AI and Sustainability: The true cost of data centers

The era of artificial intelligence (AI) has brought impressive innovations that have reshaped entire sectors but also raised deep concerns about the environmental and social impacts associated with the rapid growth of data centers. The data on energy and water consumption are alarming. For instance, a study conducted by The Washington Post in conjunction with researchers from the University of California, Riverside, revealed that using chatbots like GPT-4 can consume up to half a liter of water per generated email. If one in every ten workers in the U.S. used the chatbot twice a week, in a year, the total water consumption would be approximately 870 million liters — equivalent to the domestic consumption of the entire population of Rhode Island, on the U.S. East Coast, for three days. Additionally, data centers in cities like Santa Clara, located in the heart of Silicon Valley, California, and Goodyear, Arizona, exemplify the magnitude of environmental challenges. In Santa Clara, over 50 centers consume 60% of local energy, while in Goodyear, data centers consume over 50 million gallons of water annually.

This scenario is repeated globally, with regions like Santiago, Chile, and Guizhou, China, suffering the impacts of expanding data infrastructures. In Santiago, the water crisis exacerbated by data center consumption raises concerns about long-term sustainability. In Guizhou, economic growth driven by big data projects faces challenges in ensuring real and sustainable benefits for the local population. Such examples underscore the need for a more comprehensive ethical analysis that goes beyond energy efficiency and includes environmental and social justice.

Environmental justice and the 'Third Wave' of AI Ethics?

The "Third Wave" of AI ethics, as discussed by Aimee van Wynsberghe, Alexander von Humboldt Professor of Applied Ethics in Artificial Intelligence in Germany, proposes a more comprehensive approach that considers not only traditional aspects such as privacy and transparency but also socio-environmental impacts on vulnerable communities. This new perspective seeks to expand AI ethics to include sustainability and environmental justice concerns, highlighting the importance of assessing the broader implications of its development and implementation. Cases such as Memphis, U.S., illustrate the disproportionate impact of data center construction in historically neglected and low-income regions. The construction of a large data center in the area raised concerns about the burden on areas already affected by pollution and health issues, further exacerbating social and economic inequalities.

This phenomenon is often compared to "modern colonization," where less developed regions or those with less decision-making power are exploited to sustain the technological growth of major corporations. In Chile, the expansion of data centers in Santiago has worsened a water crisis that has lasted over a decade, while in China, the province of Guizhou faces challenges in balancing rapid technological growth with sustainable and self-sufficient economic development. Integrating social and environmental justice into impact assessments is essential for AI ethics to be truly sustainable and inclusive.

Sustainability and transparency practices in question?

Although many tech companies promote ambitious sustainability goals, practices often do not match the rhetoric. A common and controversial practice is the use of Renewable Energy Certificates (RECs). These certificates allow companies to purchase renewable energy credits without necessarily using clean energy in their operations. This creates a false impression of sustainability, allowing companies to claim carbon neutrality while continuing to use non-renewable energy sources.

The disparity between "market-based" and "location-based" emission metrics exemplifies this lack of transparency. The "market-based" approach allows companies to present reduced emissions through the purchase of RECs, while the "location-based" approach reveals the real operational impact. For instance, in 2022, Meta reported scope 2 emissions (indirect emissions from purchased energy, such as electricity from a utility) of 273 tons of CO2, but "location-based" measurements indicated emissions exceeding 3.8 million tons. This practice highlights the need for greater transparency and commitment to real sustainability, not just a superficial image of environmental responsibility.

The EU AI Act and the need for improved governance?

The European Union's AI Act represents an important step in regulating AI use and development but still has gaps, particularly regarding sustainability and environmental impacts. The analysis by Philipp Hacker, Professor of Law and Ethics of the Digital Society at the European New School of Digital Studies in Poland, suggests that while the EU AI Act includes risk assessments, it does not comprehensively address indirect greenhouse gas emissions nor establish clear transparency mechanisms for energy consumption during AI use. Improvements in this regard are necessary for the regulation to be more effective in promoting sustainable AI.

Incorporating parameters of environmental and social justice in AI impact assessments is essential. Examples of initiatives in Frankfurt, Germany, which limit the construction of new data centers, show that it is possible to create policies that balance technological development with sustainability and environmental protection. These policies can serve as a model for other regions seeking to mitigate the adverse impacts of data centers on local communities.

Pathways for Sustainable AI Governance?

Inspired by the U.S. Environmental Protection Agency's Energy Star efficiency rating program, the AI Energy Star project aims to apply a similar metric to AI models. This system would help consumers and developers choose models with lower energy consumption. The original Energy Star program has reduced more than 4 billion tons of greenhouse gas emissions over 30 years. The idea is that something similar could be done with AI models, promoting the selection of more sustainable options. Initiatives like the AI Energy Star project represent an important step in promoting energy efficiency in AI models, but that alone is not enough. It is crucial to create governance frameworks that incorporate environmental and social justice as core pillars.

The experience of regions like Santiago, Guizhou, and Memphis underscores the importance of policies that go beyond superficial sustainability goals and require genuine transparency and accountability from major corporations. Only with stronger regulations and inclusive governance can we achieve a balance between technological progress and the protection of communities and the environment.

References

Bolte, L., & van Wynsberghe, A. (2024). Sustainable AI and the Third Wave of AI Ethics: A Structural Turn. AI and Ethics, 1-12. https://doi.org/10.1007/s43681-024-00522-6

Ebert, K., Alder, N., Herbrich, R., & Hacker, P. (2024). AI, Climate, and Regulation: From Data Centers to the AI Act. arXiv. https://doi.org/10.48550/arXiv.2410.06681

Falk, S., & van Wynsberghe, A. (2023). Challenging AI for Sustainability: What Ought It Mean? AI and Ethics. https://doi.org/10.1007/s43681-023-00323-3

Los Angeles Times - California Data Centers and Clean Energy Goals. Link

Nature - Environmental Impact of AI and Data Centers. Link

Noema - China's Hinterland as a Critical Datascape. Link

Rest of World - Data Centers and Environmental Issues. Link

The Atlantic - AI Water and Climate Impact. Link

The Guardian - Data Center Gas Emissions and Tech. Link

The Washington Post - Energy AI Use: Electricity and Water Data Centers. Link

Time - Elon Musk's xAI and Grok in Memphis. Link

van Wynsberghe, A. (2021). Sustainable AI: AI for Sustainability and the Sustainability of AI. AI and Ethics, 1(213-218). https://doi.org/10.1007/s43681-021-00043-6