2024-02-01 11:30:23
European Bioethics Forum
What human for tomorrow? fileArtificial intelligence has very real consequences on the environment. Distinguishing between its positive and negative impacts is partly the work of Jacques Sainte-Marie, deputy scientific director in charge of the environment at Inria. Interview.Cloning, genome sequencing, personalized medicine, data… Technologies are changing our lives and our societies. The fourteenth edition of the European Bioethics Forum, of which Libération is a partner, will have the theme “Artificial Intelligence and us”. While waiting for the event, from February 7 to 10 in Strasbourg, Libération will publish (or put back online) in this file a series of articles on the themes covered.
Mathematician Jacques Sainte-Marie is interested in the environmental impact of artificial intelligence (AI). He directs research at Inria, the French institute specializing in digital technology. He is therefore well placed to answer our questions about the compatibility between the rapid deployment of AI and planetary boundaries.
In which areas, according to you, digital technology and AI can help with the ecological transition?
Artificial intelligence brings new results in research, in the fields of modeling in physics or biology. I am thinking of the GraphCast weather forecasting model which, in recent scientific publications, has shown better results than existing models. AI is enabling truly important scientific advances in many sectors.
That said, AI has an environmental impact. It requires infrastructure, sophisticated algorithms and a lot of data to operate. Today, general AIs like ChatGPT are getting a lot of attention. But I believe that the future belongs to very specialized AIs, which use less data and energy to do a specific task. This is not the direction taken by digital giants who often prefer to develop large general tools that can only run on their infrastructures. They thus remain in a monopoly situation.
The environmental impact of AI, precisely. Is it that well known? We often talk about the carbon impact of training AI, but less about the water requirements to run data centers or the balance sheet of a simple graphics chip…
We know much better about the direct environmental impact of digital technology than its indirect impact. Let me explain: we know that, overall for the digital sector, the manufacturing of digital tools represents 70% of emissions, data centers 20% and networks 10%. So if you keep your cell phone for five years instead of two, you almost halve its carbon footprint. Afterwards, there are of course variations depending on usage. It’s not the same thing to watch a film streaming on a 4G network or connected via wire…
We still need to progress to refine this knowledge, you are right, the water needs of data centers for example. And these questions will become more and more relevant, as we move towards an increasingly constrained world. What we still estimate very poorly, however, are the indirect effects of AI. How AI will change behavior in a sector and thus, through a rebound effect, increase or decrease emissions…
Ultimately, shouldn’t we do without digital altogether?
The future is neither all digital nor without digital. There is no golden age of humanity during which everything was going well and to which we should return. We need innovation to face new challenges. In the energy sector, it will be necessary to manage networks with a diversity of often intermittent production (solar, wind, etc.). This requires fine network management and AI can help.
I am not advocating techno-solutionism, but take away the digital from the IPCC reports, what is left? When the Club of Rome report on planetary boundaries came out in 1972, it was widely criticized. Digital modeling made it possible to objectify their results and, therefore, to convince more people. In the field of climate simulation, AI can also reduce calculations that are extremely costly to run. Today, forecasts are asked to be ever more precise. What will the precipitation be? What will be the trajectory of a cyclone? These are extremely complex calculations.
1706807315
#future #belongs #specialized #Libération