A Coruña is looking for algorithms that don’t destroy the planet: “ChatGPT’s power consumption must be brutal” | technology

A light bulb appeared in the brain of researcher Veronica Polon, an expert in artificial intelligence, one day in 2019 while she was reading a scientific article. He said that training the AI ​​language model resulted in the carbon dioxide equivalent of five cars being emitted over its lifetime. I was surprised. Artificial intelligence consumes a lot of energy when it learns, not only in using it, but even then I never thought about it, ”she recalls sitting in one of the rooms of the Research Center for Information and Communication Technologies (CITIC) at the University of La Coruña. She and seven other researchers (in Total, four men and four women) One of the teams in Spain currently immersed in the search for green algorithms; that is, in solving that great paradox created by the boom of supercomputers, the hustle of data, the cloud and ChatGPT: “Artificial intelligence helps us solve many problems , some of which are derived from climate change, but in turn they are part of the problem,” Polon warns.

Green AI seeks to design solutions that achieve the proposed goals, but in a way that is sustainable for the environment, that is, with the efficient use of computational resources. This concept, which is barely three years old, contrasts with the red artificial intelligence, which only values ​​algorithms for their performance, even if they include, for example, exceptional energy consumption. “Some importance is now being given to green algorithms, although not as much as it should be,” says Veronica Polon. He believes that behind this lack of interest from humanity is the fact that taking sustainability into account runs counter to the interests of big tech in the amazing race of artificial intelligence. He explains that Microsoft, Google and OpenAI are driving developments in this area with “significant consumption” of energy. The researcher points out that environmental scientists have spent years issuing alerts about the ecological footprint of this process. In 2010, Greenpeace already warned in a report that the farms of thousands of servers that make up the cloud “require massive amounts of energy to operate and cool themselves.”

The advent of ChatGPT, an impressive tool that millions of people interact with daily, has exacerbated the problem. “We don’t know exactly what it’s consuming, but it must be brutal, in training and using it, because it needs a lot of data and very large neural networks. And it’s not something people stop to think about, because they don’t have any information about it either. [La empresa] It’s called OpenAI, but it’s not open at all, Polon says of the company that launched the popular app and which was founded in 2015 by Sam Altman and Elon Musk, among others, as a nonprofit.

A study from the University of Copenhagen in 2020 estimated that a single training session of GPT-3 (the text generator on which ChatGPT was based) had energy consumption equivalent to 126 Danish households per year, according to the researcher. Regarding ChatGPT-4, its next version, no equations have been published yet, but there is a piece of information that gives an idea of ​​its environmental impact: trained with 1 billion variables, compared to 175,000 million trained. Used with GPT-3. I am not saying that this technological development should be stopped. But if you ask me whether it’s ethically worth all that computational effort for a model meant to entertain or help with tasks, I have mixed feelings. I admire the progress that’s been made, but I think we’re putting the focus where it’s not. It is not used on fundamental issues affecting society, he points out.

Veronica Polon, 39, set out in 2018 by investigating how algorithms could run on machines too small to have the computational power of a computer. He was not familiar with the concept of green AI, but soon realized that using fewer computing resources is more sustainable because it consumes much less energy. The same is true if the data isn’t continually moving to the cloud, he explains. In 2022, the same year he joined the Royal Academy of Exact, Physical and Natural Sciences in Spain, he was awarded a national research project that would last until 2025. His team managed to implement the algorithms, without losing precision in the results, with only 16 bits (one bit is minimum unit of information) instead of the 64-bit that common computers use today. Polon explains that they still don’t know exactly what energy drop the change entails because they’re working with simulations.

His team, consisting of Bryce Cancela, Jorge Gonzalez, Nolia Sanchez, Laura Moran, David Novoa, Eva Blanco and Samuel Suarez, is also exploring another path towards green AI, which consists of accelerated processes. They are working on trying to do the same thing in less time. “It’s all about brainstorming to improve the models, so that they work well with less data and a smaller network. Now everyone wants to advance, for example, deep learning (Deep learning) But there are problems that can be solved using much simpler models. It kills flies with cannon shots. You have to make an effort to distinguish when it’s necessary and when it’s not,” says CITIC’s Polon.

The legal brakes on AI emissions

This scientific center has about 200 researchers and 75% is devoted to the field of artificial intelligence and data science. It forms part of the university and business ecosystem for which the government has chosen A Coruña to be the headquarters of the Spanish Agency for the Supervision of Artificial Intelligence (Aesia). Green algorithms will be part of the work of this body. Spain has had a National Green Algorithms Plan since last December, which aims to promote the energy efficiency of artificial intelligence and its application to solve environmental problems. The document includes the creation of up to two chairs in this discipline, an invitation made by the University of La Coruña and that will be resolved “soon”, according to sources from the Minister of State for Digitization and Artificial Intelligence.

Of the 257.7 million euros of European Next Generation funds with which the plan has been awarded until 2025, 6.9 million will be earmarked for financial aid for research projects in this area, according to the aforementioned Secretary of State’s report. Manuel González Pinedo, director of CITIC, who has called for an improved financing structure for centers like the one he leads, emphasizes the continuity in programs and the lack of bureaucracy to attract talent from outside Spain.

Veronica Polon urges a law that regulates AI from an ethical and sustainability point of view: “Regulation is needed and must include consumption. Are we going to allow information technologies to be responsible for 20% or more of carbon dioxide emissions into the atmosphere? Well, we’ll have to rein in that , as in other industries or fields.”

You can follow country technology in Facebook And Twitter Or sign up here to receive The weekly newsletter