pith. sign in

arxiv: 2511.04776 · v3 · pith:O4KK4I2Rnew · submitted 2025-11-06 · 💻 cs.CY · cs.CL

Quantifying the Climate Risk of Generative AI: Region-Aware Carbon Accounting with G-TRACE and the AI Sustainability Pyramid

Zahida Kausar , Seemab Latif , Raja Khurram Shahzad , Mehwish Fatima This is my paper

Pith reviewed 2026-05-21 18:49 UTC · model grok-4.3

classification 💻 cs.CY cs.CL
keywords generative AIcarbon emissionsclimate riskG-TRACEAI sustainability pyramidenergy accountingsustainable technologyCO2 quantification
0
0 comments X

The pith

Region-aware accounting shows a single GenAI image trend consumed 4,309 MWh and emitted 2,068 tCO2, highlighting climate risks from decentralized inference.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that generative AI poses a measurable climate risk through its energy consumption and emissions, which vary by output type and location. It presents G-TRACE as a framework to quantify these impacts using simulations and real data for text, image, and video generation. The analysis of the Ghibli-style image trend demonstrates how viral use multiplies small costs into 4,309 MWh of energy and 2,068 tCO2 emissions. Additionally, the AI Sustainability Pyramid is introduced as a seven-level model to guide from basic carbon metrics to full stewardship and policy integration. This matters because without such tools, the environmental costs of widespread AI adoption may go unaddressed in climate strategies.

Core claim

G-TRACE quantifies training and inference emissions for generative AI across modalities and geographies with region-aware precision, showing via the 2024-2025 Ghibli-style image generation trend that decentralized participation leads to 4,309 MWh energy consumption and 2,068 tCO2 emissions; the accompanying AI Sustainability Pyramid connects these metrics across seven levels of operational readiness, optimization, and governance for sustainable deployment.

What carries the argument

G-TRACE, the GenAI Transformative Carbon Estimator, which employs real-world analytics and microscopic simulation to derive per-output energy costs and regional carbon intensities for different AI modalities, together with the AI Sustainability Pyramid that structures governance from L1 carbon accounting to L7 stewardship.

If this is right

  • Decentralized inference amplifies small per-query energy costs into large system-level carbon impacts.
  • Viral AI trends can result in tonne-scale CO2 emissions from aggregated individual actions.
  • The AI Sustainability Pyramid provides a pathway to translate emission metrics into actionable policy and operational improvements.
  • Region-specific carbon intensities must be considered for accurate assessment of AI's global climate footprint.
  • Sustainable AI deployment requires integrating quantitative metrics with governance models to support decarbonization goals.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If adopted, G-TRACE could enable AI companies to monitor and mitigate emissions based on where their users are located.
  • The pyramid model might inspire similar tiered frameworks for assessing sustainability in other emerging technologies.
  • This work suggests potential for incorporating AI emissions into international climate reporting standards.
  • Developers could test optimizations at different pyramid levels to measure reductions in energy use.

Load-bearing premise

The microscopic simulation and real-world analytics accurately capture the actual energy consumption and carbon intensities for GenAI outputs across various modalities and deployment regions.

What would settle it

Empirical data from a comparable large-scale viral GenAI usage event, such as widespread image generation, that measures total energy consumption or emissions differing markedly from the reported 4,309 MWh and 2,068 tCO2.

Figures

Figures reproduced from arXiv: 2511.04776 by Mehwish Fatima, Raja Khurram Shahzad, Seemab Latif, Zahida Kausar.

Figure 1
Figure 1. Figure 1: Overview of generative tasks and modalities (Multimodal, Video, Audio, Image, and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: GenAI Model Performance Evolution: MMLU Performance Benchmarks vs. Model [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: G-TRACE framework architecture. Three stages— [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Operationalizing G-TRACE via the AI Sustainability Pyramid. L1–L2: measurement; [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
read the original abstract

