There are no “unique market failures” associated with AI’s energy use, according to the report
Critics have been raising concerns about the energy consumption of the generative AI models that fuel chatbots like ChatGPT and Bard. But we’ve been here before, according to a new report by the Information Technology and Innovation Foundation (ITIF), a nonprofit think tank. Near the peak of the dot-com boom in the 1990s, a Forbes article lamented that “Somewhere in America, a lump of coal is burned every time a book is ordered on-line.”
The authors of that piece projected that within a decade, half of the electric grid would be powering the internet economy. As it turned out, they weren’t even close. The International Energy Agency (IEA) estimates that today’s data centers and data transmission networks “each account for about 1 to 1.5% of global electricity use.”
Generative AI models require vast amounts of computing power to create new content. As with with past technologies, though, many early claims about the energy consumption by AI have proven to be “inflated and misleading,” the ITIF report notes.
So what’s different this time around? AI can mitigate climate change, according to the authors.
It’s hard to create accurate estimates of the energy use and carbon emissions of AI systems over their lifetime because these calculations depend on many factors, including details about chips, cooling systems, data center design, and energy sources.
Most of the larger AI models require more energy than smaller ones. For instance, Google $GOOGL’s power consumption has increased, particularly from its data centers, as its business has grown. The tech giant’s data centers used about 3 terawatt-hours more electricity in 2022 than the year before. But while Google’s overall energy use has climbed, the share going to machine learning has remained constant—at between 10% and 15%.
Also, the energy requirements for inference in AI models—the process of feeding a trained model data so it can make a prediction or solve a task—have generally fallen with each new chip release, according to the ITIF report.
The report details how AI can help reduce energy consumption in several industries, including transportation, agriculture, and energy. The technology can interpret complex climate data from sensors and satellites, such as changing sea levels and rainfall, to create better forecasts and address the risks of climate change. Farmers have also long used AI for precision agriculture, reducing their use of fertilizer and water.
For their part, businesses and governments use AI to operate buildings, roads, and waterways more efficiently. In California, for example, the government has deployed it to detect and quickly respond to wildfires, which reduces carbon emissions from the blazes. Meanwhile, logistics providers use AI to optimize delivery routes, shrinking fuel consumption.
The ITIF report concludes that any activity using energy has an environmental impact, and AI is no exception. But there are no “unique market failures” associated with the technology’s power consumption that would lead to a greater impact than alternative uses, it finds. For example, a kilowatt-hour used for AI is no different than one used for watching TV or microwaving popcorn.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
