Clean while not necessarily green? Intricate consequences of the application of artificial intelligence on environmental sustainability

Abstract Background Artificial intelligence (AI) is advancing rapidly in the modern era and bringing about profound changes to human society. Meanwhile, with environmental issues severely challenging socio-economic development and human well-being, environmental sustainability has become a globally concerned issue gaining growing attention. Yet, there is a lack of systematic research into AI’s potential environmental sustainability impacts, with inconsistent findings from the few available studies. While some studies indicate that AI applications can promote environmental sustainability by boosting productivity, optimizing industrial structures, and accelerating knowledge creation, other research notes that AI is not inherently environmentally beneficial and may even pose substantial environmental risks. Results This paper empirically investigates how the application of AI affects environmental sustainability from the perspectives of carbon emissions and air pollution, employing the Quantile-on-Quantile approach. The short-term, medium-term, and long-term scales in this study are constructed based on the multiscale wavelet decomposition method. Results demonstrate that although AI applications can contribute to reducing carbon emissions to some extent, this effect exhibits significant heterogeneity at different quantiles. Especially at quantiles of higher carbon emissions, AI may even increase CO 2 . In the short term, AI applications have an overall negative effect on carbon emissions. In the middle term, the impact of AI applications on CO 2 is unstable and even positive. In the long run, at almost any quantiles of carbon emissions, AI has a positive effect on carbon emissions when the degree of AI applications is high. Comparatively, the negative effects of AI applications on air pollution are more prominent and robust. Furthermore, this effect is enhanced with higher air pollution levels. From a dynamic perspective, while the short and middle term impact of AI applications on air pollution exhibits variability, the long term effect in reducing pollution is pronounced. Conclusions AI is generally a clean while not necessarily green technology in terms of its heterogeneous impact on carbon emissions and air pollution. The findings of this research have significant implications for improving environmental sustainability in the process of reaping the technological and economic benefits of AI.