Generative Artificial Intelligence (GenAI) represents a rapidly expanding digital infrastructure whose energy demand and associated CO2 emissions are emerging as a new category of climate risk. This study introduces G-TRACE (GenAI Transformative Carbon Estimator), a cross-modal, region-aware framework that quantifies training- and inference-related emissions across modalities and deployment geographies. Using real-world analytics and microscopic simulation, G-TRACE measures energy use and carbon intensity per output type (text, image, video) and reveals how decentralized inference amplifies small per-query energy costs into system-level impacts. Through the Ghibli-style image generation trend (2024-2025), we estimate 4,309 MWh of energy consumption and 2,068 tCO2 emissions, illustrating how viral participation inflates individual digital actions into tonne-scale consequences. Building on these findings, we propose the AI Sustainability Pyramid, a seven-level governance model linking carbon accounting metrics (L1-L7) with operational readiness, optimization, and stewardship. This framework translates quantitative emission metrics into actionable policy guidance for sustainable AI deployment. The study contributes to the quantitative assessment of emerging digital infrastructures as a novel category of climate risk, supporting adaptive governance for sustainable technology deployment. By situating GenAI within climate-risk frameworks, the work advances data-driven methods for aligning technological innovation with global decarbonization and resilience objectives.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces G-TRACE (GenAI Transformative Carbon Estimator), a cross-modal, region-aware framework for quantifying training- and inference-related emissions of generative AI across modalities and geographies using real-world analytics and microscopic simulation. It applies the framework to the 2024-2025 Ghibli-style image generation trend to estimate 4,309 MWh of energy consumption and 2,068 tCO2 emissions, and proposes the AI Sustainability Pyramid, a seven-level governance model that links carbon accounting metrics (L1-L7) with operational readiness, optimization, and stewardship.

Significance. If the G-TRACE estimates are robust, the work supplies a concrete, scaled illustration of how decentralized inference can turn small per-query costs into tonne-scale system impacts from viral trends, thereby contributing quantitative methods for treating GenAI as a distinct category of climate risk. The Pyramid framework offers a structured translation of metrics into policy guidance that could support adaptive governance for sustainable AI deployment.

major comments (2)
  1. [§3] §3 (G-TRACE methodology): The per-output energy costs derived from the microscopic simulation are not shown to have been cross-validated against direct hardware power-draw measurements, model-specific optimizations, or regional GPU telemetry. Because these base unit costs are multiplied by the estimated generation volume to produce the headline 4,309 MWh and 2,068 tCO2 figures, the absence of external validation directly affects the reliability of the central empirical claim.
  2. [§4] §4 (Ghibli case study): The estimation of total generation volume and the choice of regional carbon intensities lack reported sensitivity ranges or error propagation; without these, it is difficult to assess how uncertainty in the microscopic parameters propagates into the final totals that underpin the illustration of viral-participation impacts.
minor comments (2)
  1. [Abstract] The abstract presents numerical results without any reference to data sources, exclusion criteria, or uncertainty treatment; a single sentence summarizing the validation approach would improve transparency.
  2. [§5] Notation for the seven levels of the AI Sustainability Pyramid (L1-L7) is introduced without an explicit mapping table; adding such a table would clarify how each level connects quantitative metrics to operational actions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which help clarify the strengths and limitations of our G-TRACE framework and case study. We address each major comment below and have incorporated revisions to improve methodological transparency and robustness.

read point-by-point responses

  1. Referee: [§3] §3 (G-TRACE methodology): The per-output energy costs derived from the microscopic simulation are not shown to have been cross-validated against direct hardware power-draw measurements, model-specific optimizations, or regional GPU telemetry. Because these base unit costs are multiplied by the estimated generation volume to produce the headline 4,309 MWh and 2,068 tCO2 figures, the absence of external validation directly affects the reliability of the central empirical claim.

    Authors: We acknowledge the value of direct cross-validation for strengthening empirical claims. G-TRACE's per-output costs are derived from microscopic simulation calibrated against publicly available real-world analytics and established GPU power models in the literature. Direct hardware telemetry for the exact models and regions was not feasible within the scope of this study. In the revised manuscript we add a new subsection explicitly discussing validation limitations, citing supporting benchmarks from prior work, and outlining pathways for future telemetry-based refinement. This maintains the framework's grounding in available data while transparently addressing reliability concerns. revision: yes

  2. Referee: [§4] §4 (Ghibli case study): The estimation of total generation volume and the choice of regional carbon intensities lack reported sensitivity ranges or error propagation; without these, it is difficult to assess how uncertainty in the microscopic parameters propagates into the final totals that underpin the illustration of viral-participation impacts.

