Artificial intelligence for sustainable development: assessing net carbon mitigation benefits of AI-based climate solutions with anomaly detection and life-cycle emissions accounting

Abstract

Artificial intelligence (AI) can support climate mitigation through renewable-energy forecasting, demand-side management, and low-energy transport routing, but the climate value of these applications is frequently reported without deducting the emissions generated across the AI system life cycle. This study develops an integrated, reproducible framework for estimating the net carbon mitigation benefit of AI-based climate solutions. The framework combines: (i) gross avoided emissions relative to a functionally equivalent non-AI baseline; (ii) operational emissions from model development, training, inference, data storage, and supporting infrastructure; (iii) allocated embodied emissions from computing hardware and end-of-life treatment; and (iv) an anomaly-detection layer that screens extracted parameters for unit errors, impossible values, inconsistent system boundaries, and influential outliers before carbon calculations are performed. The framework is demonstrated using three literature-derived applications concerning photovoltaic/demand-side energy management, energy-system optimization, and machine-learning-assisted electric-vehicle routing. The source studies report gross improvements of approximately 34%, 73%, and 9%, respectively; however, these percentages are not directly comparable until the functional unit, deployment scale, grid carbon intensity, and computational burden are harmonized. Within the evaluated scenarios, deployment scale and electricity-grid carbon intensity are among the factors that materially influence the estimated net benefit. Hardware manufacturing, supporting infrastructure, and end-of-life impacts could reduce the estimated net benefit further; however, these components were not quantified in the empirical demonstrations because case-specific data were unavailable. Because several computational inputs are not disclosed by the original studies, the numerical results are presented as scenario-based estimates rather than measurements. The proposed framework advances sustainable-development assessment by linking data-quality control, life-cycle emissions accounting, sensitivity analysis, and carbon break-even analysis in a single decision procedure.