Why Are People Blaming ChatGPT for the Los Angeles Fires?
In an alarming reminder of the climate crisis, wildfires raging across Los Angeles this week have destroyed thousands of buildings, killed at least five people, and forced nearly 180,000 residents to evacuate. As of Thursday afternoon, most of the fires remain uncontained, while efforts to battle the largest blaze, in the Pacific Palisades, have been hampered by temporary water shortages.
These shortages fueled a wave of social-media posts about the impact ChatGPT and other forms of artificial intelligence have on the environment: “Somewhere, the men who build AI chatbots are selecting the interiors for the rocketships they will use to leave earth and all of us burning with it,” artist and activist Matt Bernstein wrote in a viral Instagram post. “One search on ChatGPT uses 10x the amount of energy as a Google search. Training one AI model produces the same amount of carbon dioxide as 300 round trip flights between New York and San Francisco and five times the lifetime emissions of a car.”
While some social-media users suggested these AI technologies were related to the L.A. fires, the hydrants in Pacific Palisades actually ran dry due to high demand, as municipal water systems aren’t designed to fight fires as fast-moving and widespread as those in Los Angeles right now.
But as AI technology continues to expand, experts say that its energy and water usage will, too. The servers that power AI chatbots are housed in data centers and generate a lot of heat. (If you’re old enough, you may remember how hot the tower of your desktop computer used to get. Basically, the same thing is happening here.) Some data centers — like Microsoft’s in Iowa, which powered ChatGPT — rely on vast amounts of water to cool the equipment down, while others consume a lot of electricity by setting up large, air-conditioner-like units to keep the system cold enough.
A recent study by the Washington Post and researchers at the University of California broke down how much energy ChatGPT, specifically, uses to perform basic functions. The study found that an AI chatbot requires about 18 ounces of water — slightly more than a bottle — using GPT-4 to generate a single 100-word email. Generating one email per week over the course of a year could use about 27 liters of water, or about one and a half jugs. If one in ten Americans — around 16 million people — did that, ChatGPT would require more than 435 million liters of water. That’s equivalent to the water usage of all Rhode Island households for a day and a half.
AI’s electricity consumption isn’t much better, researchers found. Generating one 100-word email uses 0.14 kilowatt-hours (kWh) of electricity, which is the equivalent of powering 14 LED light bulbs for one hour. If you do that once a week for a year? You’d be consuming 7,500 kWh, or about the same energy more than nine households in D.C. use in one hour. And if 16 million Americans were doing this once a week for a year, it’d use up to 121,517 megawatt hours (MWh), or about the same electricity consumed by every household in D.C. in 20 days.
These findings track with a January 2024 report by the International Energy Agency, made up of representatives from 31 countries, which forecast that, by 2026, the energy consumed by global data centers, artificial intelligence, and cryptocurrency could more than double compared to 2022 levels. The report says the total energy usage would be comparable to the consumption of the entire country of Japan.
What does this all have to do with climate change and the current situation in Los Angeles? It’s simple: Our world is heating up more each year, which exacerbates extreme-weather events. Conserving energy and lowering the amount of greenhouse gases we’re releasing into the atmosphere can help curb that trend. Ditto with our water usage: Conservation efforts can address scarcity and help us deal more effectively with the impact of climate events. So if you care about the future of our planet, skip using AI to write that response to your boss you’ve been avoiding. We both know it’d take all of 60 seconds to do yourself.