Making AI Less "Thirsty": Uncovering and Addressing the Secret Water Footprint of AI Models

Pengfei Li,Jianyi Yang,Mohammad A. Islam,Shaolei Ren

2023-10-30

Abstract:The growing carbon footprint of artificial intelligence (AI) models, especially large ones such as GPT-3, has been undergoing public scrutiny. Unfortunately, however, the equally important and enormous water (withdrawal and consumption) footprint of AI models has remained under the radar. For example, training GPT-3 in Microsoft's state-of-the-art U.S. data centers can directly evaporate 700,000 liters of clean freshwater, but such information has been kept a secret. More critically, the global AI demand may be accountable for 4.2 -- 6.6 billion cubic meters of water withdrawal in 2027, which is more than the total annual water withdrawal of 4 -- 6 Denmark or half of the United Kingdom. This is very concerning, as freshwater scarcity has become one of the most pressing challenges shared by all of us in the wake of the rapidly growing population, depleting water resources, and aging water infrastructures. To respond to the global water challenges, AI models can, and also must, take social responsibility and lead by example by addressing their own water footprint. In this paper, we provide a principled methodology to estimate the water footprint of AI models, and also discuss the unique spatial-temporal diversities of AI models' runtime water efficiency. Finally, we highlight the necessity of holistically addressing water footprint along with carbon footprint to enable truly sustainable AI.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The paper primarily focuses on the significant consumption of water resources by artificial intelligence (AI) models during training and operation, and attempts to address the following core issues:

1. **Revealing the Hidden Water Footprint of AI Models**: While society currently pays considerable attention to the carbon footprint of AI models, there is little focus on their water footprint. The paper aims to reveal the substantial amount of water resources consumed during the training and deployment of AI models, particularly in data center cooling and electricity production.

2. **Proposing Estimation Methods**: The paper proposes a systematic method to estimate the water footprint of AI models, including direct water use (i.e., on-site cooling water) and indirect water use (i.e., water resources used for electricity generation). This method allows for a more accurate assessment of water efficiency across different regions and time points.

3. **Emphasizing the Importance of Water Resources**: As global water scarcity becomes increasingly severe, the paper emphasizes that AI models should take on social responsibility and reduce their impact on water resources. By optimizing training and deployment strategies, water consumption can be reduced without sacrificing performance.

4. **Considering Both Water and Carbon Footprints**: The paper points out that water and carbon footprints are complementary rather than substitutive in understanding environmental impact. Optimizing carbon efficiency does not necessarily improve water efficiency, so it is essential to focus on both aspects to achieve truly sustainable AI development.

In summary, the paper aims to raise awareness of the water footprint of AI models and provide a scientific method to assess and reduce this impact, thereby promoting more environmentally friendly and sustainable AI technology development.