Subscribe
There’s much to laud about the triumph of AI tools such as OpenAI’s ChatGPT in rapidly automating the business topography. But concerns prevail over their advances in artificial intelligence.
ChatGPT has been used for anything from creative writing to coding to even responding to exam questions since it was made accessible to the public for testing last November.
However, the water footprint of tools like ChatGPT has primarily been overlooked. Massive amounts of water go into maintaining the data centres of such AI tools.
For answering every 20-50 questions, ChatGPT ‘drinks’ half a litre of water, a new research titled Making AI Less ‘Thirsty:’ Uncovering and Addressing the Secret Water Footprint of AI Models has revealed.
“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,” scientists at the University of California, Riverside, said.
AI models should take social responsibility and lead by example in the collective efforts to combat the global water scarcity challenge by cutting their own water footprint, the scientists wrote in the study.
It required 700,000 litres of freshwater — about the same amount of water used in the manufacture of about 370 BMW cars or 320 Tesla electric vehicles — to train GPT-3 alone, the researchers noted.
For the recently unveiled GPT-4 AI system, which has a bigger model size, researchers predicted these statistics might grow by “multiple times”.
However, they emphasised that there is practically any publicly available data that can be used to calculate the GPT-4 water footprint reasonably.
Even though online activities like sharing, downloading and utilising ChatGPT take place digitally, the real data is physically kept in sizable data centres. These data centres produce a lot of heat, which needs cooling systems to prevent equipment failure. These centres frequently utilise water-intensive evaporative cooling towers to assist with cooling.
In order to prevent corrosion and the formation of microorganisms, the water used in this procedure must also be pure freshwater. In addition to cooling systems, data centres also require a significant amount of water for power generation.
Fortunately, AI training has scheduling flexibility. Unlike web search or YouTube streaming, which must be processed immediately, AI training can be done at almost any time of the day, said Shoalei Ren, an associate professor of electrical and computer engineering and the corresponding author of the study.
It is important to address the water use from AI because it is a fast-growing segment of computer processing demands, Ren said.
“To avoid wasteful water usage, a simple and effective solution is training AI models during cooler hours when less water is lost to evaporation,” Ren said in a university press release.
AI training is like a very big lawn that needs lots of water for cooling, Ren said. “We don’t want to water our lawns during the noon, so let’s not water our AI at noon either.”
“It is truly a critical time to uncover and address the AI model’s secret water footprint amid the increasingly severe freshwater scarcity crisis, worsened extended droughts, and quickly aging public water infrastructure,” the paper read.
Concerns over water usage have been amplified by climate change and preexisting droughts.
In 2022, nearly 300 million people were in the grip of drought in Africa, Europe, North America and Asia. East Africa was reeling under its worst drought in four decades; nearly half of the US is dry; and countries like France and Portugal were enduring the worst drought on record, Down To Earth had reported.