If you are concerned about environmental issues, reconsider your use of AI. Generative artificial intelligence consumes 30 times more energy than conventional search engines, warns researcher Sasha Luccioni, who aims to raise awareness about the environmental consequences of this emerging technology. Named one of the 100 most influential figures in AI by Time magazine in 2024, the Canadian computer scientist of Russian descent has spent years trying to measure the emissions generated by programs like ChatGPT and Midjourney. "It’s particularly disappointing that generative AI is used for web searches, " she expressed during an interview with AFP at the ALL IN artificial intelligence conference in Montreal. The language models powering these applications require vast computational resources to be trained on billions of data points, relying on sophisticated servers. Additionally, significant energy is consumed to handle each user’s specific requests. Unlike traditional search engines, which merely extract information—like identifying a country's capital—AI programs "generate new information, " making the entire process "far more energy-intensive, " she clarifies. The International Energy Agency reports that the combined consumption of the AI and cryptocurrency sectors approached 460 terawatt hours of electricity in 2022, amounting to two percent of total global energy production. **Energy Efficiency** A key researcher focused on AI's climate impact, Luccioni helped develop a tool in 2020 that aids developers in quantifying the carbon footprint associated with code execution.

"CodeCarbon" has been downloaded over a million times since its inception. As the head of climate strategy at startup Hugging Face—a platform that shares open-access AI models—she is currently working to establish a certification system for algorithms. This initiative, similar to a program from the U. S. Environmental Protection Agency that rates the energy consumption of electronic devices, would allow users and developers to understand an AI product's energy use and encourage improved decision-making. "We currently do not account for water or rare materials, " she admits, "but we can assess energy efficiency for specific tasks, allowing us to rate models with grades such as A+ or D, " she observes. **Transparency** To refine her tool, Luccioni is testing it on publicly accessible generative AI models but aims to apply it to commercial models from firms like Google and ChatGPT creator OpenAI, which have not been forthcoming in cooperating. Despite Microsoft and Google pledging carbon neutrality by the decade's end, the two tech giants witnessed their greenhouse gas emissions skyrocket in 2023, with Google’s emissions up 48 percent from 2019 and Microsoft’s increasing by 29 percent since 2020. "We are exacerbating the climate crisis, " asserts Luccioni, urging tech companies to embrace greater transparency. She believes that solutions could emerge from governments, which currently are "flying blind, " unaware of the specifics of "the data sets or the training of algorithms. " "Once we achieve transparency, we can begin legislating. " **'Energy Sobriety'** Moreover, it’s essential to "inform people about what generative AI can and cannot accomplish, and at what cost, " Luccioni states. In her latest research, she demonstrated that generating a high-definition image through AI consumes as much energy as fully charging a smartphone battery. As more companies seek to further embed this technology into daily life—through chatbots, connected devices, and online searches—Luccioni promotes the concept of "energy sobriety. " She emphasizes that the goal isn’t to reject AI but rather to selectively choose the right tools and use them wisely. © 2024 AFP