How much does ChatGPT pollute? Often and willingly we focus on the environmental impact of everything outside the network but, as evidenced by a series of studies, the internet and the training of AI also have a weight in terms of emissions
14/04/2023 by Ilaria Roncone
How much pollution and what environmental impact does ChatGPT have? In an interview with us, Ale Agostini, CEO of AvantGrade and founder of Karma Metrix, highlighted the polluting potential of artificial intelligence. Of how much pollutes internet and all that is tied to it the Italians have little awareness. Here, then, providing data and notions on how much ChatGPT pollutes is essential to understand how much it costs - in terms of environmental impact - the unbridled development of AI, thus providing further reflection on what should be regulated in this area.
READ ALSO >>> The Guarantor has presented its requests to ChatGPT in nine points
AI training requires a lot of energy
In this regard, Karma Metrix - which deals, among other things, with measuring how much the web pollutes - has tried to quantify the carbon footprint of ChatGPT. There is a contradiction that needs to be highlighted with respect to machine learning: although artificial intelligence models are also used to combat climate change, they themselves require a huge amount of energy to function.
Let’s talk about numbers: the AI sector pollutes as much as the aviation sector. We talk about the fact that training an AI system will get to issue - according to a 2019 Cornell University study - over 113 tons of carbon dioxide.
So how much does ChatGPT pollute?
Let’s get down to the specific numbers concerning the OpenAI chatbot. According to an article published in late 2022 on the online portal created by Twitter co-founder, Medium.com, ChatGPT’s carbon footprint can also reach 522 tCO2e (tonnes of CO2 equivalent).
The estimate that has been made by Chris Pointon, who deals with the development of sustainable technology, is clear: potentially ChatGPT could get to emit 3.8 tons of CO2e every day. These are estimates that some scientists have started to make in the world - other experts have calculated that training AI for ChatGPT generates CO2 equal to that of a car that travels 700 thousand kilometers - and that must be taken for what they are, or attempts to estimate a value that OpenAI itself does not provide enough information to calculate.
ChatGPT also consumes a lot of water
And not just ChatGPT, of course. This applies to all chatbots and similar technologies that - to be trained - require the consumption of a lot of energy for cooling data centers. Cooling that requires, among other things, a lot of water. A joint study by the University of Colorado Riverside and Arlington University of Texas drew attention to the numbers of this phenomenon: ChatGpt-3 training, according to estimates, led to the consumption of 700,000 liters of fresh water. To understand how much water it is, the same amount is used to make 370 BMW cars or 320 Tesla cars.
How much water does ChatGPT consume to make a conversation with a user? A bottle of water. According to the researchers, given the great success that this tool is having, the great use that it makes of it could impact very widely on water supplies (also net of the growing drought and environmental issues that increasingly affect the planet).
«The water footprint of artificial intelligence models can no longer stay under the radar, it must be treated as a priority and part of the collective efforts to combat global water challenges», the authors of the study explained, However, they pointed out that Google also consumes a lot of water: data centers in the US have drunk 12.7 billion liters of fresh water in 2021 alone and 90% of this was drinkable.