The carbon emissions of writing and illustrating are lower for AI than for humans
Open accessPublished version of a preprint
Abstract
As AI systems proliferate, their greenhouse gas emissions are an increasingly important concern for human societies. In this article, we present a comparative analysis of the carbon emissions associated with AI systems (ChatGPT, BLOOM, DALL-E2, Midjourney) and human individuals performing equivalent writing and illustrating tasks. Our findings reveal that AI systems emit between 130 and 1500 times less CO2e per page of text generated compared to human writers, while AI illustration systems emit between 310 and 2900 times less CO2e per image than their human counterparts. Emissions analyses do not account for social impacts such as professional displacement, legality, and rebound effects. In addition, AI is not a substitute for all human tasks. Nevertheless, at present, the use of AI holds the potential to carry out several major activities at much lower emission levels than can humans.
AI snapshot
Written from the paper’s full text.
TL;DR
This paper compares the carbon emissions of AI systems and humans performing equivalent writing and illustration tasks, finding that AI emits far less CO2e per page or image under the study’s assumptions. It argues that concerns about AI’s emissions should be tempered by the much larger human footprint for these tasks, while noting social, legal, and rebound-effect caveats.
Key findings
- AI writing emits 130–1500 times less CO2e per page of text than human writers: BLOOM is 1400× less impactful than a US resident and 180× less than an India resident; ChatGPT is 1100× less than a US resident and 130× less than an India resident.
- AI illustration emits 310–2900 times less CO2e per image than human illustrators: DALL-E2 is ~2500× less than a US-based artist and ~310× less than an India-based artist; Midjourney is ~2900× less than a US artist and ~370× less than an India-based artist.
- Per-query/per-image AI emissions are on the order of grams: ChatGPT ~2.2 g CO2e per query, BLOOM ~1.6 g per query, DALL-E2 ~2.2 g per query, and Midjourney ~1.9 g CO2e per image.
- Human comparison values are much larger: a US resident writing a 250-word page is ~1400 g CO2e, an India resident ~180 g; a US-based illustrator is ~5500 g CO2e per image, an India-based illustrator ~690 g.
Methods
- The study conducted numerical analyses following life cycle assessment stages: goal and scope, inventory analysis, impact assessment, and interpretation.
- It compared four AI systems—ChatGPT, BLOOM, DALL-E2, and Midjourney—against human writing and illustration, including AI hardware/energy use but excluding software development and personnel, and using human annual carbon footprints apportioned by time spent on the task.
- Data came from previously published sources including peer-reviewed articles, government reports, and databases; the authors also used ChatGPT for drafting/editing but state that data analysis, calculations, and conclusions were their own.
Limitations
- The paper states results are not generalizable to all contexts, AI is not a substitute for all human tasks, and some tasks remain more efficient for humans (e.g., an in-depth, heavily referenced original article is beyond current AI; drawing a stick figure may be faster for a human).
- The paper states it does not account for social impacts such as professional displacement, legality of training materials, and rebound effects, and notes future technology and societal changes will alter the AI/human emissions ratio.
- Evident from the text: several estimates rely on assumptions or informal sources, including Mark Twain’s 300 words/hour as average writing speed, a $200 project cost and $62.50/hour illustrator rate to infer 3.2 hours per illustration, Midjourney CEO comments about petaoperations, an informal online ChatGPT estimate, and the EPA calculator’s simplification of CO2e conversion.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
0 of 30 cited sources checked · 0 supported · 0 flagged
- Findingown data
AI systems (ChatGPT, BLOOM) emit between 130 and 1500 times less CO2e per page of text generated than human writers.
“AI systems emit between 130 and 1500 times less CO2e per page of text generated compared to human writers”
- Also stated in: § Writing: AI vs. human · plain
“BLOOM is 1400 times less impactful, per page of text produced, than a US resident writing, and 180 times less impactful than a resident of India writing. ChatGPT is 1100 times less impactful than a US resident writing, and 130 times less impactful than a resident of India writing.”
- Also stated in: § Writing: AI vs. human · plain
“AI writing (via BLOOM or ChatGPT) produces 130–1500 times less CO2e per page than a human author.”
Quote in the paperSubstantial claim - Also stated in: § Writing: AI vs. human · plain
- Findingreasoning
The combined impact of training and operation for ChatGPT is approximately 2.2 g CO2e per query.
“Consequently, the combined impact of training and operation for ChatGPT amounts to approximately 2.2 g CO2e per query.”
Quote in the paperSubstantial claim- The authors state this as an assumption: ChatGPT undergoes a full re-training of the model once per month and continues with an estimated 10,000,000 queries per day.
- Findingreasoning
Assuming a person's emissions while writing are consistent with their overall annual impact, the carbon footprint for a US resident producing a page of text (250 words) is approximately 1400 g CO2e.
“Assuming that a person’s emissions while writing are consistent with their overall annual impact, we estimate that the carbon footprint for a US resident producing a page of text (250 words) is approximately 1400 g CO2e.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
- The authors state this as an assumption: A person's emissions while writing are consistent with their overall annual impact.
- Findingown data
AI illustration systems (DALL-E2, Midjourney) emit between 310 and 2900 times less CO2e per image than human illustrators.
“AI illustration systems emit between 310 and 2900 times less CO2e per image than their human counterparts.”
Quote in the paperSubstantial claim - Findingreasoning
The impact of an AI query, encompassing both amortized training and the query itself, is on the order of a few grams CO2e.
“These figures illustrate that the impact of an AI query, encompassing both amortized training and the query itself, is on the order of a few grams CO2e.”
Quote in the paperSubstantial claim
- Findingreasoning
For BLOOM, assuming similar usage and retraining frequency as ChatGPT, the per-query impact of training is 0.10 g CO2e.
“For BLOOM, assuming a similar level of usage and frequency of retraining as for ChatGPT, the per-query impact of training is 0.10 g CO2e”
Quote in the paperSubstantial claim- The authors state this as an assumption: BLOOM has a similar level of usage and frequency of retraining as ChatGPT.
- Findingreasoning
Assuming monthly full retraining and 10,000,000 queries per day, the amortized training cost for ChatGPT is 1.84 g CO2e per query.
“Assuming that ChatGPT undergoes a full re-training of the model once per month and continues with an estimated 10,000,000 queries per day, the 552 metric tons divided by 300,000,000 queries equates to 1.84 g CO2e per query for the amortized training cost.”
Quote in the paperSubstantial claim- The authors state this as an assumption: ChatGPT undergoes a full re-training of the model once per month and continues with an estimated 10,000,000 queries per day.
- Findingown data
Embodied energy of chips, server, and end-of-life recycling are 2-5 orders of magnitude smaller than training and per-query emissions.
“We also calculated the embodied energy of the chips used to train the model and of the server used to deliver the query, and the end-of-life recycling of those devices, but all of these factors were 2–5 orders of magnitude smaller than the training and per-query emissions.”
