Prospects for using artificial intelligence to understand intrinsic kinetics of heterogeneous catalytic reactions

Heterogeneous catalysis research struggles to connect intrinsic kinetics with experimentally observed behavior due to complex multiscale models, limited observability, and a many-to-one mapping between mechanisms and data. Advances in operando experiments, atomic-scale models, microkinetic models, and reactor simulations provide rich information, but dramatically expand model complexity and uncertainty. Artificial intelligence can reduce the human time needed for modeling by enabling ‘self-driving’ multiscale models that automate model construction, refinement, and validation across scales. Increased throughput will result in large ensembles of multiscale models that better explore parameter space, yield insight into sensitivity and uncertainty, and improve quantitative agreement between theory and experiment. • Few AI models address the full multiscale nature of catalysis. • Multiscale models remain limited by assumptions and parameter uncertainty. • Operando and transient methods yield rich but heterogeneous, nonstandard data. • Generative and agentic AI can expand mechanistic search and reduce bias. • Integration of AI with modeling and data can produce higher-quality multiscale models.