Human–AI Discovery of Governing Reaction Principles for Geological Hydrogen Production

The fundamental principles governing subsurface reaction behavior remain elusive, as emergent system dynamics arise from nonlinear coupling among multiple interacting physical and chemical variables. Here, using geological hydrogen production as a model subsurface reaction network, we show that hydrogen generation is governed by an emergent and complex high-reactivity regime inaccessible to conventional experimentation. Using a closed-loop Human–AI experimental framework, we explored a nine-dimensional reaction space and identified conditions that increased hydrogen production from model brucite by up to two orders of magnitude relative to baseline rock–water reactions in fewer than 90 experiments. Explainable machine learning revealed anomalous observations which were a result of higher-order interactions dominance, with moderate temperature and bicarbonate concentration together with elevated surfactant and catalyst concentrations cooperatively driving the rate of chemistry in a narrow regime. Time-resolved characterization showed that these conditions accelerate Fe²⁺ oxidation, establishing the mechanistic basis of the high-reactivity regime. Validation using natural olivine and basalt samples demonstrates that these governing principles extend to geologically relevant rocks. More broadly, our AI-driven high-throughput workflow establishes a general strategy for uncovering governing principles across complex subsurface reaction networks, including carbon mineralization and critical mineral recovery.