Can a Machine Run Cleaner Than a Wasteful Writer?
by Maria Bustillos
People have caught on that the exploding AI industry—and particularly generative AI, which includes projects like Midjourney, Stable Diffusion and ChatGPT—is really bad for the environment. Last year, researchers estimated that training a single early large language model (LLM), ChatGPT-3, could have used as much as 700,000 liters of freshwater and produced 502 metric tons of carbon emissions. And as hype and investment ballooned, these projects have only gotten bigger and bigger.
Machine-generated text and images that simulate human work are flooding the internet, for reasons that have so far failed to justify the enormous cost of the computing power they suck up; in fact the opposite view is gaining traction, that “A.I.-generated content is upending, and often poisoning, the online information supply,” as Kashmir Hill and Tiffany Hsu wrote yesterday in the New York Times. A Pew survey late last year found that more than half of U.S. respondents were “more concerned than excited about AI in daily life.”
There’s a lot of money riding on this technology; last year’s investments in generative AI reached $25.2 billion. Any emerging public consensus, especially one that contradicts the ambitions of industry, invites a contrarian response. As if on cue, “The Carbon Emissions of Writing and Illustrating Are Lower for AI than for Humans” was published in February in Nature: Scientific Reports.
What if people were the real wasteful ones?
It’s a paywall, but a small one
Read this post and get our weekdaily newsletter for $3 a month