AI will save us. AI will doom us. AI will enslave us. AI will enlighten us.
Every week brings a new wave of hyperbolic AI headlines, each more dramatic than the last.
The discourse around AI is mostly ill-informed. Take a recent article from Fortune magazine (see comments). The headline goes: “AI doesn’t just require tons of electric power. It also guzzles enormous sums of water.”
In the article, there this statement: “In order to shoot off one email per week for a year, ChatGPT would use up 27 liters of water, or about one-and-a-half jugs… that means if one in 10 U.S. residents—16 million people—asked ChatGPT to write an email a week, it’d cost more than 435 million liters of water.”
Predictably, the article (from September 2024) has lots of likes and replies on social media talking about how AI is going to doom us all and is a waste of precious energy.
So this article assumes the amount of power required to run inference – i.e. when ChatGPT helps compose an email and maps to the amount of water required to generate that power.
Interestingly, the cost of running inference has gone down substantially over the last year. Recent research by DeepSeek (see comments) also shows how it is possible to train a state of the art model for a fraction of the cost of training foundation models.
Discourse about how AI is ruining the planet conveniently takes data from today and projects it infinitely into the future. Let me put it this way – in 1965 the average gas mileage for a small car was around 15 – 20 MPG. A modern car is 3X more fuel efficient (Toyota Prius or Honda Civic). And it is still fundamentally an internal combustion engine.
Software and AI move much, much faster than the automobile industry. So the next time you see a headline about AI’s apocalyptic resource consumption, remember – you’re probably reading tomorrow’s equivalent of “The Internet Will Crash Under Its Own Weight” articles from 1995.
