The DOI matches its Crossref record; 15 of the 15 references it cites were found in a registry or an index; 17 of the 34 sources its claims cite were checked for support.
Sustainable AI: AI for sustainability and the sustainability of AI
Open accessAbstract
While there is a growing effort towards AI for Sustainability (e.g. towards the sustainable development goals) it is time to move beyond that and to address the sustainability of developing and using AI systems. In this paper I propose a definition of Sustainable AI; Sustainable AI is a movement to foster change in the entire lifecycle of AI products (i.e. idea generation, training, re-tuning, implementation, governance) towards greater ecological integrity and social justice. As such, Sustainable AI is focused on more than AI applications; rather, it addresses the whole sociotechnical system of AI. I have suggested here that Sustainable AI is not about how to sustain the development of AI per say but it is about how to develop AI that is compatible with sustaining environmental resources for current and future generations; economic models for societies; and societal values that are fundamental to a given society. I have articulated that the phrase Sustainable AI be understood as having two branches; AI for sustainability and sustainability of AI (e.g. reduction of carbon emissions and computing power). I propose that Sustainable AI take sustainable development at the core of its definition with three accompanying tensions between AI innovation and equitable resource distribution; inter and intra-generational justice; and, between environment, society, and economy. This paper is not meant to engage with each of the three pillars of sustainability (i.e. social, economic, environment), and as such the pillars of sustainable AI. Rather, this paper is meant to inspire the reader, the policy maker, the AI ethicist, the AI developer to connect with the environment—to remember that there are environmental costs to AI. Further, to direct funding towards sustainable methods of AI.
Overview
CoChat Summary
TL;DR
This opinion paper proposes a definition of Sustainable AI as a movement to change the entire AI lifecycle toward greater ecological integrity and social justice. It distinguishes AI for sustainability from the sustainability of AI, arguing that the environmental costs of developing and using AI must be addressed alongside AI’s beneficial applications.
Key findings
- Sustainable AI is defined as covering the entire lifecycle of AI products—idea generation, training, re-tuning, implementation, and governance—and is organized into two branches: AI for sustainability and the sustainability of AI.
- The paper argues Sustainable AI should adopt sustainable development at its core, with three tensions: AI innovation versus equitable resource distribution; inter- and intra-generational justice; and environment, society, and economy.
- It cites evidence that training a single deep learning NLP model can produce approximately 600,000 lb of CO₂ emissions, roughly equal to five cars over their lifetime, and that Google’s AlphaGo Zero generated 96 tonnes of CO₂ over 40 days of research training, equal to 1,000 h of air travel or the carbon footprint of 23 American homes.
- It notes that tuning an AI model is more expensive than training it initially, and that tools such as the Machine Learning Emissions Calculator, experiment-impact-tracker, and carbontracker exist to measure or predict carbon emissions and energy consumption.
Methods
- This is an opinion/conceptual paper, not an empirical study; it proposes and argues for a definition of Sustainable AI and reviews selected literature and examples.
- No dataset, sample, or formal analysis is reported; the argument draws on prior studies, including Strubell et al. on deep learning NLP energy costs and Anthony et al. on carbontracker.
- The paper recommends policy mechanisms such as carbon emissions reporting, sustainable AI taskforces, and a proportionality framework for assessing training/tuning costs against intended applications.
Limitations
- The paper states it is not meant to engage with each of the three pillars of sustainability—social, economic, and environmental—and is instead meant to inspire readers, policymakers, AI ethicists, and developers to connect with the environment.
- The paper states that only a few studies provide numbers on AI’s environmental costs and that more studies are needed to support or refute such findings.
- Evident from the text: as an opinion paper, it offers no original empirical measurement, systematic review, or comparative analysis, and its quantitative claims rely on cited secondary sources.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
15 Valid · 2 Not valid · 6 Not checked · 1 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“the high energy demands of these models are still a concern since (1) energy is not currently derived from carbon-neutral sources in many locations”
What the sources say
- Supportedcites [17] Strubell, E., Ganesh, A., McCallum, A.: Energy and Policy Considerations for Deep Learning in NLP. In ArXiv:1906.02243 (2019). source
The Introduction states that the models’ high energy demands remain a concern because energy is not derived from carbon-neutral sources in many locations. The copy prints “carbon-neural,” an apparent typo; the Methods section’s breakdown of energy sources reinforces the intended point. (LLM, conf=0.95)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“it also allows for the model training to be 'stopped, at the user's discretion, if the predicted environmental cost is exceeded' [3, 2].”
