Climate concerns: Will A.I. Ruin the Planet or Save the Planet?

Steve Lohr

The global experiment in artificial intelligence is just beginning. But the spending frenzy by big tech companies for building and leasing of data centers, the engine rooms for A.I., is well underway. They poured an estimated $105 billion last year into these vast, power-hungry facilities. That spending spree is increasing demand for electricity and raising environmental concerns. A recent headline in The New Yorker called the energy demands of A.I. “obscene.” But there’s another perspective on A.I. and the environment, focusing not on how the technology is made but on what it can do.

A.I. has the potential to help accelerate scientific discovery and innovation in one field after another, lifting efficiency and reducing planet-warming carbon emissions in sectors like transportation, agriculture and energy production.

It’s the rise of so-called generative artificial intelligence. Generative A.I. can do a lot — not only analyze data and make predictions, but also write poetry and computer code, summarize books and answer questions, often with human-level proficiency. And that kind of computing needs a lot of energy. A query to ChatGPT requires nearly 10 times as much electricity as a regular Google search, according to a recent estimate.

Researchers had been working on generative A.I. for years, but it really burst onto the scene in November 2022 when OpenAI introduced ChatGPT, the conversational chatbot that became a sensation. Microsoft has invested more than $13 billion in OpenAI and is racing to include A.I. features in its products. So are Amazon, Google and Meta, the owner of Facebook, Instagram and WhatsApp.

How much will electricity demand increase? There are higher estimates, but experts generally forecast that energy consumption by data centers worldwide will at least double over the next few years. Goldman Sachs has estimated that electricity use by data centers will increase 160 percent by 2030. A recent forecast by the International Energy Agency projected that demand would more than double by 2026.

These predictions are all sizable increases, suggesting sharply higher greenhouse gas emissions from data centers if they obtain their power from fossil fuels like coal and natural gas. But keep in mind: The global electricity sector is huge and varied. Data centers account for about 1 percent to 2 percent of total electricity demand. That share, according to estimates, will increase to 3 percent to 4 percent by 2030.

What’s the case for A.I. as a green technology? Artificial intelligence is a general-purpose tool, experts say, that if used wisely across the economy could reduce greenhouse gas emissions by 5 percent to 10 percent by 2030, according to a study by the Boston Consulting Group that was commissioned by Google.

For example, the technology promises to “give biological design a boost,” said Drew Endy, an associate professor of bioengineering at Stanford University. The result, he said, might well be to turbocharge biology by discovering the right DNA formulas to unlock more efficient, less-polluting agriculture, for example.

A.I. could also radically transform the way we find metals that are critical not only to the tech industry but to the fight against climate change. In one case, it helped find a vast deposit of copper, a key component in electric vehicles, in Zambia. And Zanskar, a start-up in Salt Lake City, is using A.I. to try to improve the success rate of discovering geothermal energy for power plants. About 90 percent of geothermal projects started from scratch fail mainly because they drill in the wrong places, said Carl Hoiland, a co-founder and the chief executive of Zanskar. But A.I., combined with new geologic data sets like satellite and seismic sensor data, could open the door to doubling or tripling the field’s meager success rate.

In theory, that could make a big difference in the fight against global warming. Geothermal is a clean, round-the-clock energy source, but it currently accounts for less than half of 1 percent of the electric power in the United States.

The takeaway: Even though electricity demand from A.I. is expected to at least double in the coming years, the efficiency of the technology could increase at an even higher rate. There is a historical precedent. Consider what happened with cloud computing. There was a surge in energy consumption in the early 2000s. And there were concerns that the increase would continue. But while the computing output of the world’s data centers jumped sixfold from 2010 to 2018, energy consumption rose only 6 percent.

A similar trend, industry analysts say, may well emerge with A.I. “After the mania has calmed down, other incentives kick in,” said Jonathan Koomey, a former scientist at the Lawrence Berkeley National Laboratory who is now an independent researcher. “There is a huge incentive for the industry to become more efficient.”

The big tech companies are working on ways to streamline their software, hardware and cooling systems to reduce electricity consumption in their data centers. They are locating computing facilities in northern countries, pulling in cold outdoor air as a coolant to reduce electricity and water use. And they are investing in alternative energy sources.

If those efforts are successful, and if we’re smart about how we use A.I., it might eventually offer a lot of environmental bang for the buck.