Artificial intelligence and carbon emissions : impact, mechanisms and policy strategies in OECD and BRICS economies

Abstract

Climate change has become one of the most pressing issues facing humanity prompting countries to look for alternative solutions. In the light of this, AI is seen as a versatile technology that can efficiently transform and solve climate change issues. This dissertation uses different analytical frameworks to assess how AI effect carbon emissions and under which policy conditions AI is considered as an effectively decarbonization tool for OECD and BRICS economies. This first paper examined the impact of AI on carbon emissions by comparing OECD and BRICS countries and assessing the moderating role of R&D. A dynamic panel model using the GMM estimation, PCSE and DK are applied. Two categories of data, spanning from 2013 to 2023 for the AI investment variable and from 2012 to 2022 for the AI project variable, are used. In addition, we utilised AI patent data spanning from 1980 to 2023 to verify the robustness of the results. The baseline estimation of AI investment and AI impactful projects on carbon emission in the short run, as well as the long run, shows that AI adoption has a positive impact on carbon emissions. The interaction of R&D changes the positive effect into a negative one. Therefore, the effect is channelled through green innovation, clean energy adoption in AI software development. In the baseline model, AI increases emissions in both BRICS and OECD while the interaction of R&D shows a strong mitigating effect of AI on carbon emissions in OECD. Nevertheless, in BRICS, the coefficient is negative in sign but does not reach conventional significance thresholds, suggesting a weaker institutional capacity to channel R&D into green outcomes. The robustness results using AI patents also confirm that AI increases emissions. The second paper examines the transmission channels through which AI affects carbon emissions by focusing on energy systems and assessing the dynamic, nonlinear patterns of this effect across emissions quantiles by using a panel data set of 52 countries from 2012 to 2022. We computed the AI index variable using PCA. A dynamic panel with a GMM approach and panel quantile regression with nonlinear are used. The dynamic analysis shows a dynamic J-curve pattern in which AI initially increases emissions however, when the AI efficiency gains are materialized, it reduces carbon them. The mechanism analysis showed that AI increases emissions through increased energy intensity and reduces them through renewable energy development. The nonlinear effect suggests an inverted U similar to the EKC hypothesis with low emitting countries having more favourable emissions reduction effects. Using a System Dynamics (SD) modelling, the third paper evaluates policy options that potentially affect AI adoption and carbon emission trajectories in OECD and BRICS economies in three policy scenarios between the year 2024 and 2074. Findings indicate that OECD nations experience faster AI diffusion and greater emission reductions, driven by more powerful digital infrastructure and stronger absorptive capacity. Interestingly, the pre-2030 interventions have almost doubled cumulative savings compared to late action and combine policy instrument yield better results.