Some time back I got annoyed by junk mail and did some back-of-the-envelope math, and assuming that both my math and my sources are right, 5 TWh/yr is used on JUST the production of paper that will be used as junk mail, not including printing, transport, or disposal. And that's JUST in the US as well.
ChatGPT's worldwide energy usage, when I looked it up, was 17 TWh/yr. That's not all AI but it is the biggest one. I think it's reasonable to say that if we extrapolate both of those out to their full worldwide industries, they'd come out pretty similarly in terms of energy being used. But junk mail is just The Way The World Is so nobody fights that hard against it.
Likewise, a lot of the energy and water usage comparisons people pick for AI are things people ALREADY had no context for. ChatGPT uses "ten times as much energy as a Google search" sounds scary if you don't have any context for how much energy a Google search takes. Sounds a lot less scary if you phrase it as "A ChatGPT query takes 1/10 as much energy as running your TV long enough to watch a single episode of a show" or "A ChatGPT query uses less than 1/1000 as much energy as a single dryer cycle".
And like, the "1-3 bottles of water per query" thing was already massively inflated by a game of fearmongering telephone where each person in the chain increased the number to make it scarier, but even if you deflate it to the actual number (500 ml for every 5-50 queries), then some other back-of-the-envelope math I did indicated that the Average American Household could switch to a lawn irrigation system that was 15% more efficient and save enough water to use ChatGPT for literally hundreds of queries per day and still be at a net decrease in water use.
All of which is to say, lawns suck, junk mail sucks, and the amount of effort that goes into solving a problem often seems to be significantly more based on the visibility and novelty of the problem than the severity.
yousharknotpass
2087
gracieabrams
origamiaround
