https://arxiv.org/pdf/2501.16548
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 socioeconomic 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.
Ever since Deepseek made a splash at the end of 2023, podcasters and analysts have been throwing around Jevons’ Paradox while trying to figure out if more efficient AI models would result in larger or smaller resource (power in particular) requirements.
So I was looking for a paper that would analyse resource requirements and Jevons’ paradox for AI. Unfortunately I chose poorly and this paper is mostly a description of various kinds of rebound effects and how they might apply to AI. No actual analysis. In fairness the author’s do a reasonable job laying out why it’s complicated and the lack of graphs and equations in the paper should have tipped me off.
Questions
Notes
- Jevons’ Paradox: proposed in the 19th century by economist William Stanley Jevons, who observed that as coal use became more efficient, it was also paradoxically leading to an increase, and not a decrease, in the consumption of coal across different industries (W Stanley Jevons. 1866. The coal question. In The Economics of Population. Routledge, 193–204.).
- effective climate action requires grappling with how these systems reshape markets, cultural norms, and policy priorities…as they hinge not on algorithmic design but human adaptation and use patterns.
- IEA, https://www.iea.org/energy-system/buildings/data-centres-and-data-transmission-networks
- electricity demand from data centers, driven heavily by AI training and inference, is currently at 2% (1-1.5% each for DCs and networks)
- data centre electricity use in Ireland has more than tripled since 2015, accounting for 18% of total electricity consumption in 2022
- Real world examples of Jevons’ paradox include increased consumption of energy and water when production efficiency went up and prices dropped, increased traffic congestion when road throughput is improved, and increased land usage when food production efficiency goes up.
- Economic rebound effects
- Jevons’ paradox is an example of a direct rebound effect
- Indirect rebound effects: improved efficiency of one product increases the usage of another product (also known as the real income effect), eg: , money saved from more fuel-efficient vehicles can be spent on air travel or consumer products.
- New AI features in consumer hardware devices can lead to an increased rate of device replacement which has embodied carbon costs.
- Economy-wide rebound effects: an innovation provokes far-reaching changes in the production and use of other goods, producing flow-on effects in the economy at large, eg: improvements in global energy efficiency and fuel use have enabled the development of new economies on a global scale, allowing the creation of new industries. These types of rebound effects can be particularly large for so-called “general-purpose technologies” which could include AI. Hyperscaler’s purchase of PPA’s could be considered a positive example of an economy-wide rebound effect from AI.
- Societal/Behavioural rebound effects
- Induction effects: the savings in resource consumption (e.g. energy or water) gained through improved efficiency are exceeded by the increased consumption and thereby production of either the materials or the final products. eg: mass-produced furniture has reduced the quantity of wood needed to produce, but, the amount of plastics, petrochemicals and waste that are engendered by the increased accessibility of these items and therefore their increased consumption are damaging to the environment. For AI, a prime example of this is the increase in targeted advertising.
- Time rebound effects: an innovation changes consumers’ use of their time, which then frees up (or removes) time for other activities that they carry out. Hard to say what the net outcome is for AI (eg: pollution saved via better maps routing vs increased consumption from time saved by AI automations)
- Indirect policy effects: indirect geopolitical and policy effects. Eg: project Stargate shows that high-efficiency datacenter designs or AI model efficiency optimisations, are insufficient when broader regulatory and policy conditions incentivise unchecked growth.