Quote in the paperNot a substantial claim - Findingreasoning
For BLOOM, the per-query operational cost is 1.47 g CO2e.
“while the per-query operational cost is 1.47 g.”
Quote in the paperSubstantial claim- The authors state this as an assumption: BLOOM has a similar level of usage and frequency of retraining as ChatGPT.
- Findingreasoning
The total emission for BLOOM is 1.6 g CO2e per query.
“This results in a total emission of 1.6 g per query for BLOOM.”
Quote in the paperSubstantial claim- The authors state this as an assumption: BLOOM has a similar level of usage and frequency of retraining as ChatGPT.
- Findingown data
BLOOM is 1400 times less impactful per page of text produced than a US resident writing.
“BLOOM is 1400 times less impactful, per page of text produced, than a US resident writing”
Quote in the paperSubstantial claim - Findingreasoning
The carbon footprint of a desktop computer operating for the duration of a human illustrator creating an image (3.2 h) is 280 g CO2e.
“The footprint of that duration for a desktop computer is 280 g CO2e.”
Quote in the paperSubstantial claim - Findingnone given
Midjourney emits approximately 2900 times less CO2e than a US-based artist.
“Midjourney emits approximately 2900 times less CO2e than a US artist”
Quote in the paperSubstantial claim - Findingreasoning
The carbon footprint of a laptop operating for the duration of a human illustrator creating an image (3.2 h) is 100 g CO2e.
“The carbon footprint for a laptop operating for the duration of a human illustrator creating an image (3.2 h) is 100 g CO2e.”
Quote in the paperSubstantial claim - Findingreasoning
In the largest-emissions scenario, generating an image requires the AI to run on an A100 GPU for 40 s.
“therefore, the AI would need to run on that device for 40 s.”
Quote in the paperSubstantial claim - Findingown data
BLOOM is 180 times less impactful per page of text produced than a resident of India writing.
“and 180 times less impactful than a resident of India writing.”
Quote in the paperSubstantial claim - Findingcitation
The carbon footprint for a resident of India illustrator is approximately 690 g CO2e per image.
“For a resident of India, by comparison, the impact would be approximately 690 g CO2e per image22.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
- Findingown data
DALL-E2 emits approximately 310 times less CO2e than an India-based artist.
“approximately 310 times less than an India-based artist”
Quote in the paperSubstantial claim - Findingown data
ChatGPT is 130 times less impactful per page of text produced than a resident of India writing.
“and 130 times less impactful than a resident of India writing.”
Quote in the paperSubstantial claim - Findingreasoning
AI writing (via BLOOM or ChatGPT) produces substantially less CO2e than the computer usage to support humans doing that writing.
“AI also produces substantially less CO2e than the computer usage to support humans doing that writing.”
Quote in the paperSubstantial claim - Findingown data
Laptop and desktop usage supporting a human drawing an image would themselves be many times more impactful in CO2e than the AI image systems.
“both laptop and desktop usage while supporting a human drawing an image would themselves be many times more impactful than the AI systems as well.”
- Also stated in: § Illustration: AI vs. human · plain
“AI produces many times less CO2e than computer usage to support humans making images.”
Quote in the paperSubstantial claim - Also stated in: § Illustration: AI vs. human · plain
- Findingown data
Midjourney emits approximately 370 times less CO2e than an India-based artist.
“370 times less than one based in India”
Quote in the paperSubstantial claim - Findingreasoning
AI image creation (via DALL-E2 or Midjourney) produces 310-2900 times less CO2e per image than human creators.
“AI image creation (via DALL-E2 or Midjourney) produces 310–2900 times less CO2e per image than human creators.”
Quote in the paperSubstantial claim - Findingown data
DALL-E2 emits approximately 2500 times less CO2e than a US-based artist.
“Figure 2 shows that DALL-E2 emits approximately 2500 times less CO2e than a US-based artist”
Quote in the paperSubstantial claim - Findingreasoning
The carbon footprint for a US-based illustrator is approximately 5500 g CO2e per image.
“we calculate that the carbon footprint for a US-based illustrator is approximately 5500 g CO2e per image.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
- Findingown data
ChatGPT is 1100 times less impactful per page of text produced than a US resident writing.
“ChatGPT is 1100 times less impactful than a US resident writing”
Quote in the paperSubstantial claim - Premisecitation
A deployment of BLOOM generated 1.5 g CO2e per query (340 kg CO2e divided by 230,768 queries).
“In comparison, a deployment of BLOOM generated 1.5 g per query (340 kg CO2e divided by 230,768 queries)12.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [12] Luccioni, A. S., Viguier, S. & Ligozat, A.-L. Estimating the carbon footprint of BLOOM, a 176B parameter language model. https://doi.org/10.48550/ARXIV.2211.02001 (2022). source
Not checked in the free run.
- Premisecitation
Training BLOOM requires 433 MWh of energy.
“Training BLOOM requires 433 MWh of energy”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [12] Luccioni, A. S., Viguier, S. & Ligozat, A.-L. Estimating the carbon footprint of BLOOM, a 176B parameter language model. https://doi.org/10.48550/ARXIV.2211.02001 (2022). source
Not checked in the free run.
- Premisecitation
Training GPT-3 requires 1287 MWh of energy.
“training GPT-3 requires1287 MWh12”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [12] Luccioni, A. S., Viguier, S. & Ligozat, A.-L. Estimating the carbon footprint of BLOOM, a 176B parameter language model. https://doi.org/10.48550/ARXIV.2211.02001 (2022). source
Not checked in the free run.
- Premisecitation
Training BLOOM produces 50.5 metric tons of CO2e.
“Training BLOOM, a model slightly larger and substantially more energy-efficient than GPT-3, produces 50.5 metric tons of CO2e12.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [12] Luccioni, A. S., Viguier, S. & Ligozat, A.-L. Estimating the carbon footprint of BLOOM, a 176B parameter language model. https://doi.org/10.48550/ARXIV.2211.02001 (2022). source
Not checked in the free run.
- Interpretationown data
At present, the use of AI holds the potential to carry out several major activities at much lower emission levels than can humans.
“Nevertheless, at present, the use of AI holds the potential to carry out several major activities at much lower emission levels than can humans.”
Quote in the paperSubstantial claim - Premisecitation
Mark Twain's output of roughly 300 words per hour is representative of the average writing speed among authors.
“An article in The Writer magazine states that Mark Twain’s output, which was roughly 300 words per hour, is representative of the average writing speed among authors21.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [21] DiGravio, W. How many words should one write per day? (2018). https://www.writermag.com/writing-inspiration/the-writing-life/many-words-one-write-per-day/. Accessed 14 March 2023.
Not checked in the free run.
- Premisecitation
Training GPT-3 produces approximately 552 metric tons CO2e.