    Authors: The referee correctly notes the absence of formal sensitivity and error-propagation analysis. Generation volumes were estimated from public trend data and regional intensities drawn from standard IEA and grid databases. In the revised version we include a dedicated sensitivity analysis section that varies generation volume by ±20% and carbon intensities across plausible regional ranges, reporting resulting bounds on the 4,309 MWh and 2,068 tCO2 totals. This addition directly quantifies how parameter uncertainty affects the headline figures and the viral-participation illustration. revision: yes

Circularity Check

0 steps flagged

No significant circularity; estimates scale simulation outputs by external volume without self-referential reduction

full rationale

The paper's central estimate multiplies per-output energy and carbon values (obtained via G-TRACE microscopic simulation plus real-world analytics) by an independently estimated volume of Ghibli-style generations to arrive at 4,309 MWh and 2,068 tCO2. This is a direct arithmetic scaling rather than a fitted parameter that is then relabeled as a prediction. The AI Sustainability Pyramid is introduced as a seven-level governance model that maps the resulting L1-L7 metrics to policy actions; no equations or uniqueness theorems are shown that loop back to the simulation inputs or to self-citations. No self-definitional steps, ansatz smuggling, or renaming of known results appear in the derivation chain. The framework therefore remains self-contained against external benchmarks of energy telemetry and regional carbon-intensity data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

The abstract relies on unstated assumptions about the accuracy of per-query energy models and regional grid intensities; no explicit free parameters, axioms, or invented physical entities are listed.

invented entities (1)
  • AI Sustainability Pyramid no independent evidence
    purpose: Seven-level governance model linking carbon metrics to operational readiness and stewardship
    Introduced as a new framework to translate emission numbers into policy guidance

pith-pipeline@v0.9.0 · 5797 in / 1212 out tokens · 44635 ms · 2026-05-21T18:49:46.286195+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

46 extracted references · 46 canonical work pages · 13 internal anchors

  1. [1]

    GPT-4 Technical Report

    Gpt-4 technical report. arXiv preprint arXiv:2303.08774 URL:https://arxiv.org/abs/2303.08774. Act, E.A.I.,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    MusicLM: Generating Music From Text

    Musiclm: Generating music from text. arXiv preprint arXiv:2301.11325 doi:10.48550/arXiv.2301.11325. Alder, N., Ebert, K., Herbrich, R., Hacker, P.,

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2301.11325

  3. [3]

    arXiv preprint arXiv:2409.07471 URL: https://arxiv.org/abs/2409.07471

    Ai, climate, and transparency: Operationalizing and improving the ai act. arXiv preprint arXiv:2409.07471 URL: https://arxiv.org/abs/2409.07471. 22 Anthony, L.F.W., Kanding, B., Selvan, R.,

    work page arXiv

  4. [4]

    Carbontracker: Tracking and predicting the carbon footprint of training deep learning models,

    Carbontracker: Tracking and predicting the carbon footprint of training deep learning models. arXiv preprint arXiv:2007.03051 doi:10.48550/arXiv.2007.03051. Aquino-Brítez, S., García-Sánchez, P., Ortiz, A., Aquino-Brítez, D.,

    work page doi:10.48550/arxiv.2007.03051 2007

  5. [5]

    Optimizing Large Language Models: Metrics, Energy Efficiency, and Case Study Insights

    Optimizing Large Language Models: Metrics, Energy Efficiency, and Case Study Insights. Technical Report. arXiv. URL: https: //arxiv.org/html/2504.06307v1. Bai, Y ., Kadavath, S., Kundu, S., et al.,

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [6]

    Constitutional AI: Harmlessness from AI Feedback

    Constitutional ai: Harmlessness from ai feedback. arXiv preprint arXiv:2212.08073 URL: https://arxiv.org/abs/ 2212.08073. Bommasani, R., Hudson, D.A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M.S., Bohg, J., Bosselut, A., Brunskill, E., et al.,

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    On the Opportunities and Risks of Foundation Models

    On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258 URL: https: //arxiv.org/abs/2108.07258. Borsos, Z., et al.,

    work page internal anchor Pith review Pith/arXiv arXiv

  8. [8]

    IEEE/ACM Transactions on Audio, Speech, and Language Processing doi:10.1109/TASLP.2023.3288409

    Audiolm: a language modeling approach to audio gen- eration. IEEE/ACM Transactions on Audio, Speech, and Language Processing doi:10.1109/TASLP.2023.3288409. de Carvalho Souza, M.E., Weigang, L.,

    work page doi:10.1109/taslp.2023.3288409 2023

  9. [9]

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arXiv preprint arXiv:2507.06261 URL: https://arxiv. org/abs/2507.06261. Cowls, J., Tsamados, A., Taddeo, M., Floridi, L.,

    work page internal anchor Pith review Pith/arXiv arXiv

  10. [10]