What the sources say
- Supportedcites [3] Anthony LFW, Kanding B, Selvan R.: Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models. ArXiv:2007.03051 (2020). source
The Introduction (p. 2) states that carbontracker supports predictions and that model training can be stopped at the user's discretion if the predicted environmental cost is exceeded. This directly supports the attributed claim. (LLM, conf=0.95)
- Not matchedcites [2] Angwin, J., Jeff L.: Machine Bias. Text/html. ProPublica. https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing (2016) Accessed date 23 May 2016.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Not only does the 'carbontracker' tool allow for the generation of carbon impact statements”
What the sources say
- Supportedcites [3] Anthony LFW, Kanding B, Selvan R.: Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models. ArXiv:2007.03051 (2020). source
In Experiments and Results (p. 2), the authors explicitly say they used carbontracker to generate a statement reporting electricity use, CO₂-equivalent emissions, and a car-travel equivalent. This supports the claim that the tool allows users to generate carbon impact statements. (LLM, conf=0.94)
- Not matchedcites [2] Angwin, J., Jeff L.: Machine Bias. Text/html. ProPublica. https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing (2016) Accessed date 23 May 2016.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“the process of training a single, deep learning, natural language processing (NLP) model (GPU) can lead to approx. 600,000 lb of carbon dioxide emissions [17].”
- Also stated in: § 4 Environmental sustainability of AI · p. 4 · plain
“training one large NLP model (aka a transformer), with neural architecture search, resulted in over 600,000 CO2e(lbs), roughly the equivalent of carbon emissions of five cars (over the lifetime of the car) [17].”
What the sources say
- Partly supportedcites [17] Strubell, E., Ganesh, A., McCallum, A.: Energy and Policy Considerations for Deep Learning in NLP. In ArXiv:1906.02243 (2019). sourceExperimental — accuracy not yet measured
Table 3 estimates 626,155 lb of CO₂ emissions for the GPU-based NAS entry, close to the claimed 600,000 lb. But §2.1 says its 274,120 GPU-hours represent a full architecture search, not training a single model once; the citing sentence omits that distinction. (LLM, conf=0.88)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 4 Environmental sustainability of AI · p. 4 · plain
Every source the paper cites for this claim was checked and supports it.
In the paper
“(2) when renewable energy is available, it is still limited to the equipment we have to produce and store it”
What the sources say
- Supportedcites [17] Strubell, E., Ganesh, A., McCallum, A.: Energy and Policy Considerations for Deep Learning in NLP. In ArXiv:1906.02243 (2019). source
The Introduction (§1) states the attributed point almost verbatim: even where renewable energy is available, its use is limited by the equipment available to produce and store it. (LLM, conf=0.95)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
2 more claims · 1 with their cited sources checked
Create a free account to see all 24 claims and to verify the 3 references still open.
- Probably not a substantial claim. 14
- It may state two separate claims as one. 6
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Despite a growing interest in, and deployment of AI-technologies in domains critical for sustainability, few have explored possible systemic risks in depth.”
From Artificial intelligence, systemic risks, and sustainability (2021) · cites this paper as [9]
Experimental — accuracy not yet measuredIn the citing paper
“In line with the willingness to include social and environmental considerations in corporate decisions, some of these principles focus on sustainability issues such as respecting human rights, including social purposes, and the significance of shared benefits and prosperity to achieve 'common good' through AI. 156”
From Artificial Intelligence and Sustainable Decisions (2023) · cites this paper as [156]
Citations
Cites
Pigou, A. C. (1933) · The Economic Journal · doi:10.2307/2224491 · Cited by 5,137
Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., Luetge, C., Madelin, R., Pagallo, U., Rossi, F., et al. (2018) · Minds and Machines · doi:10.1007/s11023-018-9482-5 · Cited by 4,774
Energy and Policy Considerations for Deep Learning in NLP
Strubell, E., Ganesh, A., & McCallum, A. (2019) · arXiv · doi:10.48550/arxiv.1906.02243 · Cited by 3,735
Energy and Policy Considerations for Deep Learning in NLP
Strubell, E., Ganesh, A., & McCallum, A. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1355 · Cited by 3,684 · cited 9×
“[17], it is argued that there are both financial and environmental costs to Deep Learning (DL) models for natural language processing (NLP).”
