Governing AI for the climate: A political and economic analysis of G-20 regulatory thresholds

Ocean Dynamics · Springer Nature · 2026-10-20 · DOI: 10.1007/s10784-026-09742-z

Аннотация

This study examines the relationship between artificial intelligence (AI) development, proxied by AI-related scientific publications, and carbon dioxide (CO2) emissions in Group of Twenty (G-20) economies. Static and dynamic panel threshold models are estimated, with a composite index of business, labour and credit-market regulation as the threshold variable, and the analysis accounts for cross-sectional dependence and unit root properties. The results indicate a significant double-threshold effect of regulation. In the baseline models, AI development is associated with lower carbon emissions, but this association is regime-specific. In the main threshold and regime-specific fixed-effects specifications, the negative association is strongest under the Medium Regulation regime. Under the High Regulation regime, the association is weaker and not robust across static and dynamic models, which may reflect structural heterogeneity, digital rebound effects or diminishing returns to regulation. However, system generalised method of moments (GMM) and lagged-AI estimates provide a more heterogeneous picture, associating AI with higher emissions in the Low and Medium Regulation regimes and with lower emissions in the High Regulation regime; the results are therefore sensitive to the estimator. Industrial value added is positively associated with emissions in the baseline models. The findings are consistent with Ecological Modernization Theory, which holds that the environmental outcomes of technological advancement depend on institutional conditions. The medium-regulation “Goldilocks zone” is thus a result of the main threshold specification rather than an established general result, and the broader regulatory environment should be considered when assessing the climate implications of AI research capacity.

Читать у издателя →

Похожие статьи