“training GPT-3 (the system on which the popular ChatGPT chatbot is based17) produces approximately 552 metric tons CO2e11.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [11] Patterson, D. et al. The carbon footprint of machine learning training will plateau, then shrink. Computer 55, 18–28. https://doi.org/10.1109/MC.2022.3148714 (2022). source
Not checked in the free run.
- Interpretationreasoning
The environmental footprint of AI completing the two tasks studied (writing and illustrating) is substantially lower than that of humans completing those same tasks.
“The findings above demonstrate that the environmental footprint of AI completing two major tasks is substantially lower than that of humans completing those same tasks.”
- Also stated in: § Conclusion · plain
“We found that, for these two activities, at least, AI has a substantially lower carbon footprint than humans engaged in the same task.”
- Also stated in: § Human/AI collaboration · plain
“While the carbon footprint of AI is nontrivial, the footprint of humans doing the same work is far greater, and should not be discounted in the assessment of AI.”
Quote in the paperSubstantial claim- Rests on an assumption: The AI and human tasks being compared are equivalent enough in nature that their carbon footprints can be meaningfully compared.
- Also stated in: § Conclusion · plain
- Interpretationreasoning
DALL-E2's footprint is similar to the footprint of ChatGPT: 2.2 g CO2e per query.
“Given their shared reliance on GPT-3, we estimate that DALL-E2’s footprint is similar to the footprint of ChatGPT calculated above: 2.2 g CO2e per query.”
Quote in the paperSubstantial claim - Premisecitation
In the largest-emissions scenario, generating an image requires 50 petaoperations.
“In the largest-emissions scenario (from Holz’s comments), generating an image requires 50 petaoperations”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [25] Claburn, T. David Holz, founder of AI art generator Midjourney, on the future of imaging (2022). https://www.theregister.com/2022/08/01/david_holz_midjourney/. Accessed 14 March 2023.
Not checked in the free run.
- Premisecitation
A laptop computer with average power consumption of 75 W produces 27 g CO2e during the writing period.
“Assuming an average power consumption of 75 W for a typical laptop computer23, the device produces 27 g of CO2e24 during the writing period.”
Quote in the paperSubstantial claim0 of 2 sources checked- cites [23] Sibelga. How much power does a computer use and how much CO2 does that represent? (2023). https://www.energuide.be/en/questions-answers/how-much-power-does-a-computer-use-and-how-much-co2-does-that-represent/54/. Accessed 14 March 2023.
Not checked in the free run.
- cites [24] U.S. Environmental Protection Agency. Greenhouse gas equivalencies calculator (2022). https://www.epa.gov/energy/greenhouse-gas-equivalencies-calculator#results. Accessed 14 March 2023.
Not checked in the free run.
- The authors state this as an assumption: The EPA calculator captures the relationship adequately.
- Interpretationcitation
3.2 hours per illustration is a viable estimate for a professional illustrator producing a commercial piece of work based on a provided specification, across a wide range of styles and formats.
“Based on these figures, we propose that 3.2 hours per illustration is a viable estimate for a professional illustrator producing a commercial piece of work based on a provided specification, across a wide range of styles and formats.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [28] Thumbtack. How much do illustrators charge? (2020). https://www.thumbtack.com/p/illustration-rates. Accessed 14 March 2023.
Not checked in the free run.
- Premisecitation
Nvidia A100 GPUs can process 1.25 petaoperations per second while using 400 W of electricity.
“These GPUs can process 1.25 petaoperations per second while using 400 W of electricity27.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [27] NVIDIA Corporation. NVIDIA A100 GPU (2023). https://www.nvidia.com/en-us/data-center/a100/. Accessed 14 March 2023.
Not checked in the free run.
- Interpretationown data
Assuming the quality of writing produced by AI is sufficient for whatever task may be at hand, AI produces less CO2e per page than a human author.
“Assuming the quality of writing produced by AI is sufficient for whatever task may be at hand, AI produces less CO2e per page than a human author”
Quote in the paperSubstantial claim- The authors state this as an assumption: The quality of writing produced by AI is sufficient for whatever task may be at hand.
- Interpretationown data
Overall, the use of AI can significantly reduce the carbon footprint of certain tasks when compared to equivalent human activity.
“the data presented in this paper suggest that, overall, the use of AI can significantly reduce the carbon footprint of certain tasks when compared to equivalent human activity.”
Quote in the paperSubstantial claim- Rests on an assumption: The AI and human tasks being compared are equivalent enough in nature that their carbon footprints can be meaningfully compared.
- May combine two claims. 11
- The quote only partly states this claim. 1
- Classified as probably not a substantial claim. 20
May combine two claims. The quote only partly states this claim.
FindingcitationApproximating a GPU lasts 1.5 years before becoming obsolete, the amortized training cost per query for GPT-3 is 0.0001 g CO2e.
“and for GPT3 is 0.0001 g CO2e.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [20] Truly, A. When is Nvidia releasing new graphics cards? (2022). https://screenrant.com/when-is-nvidia-releasing-new-graphics-cards/. Accessed 2 May 2023.
Not checked in the free run.
- The authors state this as an assumption: A GPU lasts 1.5 years before becoming obsolete.
May combine two claims.
Findingown dataA GPT-3 query takes approximately 3.8 s per page of text (4.4 s for ChatGPT to produce 292 words).
“Since, based on our own measurement, a GPT3 query takes approximately 3.8 s per page of text (4.4 s for ChatGPT to produce 292 words)”
Quote in the paperNot a substantial claimMay combine two claims.
FindingcitationApproximating a GPU lasts 1.5 years before becoming obsolete, the amortized training cost per query for BLOOM is 0.00004 g CO2e.
“Approximating that a GPU lasts 1.5 years before becoming obsolete20, the amortized training cost per query for BLOOM is 0.00004 g CO2e”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [20] Truly, A. When is Nvidia releasing new graphics cards? (2022). https://screenrant.com/when-is-nvidia-releasing-new-graphics-cards/. Accessed 2 May 2023.
Not checked in the free run.
- The authors state this as an assumption: A GPU lasts 1.5 years before becoming obsolete.
Probably not a substantial claim: may not pass S3 (load-bearing).
FindingreasoningThe embodied energy of the server per query is 0.03 g CO2e.
“the embodied energy of the server per query is 0.03 g CO2e.”
Quote in the paperNot a substantial claimMay combine two claims.
FindingreasoningA resident of India has an annual impact of 1.9 metric tons CO2e, equating to around 180 g CO2e per page.
“In contrast, a resident of India has an annual impact of 1.9 metric tons22, equating to around 180 g CO2e per page.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
May combine two claims.
FindingreasoningIf a person takes 1 min to write a query and waits 4 s for the query to be handled, at 15 metric tons CO2e per year for a US resident, 1.06 min has a footprint of 30 g CO2e, approximately 15 times greater than the AI itself.
“If a person takes 1 min to write a query, and needs to wait 4 s (0.07 min) for the query to be handled: at 15 metric tons CO2e per year for a US resident, 1.06 min has a footprint of 30 g CO2e, approximately 15 times greater than the AI itself.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
May combine two claims.