    AI & Society 36, 1057–1081

    The ai gambit: Leveraging artifi- cial intelligence to combat climate change—opportunities, challenges, and recommen- dations. AI & Society 36, 1057–1081. doi:10.1007/s00146-021-01294-x. Dodge, J., Prewitt, T., Tachet des Combes, R., Odmark, E., White, R., Mullins, J., Schwartz, R., Smith, N.A., Strubell, E.,

    work page doi:10.1007/s00146-021-01294-x

  11. [11]

    Measuring the carbon intensity of ai in cloud instances, in: Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, ACM. pp. 1877–1894. doi:10.1145/3531146. 3533234. 23 FADIYA, P., PS, S.L., SHEENA, K.,

    work page doi:10.1145/3531146 2022

  12. [12]

    Authorea Preprints doi:10.22541/au.173656616.35086333/v1

    Sora ai: The future of video generation. Authorea Preprints doi:10.22541/au.173656616.35086333/v1. Gatys, L.A., Ecker, A.S., Bethge, M.,

    work page doi:10.22541/au.173656616.35086333/v1

  13. [13]

    In: 2016 IEEE Conference on Computer Vision and Pattern Recog- nition (CVPR)

    Image style transfer using convolutional neural networks, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 2414–2423. doi:10.1109/CVPR.2016.265. Hacker, P.,

    work page doi:10.1109/cvpr.2016.265 2016

  14. [14]

    arXiv preprint arXiv:2310.04072 URL: https://arxiv.org/abs/ 2310.04072

    Ai regulation in europe: from the ai act to future regulatory chal- lenges. arXiv preprint arXiv:2310.04072 URL: https://arxiv.org/abs/ 2310.04072. Han, S., Mao, H., Dally, W.J.,

    work page arXiv

  15. [15]

    Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding

    Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding. arXiv preprint arXiv:1510.00149 doi:10.48550/arXiv.1510.00149. Henderson, P., Hu, J., Romoff, J., Brunskill, E., Jurafsky, D., Pineau, J.,

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1510.00149

  16. [16]

    Advances in neural information processing systems 35, 8633–8646

    Video diffusion models. Advances in neural information processing systems 35, 8633–8646. URL: https://proceedings.neurips.cc/paper_files/ paper/2022/hash/39235c56aef13fb05a6adc95eb9d8d66. Jiang, Y ., Liu, X., Wang, L., Zhang, M.,

    work page 2022

  17. [17]

    Kaack, L.H., Donti, P.L., Strubell, E., Kamiya, G., Creutzig, F., Rolnick, D.,

    doi:10.1007/s11783-024-1918-y. Kaack, L.H., Donti, P.L., Strubell, E., Kamiya, G., Creutzig, F., Rolnick, D.,

    work page doi:10.1007/s11783-024-1918-y 1918

  18. [18]

    Nature Climate Change 12, 518–527

    Aligning artificial intelligence with climate change mitigation. Nature Climate Change 12, 518–527. doi:10.1038/s41558-022-01377-7. Karvonen, F.,

    work page doi:10.1038/s41558-022-01377-7

  19. [19]

    & Adi, Y

    Audiogen: Textually guided audio gen- eration. arXiv preprint arXiv:2209.15352 doi:10.48550/arXiv.2209.15352. Lacoste, A., Luccioni, A., Schmidt, V ., Dandres, T.,

    work page doi:10.48550/arxiv.2209.15352

  20. [20]

    Quantifying the Carbon Emissions of Machine Learning

    Quantifying the carbon emissions of machine learning. arXiv preprint arXiv:1910.09700 doi:10.48550/ arXiv.1910.09700. Lewis, D., Lasek-Markey, M., Golpayegani, D., Pandit, H.J.,

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  21. [21]

    arXiv preprint arXiv:2503.05787 URL: https://arxiv.org/abs/2503.05787

    Mapping the regulatory learning space for the eu ai act. arXiv preprint arXiv:2503.05787 URL: https://arxiv.org/abs/2503.05787. Lin, B., Tang, Z., Ye, Y ., Cui, J., Zhu, B., Jin, P., Huang, J., Zhang, J., Pang, Y ., Ning, M., et al.,

    work page arXiv

  22. [22]

