The Green AI versus AI for Green Paradox
TL;DR
It is argued that AI sustainability should be evaluated from a system-level perspective that integrates computational efficiency, environmental utility, deployment scale, and governance, and that improvements in relative efficiency do not translate into reductions in absolute environmental impact.
Abstract
Artificial intelligence (AI) plays a dual role in the environmental sustainability transition: while AI can support decarbonization and resource optimization, its development and deployment also impose computational and environmental demands. This review synthesizes 281 peer-reviewed studies retrieved from EBSCOhost and published between July 2024 and July 2026. Following PRISMA guidelines, bibliometric and thematic analyses were conducted to address three research questions: computational efficiency and the environmental footprint of AI (RQ1); AI applications for decarbonization and green transformation (RQ2); and rebound effects, structural trade-offs, and governance (RQ3). Here, three thematic clusters were identified. Green AI: Computational Footprint & Model Efficiency dominates the literature (n = 175; 62.28%), followed by AI for Green: AI-Driven Decarbonization & Green Transition (n = 59; 21.00%) and The Paradox, Rebound Effects & Governance (n = 47; 16.73%). The findings show progress in reducing model-level computational demands and applying AI to energy, industrial, mobility, and resource management systems. However, they also indicate that improvements in relative efficiency do not translate into reductions in absolute environmental impact, particularly when deployment scale, rebound effects, life-cycle emissions, and institutional conditions are taken into account. The paper argues that AI sustainability should be evaluated from a system-level perspective that integrates computational efficiency, environmental utility, deployment scale, and governance.