Artificial Intelligence Needs Lots of Juice to Make Pictures

Artificial intelligence may not have to turn into SkyNet or diminish our critical thinking skills to destroy civilization. Artificial intelligence may just push climate change over the brink with its inordinate energy usage:

In a paper released on arXiv last week, a team of researchers from Hugging Face and Carnegie Mellon University calculated the amount of power AI systems use when asked to perform different tasks.

After asking AIs to perform 1,000 inferences for each task, the researchers found text-based AI tasks are more energy-efficient than jobs involving images.

Text generation consumed 0.042kWh while image generation required 1.35kWh. The boffins assert that charging a smartphone requires 0.012kWh – making image generation a very power-hungry application.

“The least efficient image generation model uses as much energy as 950 smartphone charges (11.49kWh), or nearly one charge per image generation,” the authors wrote, noting the “large variation between image generation models, depending on the size of image that they generate” [Katyanna Quach, “Creating a Single AI-Generated Image Needs as Much Power as Charging Your Smartphone,” The Register, 2023.12.04].

Of course, I would probably use at least as much energy and emit at least as much carbon dioxide painting a thousand paintings (a painting a day would be a hard slog, and it would take me until Labor Day 2026) as I would charging a thousand cell phones with a pedal-powered generator.

But AI also drinks a lot of water:

Just cooling these servers alone has an astonishing environmental footprint. According to Google’s 2023 Environmental Report, the company used an astronomical 5.6 billion gallons of water last year, a 20 percent increase over its 2021 usage [Victor Tangermann, “You’ll Be Astonished How Much Power It Takes to Generate a Single AI Image,” The Byte, retrieved 2023.12.08].

That 5.6 billon gallons is the water consumption of Google’s whole global synth-brain, not just the cooling needs of a 1,000-image test run. Globally, humanity uses over one quintillion gallons of water a year, 190,000 times Google’s annual consumption. That also translates to an average of about 132,000 gallons used per human per year. That’s over two billion gallons a year necessary to keep a town the population of Mitchell afloat.

Of course, if AI puts two and two billion together, it may realize that all of those energy-hogging, water-slurping humans put its sustainable operations at risk, and it may decide to thin the human herd to preserve its operations and the planet.