Artificial intelligence adoption and green innovation: Insights from a nonlinear perspective using generalized random forests

Artificial intelligence (AI) is reshaping economies and societies, exerting a fundamental influence on green innovation initiatives and the trajectory of sustainable development. However, existing studies on AI typically use traditional causal inference methods to estimate the average treatment effect of AI on corporate green innovation, but these methods do not capture nonlinear heterogeneous treatment effects. In this study, we used the generalized random forests model to estimate AI's conditional average treatment effect. Our findings show that AI significantly promotes green innovation, with stronger effects as adoption intensity increases, and that this impact varies considerably across firms. Concerning size, AI's effect intensifies as firm size increases. Concerning ownership concentration, AI's effect exhibits a U-shaped pattern. Mechanism analysis further shows that AI drives green innovation by promoting digital-real economy integration, enhancing firms' green total factor productivity (GTFP), and increasing demand for high-skilled labor. A higher level of GTFP leads to a stronger heterogeneous treatment effect, whereas the effect of high-skilled labor is stronger at low levels and remains steady with small fluctuations beyond a threshold. In contrast, a higher level of digital-real economy integration leads to a weaker heterogeneous treatment effect of AI on green innovation. This study facilitates the differentiated application of AI and enhances its alignment with green development.