Will Generative AI exacerbate the climate crisis? ADOBE Stock Photo.

Dear Commons Community,

In a letter to “Science”, three Chinese scientists, Qiong Chen¹, Jinghui Wang², and Jialun Lin¹∗, raise concerns about the effects of AI expansion on climate. They comment that the explosion of generative AI is exacerbating the global climate crisis. As reported by Science.

“The economic benefits of generative artificial intelligence (AI) could reach US$7.9 trillion annually (¹). The emergence of groundbreaking generative AI tools has spurred development (²). However, the explosion of generative AI is exacerbating the global climate crisis. The scientific community, industry, and policy-makers must urgently address its effects.

ChatGPT, a natural language processing tool released by OpenAI in 2022 (³), has sparked a global wave of large language models, driving a surge in demand for intelligent computing power. Concurrently, climate challenges continue unabated. The combined electricity consumption of the AI and cryptocurrency industries reached 460 terawatt-hours (TWh) in 2022, accounting for about 2% of the global total energy consumption. Consumption is projected to reach 1000 TWh by 2026 (⁴). In 2024, increased energy consumption by data centers drove a surge in carbon emissions (⁵). As the generative AI industry expands, its electricity consumption is expected to grow rapidly (⁵).

The energy consumption of generative AI stems from the huge computational power required during model training and the resources expended in responding to user queries (⁶). ChatGPT will likely consume more than half a million kilowatt-hours of electricity daily to handle about 200 million user requests (⁷). The latest model, OpenAI o1, incurs even greater carbon emissions when tackling complex reasoning tasks (⁸). High-performance hardware, such as graphics and tensor processing units, requires additional electricity to power cooling systems and consume substantial amounts of water (⁹).

Principles such as the Montreal Declaration for a Responsible Development of Artificial Intelligence (¹⁰) have been widely accepted in pursuit of sustainable generative AI. Since 2023, internet technology companies around the world have invested heavily in purchasing clean energy (¹¹). However, greenhouse gas emissions have still risen substantially because of the surge in generative AI research and development (⁴).

Researchers must explore methods to reduce generative AI’s energy consumption. For example, measuring AI’s total footprint on the environment throughout its life cycle can pinpoint opportunities to reduce its negative effects. Transferring knowledge from a complex model to a simpler, smaller, and more efficient model can reduce computational costs while maintaining model performance. Transforming highdimensional data into low-dimensional data while maintaining important information can help simplify the data, reduce computational costs, and improve model performance (⁶). In addition, enterprises need to publicly disclose the actual energy consumption of their generative AI projects. Nations should regulate generative AI technologies and refine sustainable generative AI regulations. Strengthening regional communication will facilitate the creation of international standards and norms as well as knowledge sharing and innovation (¹²). Countries should also bolster support for renewable energy sources and raise public awareness of generative AI’s environmental impact through education and advocacy.

∗Corresponding authors.

¹School of Biomedical Information and Engineering, Hainan Medical University (Hainan Academy of Medical Sciences), Haikou, China.

²Modern Education Technology Center, Hainan Medical University (Hainan Academy of Medical Sciences), Haikou, China.

References

1. “The economic potential of generative AI: The next productivity frontier” (McKinsey & Company, 2023); https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-ai-the-nextproductivity-frontier#introduction.
2. “Generative AI: Steam engine of the fourth industrial revolution?” (World Economic Forum, 2024); https://www.weforum.org/events/worldeconomic-forum-annual-meeting-2024/sessions/industry-applications-of-generative-ai/.
3. “Introducing ChatGPT” (OpenAI, 2022); https://openai.com/index/chatgpt/.
4. “Electricity 2024: Analysis and forecast to 2026” (International Energy Agency, 2024); https://www.iea.org/reports/electricity-2024.
5. “Powering the AI revolution” (Morgan Stanley, 2024); https://www.morganstanley.com/ideas/ai-energy-demand-infrastructure.
6. J. An, W. Ding, C. Lin, Nature 615, 586 (2023).
7. A. de Vries, Joule 7, 2191 (2023).
8. M. Zeitlin, “What does OpenAI’s new breakthrough mean for energy consumption?” (HEATMAP, 2024); https://heatmap.news/technology/openai-o1-energy.
9. Z. Wang, “How AI consumes water: The unspoken environmental footprint” (Deepgram, 2024); https://deepgram.com/learn/how-ai-consumes-water.
10. “Montreal Declaration for a responsible development of artificial intelligence” (Université de Montréal, 2018); https://www.montrealdeclaration-responsibleai.com.
11. K. Harrison, “Amazon is top green energy buyer in a market dominated by US” (BloombergNEF, 2024); https://about.bnef.com/blog/amazon-is-top-greenenergy-buyer-in-a-market-dominated-by-us/.
12. R. Raper et al., Sustainability 14, 4019 (2022).

Tony

This entry is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license.