AI Chatbots: Thirsty for Attention? Unveiling Water Footprint!
ChatGPT, a revolutionary AI chatbot, has transformed communication since its debut seven months ago.
Critics voice concerns about generative AI, but the real threat lies in the massive amount of water these bots consume
Researchers from the University of Colorado Riverside and the University of Texas Arlington reveal that training GPT-3 in Microsoft's US data centers consumes 700,000 liters of freshwater, equivalent to producing 370 BMW cars or 320 Tesla electric vehicles.
For every 20-50 questions answered, ChatGPT drinks 500ml of water, raising concerns due to its vast user base of over 100 million.
Google's AI chatbot, Bard, also contributes to water consumption, with global data centers guzzling 15 billion gallons of water in 2022.
The trade-off between carbon efficiency and water efficiency poses a challenge for data centers, requiring a balance in engineering solutions.
"Spatio-temporal diversity" is a proposed solution to reduce water footprint by scheduling AI model training and inference in different places and times.
Shifting data centers to colder countries or underwater is another potential solution, consuming less energy and water.
Google invests in renewable energy and collaborates with communities to mitigate water usage impacts, while Microsoft aims to cut data center water use by 95% by 2024 and plans to submerge servers in fluorocarbon-based liquid baths.
As climate change intensifies, finding sustainable ways to cool data centers becomes essential to address water scarcity issues.