AI Has Been Using A Secret Resource.
Welcome back to our little corner of the internet! Today we’ll be learning about a precious resource that AI has been using all this time, pretty much unbeknownst to the general public. Let’s dive right in.
Water. Did you know that to produce just one cup of coffee it takes up to 140 litres of water? Think of that as approximately 9,500 (US) tablespoons.
But even worse — did you know ChatGPT takes up 500ml (34 tablespoons) of water for every 20–50 commands inputted? And that’s not even all of it, because the website has over 60 million visits per day.
Now according to the eco-empowered EuroNews website, scientists believe that whilst training GPT-3 alone, Microsoft consumed over 700,000 litres of water.
47,339,631. Now that’s a lot of tablespoons.
That’s GPT-3, but as technology like GPT-4 arises we can only imagine how much fresh water would be consumed (this is partly because there is little to no data available to the public on this topic, which is all the more reason to speak up about it).
As a study in March showed, over two billion people worldwide don’t have access to clean drinking water. There’s approximately eight billion people on Earth, so that means 1/4 of the population don’t have good water to drink. So is it fair that we can consume this much in the name of technology? Well, I want to say maybe.
And to make things worse, water isn’t all. AI is leaving major carbon footprints too. According to a study, GPT-3’s daily carbon footprint was been estimated to be equivalent to 50 pounds of CO2 or 8.4 tons of CO2 in a year. The energy used to simply train GPT-3 could power an average American household for not mere days or months — but for years!
While being excited about AI is understandable (as the technology is fascinating!), it’s important to remember that “a single request in ChatGPT can use 100x more energy than one Google Search”.
How is ChatGPT using this water?
According to EuroNews, “most of the chatbot’s cloud computing relies on thousands of servers inside data centres around the world”. The wild and wacky figures refer to the “fresh, clean water” used by these data centres to generate electricity and cool the innumerable racks of servers which help applications like ChatGPT run.
Cooling the racks of servers is harder than it sounds because server rooms have to be kept at a constant 50–80 degrees Fahrenheit to prevent malfunctioning, but the servers convert their electricity into heat. Cooling towers are leveraged in these special rooms to act against the heat by evaporating cold water (which is why AI consumes so much water).
But OpenAI isn’t the only one responsible. Google requested over 2.3 billion gallons of water to cool servers in data centres (which were in just 3 states) in 2019.
Authors of the paper “Making AI Less Thirsty” agreed that the same amount of freshwater used to train GPT-3 “is equivalent to the amount needed to fill a nuclear reactor’s cooling tower”.
AI is an amazing tool which could allow us to become an even more advanced civilisation as a whole. However, I would still bear in mind one question: “is it worth the earthly cost?”
A huge thank you to Clive Thompson for introducing this important fact to his Medium audience, and for inspiring this post.
Please bear in mind that this brief blog post was not intending to blame or scare anyone, I solely wrote it to draw awareness to this important matter. Thank you so much for reading this, please do leave a few claps and comment if you enjoyed it. If you want tech-savvy stories, then follow me for more!
See you in the next one!
