BURSTWith the rise of deep learning since 2010, the number of computations needed to train artificial intelligence (AI) models has doubled about every six months. Enabling that growth requires a lot of energy, and in practice most of that comes from fossil fuels.
On the other hand, there are also countless ways AI can help save energy, which raises the pressing question of whether AI will lead to a net increase or decrease in energy consumption and, subsequently, what effect that will have on climate change.
That question was at the center of the session “AI in the Era of Climate Change: Solution or Problem?” at the AAAS Annual Meeting in Boston in February.
Massachusetts Institute of Technology (MIT) professor Priya Donti, co-founder of the global nonprofit Climate Change AI, discussed the various ways AI can have a positive or negative impact on climate change (also the topic of a 2020 article in Nature Climate Change to which she contributed).
One of the ways AI can have a positive impact, Donti said, is through smarter analysis of raw data. Such AI use would include “Assessing greenhouse gas emissions using satellite imagery, for example, or deforestation, crop yields, or the energy efficiency of buildings,” she said.
AI can also be used to improve predictions of solar and wind energy output, and demand for electricity and transportation. AI can optimize complex heating and cooling systems and power grids, and can be used in predictive maintenance. Said Donti, “The German rail operator Deutsche Bahn uses AI fairly widely to detect where there might be issues in the switching infrastructure underlying the rail system. Another example is detecting methane leaks in natural gas infrastructure.”
In the scientific domain, AI can be used to accelerate the discovery of new batteries, electrofuels, or carbon-absorbing materials. Furthermore, she said, AI can replace computationally intensive models of the Earth’s climate or energy systems with surrogate models that perform equally well but with much less computation.
On possible negative impacts of AI on climate change, Donti addressed the notion of “greening the grid” by powering AI applications with green energy (energy derived from naturally replenished resources like solar, wind, and water) so they do not contribute to climate change. “Greening the grid is important,” Donti said, “but it’s not sufficient. When something becomes more efficient, it often causes it to be used more, which means that the overall impact may increase. Further, it’s not just about energy use, but also about the use of water for datacenters, and materials for the underlying hardware.”
Eric Masanet, a professor and Mellichamp Chair in Sustainability Science for Emerging Technologies at the Bren School of Environmental Science & Management at the University of California, Santa Barbara, has been studying the energy consumption of datacenters for 20 years. He said that from the late 1990s until around 2010 the energy consumption of datacenters increased rapidly. Then a new period dawned; until about 2018, both datacenter software and hardware became much more efficient, and energy consumption leveled off.
Credit: Robb Cohen Photography & Video
However, with large language models, we are now in a new era where we lack data, Masanet said. “How many servers are out there? What type of datacenters? What kind of cooling technologies? and so forth . . . . Companies aren’t very transparent. There are a lot of assumptions that go into modelling the energy use of datacenters.”
Masanet, working with colleagues from Lawrence Berkeley National Laboratory, released a study in December 2024 of where AI might go, in terms of its U.S. energy demand. What they found, Masanet said, was that “it is very likely that in the near term, up to about 2028, we are going to see increasing energy use from AI datacenters.” In numeric terms, the study projected power for AI datacenters on an annual basis growing from 4.4% of total U.S. energy use in 2024 to between 6.7% and 12.0% of total U.S. energy consumption in 2028.
Still, the data is far from clear. The Intergovernmental Panel on Climate Change (IPCC), in its report “Climate Change 2022: Mitigation of Climate Change,” assessed various digital solutions that had been proposed to reduce energy consumption, but Masanet showed that their impacts varied widely, ranging from positive to even negative effects.
“We need better data, we need better analysis,” said Masanet, as he and colleagues advocated last year in an article in Nature. In addition, he said, “We need to look not just at energy and carbon, but also at water, land, and air pollution. Most studies are at the national or global level, but there are local communities where datacenters are rapidly expanding, where energy prices are going up and local air quality is being diminished.”
Elke Weber, the Gerhard R. Andlinger Professor in Energy and the Environment at Princeton University, shifted the discussion towards psychology, focusing on how AI could help or hinder with addressing climate change on the level of people as citizens, consumers, and employees. Said Weber, “Even well-informed citizens might not know, in fact do not know, we have research on that, what effective actions they can take in everyday life. There are also cognitive and motivational deficits. We discount future costs and benefits way too much. We don’t like trade-offs. We don’t like risks. We have a single-action bias.”
AI could help make related decisions more rational, Weber said. “AI could help people by providing information at crucial points in time when they make decisions. AI can facilitate an individualized choice architecture, for example in the form of a smart home, or default settings when it comes to consumption decisions.”
In addition, AI could facilitate and coordinate collective action at home or at work, Weber said. “Reminding us that 80% of Americans care about climate change, AI could help overcome our biases,” Weber said. “A little prompt on your smartphone might already help.”
Weber said generative AI can affect personal decision-making in two ways; it can help with coordinating and collective action, but it can also amplify misinformation. “We have to be aware of either hype, or gloom and doom. Let’s employ the summary and integration strengths of AI selectively and wisely and counteract uses of AI that are designed to spread misinformation and prevent rational problem-solving.”
Bennie Mols is a science and technology writer based in Amsterdam, the Netherlands.