    MoE-LLaVA: Mixture of Experts for Large Vision-Language Models

    Moe-llava: Mixture of experts for large vision-language models. arXiv preprint arXiv:2401.15947 URL:https://arxiv.org/abs/2401.15947. 24 Liu, A., Feng, B., Wang, B., Wang, B., Liu, B., Zhao, C., Dengr, C., Ruan, C., Dai, D., Guo, D., et al.,

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model

    Deepseek-v2: A strong, economical, and efficient mixture-of- experts language model. arXiv preprint arXiv:2405.04434 URL: https://arxiv. org/abs/2405.04434. Luccioni, A.S., Jernite, Y ., Strubell, E., 2023a. Power hungry processing: Watts driving the cost of ai deployment?, in: Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transp...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1145/3593013.3594097 2023

  24. [24]

    Procedia CIRP 122, 59–64

    Estimating the environmental impact of generative-ai services using an lca-based methodology. Procedia CIRP 122, 59–64. doi:10.1016/j.procir.2024.01.009. Marcus, G., Davis, E., Aaronson, S.,

    work page doi:10.1016/j.procir.2024.01.009 2024

  25. [25]

    McKinsey & Company,

    arXiv preprint arXiv:2204.13807 URL:https://arxiv.org/abs/2204.13807. McKinsey & Company,

    work page arXiv

  26. [26]

    Technical Report

    The State of AI in Early 2024: Gen AI Adoption Spikes and Starts to Generate Value. Technical Report. McKin- sey & Company. URL: https://www.mckinsey.com/capabilities/ quantumblack/our-insights/the-state-of-ai-2024. McKinsey Global Institute,

    work page 2024

  27. [27]

    Technical Report

    Rapid Development and Deployment of Powerful Gen- erative AI Models Comes with Environmental Consequences. Technical Report. Massachusetts Institute of Technology. URL: https://news.mit.edu/2025/ environmental-consequences-generative-ai. Patterson, D., Gonzalez, J., Le, Q., Liang, C., Munguia, L.M., Rothchild, D., So, D., Texier, M., Dean, J.,

    work page 2025

  28. [28]

    Carbon Emissions and Large Neural Network Training

    Carbon emissions and large neural network training. arXiv preprint arXiv:2104.10350 URL:https://arxiv.org/abs/2104.10350. Qiu, J., Wang, X., Zhang, P., Chen, W.,

    work page internal anchor Pith review Pith/arXiv arXiv

  29. [29]

    IEEE Access 8, 218841–218864

    Green ai: A comprehensive survey on energy-efficient machine learning. IEEE Access 8, 218841–218864. doi: 10.1109/ ACCESS.2020.3041831. del Rey, S., Martínez-Fernández, S., Cruz, L., Franch, X.,

    work page arXiv 2020

  30. [30]

    arXiv preprint arXiv:2307.05520 URL: https://arxiv.org/abs/ 2307.05520

    How to use model architecture and training environment to estimate the energy consumption of dl training. arXiv preprint arXiv:2307.05520 URL: https://arxiv.org/abs/ 2307.05520. 25 Rolnick, D., Donti, P.L., Heim, L.H., Battiston, G., Settle, S., Guo, T.S., Kakodkar, A., Kalia, K., Milojevic-Dupont, N., Mustafa, N., et al.,

    work page arXiv

  31. [31]

    High-resolution image synthesis with latent diffusion models, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, IEEE. pp. 10684–10695. doi:10.1109/CVPR52688.2022.01042. Samsi, S., Zhao, D., McDonald, J., et al.,

    work page doi:10.1109/cvpr52688.2022.01042 2022

  32. [32]

    From Words to Watts: Benchmarking the Energy Costs of Large Language Model Inference,

    From words to watts: Benchmarking the energy costs of large language model inference. arXiv preprint arXiv:2310.03003 URL:https://arxiv.org/abs/2310.03003. Sanh, V ., Debut, L., Chaumond, J., Wolf, T.,

    work page arXiv

  33. [33]

    DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter

    Distilbert, a distilled version of bert: Smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108 doi:10. 48550/arXiv.1910.01108. Schwartz, R., Dodge, J., Smith, N.A., Etzioni, O.,

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  34. [34]

    Smith, and Oren Etzioni

    Green ai. Communications of the ACM 63, 54–63. doi:10.1145/3381831. Shukla, P.R., Skea, J., Slade, R., Al Khourdajie, A., van Diemen, R., McCollum, D., Pathak, M., Some, S., Vyas, P., Fradera, R., et al.,

    work page doi:10.1145/3381831

  35. [35]

