Artificial intelligence in the climate service: a real solution?

The masses of data manipulated by artificial intelligence can be useful in many areas related to the ecological transition: from the heat waves’ prevention to the fires’ management, through the carbon emissions’ monitoring, the tool seems promising. In reality very polluting, it may not ultimately be the miracle cure for the climate crisis.

A not-so-miracle solution

Could artificial intelligence (AI) help us effectively fight climate change? Since 2010, this computer program coded to automatically perform tasks from collection and processing data, has gradually interfered in our lives: it is used in connected speakers, GPS or even on streaming platforms. Today, several researchers claim that this technology could accelerate the ecological transition. AI, at the heart of research on climate change for several years, would, among other things, make it possible to anticipate extreme events and improve the exploration of difficult environments (seabed, South and North poles) as well as harvesting climate data.

Could it be the climate magic bullet? Canadian author Naomi Klein is not sure about that. In an op-ed published in the Guardian newspaper, she writes: “ The climate crisis is not, in fact, a mystery or a riddle we haven’t yet solved due to insufficiently robust data sets. We know what it would take, but it’s not a quick fix – it’s a paradigm shift. Waiting for machines to spit out a more palatable and/or profitable answer is not a cure for this crisis, it’s one more symptom of it.“ “AI is absolutely not the solution”, this was confirmed by researchers, interviewed by the independent French newspaper Reporterre. “Saying that AI will solve everything is dangerous techno-solutionist talk. On the contrary, these neural networks are precipitating the collapse,” says Frédéric Bordage, president of the Green IT collective, which campaigns for sustainable computing. AI has, indeed, big flaws that cannot be ignored: from its manufacture to its obsolescence, it has big environmental consequences. For example, its calculation speed is made possible thanks to microprocessors, which consume a lot of rare and exhaustible materials: « We only have a few decades of reserves left », warns Frédéric Bordage.

The AI prediction process is based on the deep learning method. To popularize, you can imagine that several hundred artificial neural networks (like a human brain) learn to read, write, recognize faces or sounds. The more different information and experiences there are, the better the AI algorithm becomes. These periods of deep learning consume a lot of electricity. According to a 2019 study by the Massachusetts Amherst’s University, the amount of energy used to train certain particularly complex neural networks would be equivalent to that of a human in fifty-seven years.

More recently, Greenly, a French platform which allows companies to assess their emissions in real time, assessed the carbon footprint of ChatGPT – software based on artificial intelligence. The result is impressive: 240 tonnes of CO2 used. This is the equivalent of 136 round trips between Paris and New York.

In addition, training an AI also requires storing millions of data. This therefore requires building ever more data centers, the ecological cost of which is high.

 

« Analyze an amount of data and links impossible for humans »

However, some uses of AI, especially in science labs since the early 2000s, are promising. Until then, the use of AI was limited, due to not powerful enough computers, to allow it to be used to its maximum capacity. Now, thanks to technological progress, it is able « to analyze a quantity of data impossible for humans and to infer links that we cannot make with our brain », summarizes Frédéric Bordage. “What we get to do today is much more important than what we did before. And in five years, we will have made further progress”, confirms Freddy Bouchet, researcher at the CNRS. He and his team recently developed a method to predict heat waves almost a month in advance.

Indeed, by transmitting climate data spread over 8,000 years – as CNRS researchers did – the AI can, from the environmental conditions, detect the arrival of an extreme heat wave.

Still in the same field, it also makes it possible to guide in real time « firefighters on the ground in the event of natural disasters and thus limit human losses », teaches us Virginie Mathivet, French doctor in artificial intelligence. For his part, Jim Bellingham, executive director of the Johns Hopkins Institute, sees it as a revolution for the exploration of areas where the consequences of global warming are not yet fully measured. According to him, future expeditions could be carried out entirely by AI-powered robots, capable of adapting to unstable conditions.

Mostly Recreational AI

In recent years, some companies have highlighted the use of AI in the service of ecological transition: it would make it possible to produce less polluting concrete or to lower emissions from the aviation sector… Thanks to AI, we could even imagine, adds Jim Bellingham, creating « an electricity grid based solely on renewable energy » or make materials that capture CO2, continues Virginie Mathivet. Latest announcement: At the end of May, Google launched Flood Hub, a tool designed to identify and report flood risks.

But apart from these few programs targeting ecological transition, this type of powerful tool unfortunately remains mainly used for « recreational uses », points out Frédéric Bordage. If we can consider that AI is not used wisely (to fight cancer or the disappearance of biodiversity), some large companies use it to increase their profits, thus perpetuating a deleterious system for the planet. “At the moment, AI serves the economic interests of multinationals,” notes Frédéric Bordage, president of Green IT. These companies refine advertising and marketing by precisely analysing consumer practices, practices that are neither very ecological nor very ethical in short.

Naomi Klein, « AI machines aren’t ‘hallucinating’. But their makers are », The Guardian, May 8, 2023 : https://www.theguardian.com/commentisfree/2023/may/08/ai-machines-hallucinating-naomi-klein

Emma Strubell, Anaya Ganesh, Andrew McCallum, « Energy and Policy Consideration for Deep Learning in NLP » College of Information and Computer Sciences University of Massachusetts, 2019 : https://aclanthology.org/P19-1355.pdf

Washington Post Live, « Future of Flight with Alaska Air Group CEO Ben Minicucci and Neste US President Jeremy Baines », The Washington Post, July 1st, 2021 : https://www.washingtonpost.com/washington-post-live/2021/06/30/future-flight-with-alaska-air-group-ceo-ben-minicucci-neste-us-president-jeremy-baines/

Freddy Bouchet, « Changements climatiques : une meilleure prédiction des canicules grâce à l’IA », CNRS, April 3, 2023 : https://www.cnrs.fr/fr/changements-climatiques-une-meilleure-prediction-des-canicules-grace-lia

Nina Guerineau, « Intelligence artificielle et climat : quelques espoirs, beaucoup de pollution », Reporterre, June 5, 2023 : https://reporterre.net/Intelligence-artificielle-et-climat-quelques-espoirs-beaucoup-de-pollutions

Marc Zaffagni, « Meta utilise l’IA pour élaborer de nouvelles formules de béton moins polluantes », CNET, April 28, 2023 : https://www.cnetfrance.fr/news/meta-utilise-l-ia-pour-elaborer-de-nouvelles-formules-de-beton-moins-polluantes-39941185.htm

Emilie Echaroux, « L’AI est-elle vraiment la solution pour limiter l’impact carbone des avions ? », Usbek&Rica, July 13, 2021 : https://usbeketrica.com/fr/article/l-ia-est-elle-vraiment-la-solution-pour-limiter-l-impact-carbone-des-avions

Daniel Rousseaux, « Pour une vraie sobriété, limitons les data centers », Reporterre, january 16, 2023: https://reporterre.net/Pour-une-vraie-sobriete-encadrons-les-data-centers