BackgroundInfluential(2018) · Cited by 3,510
Mensah, J. (2019) · Cogent Social Sciences · doi:10.1080/23311886.2019.1653531 · Cited by 1,876
Emotional expressions reconsidered: Challenges to inferring emotion from human facial movements
Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019) · Psychological Science in the Public Interest · doi:10.1177/1529100619832930 · Cited by 1,748
The ethics of artificial intelligence
Bostrom, N., & Yudkowsky, E. (2014) · The Cambridge Handbook of Artificial Intelligence · doi:10.1017/cbo9781139046855.020 · Cited by 918 · cited 2×
“Further, sustainable development is a “development paradigm as well as the concept that calls for improving living standards without jeopardising the earth’s ecosystems or causing environmental challenges such as deforestation and water and air pollution” [13, 6].”
BackgroundEconomic, social, and environmental sustainability in development theory and urban planning practice
Basiago, A. D. (1998) · Environmentalist · doi:10.1023/a:1006697118620 · Cited by 521
Quantifying the Carbon Emissions of Machine Learning
Lacoste, A., Luccioni, A., Schmidt, V., & Dandres, T. (2019) · arXiv · doi:10.48550/arxiv.1910.09700 · Cited by 408 · cited 2×
“With tools like the ‘machine learning emissions calculator’ [12] and the ‘experiment-impact-tracker’ [10] this should no longer be the case.”
Method
Cited by
Ooi, K.-B., Tan, G. W.-H., Al-Emran, M., Al-Sharafi, M. A., Capatina, A., Chakraborty, A., Dwivedi, Y. K., Huang, T.-L., Kar, A. K., Lee, V.-H., et al. (2025) · Journal of Computer Information Systems · doi:10.1080/08874417.2023.2261010 · Cited by 746
“141 On the other hand, managing IT resources in a way that balances environmental, social, and economic factors is referred to as sustainable IT.”
Artificial intelligence, systemic risks, and sustainability
Galaz, V., Centeno, M. A., Callahan, P. W., Causevic, A., Patterson, T., Brass, I., Baum, S., Farber, D., Fischer, J., Garcia, D., et al. (2021) · Technology in Society · doi:10.1016/j.techsoc.2021.101741 · Cited by 497
“[7,9, 10], yet their potential allocative harms [11] and unexpected social and ecological effects [12] are poorly elaborated, and more often than not, overlooked.”
BackgroundShaping the future of healthcare: ethical clinical challenges and pathways to trustworthy AI
Goktas, P., & Grzybowski, A. (2025) · Journal of Clinical Medicine · doi:10.3390/jcm14051605 · Cited by 336
“Optimizing algorithms for energy efficiency , reducing computational complexity while maintaining performance.”
Pesaresi, M., Schiavina, M., Politis, P., Freire, S., Krasnodębska, K., Uhl, J. H., Carioli, A., Corbane, C., Dijkstra, L., Florio, P., et al. (2024) · International Journal of Digital Earth · doi:10.1080/17538947.2024.2390454 · Cited by 292
Abou-Foul, M., Ruiz-Alba, J. L., & López-Tenorio, P. J. (2023) · Journal of Business Research · doi:10.1016/j.jbusres.2022.113609 · Cited by 253
Metadata
- Authors
- van Wynsberghe, A.
- Venue
- AI and Ethics
- Published
- 2021-02-26
- Volume
- 1
- Issue
- 3
- Pages
- 213-218
- ISSN
- 2730-5953 (ISSN-L)
- 2730-5961
- OpenAlex
- W3135539146
- Cited by
- 939
- References
- 13
- Citations by year
- 2026: 230 · 2025: 295 · 2024: 210 · 2023: 122 · 2022: 64 · 2021: 18
- FWCI
- 129.35
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- AI and Ethics (publishedVersion, cc-by)
- Funders
- Alexander von Humboldt-Stiftung
- Rheinische Friedrich-Wilhelms-Universität Bonn
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Environmental Science › Management, Monitoring, Policy and Law › Innovation, Sustainability, Human-Machine Systems
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Green IT and Sustainability
- Keywords
- sustainable AI, AI for sustainability, AI sustainability, environmental costs of AI, CO2 emissions, Sustainable Development Goals, social justice, intergenerational justice, intragenerational justice, AI lifecycle, AI governance, resource allocation, environmental ethics
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Sustainable AI: AI for sustainability and the sustainability of AI. https://app.cochat.ai/papers/10.1007/s43681-021-00043-6
- The paper
- van Wynsberghe, A. (2021). Sustainable AI: AI for sustainability and the sustainability of AI. AI and Ethics. https://doi.org/10.1007/s43681-021-00043-6
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.