Analyzing the green effects of AI policies: a causal random forest approach with SHAP value interpretation

Abstract

In the context of global climate change, China has been actively advancing the development of artificial intelligence (AI) technology to facilitate the green and low-carbon transformation of its economy. Drawing upon panel data of Chinese prefecture-level cities from 2012 to 2023, this study employs panel regression models and causal random forest models to systematically evaluate the impact of the National New Generation Artificial Intelligence Innovation and Development Pilot Zone policy on urban carbon emissions. Our findings reveal that the establishment of AI pilot zones significantly reduces urban carbon emission indices, exerting emission reduction effects through the promotion of industrial structure upgrading and the enhancement of energy utilization efficiency. Furthermore, SHAP value analysis identifies pronounced geographical heterogeneity in the policy impact, with more substantial emission reduction effects observed in western regions, non-urban cluster cities, non-environmental protection priority cities, and non-old industrial base cities. These results indicate that the emission reduction effects of AI technology exhibit context-dependent characteristics, with their magnitude constrained by multiple factors including regional economic development stages, industrial structures, and policy environments. This research provides a theoretical foundation for formulating differentiated AI technology promotion policies, suggesting that priority should be accorded to regions with greater emission reduction potential for AI pilot zone implementation, and that customized policies should be developed according to distinct regional characteristics to maximize the positive effects of AI technology in advancing urban green and low-carbon development