Source-linked AI summary
How can AI accelerate the green transition?
Jacques Sainte-Marie, Marie-Fleur Simmet, Fanny Terrier
TL;DR
Climate action requires structural change and long-term investment, while AI and digital systems also carry environmental costs. This paper examines how they may contribute to the ecological transition, reporting modest net carbon gains in initial digital use cases alongside resource-impact shifts and rising data-centre electricity demand.
Problem
Climate transition is constrained by infrastructure inertia and large investments amid shrinking budgetary margins, while the costs of inaction are rising.
Source: Infrastructure inertia and the… · p. 2Method
The document examines barriers to ecological transition and AI’s environmental effects while considering how digital technology and AI can contribute to these transformations.
Source: Infrastructure inertia and the… · p. 2Results
Initial digital decarbonization use cases yield modest net carbon gains versus sector footprints, sometimes shifting impacts to abiotic-resource depletion; data centres used 1.5% of global electricity in 2025, potentially 3% by 2030.
Source: A growing environmental footprint,… · p. 3Takeaways & Limitations
Managing environmental impacts includes training, educational content and labels, while avoiding unnecessary generation when information is already available.
Source: Specialisation and restraint in usage… · p. 6Assessing digital solutions’ net environmental impacts remains difficult, particularly because of potential negative externalities.
Source: A growing environmental footprint,… · p. 3
Abstract
from arXiv · showhide
In his essay The Gift, published in 1925, the sociologist Marcel Mauss emphasised the 'total' scope of the phenomenon of gift-and counter-gift-giving, noting that it 'expresses, simultaneously and all at once, all manner of institutions: religious, legal and moral (...)\,; economic (...)\,; not to mention the aesthetic phenomena to which these facts give rise (...).'' The emergence of artificial intelligence (AI), which is suddenly and profoundly disrupting individual and collective lives, bears a resemblance to the ''total social fact'' described a hundred years ago. AI refers to technologies designed to simulate human intelligence, namely an ''automated system that (...) deduces, from received inputs, how to generate output results such as predictions, content, recommendations or decisions that may influence physical or virtual environments.'' 4 (OECD). Having become central to human relations, AI relies on physical and globalised infrastructure and therefore, also has an impact on the environment. This paper analyses some of these effects and how they are perceived by society. This text offers an exploration of the links between AI and the ecological transition, seeking to identify new opportunities and constraints. To this end, it will identify the barriers to the ecological transition (1) and the environmental impacts attributable to AI (2). It will then go on to describe how to live in a constrained world (3) while AI accelerates scientific discovery (4) and profoundly transforms society (5).
1. Some barriers to the ecological transition
The ecological transition is constrained by the limited reach of individual action, the need for structural change, and competing economic and time pressures. Long-lived infrastructure and rising climate costs make it difficult to balance immediate adaptation with long-term investment.
The limits of individual and collective action in the face of climate change
- Individual and small-group action has limited reach, while reaching carbon neutrality requires reducing a French person’s footprint from around 8 to 2 tonnes of CO₂ equivalent per year and structural change.Environmental harms are also unpriced, while distant climate impacts can make public awareness and good intentions short-lived.Source: The limits of individual and… · pp. 1–2
Infrastructure inertia and the budgetary dilemma of adaptation
- Infrastructure timelines create inertia: France’s 2050 carbon-neutrality target is medium-term for people but short-term for transport, housing, industry, and energy infrastructure.The average tractor is 28 years old, and nuclear power stations operate for around 50 years, requiring long-term planning alongside shorter-term constraints.Source: Infrastructure inertia and the… · p. 2
- Immediate climate responses risk displacing long-term measures, such as planting forest species suited to future conditions that take decades to have significant effects.The passage contrasts such measures with shorter-term responses including disaster compensation, geoengineering, and climate interventions.Source: Infrastructure inertia and the… · p. 2
- The cost of climate-related events is rising as France’s budgetary margins shrink, leaving a dilemma between the cost of inaction and considerable investment across many sectors.The required investments span mobility, housing, industry, energy, agriculture, and food.Source: Infrastructure inertia and the… · p. 2
- The document explores how digital technology and AI may contribute to ecological transition while also considering their negative aspects.Source: Infrastructure inertia and the… · p. 2
A growing environmental footprint, whose scale remains uncertain
Digital technology and AI have environmental costs, while digital applications may also yield environmental gains. Their scale and net effects remain uncertain, and efficiency improvements can coexist with growth in model size.
