Will AI cost us the earth?

In 2023, the federal Department of Education created the Australian Framework for Generative Artificial Intelligence in Schools, seeking to guide the use of AI in an ‘ethical and responsible’ way.

Developed in consultation with the states and relevant agencies, the focus was on embracing AI in a way that benefited students and the school communities, while ensuring its use was fair, safe and accountable.

Read the latest print edition of School News online HERE.

But while much of the debate and discussion around AI has been about how to prevent plagiarism and the integrity of student-generated work, one aspect that seemingly hasn’t been given much consideration is the environmental cost of such technology.

Putting aside the monetary costs of developing AI tools (Forbes reports that the technical cost of the latest edition of ChatGPT was up to $US78 million), the unseen costs might end up costing us so much more.

Environmental costs of AI

While it may seem innocuous—or even ‘free’—to type a question into ChatGPT, there are hidden costs we likely don’t consider.

After years of silence on the topic, OpenAI chief executive Sam Altman was reported as saying during the recent World Economic Forum’s annual meeting in Davos that the future of AI will consume ‘vastly more’ power than people expected.

Researcher and Professor Kate Crawford wrote in Nature that she’s glad he has finally admitted it after “consistent downplaying and denial about the AI industry’s environmental costs since… 2018”.

“It’s estimated that a search driven by generative AI uses four to five times the energy of a conventional web search. Within years, large AI systems will likely need as much energy as entire nations,” she adds.

With the demand for AI increasing, there is a corresponding increasing demand for land, power and water, all essential for the data centres required to house the enormous infrastructure, and research by Li et al. from the University of California and The Washington Post has calculated the amount of water and electricity used in AI data generation.

Water is used in three ways: it is required for cooling the servers, consumed for electricity generation and also used in the manufacturing process of servers.

“Training the GPT-3 language model in Microsoft’s state-of-the-art U.S. data centres can directly evaporate 700,000 litres of clean freshwater, but such information has been kept a secret,” the researchers wrote in their article, Making AI Less Thirsty.

As a comparison, freshwater use by Google in 2023 rivalled that of PepsiCo, but at least Pepsi consumers could see (and drink) the outcome of the water use.

The newer GPT-4 model, released in 2024, is expected to consume substantially more water and energy than the now outdated GPT-3 model, with the researchers estimating a single 100-word email written with GPT-4 would use around 519ml water, or just less than a standard bottle of water.

“Critically, the global AI demand is projected to account for 4.2 to 6.6 billion cubic metres of water withdrawal in 2027, which is more than the total annual water withdrawal of Denmark or half the United Kingdom.”

The researchers conclude: “If not properly addressed, AI’s water footprint can potentially become a major roadblock to sustainability and create social conflicts as freshwater resources suitable for human use are extremely limited and unevenly distributed. AI’s water footprint can no longer stay under the radar and must be addressed as a priority as part of the collective efforts to combat global water challenges.”

Australia is no stranger to water conservation. After all, we are the land of sprinkler days, two-minute showers and turning the tap off when you brush your teeth. Knowing what we do now about the incredible water and power demands of generative AI, it will be interesting to see whether our love affair with ChatGPT continues.

Do you, your students and/or your school use generative AI? Take our anonymous survey here! Results will be used to inform our upcoming term two special report on AI in education.