FindingreasoningGenerating an image in the largest-emissions scenario requires 4.5 Wh to process, emitting approximately 1.9 g CO2e.
“This work would require 4.5 Wh to process, emtting approximately 1.9 g CO2e24.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [24] U.S. Environmental Protection Agency. Greenhouse gas equivalencies calculator (2022). https://www.epa.gov/energy/greenhouse-gas-equivalencies-calculator#results. Accessed 14 March 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).
Interpretationnone givenAI is not a substitute for all human tasks.
“In addition, AI is not a substitute for all human tasks.”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationThe emission footprint of a US resident is approximately 15 metric tons CO2e per year, translating to roughly 1.7 kg CO2e per hour.
“the emission footprint of a US resident is approximately 15 metric tons CO2e per year22, which translates to roughly 1.7 kg CO2e per hour.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Study design and conductnone givenThe emissions analyses do not account for social impacts such as professional displacement, legality, and rebound effects.
“Emissions analyses do not account for social impacts such as professional displacement, legality, and rebound effects.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe embodied energy footprint of an A100 GPU is 150 kg CO2e.
“The embodied energy footprint of an A100 GPU is 150 kg CO2e12.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [12] Luccioni, A. S., Viguier, S. & Ligozat, A.-L. Estimating the carbon footprint of BLOOM, a 176B parameter language model. https://doi.org/10.48550/ARXIV.2211.02001 (2022). source
Not checked in the free run.
May combine two claims.
PremisecitationChatGPT produces 0.382 g CO2e per query, based on 3.82 metric tons CO2e per day divided by 10,000,000 queries per day.
“An informal online estimate for ChatGPT indicates that it produces 0.382 g CO2e per query18, based on 3.82 metric tons CO2e per day divided by 10,000,000 queries per day.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [18] Pointon, C. The carbon footprint of ChatGPT (2022). https://medium.com/@chrispointon/the-carbon-footprint-of-chatgpt-e1bc14e4cc2a. Accessed 14 March 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe energy footprint for recycling an equivalent device is 1 kg CO2e.
“The energy footprint for recycling an equivalent device is 1 kg CO2e19.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [19] Circular Computing. What is the carbon footprint of a laptop? (2021). https://circularcomputing.com/news/carbon-footprint-laptop/. Accessed 2 May 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe embodied energy of the server used to conduct a query is approximately 2500 kg CO2e.
“Similarly, the embodied energy of the server used to conduct the query is approximately 2500 kg CO2e12.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [12] Luccioni, A. S., Viguier, S. & Ligozat, A.-L. Estimating the carbon footprint of BLOOM, a 176B parameter language model. https://doi.org/10.48550/ARXIV.2211.02001 (2022). source
Not checked in the free run.
May combine two claims.
InterpretationreasoningAI can play an important role in various sectors of society without, at present, running afoul of problematic carbon emissions.
“AI can play an important role in various sectors of society without, at present, running afoul of problematic carbon emissions.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).
InterpretationreasoningFuture changes in technology and society will change the relative environmental impact of AI compared with human activity.
“future changes in technology and society will undoubtedly change the relative environmental impact of AI as well.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Interpretationnone givenDrawing a stick figure is likely faster for a human than for an AI at present.
“drawing a stick figure is likely faster for a human than an AI at present”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).
Interpretationnone givenBoth human- and AI-produced text will likely need to be revised and rewritten based on the human authors' sense for how effectively the text expresses the desired content.
“In addition, we note that there is significant complexity to writing processes: both human- and AI-produced text will likely need to be revised and rewritten based on the human authors’ sense for how effectively the text expresses the desired content.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
PremisecitationAI data centers often run on Nvidia A100 GPUs.
“AI data centers, such as Google’s Compute Engine26, often run on Nvidia A100 GPUs27.”
Quote in the paperSubstantial claim0 of 2 sources checked- cites [26] Kleban, C. & Parthasarathy, B. Announcing the Google Cloud A2 VM family based on NVIDIA A100 GPU (2020). https://cloud.google.com/blog/products/compute/announcing-google-cloud-a2-vm-family-based-on-nvidia-a100-gpu. Accessed 14 March 2023.
Not checked in the free run.
- cites [27] NVIDIA Corporation. NVIDIA A100 GPU (2023). https://www.nvidia.com/en-us/data-center/a100/. Accessed 14 March 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.
Study design and conductreasoningThe revision process exists in both human and AI-assisted writing and is beyond the scope of this analysis.
“Since this revision process exists in both human and AI-assisted writing, we feel it is beyond the scope of this analysis.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S2 (checkable), S4 (based).
Interpretationnone givenAI is poised to take over roles once thought to be solely the domain of humans, namely roles requiring creativity and the ability to integrate across multiple intellectual domains.
“AI is poised to take over roles once thought to be solely the domain of humans—those requiring creativity and the ability to integrate across multiple intellectual domains to synthesize concepts from each.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe per capita environmental impact of a resident of India is still only one seventh the impact of a US resident.
“the impact of a resident of India is still only one seventh the impact of a US resident”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Interpretationnone givenFor a complex illustration such as one resembling an oil painting, an AI is faster than a human at present.
“whereas the reverse is true for a complex illustration such as one resembling an oil painting”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningThe trends in per capita human environmental impact in the US and India are highly likely to affect the human side of the AI/human ratio of environmental costs for the activities addressed in the article.
“they are highly likely to affect the human side of the AI/human ratio of the environmental costs related to the activities addressed in this article.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Interpretationnone givenWriting an in-depth, heavily-referenced, original article on a niche scientific topic is currently beyond the capabilities of an AI.
“writing an in-depth, heavily-referenced, original article on a niche scientific topic is currently beyond the capabilities of an AI”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe per capita environmental impact of a human in India has been rising almost continuously since the 1940s.
“the per capita impact of a human in India has been rising almost continuously since the 1940s”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
May combine two claims.
PremisecitationThe average cost for an illustration project is $200 and the average hourly rate of pay for an illustrator is $62.50/h.
“we combined the average cost for an illustration project ($20028) and the average hourly rate of pay for an illustrator ($62.50/h28)”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [28] Thumbtack. How much do illustrators charge? (2020). https://www.thumbtack.com/p/illustration-rates. Accessed 14 March 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe per capita environmental impact of a human in the US has been mostly falling since it peaked in the 1970s.
“the per capita impact of a human in the US has been mostly falling since it peaked in the 1970s”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [22] H. Ritchie, M. R. & Rosado, P. CO2 and Greenhouse Gas Emissions (2020). https://ourworldindata.org/co2-and-other-greenhouse-gas-emissions. Accessed 14 March 2023.
Not checked in the free run.
Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).
InterpretationreasoningAny accounting for the return on energy expenditure is confounded by the impact to the rest of the network in which it is embedded.
