The dizzying rise of AI and machine-learning in the last year, made manifest in the popular consciousness by ChatGPT, has followed a familiar pattern. Years (decades!) of specialized research grinds forward, yielding interesting but not necessarily groundbreaking results, until the moment someone realizes there’s money to be made at scale. This moment, when research becomes product, is usually very brief, as the measured cadence of the lab explodes forward into the heedless pace of the marketplace.
Without checking, how long would you say ChatGPT has been around? The answer is seven months.
And in that brief time ChatGPT has come to represent all our fears and hopes about an AI-determined future. Will it make our lives easier, alleviating our daily decision fatigue, freeing our big human brains for more worthwhile preoccupations? Or will it one day decide that humanity has had its turn, fucked up royally, and needs to go?
These are questions that have been written about endlessly for a generation and will no doubt serve as daily op-ed fodder for the next decade. But here’s the question we should be asking today:
Are you fucking telling me that each adorable little ChatGPT interaction uses a half-liter of fresh water?
According to this article in Forbes,
ChatGPT “drinks” the equivalent of a 500ml bottle of water for a simple conversation of 20 to 50 questions and answers. Which may not seem like much… until you consider that the chatbot has more than 100 million active users, each of whom engages in multiple conversations.
And to put that in perspective, the article continues:
In 2021, [Google’s] data centers in the US alone will consume 12.7 billion liters of freshwater for on-site cooling, about 90% of which will be potable water.
That’s right, potable water. This is happening in a country that allows its cities to go years without decent municipal drinking water; in a world where “771 million people lack access to clean water,” and where “women and girls spend an estimated 200 million hours carrying water every day.”
All so we can make it easier for high schoolers to write their book reports, or college grads to write resumes, or for… Book websites to publish cheeky listicles?
This last one is what set me off. The website 5 Books has always been an interesting repository of expert-lead book recommendations, much better than any of the algorithmically driven recommendation engines created by big tech (because, you know, human editors). And though I understand this is a cheeky* little foray into AI, it was essentially the equivalent of pouring a liter of fresh, potable water into the Serpentine.
Don’t do this. Don’t outsource things you’re perfectly capable of doing to a techno-capitalist’s water-wasting machine.
Your experiments aren’t that interesting. And they certainly aren’t worth the water you’re pissing away to cool a server farm while half the planet struggles to adjust to deeper, longer cycles of drought.
(And if you’re desperate for book recommendations, we have plenty of those.)
*It’s a British website, people.
UPDATE: Here’s the original, non-peer-reviewed paper that does a fine-grained breakdown of ChatGPT water usage within a data center as a whole.