On top of all the reasons that we shouldn’t use generative artificial intelligence (AI) models for our college work, it has recently been coming to light across social media that the most popular model, ChatGPT, is also bad for the environment. We’ve all seen those posts recently claiming that one ChatGPT search ‘kills three trees’ but how could this really happen? It doesn’t seem obvious to those like myself without a background in computers, but it’s clear we have more to worry about than inaccuracies and made-up studies when it comes to using AI for our homework.
“ChatGPT generates 8.4 metric tons of carbon dioxide per year to run the data centres, more than double the 4 metric tons that each human emits annually”
An early version of ChatGPT first came out in 2018 and the version we know so well today was released in 2022, meaning that this field is still very new. However, with how fast it caught on and with no plans for stopping, researchers have taken a key interest in its effects and many studies have focused on its environmental impact almost from the beginning, unbeknownst to the general public. In a book called Applied Data Science and Smart Systems, Priyanka Bhaskar and Neha Seth dedicate a whole chapter to the environment and sustainability from a ChatGPT perspective, stating that ‘ChatGPT generates 8.4 metric tons of carbon dioxide per year to run the data centres, more than double the 4 metric tons that each human emits annually’. To look into where this comes from, we must get one fact clear; it is not actually ChatGPT itself that creates this waste, it is all the technology used to ‘train’ it and support it.
The reason ChatGPT needs to be ‘trained’ is that as an AI model, it runs off of pre-learned patterns and information that has been fed to it by its developers. In other words, ChatGPT isn’t ‘answering’ your questions, it is merely spitting out what it predicts to be the correct answer based on patterns that it has been taught before, which accounts for its inaccuracies, but that’s another story. To do this, a large amount of technology and therefore a large amount of power is required, which is the first way that the use of ChatGPT damages the environment even before it is ready for use, especially if this power comes from nonrenewable resources such as fossil fuels. As we know, this releases carbon dioxide, a greenhouse gas, into the atmosphere and so contributes to global warming.
“The water itself often comes from valuable freshwater habitats and its waste mightn’t always be disposed of responsibility”
ChatGPT also has a large water footprint, as the data centres which operate it behind the scenes need water to cool down their mechanical parts to prevent damage, much like how the fan on your laptop switches on while you’re grinding out a last minute article in the library. If you’re wondering what a data centre is, it’s a dedicated space or building for housing computers and their associated parts to store and process large amounts of information, using vast amounts of water and electricity as stated above. Of course, a part of Ireland’s short sighted future plans for growth are to allow the building of more of these centres. On top of the water used to train ChatGPT, which according to the University of California took 700,000 gallons for Microsoft to do, and that used to power one conversation with ChatGPT, which is estimated to be the equivalent of a 500ml bottle of water, the water itself often comes from valuable freshwater habitats and its waste mightn’t always be disposed of responsibility, further contributing to biodiversity loss and therefore climate change.
On top of all this, there is of course the materials which go into the building of the computer systems themselves. It’s easy to forget that a simple laptop requires the use of many materials which are difficult to dispose of when needed, including hard plastics, which are just the tip of the iceberg. Of course, these things can be recycled theoretically but there is also the mining for precious materials such as gold for the circuit boards of computers and the use of non- rechargeable lithium batteries which have a heavy impact on the environment, whether the effects are right under our noses or not.
“Is the blame not being passed once again from big companies to largely innocent members of the public?”
Many solutions are offered to offset the use of AI models, in particular ChatGPT. We can unplug unused items at home and turn off our taps and recycle our milk cartons, but when we look critically at these suggestions, is the blame not being passed once again from big companies to largely innocent members of the public? It is said that we can also limit our use of ChatGPT to only the essentials, but are there really any ‘essential’ needs for these models? Call me a luddite, but would it not be better for us all to relearn how to use google or even better, our brains? With its negative environmental impacts on top of all the many other ethical reasons we shouldn’t depend on ChatGPT to write our essays, I invite you to figure out for yourself whether it’s really necessary. Sure it might be useful for some quick and specific answers, but which will kill us faster: a longer than anticipated google search or a world on fire?