“Authorship does not exist in a vacuum, and any accounting for the return on energy expenditure is confounded by the impact to the rest of the network in which it is embedded.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Interpretationnone givenThe carbon footprint estimates for human illustration may vary depending on factors such as location, individual energy consumption, and illustration complexity.
“These estimates may vary depending on factors such as location, individual energy consumption, and illustration complexity.”
Quote in the paperNot a substantial claim
The authors state this as an assumption.
ChatGPT undergoes a full re-training of the model once per month and continues with an estimated 10,000,000 queries per day.
“Assuming that ChatGPT undergoes a full re-training of the model once per month and continues with an estimated 10,000,000 queries per day”
The authors state this as an assumption.
A person's emissions while writing are consistent with their overall annual impact.
“Assuming that a person’s emissions while writing are consistent with their overall annual impact”
The paper relies on this without support.
The US and India are examples of countries with the highest and lowest per capita impact among large countries (over 300 M population).
“In this analysis, we use the US and India as examples of countries with the highest and lowest per capita impact among large countries (over 300 M population).”
The authors state this as an assumption.
The EPA calculator captures the relationship adequately.
“we assume that the EPA calculator captures the relationship adequately.”
The authors state this as an assumption.
The quality of writing produced by AI is sufficient for whatever task may be at hand.
“Assuming the quality of writing produced by AI is sufficient for whatever task may be at hand”
The authors state this as an assumption.
BLOOM has a similar level of usage and frequency of retraining as ChatGPT.
“assuming a similar level of usage and frequency of retraining as for ChatGPT”
The authors state this as an assumption.
A GPU lasts 1.5 years before becoming obsolete.
“Approximating that a GPU lasts 1.5 years before becoming obsolete20”
The paper relies on this without support.
A desktop computer consumes 200 W and generates 72 g CO2e in the same amount of time as the writing period.
“In comparison, a desktop computer consumes 200 W, generating 72 g CO2e in the same amount of time.”
The paper relies on this without support.
The AI and human tasks being compared are equivalent enough in nature that their carbon footprints can be meaningfully compared.
“The results for each specific task reflect an array of assumptions about the nature of these tasks and the people and AIs engaged in such tasks.”
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?
AI colonialism represents a new form of digital and environmental exploitation, where benefits accrue to the Global North's super-rich class while the Global South suffers the externalities.
“AI colonialism thus represents a new form of digital and environmental exploitation, where the benefits accrue to the Global North's super-rich class while the Global South remarkably suffers the externalities.10,11,12,13,14,15,16”
From Artificial Intelligence Colonialism: Environmental Damage, Labor Exploitation, and Human Rights Crises in the Global South (2024) · cites this paper as [15]
The Discussion (Benefits and drawbacks; page unavailable) describes OpenAI outsourcing disturbing content-labeling work to workers in Kenya, which supports a limited concern about burdens borne by workers there. It does not characterize this as AI colonialism or say that environmental externalities fall on the Global South while benefits accrue to a super-rich class in the Global North. (LLM, conf=0.72)
Experimental — accuracy not yet measuredOne study suggests carbon emissions of writing and illustrating are lower for AI than for humans.
“One study, for example, suggests the carbon emissions of writing and illustrating are lower for AI than for humans (71).”
From Can Generative AI improve social science? (2024) · cites this paper as [71]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
BLOOM's amortized training emissions are understated
MinorAt the stated usage of 300 million queries per month, 50.5 metric tons of training emissions amount to about 0.168 g per query, not 0.10 g. In the check, the denominator is expressed in millions of queries, so metric tons divided by millions of queries gives grams per query.
Reported 0.1 · recomputed 0.168333
“For BLOOM, assuming a similar level of usage and frequency of retraining as for ChatGPT, the per-query impact of training is 0.10 g CO2e”
Cites
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021) · Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3442188.3445922 · Cited by 6,988 · cited 1× · see citing works
“Many relevant issues have been discussed elsewhere 29,30 ; here, we touch on a few items of most relevance to AI for writing and illustrating.”
BackgroundArtificial Intelligence in Education: A Review
Chen, L., Chen, P., & Lin, Z. (2020) · IEEE Access · doi:10.1109/access.2020.2988510 · Cited by 4,133 · cited 1× · see citing works
“development of new approaches to sustainable futures (26); they could lead to benefits in medicine (27); and they could improve human educational systems (28).”
BackgroundEnergy 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 · see citing works
Automation and New Tasks: How Technology Displaces and Reinstates Labor
Acemoglu, D., & Restrepo, P. (2019) · Journal of Economic Perspectives · doi:10.1257/jep.33.2.3 · Cited by 2,612 · see citing works
Tools such as ChatGPT threaten transparent science; here are our ground rules for their use
(2023) · Nature · doi:10.1038/d41586-023-00191-1 · Cited by 605 · see citing works
Cowls, J., Tsamados, A., Taddeo, M., & Floridi, L. (2023) · AI & SOCIETY · doi:10.1007/s00146-021-01294-x · Cited by 552 · see citing works
Cowls, J., Tsamados, A., Taddeo, M., & Floridi, L. (2021) · doi:10.2139/ssrn.3804983 · Cited by 552 · see citing works
The carbon impact of artificial intelligence
Dhar, P. (2020) · Nature Machine Intelligence · doi:10.1038/s42256-020-0219-9 · Cited by 509 · see citing works
Nikitas, A., Michalakopoulou, K., Njoya, E. T., & Karampatzakis, D. (2020) · Sustainability · doi:10.3390/su12072789 · Cited by 495 · see citing works
The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink
Patterson, D., Gonzalez, J., Holzle, U., Le, Q., Liang, C., Munguia, L.-M., Rothchild, D., So, D. R., Texier, M., & Dean, J. (2022) · Computer · doi:10.1109/mc.2022.3148714 · Cited by 400 · see citing works
Self-citationBrand, C., Götschi, T., Dons, E., Gerike, R., Anaya-Boig, E., Ávila-Palència, I., de Nazelle, A. J., Gascón, M., Gaupp-Berghausen, M., Iacorossi, F., et al. (2021) · Global Environmental Change · doi:10.1016/j.gloenvcha.2021.102224 · Cited by 269 · see citing works
Artificial intelligence techniques in finance and financial markets: A survey of the literature
Milana, C., & Ashta, A. (2021) · Strategic Change · doi:10.1002/jsc.2403 · Cited by 228 · see citing works
Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model
Luccioni, A. S., Viguier, S., & Ligozat, A.-L. (2022) · arXiv · doi:10.48550/arxiv.2211.02001 · Cited by 192 · cited 11× · see citing works
“For BLOOM, assuming a similar level of usage and frequency of retraining as for ChatGPT, the per-query impact of training is 0.10 g CO2e, while the per-query operational cost is 1.47 g.”