    Contribution of working group III to the sixth assess- ment report of the Intergovernmental Panel on Climate Change 10, 9781009157926

    Climate change 2022: Mitigation of climate change. Contribution of working group III to the sixth assess- ment report of the Intergovernmental Panel on Climate Change 10, 9781009157926. doi:10.1017/9781009157926. Siau, K., Wang, W.,

    work page doi:10.1017/9781009157926 2022

  36. [36]

    Journal of Database Management (JDM) 31, 74–87

    Artificial intelligence (ai) ethics: ethics of ai and ethi- cal ai. Journal of Database Management (JDM) 31, 74–87. doi: 10.4018/JDM. 2020040105. Somani, A., Sunny, M.,

    work page doi:10.4018/jdm

  37. [37]

    Frontiers in Communication 9, 1535156

    Digital transformation in journalism: Mini review on the impact of ai on journalistic practices. Frontiers in Communication 9, 1535156. doi:10.3389/fcomm.2025.1535156. Souza, A., Jasoria, S., Chakrabarty, B., Bridgwater, A., Lundberg, A., Skogh, F., Ali-Eldin, A., Irwin, D., Shenoy, P.,

    work page doi:10.3389/fcomm.2025.1535156 2025

  38. [38]

    Casper: Carbon-aware scheduling and pro- visioning for distributed web services, in: Proceedings of the 14th International Green and Sustainable Computing Conference, pp. 67–73. doi: 10.1109/IGSC59603. 2023.00017. Strubell, E., Ganesh, A., McCallum, A.,

    work page doi:10.1109/igsc59603 2023

  39. [39]

    13693–13696

    Energy and policy considerations for modern deep learning research, in: Proceedings of the AAAI conference on artificial intelligence, pp. 13693–13696. doi:10.1609/aaai.v34i09.7123. 26 Torres, A.J., Alberto, J.M.C., Guieb, A.P.J., Paray, A.D., Villarama, J.A.,

    work page doi:10.1609/aaai.v34i09.7123

  40. [40]

    Language, Technology, and Social Media 3, 17–29

    Language, identity, and ethics in ai-driven art: Perspectives from human artists in digital environments. Language, Technology, and Social Media 3, 17–29. doi:10.26858/ltsm.v3i1.74447. Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al.,

    work page doi:10.26858/ltsm.v3i1.74447

  41. [41]

    LLaMA: Open and Efficient Foundation Language Models

    Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 URL: https:// arxiv.org/abs/2302.13971. Tripp, C.E., Perr-Sauer, J., Gafur, J., Nag, A., Purkayastha, A., Zisman, S., Bensen, E.A.,

    work page internal anchor Pith review Pith/arXiv arXiv

  42. [42]

    arXiv preprint arXiv:2403.08151 URL:https://arxiv.org/abs/2403.08151

    Measuring the energy consumption and efficiency of deep neural networks: An empirical analysis and design recommendations. arXiv preprint arXiv:2403.08151 URL:https://arxiv.org/abs/2403.08151. U.S. Government Accountability Office,

    work page arXiv

  43. [43]

    arXiv preprint arXiv:2407.16894 URL: https://arxiv

    Estimating the increase in emissions caused by ai- augmented search. arXiv preprint arXiv:2407.16894 URL: https://arxiv. org/abs/2407.16894. Verdecchia, R., Sallou, J., Cruz, L.,

    work page arXiv

  44. [44]

    WIREs Data Mining and Knowledge Discovery 13, e1507

    A systematic review of green ai. WIREs Data Mining and Knowledge Discovery 13, e1507. doi:10.1002/widm.1507. Vives, A.B., Morales, F.,

    work page doi:10.1002/widm.1507

  45. [45]

    Joule 7, 2191–2194

    The growing energy footprint of artificial intelligence. Joule 7, 2191–2194. doi:10.1016/j.joule.2023.09.004. Wu, C.J., Raghavendra, R., Gupta, U., Acun, B., Ardalani, N., Maeng, K., Chang, G., Aga, F., Huang, J., Bai, C.,

    work page doi:10.1016/j.joule.2023.09.004 2023

  46. [46]

    Reducing the carbon impact of generative ai inference (today and in 2035), in: Proceedings of the 2nd Workshop on Sustainable Computer Systems, ACM. pp. 1–7. doi:10.1145/3604930.3605705. xAI Team,

    work page doi:10.1145/3604930.3605705 2035