'May drink a 500 ml bottle of water for 20-50 questions': ChatGPT data centres consumes a lot of water, warns study

A new study has revealed that the water consumption of data centres required to power the artificial intelligence chatbot ChatGPT by billions of users worldwide is "extremely large".

ChatGPT is an AI program that has gained attention for its ability to generate quick and comprehensive answers to a wide range of queries. With over 100 million monthly active users, it is the fastest-growing consumer application in history and it has also cleared a few exams.

Meanwhile, experts noted that these achievements may have come at the expense of the chatbot consuming massive amounts of water.

While earlier studies highlighted the carbon footprint of such AI models, scientists claimed that water usage to run them on a big scale has "remained under the radar".

The new research published as a pre-print on arXiv, suggests that a conversation with an AI chatbot in a single system may "consume" a "500ml bottle of water" for 20-50 questions.

For the research, a new framework was created to know the amount of clean freshwater consumed for bringing out electricity to power data centre servers and for cooling servers to run AI models.

Scientists explained it through an example by saying that the training of GPT-3 alone, Microsoft is said to have consumed 700,000 litres (185,000 gallons) of water, which has the power to produce 370 BMW cars.

Notably, Google’s LaMDA is said to consume a “stunning” amount of water in the order of millions of litres, scientists added.

The researchers recommended companies using AI models to "take social responsibility" and address their own water footprint in response to the world's water problems.

“AI models can, and also 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,” scientists wrote in the study.

“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,” they said.

Researchers also hinted that these numbers may rise by “multiple times” for the recently launched GPT-4 AI system with a larger model size.

Scientists demand an abundant amount of operational data and the efficiency of water usage during the operation of such systems in order to increase transparency of the water footprint of AI models.

Also Read: ChatGPT performs poorly in JEE Advanced; solves only 11 questions in both papers