Carbon Footprint: The Great Ecological Challenge Of Green AI

Behind the great promises of artificial intelligence, the ecological carbon footprint of this technology raises questions. In a world seeking an environmental transition, finding more sustainable solutions regarding AI is crucial.

Today, imagining our world without artificial intelligence (AI) is becoming more and more complicated as this technology interferes a little more in our daily lives every day whether it’s personalized recommendations on streaming platforms, accurate weather predictions, virtual assistants that help us organize our day, or ChatGPT.

Yet, while we marvel at these technological feats, one detail, which is not a detail, is often left aside: the environmental impact of AI. During the learning phase, but also to provide the desired answers, each artificial intelligence consumes a lot of energy. Like blockchains and cryptocurrencies, the intensive calculations required to train and run these systems are not negligible.

Table of Contents

AI: High Energy Consumption

The carbon footprint of this technology is far from negligible. A study from the University of California estimated that the training phase alone of ChatGPT, by feeding it millions of data from the web, had a significant energy weight. She judges that this required 1,287 MWh of electricity or the production of 552 tonnes of CO2 equivalent.

This corresponds to the average annual consumption of just over 123 diesel vehicles. It is also comparable to the yearly energy consumption of a small town. Another study, this time at the University of Massachusetts, found that learning a single complex AI model can produce as much carbon dioxide as five cars over their entire lifespan.

Furthermore, every time we do a Google search, ask Alexa to play our favorite song or let Netflix choose our next binge-watch, we engage AI, and thus, we are consuming energy. Therefore, behind the comfort and facilities that AI brings to our daily lives, significant energy consumption is often underestimated.

Data Centers, Symbols Of The Ecological Challenge Of AI

Data centers are essential for storing and processing the vast data we generate daily. Bringing together thousands of servers, these infrastructures consume an astronomical amount of energy to operate and to be cooled. According to Greenpeace’s “Clicking Clean” report, the energy consumption of data centers and information networks represented approximately 7% of global electricity consumption in 2022.

And with the constant increase in the volume of data processed, this trend is only accelerating. Furthermore, these data centers are not only energy consumers but also electronic waste producers. Frequently renewing servers to stay on the cutting edge and ensure optimal performance generates enormous amounts of garbage, adding another dimension to AI’s ecological footprint.

Can AI Be “Green”?

Despite the gloomy picture that could be painted of the ecological carbon footprint of AI, there are encouraging signs. Researchers and companies worldwide are starting to look seriously at the idea of ”greener” AI. Both for the good of the planet and that of their wallets. OpenAI, ChatGPT’s parent company, spends a fortune supporting the number of requests site users send.

Among the solutions considered are optimizing algorithms to reduce energy consumption or the development of more eco-efficient hardware. Furthermore, some companies strive to make their data centers more sustainable, particularly by using renewable energies and implementing natural cooling systems. It is clear that the road to genuinely green AI is still long, but these initiatives show that we have already started to travel it.