BackgroundMethodInfluentialMuralikrishna, I. V., & Manickam, V. (2017) · Environmental Management · doi:10.1016/b978-0-12-811989-1.00005-1 · Cited by 172 · see citing works
Rebound Effects and ICT: A Review of the Literature
Gossart, C. (2015) · Advances in Intelligent Systems and Computing · doi:10.1007/978-3-319-09228-7_26 · Cited by 154 · see citing works
Economics for the future – Beyond the superorganism
Hagens, N. J. (2019) · Ecological Economics · doi:10.1016/j.ecolecon.2019.106520 · Cited by 143 · see citing works
Computational sustainability: computing for a better world and a sustainable future
Gomes, C., Dietterich, T., Barrett, C., Conrad, J., Dilkina, B., Ermon, S., Fang, F., Farnsworth, A., Fern, A., Fern, X., et al. (2019) · Communications of the ACM · doi:10.1145/3339399 · Cited by 126 · see citing works
Forecasting extreme labor displacement: A survey of AI practitioners
Gruetzemacher, R., Paradice, D., & Lee, K. B. (2020) · Technological Forecasting and Social Change · doi:10.1016/j.techfore.2020.120323 · Cited by 105 · see citing works
Li, Q., Ren, Z., Cao, K., Li, M. M., Wang, K., & Zhou, Y. (2022) · Science Advances · doi:10.1126/sciadv.abj1624 · Cited by 66 · see citing works
Benefits of Artificial Intelligence in Medicine
Yeasmin, S. (2019) · International Conference on Computer Applications & Information Security (ICCAIS) · doi:10.1109/cais.2019.8769557 · Cited by 58 · see citing works
(2023) · Nature Machine Intelligence · doi:10.1038/s42256-023-00613-9 · Cited by 45 · see citing works
ChatGPT and Works Scholarly: Best Practices and Legal Pitfalls in Writing with AI
Tomlinson, B., Torrance, A. W., & Black, R. W. (2023) · SMU Law Review Forum · doi:10.25172/slrf.76.1.5 · Cited by 24 · see citing works
Self-citation
Flagged first, then by citations
Research works printed but not linked: 1
- IPCC Summary for policymakers. In Climate Change 2021 (eds Masson-Delmotte, V. et al.) (Cambridge University Press, 2021).
Cited by
Sorted by citations
Can Generative AI improve social science?
Bail, C. A. (2024) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2314021121 · Cited by 369 · see citing works
Mapping the ethics of generative AI: a comprehensive scoping review
Hagendorff, T. (2024) · Minds and Machines · doi:10.1007/s11023-024-09694-w · Cited by 141 · see citing works
Xiao, T., Nerini, F. F., Matthews, H. D., Tavoni, M., & You, F. (2025) · Nature Sustainability · doi:10.1038/s41893-025-01681-y · Cited by 121 · see citing works
Reconciling the contrasting narratives on the environmental impact of large language models
Ren, S., Tomlinson, B., Black, R. W., & Torrance, A. W. (2024) · Scientific Reports · doi:10.1038/s41598-024-76682-6 · Cited by 68 · see citing works
Self-citationLuccioni, A. S., Strubell, E., & Crawford, K. (2025) · Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3715275.3732007 · Cited by 71 · see citing works
Regilme, S. S. F. (2024) · SAIS Review of International Affairs · doi:10.1353/sais.2024.a950958 · Cited by 65 · see citing works
“10,11,12,13,14,15,16 Neoliberal capitalism drives technological innovation for profit and efficiency, often neglecting labor rights, environmental protections, and socioeconomic justice.”
Integrating ESG and AI: a comprehensive responsible AI assessment framework
Lee, S. U., Perera, H., Liu, Y., Xia, B., Lu, Q., Zhu, L., Cairns, J., & Nottage, M. (2025) · AI and Ethics · doi:10.1007/s43681-025-00741-5 · Cited by 67 · see citing works
Designing Home Automation Routines Using an LLM-Based Chatbot
Giudici, M., Padalino, L., Paolino, G., Paratici, I., Pascu, A. I., & Garzotto, F. (2024) · Designs · doi:10.3390/designs8030043 · Cited by 40 · see citing works
Karjus, A. (2025) · Humanities and Social Sciences Communications · doi:10.1057/s41599-025-04503-w · Cited by 68 · see citing works
Advances and challenges in energy and climate alignment of AI infrastructure expansion
Lal, A., & You, F. (2025) · Advances in Applied Energy · doi:10.1016/j.adapen.2025.100243 · Cited by 35 · see citing works
Ji, Z., & Jiang, M. F. (2025) · Renewable and Sustainable Energy Reviews · doi:10.1016/j.rser.2025.116159 · Cited by 35 · see citing works
Competing narratives in AI ethics: a defense of sociotechnical pragmatism
Watson, D., Mökander, J., & Floridi, L. (2024) · AI & Society · doi:10.1007/s00146-024-02128-2 · Cited by 36 · see citing works
Artificial Intelligence in Ecology: A Commentary on a Chatbot's Perspective
Haghighi, S. R., Saqalaksari, M. P., & Johnson, S. N. (2023) · Bulletin of the Ecological Society of America · doi:10.1002/bes2.2097 · Cited by 35 · see citing works
The barriers to uptake of artificial intelligence in hepatology and how to overcome them
Clusmann, J., Balaguer‐Montero, M., Bassegoda, O., Schneider, C. V., Seraphin, T. P., Paintsil, E. K., Luedde, T., Pérez-López, R., Caldéraro, J., Gilbert, S., et al. (2025) · Journal of Hepatology · doi:10.1016/j.jhep.2025.07.003 · Cited by 29 · see citing works
RetractedBeyond the stereotypes: Artificial Intelligence image generation and diversity in anesthesiology
Gisselbaek, M., Minsart, L., Köselerli, E., Suppan, M., Meço, B. C., Seidel, L., Albert, A., Chang, O. L. B., Saxena, S., & Berger‐Estilita, J. (2024) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2024.1462819 · Cited by 21 · see citing works
Mésároš, P., Smetanková, J., Behúnová, A., & Krajníková, K. (2024) · Mobile Networks and Applications · doi:10.1007/s11036-024-02368-y · Cited by 22 · see citing works
The impact of artificial intelligence on the energy transition: evidence from Chinese cities
Xu, J., Chen, Y., Yang, N., & Shao, S. (2025) · World Development · doi:10.1016/j.worlddev.2025.107126 · Cited by 20 · see citing works
Machine learning for a sustainable energy future
Oral, B., Çoşgun, A., Kilic, A., Eroğlu, D., Günay, M. E., & Yıldırım, R. (2024) · Chemical Communications · doi:10.1039/d4cc05148c · Cited by 16 · see citing works
Grassini, S. (2024) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2024.2422755 · Cited by 19 · see citing works
