The emergence ofgenerative artificial intelligence (AI) alerts experts about the water consumption of technology giants. So much so that some are calling for strong measures to limit “ ecological impact » of AI.
AI drives up water consumption in data centers
With the boom in generative AI, big tech’s water consumption continues to rise. In 2022, Microsoft increased its water consumption by 34%, Google by 22% and Meta by 3%. According to Shaolei Ren, a professor at the University of California, achieving between 10 and 50 responses to ChatGPT, powered by GPT-3.5, would be equivalent to “ drink » a 500 ml bottle of water.
Growing demand for AI is likely to lead to an increase in water extraction from 4.2 to 6.6 billion cubic meters by 2027, researchers estimate. This is equivalent to half the amount consumed by the UK each year. “ Within a few years, large artificial intelligence systems will likely need as much energy as entire countries “, warns Kate Crawford, AI specialist and professor at the University of Southern California.
The problem is all the more serious in a context of increasing fresh water scarcity, worsening prolonged droughts and rapid aging of public water supply infrastructure “.
Calls for large-scale collaboration
The companies concerned recognize the situation. “ We recognize that training large models can be water intensive, and that’s one of the reasons we’re constantly working to improve efficiency », indicated OpenAI, whose models work using Microsoft servers, at Financial Times.
Data center operators aim to reinject more water than they consume by 2030 into systems such as aquifers. The authorities are also aware of the problem. At the beginning of February, American Democratic elected officials presented a law on the environmental impacts of artificial intelligence. It includes, in particular, the creation of standards to assess the effects of technology on the environment.
But this is not enough according to experts. “ To truly address the environmental impacts of AI, a multi-faceted approach including the AI industry, researchers and policymakers is needed. », recommends Kate Crawford. It also calls on AI providers to collaborate with “ specialists in social and environmental sciences to guide technical designs towards greater ecological sustainability “.
In addition to water, data centers also require large energy capacities. By 2026, electricity consumption of data centers in Europe will increase by 50%, according to the International Energy Agency (IEA).
Source : The Financial Times, Nature
7