Chapter 31: Further Reading — The Environmental Cost of AI

Foundational Research Papers

1. Strubell, E., Ganesh, A., & McCallum, A. (2019). "Energy and Policy Considerations for Deep Learning in NLP." ACL 2019 Proceedings. The paper that made AI training energy costs visible to the research community. Estimated training costs for BERT, GPT-2, and other NLP models; proposed mandatory reporting of computational cost in publications. Essential starting point.

2. Patterson, D., Gonzalez, J., Le, Q., Liang, C., Munguia, L. M., Rothchild, D., ... & Dean, J. (2021). "Carbon Emissions and Large Neural Network Training." arXiv preprint arXiv:2104.10350. Google and Berkeley researchers' systematic analysis of training carbon costs for multiple large models, establishing the methodology and providing the baseline carbon cost estimates. The most comprehensive quantitative study available.

3. Li, P., Yang, J., Islam, M. A., & Ren, S. (2023). "Making AI Less 'Thirsty': Uncovering and Addressing the Secret Water Footprint of AI Models." arXiv preprint arXiv:2304.03271. The first systematic study of AI water consumption, estimating GPT-3 training and ChatGPT inference water footprints. Foundational for understanding water as a dimension of AI environmental impact.

4. Schwartz, R., Dodge, J., Smith, N. A., & Etzioni, O. (2020). "Green AI." Communications of the ACM, 63(12), 54–63. The defining paper of the Green AI research movement, arguing for reporting computational cost alongside accuracy and investing in efficient AI research. Essential for the research community and policy audience alike.

Energy and Carbon

5. International Energy Agency. (2024). "Electricity 2024: Analysis and Forecast to 2026." IEA. The IEA's current assessment of global electricity demand, including data center growth driven by AI. Updated annually; the current year version is the most relevant. Available free at iea.org.

6. Goldman Sachs Research. (2024). "AI Is Poised to Drive 160% Increase in Data Center Power Demand." Goldman Sachs. The investment bank's analysis of AI's data center power demand trajectory. Widely cited in business press; provides useful commercial projections alongside environmental analysis.

7. Google. (Annual). "Google Environmental Report." Google Sustainability. Google's most detailed corporate environmental report — the most comprehensive public sustainability disclosure from a major AI company. Available annually at sustainability.google.

8. Microsoft. (Annual). "Environmental Sustainability Report." Microsoft. Microsoft's environmental report, notable for acknowledging the tension between AI growth and carbon commitments. Available annually at microsoft.com/en-us/corporate-responsibility.

Water

9. World Resources Institute. (Continuously updated). "Aqueduct Water Risk Atlas." WRI. The standard tool for assessing water stress levels by geography. Available free at wri.org/aqueduct. Essential for data center siting analysis and water justice assessment.

10. Mytton, D. (2021). "Data Centre Water Consumption." npj Clean Water, 4(1), 11. Academic analysis of data center water consumption methodologies, variations by cooling technology and climate, and disclosure gaps. More technical than practitioner-oriented but important for rigorous analysis.

11. NRDC. (2022). "Data Center Surge." Natural Resources Defense Council. NRDC's analysis of data center water and energy use in the American West, specifically addressing Arizona and Colorado River basin impacts. Combines technical analysis with policy advocacy.

Hardware Supply Chain

12. Sovacool, B. K., Ali, S. H., Bazilian, M., Radley, B., Nemery, B., Okatz, J., & Mulvaney, D. (2020). "Sustainable Minerals and Metals for a Low-Carbon Future." Science, 367(6473), 30–33. Analysis of the critical minerals required for clean energy and digital technology transitions, including supply chain sustainability challenges. Contextualizes AI hardware mineral requirements within the broader low-carbon technology minerals challenge.

13. Huang, K., et al. (2022). "Semiconductor Supply Chain: Rare Earth Materials and Geopolitics." MIT Energy Initiative. Analysis of rare earth dependencies in semiconductor manufacturing — the upstream material basis of AI computing infrastructure.

14. United Nations. (2022). "Global E-Waste Monitor 2022." UNITAR / ITU. The standard global assessment of e-waste generation, composition, recycling rates, and trends. Essential for understanding the end-of-life dimension of AI hardware.

Green AI Technical Research

15. Tan, M., & Le, Q. V. (2019). "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks." ICML 2019. The research introducing systematic efficient scaling of neural networks, demonstrating that thoughtful architecture design can achieve state-of-the-art accuracy at dramatically lower computational cost. Foundational technical reference for Green AI.

16. Sanh, V., Debut, L., Chaumond, J., & Wolf, T. (2019). "DistilBERT, a Distilled Version of BERT: Smaller, Faster, Cheaper and Lighter." arXiv preprint arXiv:1910.01108. The paper introducing knowledge distillation for BERT, demonstrating 40% size reduction with 97% accuracy retention — a key technical result for the efficient AI agenda.

17. Fedus, W., Zoph, B., & Shazeer, N. (2022). "Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity." Journal of Machine Learning Research, 23(120), 1–39. The influential paper on sparse mixture-of-experts architectures that achieve large model capacity with lower per-token computation — the architectural approach underlying some of the most computationally efficient frontier models.

Justice and Regulatory Dimensions

18. Brevini, B. (2022). Is AI Good for the Planet? Polity Press. Critical assessment of AI's environmental impact from a political economy perspective, including attention to global justice dimensions. Accessible and direct.

19. European Commission. (2024). "EU AI Act." Official Journal of the European Union. The full text of the AI Act, including environmental sustainability provisions. Available at eur-lex.europa.eu.

20. Dodge, J., Prewitt, T., des Combes, R. T., Odmark, E., Schwartz, R., Strubell, E., ... & Luccioni, S. (2022). "Measuring the Carbon Intensity of AI in Cloud Instances." ACM FAccT 2022. Research on measuring actual carbon intensity of AI cloud computing by time and location, providing the technical methodology for more accurate carbon accounting than average-grid approaches. Important for organizations seeking to measure their AI inference carbon footprint.