Generative AI uses more energy, ‘accelerates climate crisis’: experts

Montreal, Canada:

If you care about the environment, think twice before using AI.

Generative AI uses 30 times more energy than a traditional search engine, warns researcher Sasha Luccioni, who has set herself the goal of raising awareness of the impact of this new technology on the environment.

Recognized by the American magazine Time as one of the 100 most influential people in the world of artificial intelligence in 2024, the Canadian computer scientist of Russian origin has been trying to determine the broadcast of programs such as ChatGPT and Midjourney for several years.

“What is particularly disappointing to me is that generative AI is being used to search the internet,” laments the researcher, who spoke to AFP on the sidelines of the ALL IN conference on artificial intelligence held in Montreal.

The language models on which the programs are based require enormous computing power to train on billions of data points, which in turn requires powerful servers.

Then there is the energy used to respond to each individual user’s requests.

Instead of simply extracting information, “like a search engine does to find the capital of a country, for example,” AI programs “generate new information,” making the whole endeavor “significantly more energy-intensive,” he explains.

According to the International Energy Agency, the artificial intelligence and cryptocurrency sectors combined consumed almost 460 terawatt-hours of electricity in 2022, representing two percent of total global production.

Energy efficiency

Luccioni, a leading researcher on AI’s impact on climate, helped create a tool in 2020 for developers to quantify the carbon footprint of running a piece of code. “CodeCarbon” has since been downloaded more than a million times.

The head of climate strategy at the startup Hugging Face, a platform that provides open artificial intelligence models, is currently working on creating a certification system for algorithms.

Similar to the U.S. Environmental Protection Agency’s program that awards points based on the energy use of electronics and household appliances, this would allow AI product energy usage to be understood to encourage users and developers to “make better decisions.”

“We don’t take into account water or rare materials,” he admits, “but at least we know that for a specific task we can measure energy efficiency and say this model is A+ and that model is D,” he says.

Transparency

To develop her tool, Luccioni is experimenting with it on generative AI models that anyone has access to, or on open-source models, but she would also like to do this work on commercial models from Google or ChatGPT’s creator OpenAI, which have reluctantly agreed to it.

Although Microsoft and Google have pledged to achieve carbon neutrality by the end of the decade, the US tech giants’ greenhouse gas emissions will increase dramatically in 2023 due to AI: 48% for Google compared to 2019 and 29% for Microsoft compared to 2020.

“We are accelerating the climate crisis,” Luccioni says, calling on tech companies to be more transparent.

The solution, he says, could come from governments, which for now are “acting blindly,” not knowing what is “in the data sets or how the algorithms are trained.”

“Once we have clarity, we can start legislating.”

“Energy sobriety”

According to Luccioni, it is also necessary to “explain to people what generative AI can and cannot do and at what cost.”

In her latest study, the researcher showed that generating a high-resolution image using AI consumes as much energy as fully charging a cell phone battery.

At a time when more and more companies are looking to integrate technology even more deeply into our lives — through conversational bots and connected devices or internet search engines — Luccioni advocates for “energy sobriety.”

As he emphasizes, the idea is not to oppose artificial intelligence, but to choose the right tools and use them wisely.

(Except for the headline, this story has not been edited by NDTV staff and is published from a syndicated feed.)