlight ML-guided discovery loops that couple interpretable models, generative design, autonomous experimentation, and expert intervention to identify new catalysts, interfacial optimization, and operating conditions. We hope this review helps CO₂RR researchers to use AI-assisted approaches to designing, understanding, and optimizing their own catalyst systems.