The untapped power of AI to solve complex climate issues is overshadowed by energy fears
Editor’s note: This is the latest in our Emerging Voices series featuring work by new writers including university students on issues involving climate and capital. -Barclay Palmer
As artificial intelligence gobbles up more electricity, it will significantly worsen our climate crisis, right? Not so fast. While wildfires rage and rising seas batter coastlines, the immense power of AI may lie untapped, overshadowed by fears of its energy use. What if AI is not the enemy but an emerging ally in the transition toward a sustainable world?
Let’s check some facts. Global AI servers do consume massive amounts of electricity, with newly manufactured devices consuming 85 to 134 terawatt hours annually by the year 2027, according to a study by Digiconomist founder Alex de Vries. That’s about as much as the entire Netherlands. Statista projects the AI market will grow a compounding 28.46% annually through 2030. That will require a lot of energy. But for perspective, it’s less than half a percent of global consumption, de Vries calculates.
What if AI can bring solutions and efficiencies that far outweigh its demand?
Accelerating climate solutions
Most AI applications don’t require a great deal of energy to operate. “A lot of models…can run on a laptop,” Priya Donti, an MIT professor and Climate Change AI co-founder, said at this year’s ClimateTech conference. “While the amount of computing needed for each of the largest ML training runs is growing rapidly,” Donti and others wrote in Nature.com, “the extent to which efficiency improvements in computing (doubling every 2–3 years), can limit overall ML-related energy use in data centres is uncertain.”
AI’s true energy demand will depend on how widely AI is adopted, de Vries noted, adding that ChatGPT use is already demanding more energy than its data gathering phase did. “It would be wise to keep Jevons’ paradox in mind. Benefits such as cost reductions in certain goods and services may lead to rebound effects due to increased consumer demand, increasing resource consumption at a total level rather than decreasing it,” de Vries said. “This effect has been observed throughout all technological change and automation over the past 160 years. For example, we now have more jobs than ever before, despite all those advances.”
AI brings us tools that can accelerate our search for climate solutions.
More importantly, AI brings us tools that can accelerate our search for climate solutions. In “Tackling Climate Change With Machine Learning,” a paper Donti co-authored with 21 fellow experts to advocate the climate applications of machine learning, she notes we now have more climate-related data than humans can fully analyze.
AI can help with that. One example involves training AI models to make buildings more energy efficient. “By considering various factors such as solar radiation and weather forecasts,” the European energy company Vattenfall says, its system “can stay one step ahead and predict the amount of heat the different parts of the building need to maintain the right temperature with the least possible heat supply.”
Vattenfall said its system reduced energy consumption in two Swedish apartment buildings by about 20 percent. AI experts say such technology can improve efficiency in nearly all buildings. Given that buildings consume about 30% of global “final energy consumption” and 26% of global energy-related emissions, according to the International Energy Agency (IEA), such reductions could take a significant step toward decarbonization.
More specifically, heating spaces inside buildings uses about one third of energy consumed by U.S. building operations, the U.S. Energy Information Administration (EIA) reports. Assuming a similar figure globally, space heating accounts for about 10% (ie. one third of 30%) of the world’s “final” electricity use.
If Vattenfall’s 20 percent reduction in Sweden were applied globally, AI could save nearly 2% of the world’s annual electricity. That’s equivalent to reducing about 457 terawatt hours of electricity, according to the International Energy Agency (IEA) — more than 3 times the annual consumption of the entire global AI sector on the conservative end as estimated by the de Vries study cited in The New York Times and The Verge. And that’s only one of the many climate-related AI applications.
Better data
Another example of the power AI offers in supporting efforts against climate change is its ability to analyze complex imagery accurately. Open Climate Fix, a non-profit using computing to reduce carbon emissions, is training AI models to read cloud cover from satellite data to form estimates that better match electricity demand with short-term solar energy production.
With improved supply-demand estimates, fewer natural gas generators must be kept on for “spinning reserves,” or excess capacity needed to compensate for power fluctuations. Open Climate Fix says its AI forecasts could save £1 million to £10 million ($1.27 million to $12.7 million) and about 100,000 tonnes of CO2 per year in the UK alone.
AI forecasts could save £1 million to £10 million ($1.27 million to $12.7 million) and about 100,000 tonnes of CO2 per year in the UK alone.
University of Leeds researchers have also trained AI models to measure changes in icebergs 10,000 times faster than a human could. Their groundbreaking development uses a neural network that accurately charts Antarctic icebergs in satellite images in just 0.01 seconds. The neural network has proven to be over 99% accurate with radar images from Sentinel-1, regardless of cloud cover. Previously, such data collection required time-consuming manual efforts.
Perfection is the enemy of the good
This AI-collected data helps scientists understand and quantify how much meltwater and nutrients these icebergs release into the ocean, which is important for ocean physics, biology, and maritime operations. These vastly improved methods of data collection might also help give climate change skeptics a new perspective, or better communicate the urgency of our climate crisis to those living near coastlines.
Still, such examples barely scratch the surface. AI provides the tools for precise climate modeling and improved resource efficiencies across nearly all sectors of the economy. While AI demand does challenge our aged and vulnerable grid, it has begun to show its potential for optimizing energy use and understanding our critical situation.
As we stand at a pivotal moment in our environmental crisis, it is essential that we leverage every tool available. That means including AI, not just railing at it.
Featured photo: Icebergs in the Amundsen Sea captured by the Copernicus Sentinel-1 radar satellite mission.
