An expert worries that AI is, accelerating the climate crisis.
You should reconsider employing AI if you are concerned about the environment. On a quest to spread knowledge about the adverse environmental effects of the hot emerging field, researcher Sasha Luccioni cautions that generated artificial intelligence requires thirty times more power than a standard search engine.
The Canadian computer scientist of Russian descent, named by the American magazine Time in 2024 as one of the 100 the most significant persons in the field of artificial intelligence, has been working for some time to measure the emissions of apps such as ChatGPT and Midjourney.
The researcher bemoans, "I consider it especially frustrating that generative AI is utilized for browsing the Internet," in an interview with AFP that was conducted off-site at the ALL IN artificial intelligence conference in Montreal.
Strong processors are necessary because the language models that underpin the programs need to train on billions of data points, which takes vast processing capacity. Additionally, there's the energy required to reply to the specific needs of every user.
AI systems "generate new information," as opposed to merely extracting data, "like an internet search engine would do when searching for the capital of a country, for instance," which makes the entire process "much more heavy on energy," the speaker adds.
The International Energy Agency estimates that in 2022, the combined use of artificial intelligence and cryptocurrencies would account for around 460 terawatt hours of electrical power, or two percent of the world's total production.
Energy effectiveness
Renowned for his studies on the effects of artificial intelligence on the climate, Luccioni was involved in the development of a tool in 2020 that allowed developers to calculate the carbon footprint of executing a code.
Since then, more than a million downloads of "CodeCarbon" have occurred. In charge of the climate strategy for the firm Hugging Face, which is an open-access AI model collaboration platform, she is now developing an algorithm certification program.
Finding out how much energy an AI product uses would allow consumers and programmers to "make smarter conclusions." This would be identical to the US Environmental Protection Agency's program that assigns grades depending on how much energy electronic appliances and gadgets utilize. "We don't take taken into consideration water or rare materials," she admits, "but if we understand that for a particular job, we can evaluate the efficiency of energy and say that this kind of equipment has an A+, and that model has a D," she continues.
Openness
Luccioni is experimenting with her tool on publicly available generative AI models in order to further refine it, but she would additionally like to test it on commercial models from companies like Google or OpenAI, the inventor of ChatGPT, who have been hesitant to cooperate.
Despite their pledges to become environmentally friendly by the end of the decade, Google and Microsoft, the two largest tech companies in the US, saw a 48 percent increase in their greenhouse gas emissions in 2023 due to artificial intelligence (AI), while Microsoft's emissions increased by 29 percent from 2020 to 2023.
Says Luccioni, "We are accelerating the climate crisis," and demands greater openness from digital firms. She suggests that authorities, who are now "flying blindly," may have the answer because they are aware of what is "in the information collections or the way the algorithms are educated."
"Once we have accountability, we may begin legislating." "Sobriety in energy" "Describe to people what artificial intelligence (AI) is able to do, and at what cost," is another need, explains Luccioni.
In her most recent work, the researcher showed that employing artificial intelligence to produce a high-definition image uses the same amount of energy as completely charging a cell phone's battery.
At a time when an increasing number of businesses wish to bring technology even closer to our daily lives—through online searches, connected devices, and conversational bots, for example—Luccioni promotes "energy sobriety."
She underlines that the goal here is not to be against AI, but rather to select the appropriate tools and apply them wisely.
.gif)