ML@ChemE: Past, Present, and Future of Machine Learning in Chemical Engineering
Özdemir, P., & Yıldırım, R. (2025) · ChemBioEng Reviews · doi:10.1002/cben.70012 · Cited by 14 · see citing works
A systems approach to AI and education in a post-digital world
Sharples, M. (2025) · Theory Into Practice · doi:10.1080/00405841.2025.2528549 · Cited by 12 · see citing works
Chen, H., Shen, Q., Li, X., Xu, W. H., & Liu, Y. (2025) · Energy · doi:10.1016/j.energy.2025.137311 · Cited by 11 · see citing works
Osmanlliu, E., Senkaiahliyan, S., Eisen-Cuadra, A. M., Kalla, M., Kalema, N. L., Teixeira, A. R., & Celi, L. A. (2025) · NEJM Catalyst · doi:10.1056/cat.24.0501 · Cited by 11 · see citing works
Lee, C., Kumar, S., Vogt, K. A., Munshi, M., Tallapudi, P., Vogt, A., Awad, H., & Khan, W. S. (2025) · Scientific Reports · doi:10.1038/s41598-025-10451-x · Cited by 13 · see citing works
A Critical Analysis of Machine Learning Eco-feedback Tools through the Lens of Sustainable HCI
Görücü, S., Morais, L., & Panagiotidou, G. (2025) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3706598.3713198 · Cited by 10 · see citing works
A Water Efficiency Dataset for African Data Centers
Shumba, N., Tshekiso, O., Li, P., Fanti, G., & Ren, S. (2025) · Proceedings of the ACM SIGCAS/SIGCHI Conference on Computing and Sustainable Societies · doi:10.1145/3715335.3735483 · Cited by 10 · see citing works
Beyond AI automation: Redesigning organizations for human-AI synergy
Krotov, V. (2025) · Business Horizons · doi:10.1016/j.bushor.2025.06.007 · Cited by 9 · see citing works
Lal, A., & You, F. (2025) · Environmental Science & Technology · doi:10.1021/acs.est.4c11477 · Cited by 8 · see citing works
Exploring Anthropomorphism in Conversational Agents for Environmental Sustainability
Giudici, M., Scherini, S., Chaussumier, P., Ginocchio, S., & Garzotto, F. (2025) · Designing Interactive Systems Conference · doi:10.1145/3715336.3735700 · Cited by 9 · see citing works
The climate impacts of healthcare digitalization: A scoping review
Nissinen, A., Pitkänen, A., Барчук, А. А., Hosseinian, A., Johansson, A., Kaisti, M., Karvonen, J., Marttinen, P., Moen, H., Peltola, E., et al. (2025) · Digital Health · doi:10.1177/20552076251364666 · Cited by 7 · see citing works
Spillo, G., Filippo, A. D., Fontana, E., Milano, M., & Semeraro, G. M. (2025) · ACM Conference on User Modeling, Adaptation and Personalization (UMAP) · doi:10.1145/3699682.3728336 · Cited by 6 · see citing works
The Carbon Cost of Intelligence: A Domain-Specific Framework for Measuring AI Energy and Emissions
Kaur, R., Kundu, T., Park, K., & Pinsky, E. (2026) · Energies · doi:10.3390/en19030642 · Cited by 6 · see citing works
Zhu, J., & Molnár, A. (2026) · Computers in Human Behavior · doi:10.1016/j.chb.2026.108929 · Cited by 10 · see citing works
Fake News Detection in Portuguese Under Large Language Model-Generated Content
Silva, R. M., Amamou, H., Ferraz, L. B. S., Silva, F. K. A. D., & Avila, A. R. (2025) · Journal of the Brazilian Computer Society · doi:10.5753/jbcs.2025.5525 · Cited by 5 · see citing works
Integrating Virtual Reality, Chatbots, and BDI Agents: VEsNA Goes Fast!
Ferrando, A., Gatti, A., & Mascardi, V. (2026) · Springer eBooks · doi:10.1007/978-3-032-01082-7_11 · Cited by 5 · see citing works
Artificial intelligence applications for advancing sustainable green finance
Nurmasari, N. D., & Bilad, M. R. (2026) · Discover Sustainability · doi:10.1007/s43621-026-02980-w · Cited by 4 · see citing works
Qiao, Y., Min, X., Ge, S., Zhu, X., & Yang, N. (2026) · Journal of environmental chemical engineering · doi:10.1016/j.jece.2026.122522 · Cited by 7 · see citing works
Marimekala, D., & Lamb, J. P. (2024) · IEEE Integrated STEM Education Conference (ISEC) · doi:10.1109/isec61299.2024.10664756 · Cited by 4 · see citing works
Zesch, T., Hanses, M., Seidel, N., Aggarwal, P., Veiel, D., & de Witt, C. (2024) · International Conference on Information Technology Based Higher Education and Training (ITHET) · doi:10.1109/ithet61869.2024.10837635 · Cited by 5 · see citing works
Haworth, M. (2026) · Transactional Analysis Journal · doi:10.1080/03621537.2026.2595388 · Cited by 4 · see citing works
Radovan, M., Košmerl, T., & Radovan, D. M. (2026) · Sustainability · doi:10.3390/su18147213 · Cited by 3 · see citing works
Life Cycle Carbon Emissions Accounting of China’s Physical Publishing Industry
Xu, R., Yang, Y., Zhang, L., Li, Q., Qian, F., Song, L., & Xie, B. (2025) · Sustainability · doi:10.3390/su17041664 · Cited by 1 · see citing works
CARBON FOOTPRINT THROUGH THE LENS OF INFORMATION AND COMMUNICATION TECHNOLOGY
Gençtürk, F., & Biroğul, S. (2025) · Uluslararası sürdürülebilir mühendislik ve teknoloji dergisi. · doi:10.62301/usmtd.1712846 · Cited by 1 · see citing works
Lin, H.-J., Lin, H.-P., & Hsieh, I.-Y. L. (2026) · Transportation Letters · doi:10.1080/19427867.2026.2623085 · Cited by 1 · see citing works
Toth, C., Nastal, J., Barney, T. B., Messer, K., & Godfrey, J. (2026) · The Journal of Writing Analytics · doi:10.37514/jwa-j.2026.9.1.01 · Cited by 1 · see citing works
Gow, S., & Illingworth, S. (2026) · Artificial Intelligence in Education · doi:10.1108/aiie-06-2025-0151 · Cited by 1 · see citing works
Li, P., De, G., Zhang, J., 邹绍辉, Liu, H., & Li, Y. (2026) · Energy Economics · doi:10.1016/j.eneco.2026.109346 · Cited by 1 · see citing works
Kelters, B. (2026) · AI & Society · doi:10.1007/s00146-026-03055-0 · Cited by 1 · see citing works
Ramaditya, M., Ikhsan, M., LUBIS, M. F. S., & Riantoputra, C. D. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2026.1784986 · Cited by 1 · see citing works
Drivers of carbon emissions in the era of artificial intelligence: case of Japan
Kraujalienė, L., Yaseen, A., Kromalcas, S., & Kauzonė, H. M. (2026) · Technological and Economic Development of Economy · doi:10.3846/tede.2026.26996 · Cited by 1 · see citing works
On the Datafication of Future AI Agents
Farina, M., Wang, Y., Yu, X., & Pedrycz, W. (2026) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2026.2692018 · Cited by 1 · see citing works
