Tag: artificial-intelligence

Identifying Trustworthiness Challenges in Deep Learning Models for Continental-Scale Water Quality Prediction

Xiaobo Xia, Xiaofeng Liu, Jiale Liu, Kuai Fang, Lu Lu, Samet Oymak, William S. Currie, Tongliang Liu

Water quality is foundational to environmental sustainability, ecosystem resilience, and public health. Deep learning offers transformative potential for large-scale water quality prediction and scientific insights generation. However, their widespread adoption in high-stakes operational decision-making, such as pollution mitigation and equitable resource allocation, is prevented by unresolved trustworthiness challenges, including performance disparity, robustness, uncertainty, interpretability, generalizability, and reproducibility. In this work, we present a multi-dimensional, quantitative evaluation of trustworthiness benchmarking three state-of-the-art deep learning architectures: recurrent (LSTM), operator-learning (DeepONet), and transformer-based (Informer), trained on 37 years of data from 482 U.S. basins to predict 20 water quality variables. Our investigation reveals systematic performance disparities tied to process complexity, data availability, and basin heterogeneity. Management-critical variables remain the least predictable and most uncertain. Robustness tests reveal pronounced sensitivity to outliers and corrupted targets; notably, the architecture with the strongest baseline performance (LSTM) proves most vulnerable under data corruption. Attribution analyses align for simple variables but diverge for nutrients, underscoring the need for multi-method interpretability. Spatial generalization to ungauged basins remains poor across all models. This work serves as a timely call to action for advancing trustworthy data-driven methods for water resources management and provides a pathway to offering critical insights for researchers, decision-makers, and practitioners seeking to leverage artificial intelligence (AI) responsibly in environmental management.

Comments:Accepted by Nexus (Cell Press). 61 pages, 24 figures, 2 tables
Subjects:Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
Cite as:arXiv:2503.09947 [cs.LG]
 (or arXiv:2503.09947v3 [cs.LG] for this version)
 https://doi.org/10.48550/arXiv.2503.09947Focus to learn more

Submission history

From: Xiaobo Xia [view email]
[v1] Thu, 13 Mar 2025 01:50:50 UTC (9,275 KB)
[v2] Sun, 15 Jun 2025 11:47:43 UTC (9,274 KB)
[v3] Sat, 25 Oct 2025 01:57:51 UTC (22,023 KB)

CLICK HERE TO READ MORE

https://arxiv.org/abs/2503.09947?

ChatGPT ‘drinks’ a bottle of fresh water for every 20 to 50 questions we ask, study warns

By Rosie Frost

Published on 20/04/2023 – 13:13 GMT+2•Updated 21/04/2023 – 8:19 GMT+2

There is still very little research into the environmental impact of AI.

For every 20 to 50 questions ChatGPT is asked, it “drinks” a bottle of water according to new research.

OpenAI’s AI chatbot has soared in popularity thanks to its uncanny ability to accurately answer our questions. After being made available to the public for testing last November, it has been used for everything from poetry to coding and even answering exam questions meant for medical students.

But despite billions of users around the world, there’s still very little research on what environmental impact AI like this is having.

A new study from researchers at the University of Colorado Riverside and the University of Texas Arlington in the US gives some insight into its water consumption. The paper has not yet been peer-reviewed and has been shared ahead of its publication.

Its authors say that the “water footprint” of these AI models has so far “remained under the radar”.

How do AI chatbots use water?

The study’s water consumption figures refer to fresh clean water used by data centres to generate electricity and cool the racks of servers.

Most of the prominent chatbots’ cloud computing relies on thousands of servers inside data centres around the world. Computers are used to train algorithms known as ‘models’ to perform tasks like answering questions from users.

During a 20 to 50 question conversation with the AI chatbot, they estimate it could “drink” a 500ml bottle of water.

A bottle of water might not seem like much but ChatGPT has billions of users.

“While a 500ml bottle of water might not seem too much, the total combined water footprint for inference is still extremely large, considering ChatGPT’s billions of users,” the researchers say.

Scientists believe that while training GPT-3 alone, Microsoft may have consumed an incredible 700,000 litres of water.

More complex next-generation models like GPT-4 could consume even more during training, they say, but there is hardly any publicly available data with which to make an accurate estimate.

Companies need to ‘take responsibility and lead by example’

The study’s authors have urged companies to “take social responsibility and lead by example” to address their water footprint in the face of global shortages.

Earlier this year, a landmark report on water economics said that demand is expected to outstrip the supply of fresh water by 40 per cent by the end of this decade. The report from the Global Commission on the Economics of Water said that all industries need to overhaul their wasteful practices.

The study’s authors are also asking for more data transparency so that the environmental impact of these AI systems can be better assessed through research like this.

“AI models’ water footprint can no longer stay under the radar – water footprint must be addressed as a priority as part of collective efforts to combat global water challenges,” they conclude.

OpenAI didn’t immediately respond to Euronews’ request for comment and Microsoft declined to comment on the study.

CLICK HERE FOR MORE INFORMATION

https://www.euronews.com/green/2023/04/20/chatgpt-drinks-a-bottle-of-fresh-water-for-every-20-to-50-questions-we-ask-study-warns