Mitigating environmental public health risks via artificial intelligence: mechanisms and boundary conditions

Background: Environmental pollution threatens population health through multiple and overlapping pathways, including gaseous emissions, wastewater discharge, and industrial solid waste. Artificial intelligence (AI) can improve environmental monitoring, energy management, and production optimization, but its broader relationship with multidimensional environmental public health risks remains insufficiently understood. This study examines whether and under what conditions artificial intelligence is associated with lower pollution-related environmental public health risks. Methods: Provincial panel data covering 30 regions in China from 2014 to 2023 were analyzed. A multidimensional environmental public health risk index was constructed from carbon dioxide emissions, sulfur dioxide emissions, nitrogen oxide emissions, industrial wastewater discharge, and industrial solid waste. Two-way fixed-effects models were combined with mediation analysis, heterogeneity and marginal-effect analysis, alternative measurement, additional control and winsorization tests, dynamic panel estimation, and a panel threshold model. Results: Higher levels of artificial intelligence were associated with lower environmental public health risks in the principal fixed-effects models, and the negative relationship remained consistent across alternative measurement, additional digital-infrastructure controls, winsorization, and concurrent policy specifications. Technological expenditure emerged as an implementation pathway through which digital capability can be translated into monitoring systems, cleaner equipment, and environmental management infrastructure, while green patents reflected a longer-horizon innovation process. Environmental investment strengthened the negative association by providing the financial and physical capacity required for artificial intelligence deployment. Electricity consumption identified greater potential for energy and production optimization, although the contribution of artificial intelligence varied across energy-use conditions. Artificial intelligence remained negatively associated with environmental public health risks across environmental regulation regimes, while its marginal contribution changed non-linearly with regulatory intensity. Conclusion: Artificial intelligence functions as a conditional environmental capability rather than an automatic technological solution. Its public health value is more likely to emerge when digital development is supported by technological expenditure, environmental investment, operational implementation, and coordinated regulatory design. These findings provide a multidimensional framework for understanding how artificial intelligence can contribute to pollution-related environmental public health risk mitigation.