A growing environmental footprint, whose scale remains uncertain
- 4.4% of France’s carbon footprint came from digital technology in ADEME’s 2022 estimate, compared with 30% for transport and 20% for food.Source: A growing environmental footprint,… · p. 3
- Initial digital use cases show modest net carbon gains relative to the sectors involved, sometimes accompanied by shifts in impact, particularly greater depletion of abiotic resources.ADEME cautions that these results cannot be generalised and says potential negative effects should be anticipated before implementation.Source: A growing environmental footprint,… · p. 3
- Data centres accounted for 1.5% of global electricity demand in 2025, a share that could rise to 3% by 2030; AI also increases demand for water, electricity, and mineral resources.Source: A growing environmental footprint,… · p. 3
- AI is expected to speed energy-sector innovation through improved grid management and maintenance and more accurate forecasts of generation and consumption.The IEA notes that indirect effects could offset gains and adoption may be limited by data access, infrastructure, or societal barriers.Source: A growing environmental footprint,… · p. 3
- Performance gains in AI hardware and software can reduce electricity consumption while enabling increasingly large models.Source: A growing environmental footprint,… · p. 3
- AI’s environmental impacts can be difficult to separate from digital technology’s because AI is embedded in many devices and tools; focusing on a few language models gives a narrow view.Source: A growing environmental footprint,… · pp. 3–4
The need for transparency and environmental indicators
Assessing AI’s environmental performance is difficult when impacts are incompletely disclosed and resource use varies across tasks and systems. The paper points to lifecycle transparency, quantitative indicators, and voluntary reporting as ways to inform responsible use.
The need for transparency and environmental indicators
- Energy use can differ by several orders of magnitude between a simple text query and an image-generation query, while system-specific water-use comparisons cannot easily be generalised.Rarely disclosed information includes cooling methods, training frequency, processor lifespan, and processor counts.Source: The need for transparency and… · p. 4
- Lifecycle transparency should quantify water, minerals, and energy from manufacture through end-of-life, with algorithmic-complexity information for queries or development stages.Source: The need for transparency and… · p. 4
- Environmental labels and indicators could encourage good practices and voluntary collaboration, including through the Coalition for Sustainable AI’s measurement and reporting tools.Source: The need for transparency and… · p. 4
The increasing scarcity of resources as a driver of a chosen frugality, for the benefit of society
Resource constraints may make efficiency an asset for competitiveness and sovereignty. The section argues for frugality across digital technologies and an interdisciplinary view of AI’s societal effects.
The increasing scarcity of resources as a driver of a chosen frugality, for the benefit of society
- Demand for some metals is rising while supply is limited, and rare-earth permanent magnets are essential to digital and energy-transition technologies.In 2024, China refined 91% of rare earths used in permanent magnets and produced 94% of these magnets, giving it leverage over other countries’ technological development.Source: The increasing scarcity of resources… · p. 5
- Resource efficiency may become both a constraint and an asset for competitiveness and strengthening sovereignty.The paper presents an eco-responsible digital sector as a way to combine economic and environmental performance.Source: The increasing scarcity of resources… · p. 5
- Low-tech means digital technology that is appropriate, frugal, and resilient; sustainable digital systems require low-tech components from devices to software, networks, and sensors.Many everyday computer and phone tasks do not require large data volumes, high processing speeds, or sophisticated features.Source: The increasing scarcity of resources… · pp. 5–6
- AI’s development trajectory depends not only on research and technology but also on adoption and use, making interdisciplinary analysis relevant to public policy and rights.Source: The increasing scarcity of resources… · p. 6
Specialisation and restraint in usage as the keys to frugal AI
The section presents specialisation, frugal design and more deliberate use as ways to reduce AI’s environmental impact. It also frames AI as an opportunity for science and the ecological transition.
Specialisation and restraint in usage as the keys to frugal AI
- Specialised AI models can be smaller and less environmentally impactful, easier to build and manage, and supportive of innovation and legal security.The passage contrasts domain-specific systems with an oligopoly of large models and notes that limited, quality-assured data can provide security.Source: Specialisation and restraint in usage… · p. 6
- Developing frugal AI is presented as a promising strategy requiring alternative approaches and doing more with less.Source: Specialisation and restraint in usage… · p. 6
- Training, educational content and labels can inform users’ choices, while existing information need not be regenerated by AI.The passage gives recipes and poems as examples of information already available online.Source: Specialisation and restraint in usage… · p. 6
- AI also offers scientific and knowledge-related opportunities that can support the ecological transition.Source: Specialisation and restraint in usage… · p. 7
A new way of interacting with data to drive scientific progress
The section describes AI as a way to extract meaningful information from abundant observations and advance scientific understanding. It can complement traditional models and offer new ways to conceptualise difficult-to-model phenomena.