Artificial intelligence data centers could reach one percent of global electricity demand by 2030
Chen, D., Zhou, Z., Cai, Y., Qin, J., Katchova, A. L., & Chen, L. (2026) · Communications Sustainability · doi:10.1038/s44458-026-00152-5 · Cited by 1 · see citing works
AI Innovations in Global Health in the Context of Climate Change
Papadopoulou, P., Apostolaki, S., Lytras, M. D., & Konstantinopoulou, S. (2024) · Advances in healthcare information systems and administration book series · doi:10.4018/979-8-3693-4402-6.ch001 · Cited by 1 · see citing works
Künstliche Intelligenz, Klima und Bildung
Autenrieth, D., & Schluchter, J.-R. (2025) · merz | medien + erziehung · doi:10.21240/merz/2025.2.12 · Cited by 1 · see citing works
Pospieszny, P., & Brodowicz, D. P. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5180067
Growing Through Algorithms: Reimagining Plant Life with AI Art
Novaković, J. (2025) · AM Journal of Art and Media Studies · doi:10.25038/am.v0i28.619
Creating space for climate justice in library and information science
Oliphant, T., Youngman, T., Hackborn, D., Nathan, L. P., & Patin, B. J. H. (2025) · Information Research · doi:10.47989/ir30colis52330
Rethinking Green Ai for Sustainable ComputingPreprint
Zadid, T. H., Tanim, S. A., Hossen, M. R., Biswas, T., & Sattar, F. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5367962
May AI be sustainable? An outlook on sustainability and technology
Kondratenko, A., Kondratenko, A. A., & Kamberov, K. H. (2025) · AI & Society · doi:10.1007/s00146-025-02804-x
Urama, C. E., Oboodoechi, D. N., Chukwu, N. O., Omeje, A. N., Asogwa, H. T., & Ukwueze, E. R. (2025) · Emerald Group Publishing Limited eBooks · doi:10.1108/978-1-83662-570-420251013
CIJENA KLIKA: BIOETIČKI OSVRT NA ODRŽIVOST DIGITALNOG DRUŠTVA
Kelam, I., & Dulemba, D. (2025) · ARHE · doi:10.19090/arhe.2025.43.129-155
GenAI Paradox: Balancing Creative Innovation with Authenticity, Ethics, and Legal Compliance
Perseghin, E., & Foresti, G. L. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-11381-8_33
Yang, Y. B., & Suh, E.-K. (2026) · IEEE Access · doi:10.1109/access.2026.3661618
The Climate Impact Paradox of Artificial Intelligence: Literature Review and Research Prospects
Chen, Z., Xiong, Q., Hu, Z., Liu, S., & Ma, J. (2026) · Journal of Energy and Climate Change · doi:10.3724/j.issn.2097-4981.jecc-2025-0385
Generative AI and Libraries: Claiming Our Place in the Center of a Shared Future
Gehred, A. P. (2026) · Medical Reference Services Quarterly · doi:10.1080/02763869.2026.2619769
Richards, D. R., Worden, D., & Lavorel, S. (2026) · Landscape and Urban Planning · doi:10.1016/j.landurbplan.2026.105615
Sherly, K., Veeresha, P., & Başkonuş, H. M. (2026) · Fractal and Fractional · doi:10.3390/fractalfract10030184
Mapping the Responsible Democratization of Generative AI through Participatory Futuring
Jansen, A., Sankaran, S., Kalving, M., Väänänen, K., Geurts, E., Rovelo, G., Munteanu, C., Johry, A., & Häkkilä, J. (2026) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/3772363.3778759
Ye, L. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6591685
Framing AI and SDGs Relationship: A Comparative Media Study Across Five European Countries
López-Carrión, A. E. (2026) · Springer eBooks · doi:10.1007/978-3-032-16800-9_13
AI Will Determine Next Generation National Power and Economic Security
Shibuya, K. (2026) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3373-7205-1.ch008
Closed Rooms and Oracles: the poetics of vector space in Sam Riviere’s Conflicted Copy
Farrier, D. S. (2026) · Journal of British and Irish innovative poetry · doi:10.16995/bip.25843
Willem, T., & Piraud, M. (2026) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3805689.3812295
Assheton, G., Kioupi, V., & Michel‐Villarreal, R. (2026) · World sustainability series · doi:10.1007/978-3-032-22076-9_38
Alkan, F., Özülüş, E., & Yücel, A. S. (2026) · IGI Global eBooks · doi:10.4018/979-8-3373-8490-0.ch009
AI ethics after generative AI: from consequences to effects
Gianni, R. (2026) · AI & Society · doi:10.1007/s00146-026-03308-y
Gula: consumir recursos y producir contenido sin medida
Navia, J. M. A., Zapata, E. M. J., & Castillo, V. S. (2026) · doi:10.62486/978-9915-9928-6-0.ch05
Key Drivers in the Development of the Green Economy in the Gulf and Levant States
Lopatnikov, D. L., & Gladenkova, T. A. (2026) · GEOGRAPHY ENVIRONMENT SUSTAINABILITY · doi:10.24057/2071-9388-2026-3945
- Authors
- Tomlinson, B., Black, R. W., Patterson, D. J., & Torrance, A. W.
- Affiliations
- University of California, Irvine (US)
- University of Kansas (US)
- Corresponding author
- Tomlinson, Bill
- Venue
- Scientific Reports
- Published
- 2024-02-14
- Volume
- 14
- Issue
- 1
- Pages
- 3732-3732
- ISSN
- 2045-2322 (ISSN-L)
- PMC
- PMC10867074
- OpenAlex
- W4391822177
- Semantic Scholar
- d9a0ce8a72e2858780b6a83a7dc198602ce806b6
- Cited by
- 77
- References
- 20
- Citations by year
- 2026: 33 · 2025: 31 · 2024: 12 · 2023: 1
- FWCI
- 7.85
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- Scientific Reports (publishedVersion, cc-by)
- Research Square (acceptedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- eScholarship (California Digital Library) (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- Funders
- National Science Foundation (DUE-2121572)
- Topics
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Green IT and Sustainability
- Physical Sciences › Computer Science › Hardware and Architecture › Energy Efficiency in Computing
- Physical Sciences › Environmental Science › Environmental Engineering › Environmental Impact and Sustainability
- Keywords
- CO2 emissions, ChatGPT, bloom, DALL-E 2, Midjourney, human-AI comparison, text generation, image generation, environmental impact of AI
- MeSH terms
- Greenhouse Gases, Carbon, Humans
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.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.
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
