[Paper Review] From Efficiency Gains to Rebound Effects: The Problem of Jevons' Paradox in AI's Polarized Environmental Debate

Alexandra Sasha Luccioni, Emma Strubell+1 more|ArXiv.org|Jan 27, 2025

Energy, Environment, and Transportation Policies4 citations

TL;DR

The paper argues that AI’s environmental impact includes significant indirect rebound effects, so efficiency gains may spur higher overall consumption; it advocates interdisciplinary lifecycle and socio-economic analyses to address this.

ABSTRACT

As the climate crisis deepens, artificial intelligence (AI) has emerged as a contested force: some champion its potential to advance renewable energy, materials discovery, and large-scale emissions monitoring, while others underscore its growing carbon footprint, water consumption, and material resource demands. Much of this debate has concentrated on direct impacts -- energy and water usage in data centers, e-waste from frequent hardware upgrades -- without addressing the significant indirect effects. This paper examines how the problem of Jevons' Paradox applies to AI, whereby efficiency gains may paradoxically spur increased consumption. We argue that understanding these second-order impacts requires an interdisciplinary approach, combining lifecycle assessments with socio-economic analyses. Rebound effects undermine the assumption that improved technical efficiency alone will ensure net reductions in environmental harm. Instead, the trajectory of AI's impact also hinges on business incentives and market logics, governance and policymaking, and broader social and cultural norms. We contend that a narrow focus on direct emissions misrepresents AI's true climate footprint, limiting the scope for meaningful interventions. We conclude with recommendations that address rebound effects and challenge the market-driven imperatives fueling uncontrolled AI growth. By broadening the analysis to include both direct and indirect consequences, we aim to inform a more comprehensive, evidence-based dialogue on AI's role in the climate crisis.

Motivation & Objective

• Motivate a more nuanced AI climate debate by highlighting indirect and rebound effects beyond direct energy use.
• Synthesize insights from lifecycle assessment, economics, and social sciences to model AI’s full environmental footprint.
• Develop a taxonomy of indirect environmental impacts and rebound mechanisms relevant to AI technologies.
• Provide actionable recommendations to mitigate rebound effects and influence policy and governance.”

Proposed method

• Review existing literature on AI environmental impacts, including direct emissions and resource use.
• Adapt and extend a qualitative taxonomy of second-order ICT environmental effects for AI (Table 1 discussion).
• Conceptualize AI rebound effects through material, economic, and societal dimensions (Sections 3.1–3.3).
• Discuss interdisciplinary approaches combining lifecycle assessment with socio-economic analysis.
• Offer policy and governance recommendations to address rebound effects.”

Experimental results

Research questions

• RQ1 What are the indirect environmental impacts and rebound effects associated with AI beyond direct energy and resource use?
• RQ2 How do social, economic, and governance contexts shape AI’s climate footprint?
• RQ3 What methods can track and mitigate AI-induced rebound effects across material, economic, and societal domains?
• RQ4 What policy directions can reduce AI’s total environmental harm by addressing systemic rebound mechanisms?

Key findings

• Direct efficiency gains in AI do not guarantee reduced overall resource use due to rebound effects.
• AI’s indirect impacts span material, economic, and societal domains, influencing consumption patterns and infrastructure demands.
• Lifecycle and interdisciplinary analyses are needed to capture the full climate footprint of AI, including data centers, hardware, and usage patterns.
• Policy and governance strategies should address market incentives, business models, and social norms to counter rebound effects and curb AI growth if necessary.