A new way of interacting with data to drive scientific progress
- AI can advance science by offering new ways to conceptualise studied objects, beyond building on existing work.Source: A new way of interacting with data to… · p. 7
- Traditional models have grown increasingly complex and difficult to handle, parametrise and study, while validation and information extraction remain challenging.Source: A new way of interacting with data to… · p. 7
- Abundant, varied digital measurements let AI and data science extract meaningful information that characterises phenomena and predicts their evolution.The passage notes that data-driven forecasting models in meteorology produce high-quality results.Source: A new way of interacting with data to… · p. 7
- Data-driven models can replace traditional approaches or combine with mechanistic models, which provide low-frequency trends while data-driven models capture high-frequency components.Source: A new way of interacting with data to… · p. 7
AI can be a tool for the ecological transition
The section identifies fields where AI can support ecological transition efforts, particularly through optimisation, forecasting, monitoring and decision support. These opportunities are presented in domains involving complex phenomena that are difficult to model.
AI can be a tool for the ecological transition
- The passage frames these application fields as suitable for AI because their phenomena involve many parameters, complex modelling, variable behaviour or living organisms.Source: AI can be a tool for the ecological… · p. 8
- Renewable-energy applications include production optimisation, demand forecasting, grid balancing and predictive maintenance.Source: AI can be a tool for the ecological… · p. 8
- In agriculture, AI can support decisions, help develop genetic diversity for climate adaptation and enable robots to perform arduous tasks.Source: AI can be a tool for the ecological… · p. 8
- Language models applied to consultancy and expertise can analyse text and data corpora, identify trends and answer complex questions.Source: AI can be a tool for the ecological… · p. 8
- Home automation can support smart energy management, anomaly detection and monitoring of air quality, humidity and temperature.Source: AI can be a tool for the ecological… · p. 8
- For biodiversity, AI can monitor ecosystem functioning and interactions, predict biodiversity changes and develop indicators.Source: AI can be a tool for the ecological… · p. 8
5. The transformations brought about by AI
AI reshapes how knowledge is shared, decisions are made, and intellectual work is produced, creating opportunities to influence societal choices. The paper argues that ecological objectives should guide AI design, while noting that its proposals are not quantitatively assessed.
5. The transformations brought about by AI
- Digital tools can influence habits, behaviours and values, while disruptive innovation can replace established ways of doing things.The paper presents this influence as a possible lever for environmental protection.Source: §5 · p. 8, A driver of knowledge dissemination… · p. 9
A driver of knowledge dissemination and access
- Scientific literature often remains confined to specialists because it is technically complex, fragmented, costly to access, and difficult to grasp at scale.The paper reports around 2.8 million articles published in 2022.Source: A driver of knowledge dissemination… · pp. 8–9
- AI and chatbots can tailor and synthesise scientific knowledge and let users refine answers through natural-language dialogue.The paper describes knowledge dissemination and interaction as a domain profoundly changed by AI.Source: A driver of knowledge dissemination… · pp. 8–9
- Conversational agents have become a primary gateway to knowledge, although their development raises questions about bias, hallucinations, intellectual property and liability.The paper says answer quality heralds a new era in interaction with knowledge.Source: A driver of knowledge dissemination… · p. 9
Automation as a driver of transformation in decision-making
- AI’s statistical, extrapolative reasoning may favour short-term decisions over long-term planning, while autonomous agents make oversight and decision reconstruction harder.The paper calls for ecological-transition objectives to be incorporated into AI systems at the design stage.Source: Automation as a driver of… · p. 9
Changes to the value chain of work and reasoning
- AI can produce verified results, including mathematical proofs, and may disrupt the value chain by industrialising tasks involving analysis, synthesis, data collection and consultancy.The paper says AI had produced numerous proofs of problems open for decades by 2026.Source: In the realm of intellectual… · p. 10
- The shift toward abundant, easier-to-produce results may reduce the value assigned to each, while re-humanising some professions could give greater value to environmental-protection work.The paper frames re-humanisation of professions and skilled trades as an appropriate response to continuing labour-market changes.Source: In the realm of intellectual… · p. 10
- The paper calls for shaping AI’s next societal cycle toward a desired trajectory, but offers no specific formula or quantitative assessment.It describes its contribution as an analysis of possible developments and directions for promoting the ecological transition.Source: In the realm of intellectual… · pp. 10–11