The ecological impact of AI trends

Unless you've been living under a rock, you haven't missed the new trends of filters based on artificial intelligence that have taken over social media in recent weeks. Among the most popular is the Hayao Miyazaki filter, capable of transforming any photo into an illustration with the unique style of Studio Ghibli. Another viral trend is the Starter Pack, which generates an image representing a person or brand as a doll, surrounded by iconic elements that define them. These phenomena, beyond their carefree appearance, have raised ethical questions about copyright concerning the Miyazaki filter, as well as concerns about environmental impact. Indeed, the use of generative artificial intelligence consumes much more energy than simple web browsing. "What is different in generative AI is the power density it requires. Fundamentally, it's just computation, but a generative AI training cluster can consume seven to eight times more energy than a typical computing workload", explains Noman Bashir, lead author of the impact article, who is a Computing and Climate Impact Fellow at the MIT Climate and Sustainability Consortium (MCSC).

Thus, a query on ChatGPT consumes as much energy as ten Google searches, according to the International Energy Agency. Moreover, the environmental impact is even greater when it comes to generating images via artificial intelligence. These operations mobilize significantly heavier computing resources, resulting in much higher energy consumption. According to a study by Carnegie Mellon University, an image requires between two to five liters of water. In terms of greenhouse gases, generating 1,000 AI images is equivalent to driving 6.5 kilometers in a gasoline-powered car. Although it is difficult to quantify the exact environmental consequences of these specific AI filters, it is clear that these images have been a disaster for the Earth. The buzz around the Studio Ghibli filter was so intense during its launch week—with 700 million queries—that Sam Altman, president and CEO of OpenAI, stated on X that the company’s servers were "melting."

Generative technologies require an astronomical amount of water, mainly to cool the infrastructures housing the servers used by AI platforms. For instance, ChatGPT is estimated to use 148 million liters of water daily according to Business Energy UK. Furthermore, AI-related water consumption could reach between 4.2 and 6.6 billion cubic meters by 2027, according to a Washington Post study. A true ecological aberration in an age of climate disruption. It is hard to imagine a sustainable future for AI given the fierce competition among companies (OpenAI, Microsoft, Google, Meta, etc.) that will only accelerate the race. Today, ChatGPT averages 125 million users per day according to DemandStage. A number that will likely grow exponentially over the years. It is urgent for the multinationals developing these technologies to adopt a sustainable approach to prevent an ecological catastrophe that is already underway.