AI increases water consumption in big techs – 02/27/2024 – Tech

The world’s biggest technology companies have substantially increased their use of water to cool data centers, raising concerns about the environmental impact of the boom in generative artificial intelligence.

Microsoft, Google and Meta, owner of WhatsApp and Instagram, have increased water use in recent years, with millions of users connecting to their online services.

Academics suggest that demand for AI would increase water withdrawals — where it is removed from underground or surface sources — to between 4.2 billion and 6.6 billion cubic meters by 2027, or about half the volume consumed by the UK per year. each year.

Researchers at the University of California in the US wrote in a paper cited in the journal Nature last week that it was a “critical time to discover and map the steps of AI models amid the increasingly serious freshwater scarcity crisis, worsened by prolonged droughts and lagging public water infrastructure”.

This concern has grown over the past year as major technology companies compete to launch products that use generative AI, achieve language models capable of processing and generating large amounts of text, numbers and other data.

These models require a large amount of computer power to operate, requiring the use of huge server farms that use ice water to cool the equipment by absorbing heat from the air. Some of the water evaporates in the cooling process, while some can be reused.

Water is used in most forms of fuel and power generation, for example to pump oil and gas or to produce steam in thermal power plants. It also evaporates from the surface of reservoirs used for hydroelectric energy.

In 2022, the most recent period with data available, Microsoft increased its water consumption by 34%, Google by 22% and Meta by 3% as a result of increased data center usage.

These companies aim, by 2030, to return more water to systems — through the use of aquifers, for example — than they consume, financing work to improve irrigation infrastructure or restoring wetland systems.

A month before OpenAI, owner of ChatGPT, completed training its most advanced model (the GPT-4), a data center in West Des Moines, Iowa (USA), consumed 6% of the district’s water, according to with a lawsuit filed by its residents.

Shaolei Ren, an associate professor at UC Riverside, suggested that asking for between 10 and 50 responses from ChatGPT, running on its previous model GPT-3, would be equivalent to “drinking” a 500 ml bottle of water, depending on when and where it is deployed.

GPT-4 has more parameters and requires more power, so it will likely use more water, Ren said. Detailed information about the model’s energy usage was not available.

Researchers have called for more comprehensive data and greater transparency from AI companies, such as making available how much they consume, for example, search engines and AI services.

“We recognize that training large models can require a lot of water, which is one of the reasons we are constantly working to improve efficiency,” said Open AI when asked about the topic. “We also believe that large language models can be useful in accelerating scientific collaboration and the discovery of climate solutions.”

Microsoft stated that “Currently, the AI ​​division represents only a fraction of the electricity used by data centers, which collectively use about 1% of the global electricity supply. How much this increases and how the growth of AI affects this race will depend on many factors “. Google declined to comment.

Kate Crawford, a research professor at USC Annenberg who specializes in the social impacts of AI, said: “Without greater transparency and more reporting on the subject, it is impossible to track the true environmental impacts of AI models. And that’s important in a At a time when many parts of the planet are experiencing deep and prolonged droughts, and drinking water is already a scarce resource.”

Therefore, she warned about caution that must be taken. “We don’t want to blindly use generative AI tools without knowing their true impacts at a time when the planet is already facing a climate crisis,” she said.