“With AI, our highways and parking lots will become obsolete” – L’Express

an increase less than 3 in 2024 – LExpress

AI is not afraid of Kafkaesque problems. And the climate crisis is the most complex puzzle facing humanity today. It is necessary to analyze an incredible quantity of variables, in all sectors of the economy. Optimize what can be. Rethink the way we operate. A colossal project in which artificial intelligence can play a central role, explains Gilles Babinet, president of the National Digital Council, in his latest book, Green AI. Artificial intelligence at the service of the climate, published at the end of March by Odile Jacob. Interview.

L’Express: For several years now, AI has been presented as an interesting tool in the fight against climate change. This doesn’t seem so simple to put into practice. What’s blocking?

Gilles Babinet: AI is an old field of research, but its integration into productive systems is only a few years old. For artificial intelligence to make a profound difference, there needs to be a very significant change in logic and scale. If we want to reduce the carbon footprint of transport, for example, it is not enough to build more efficient vehicles. We must be able to encourage citizens to travel at different times. It is the entire experience that must be transformed.

How can AI actually change transportation?

AI is very good at managing hundreds of variables, which is precisely the challenge of driving, with signals, pedestrians, the weather, other vehicles… Autonomous driving has long left me skeptical. I am now convinced that it will gradually become more widespread. And that will change everything.

READ ALSO: How to make France an AI giant: instructions for use

If you have car sharing services that can send a vehicle to the desired address and time, which parks itself or leaves when you get out, you make the individual car and parking burdensome, in addition to ‘be expensive. This also opens the way to new semi-collective transport, for example taxis or autonomous minibuses accommodating six to ten people, which would simplify travel in peri-urban areas or towns. Autonomous vehicles also do not need to be spaced 70 meters apart on the highway to follow each other in complete safety; 60 centimeters may be enough, because they react in milliseconds. All this will put an end to traffic jams. And many of our highways and parking lots will become obsolete.

Degrowthists, techno-solutionists… In your book, you affirm that both of these currents are wrong, why?

I don’t see many solutionist technologists in France… Perhaps we need a little more! [Rires.] The degrowthists, on the other hand, are taking an increasingly important place in our country. The problem is that you shouldn’t have a binary vision. The most interesting thing about degrowthists is their ability to consider radically changing our way of living, to invent new uses to reduce our carbon footprint. They are right to encourage this. It will have to be done. On the other hand, they are wrong to judge that technology is, in essence, opposed to workers. The techno solutionists, for their part, are right to think that there are enormous productivity gains to be made thanks to technology. Their mistake is to refuse to change some of their uses. In the United States, citizens hooked onAmerican way of life and their big car who think that technology will solve everything are making a mistake. If we synthesize these two trends while keeping only the best – productivity and new uses – we can decarbonize very quickly. If we lock ourselves into one or the other, we will not solve the climate crisis.

How has the left’s relationship to technological progress evolved?

François Ruffin’s book Their progress and ours is interesting, but I do not share its initial premise, which is to say that technology is structurally opposed to workers. During the decades of expansion following World War II, productivity gains were evenly distributed between capital and labor. It is a short memory to forget this and think that technology is inherently alienating. As long as the left does not overcome this bias, it will not formulate modern proposals. The real subject is the reappropriation of productivity gains, which have certainly not been significant in recent years, but which will become so with AI.

And the one on the right?

In the United States, the MAGA (“Make America Great Again”) right is incapable of thinking about future projects. In France, the right is becoming anti-globalization. There is a monumental misunderstanding of the great lessons of economics, the specialization of nations. The political world does not perceive the new paradigm that technology has given rise to. Of course, these complex subjects are not the easiest to raise in an electoral context. But it makes me think of this television performance by Valéry Giscard d’Estaing, in 1972, brilliantly exposing the thesis of the degrowthists. He explains that he does not agree with it, without belittling it. And he analyzes it with great clarity. These are object lessons that we are no longer capable of doing today.

Should we rethink the imagination of progress?

Progress, no, because progress refers precisely to the benefits of technology for the benefit of all. But we need to rethink what we call innovation. Is TikTok progress? This is subject to debate. As a liberal economist, I would call this the management of externalities. If I create a service that is popular but intoxicates kids and takes years of their attention, that poses a problem. I don’t like excessive control, but when the market creates negative externalities, we have to see them. The specialization of nations is very good. The market is efficient. But we must do this in the service of the common good. And when negative externalities appear, they must be controlled.

France, you write, has a great opportunity to seize in environmental AI. Why that ?

France is an unknown giant of the environment and carbon-free energy. We have world leaders in electricity at all levels: production with EDF and Engie, distribution with Sonepar and Rexel, manufacturing with Schneider and Legrand. The population is very aware of the climate issue. Besides, our decarbonization strategy is pretty good, whatever the Earth Uprisings say. There is also a dynamic network of French start-ups in the environmental field. Finally, France is a Colbertist country. This has its drawbacks, but constitutes a decisive advantage when seeking to have an impact on a country scale.

Are AI players taking up the climate issue vigorously?

All venture capitalists will tell you, there is a marked interest in impact technologies. Funds specializing in sustainable development have flourished. The stories themselves have changed. The startuper who shakes up the rules without worrying about the consequences is no longer the model to follow. Today, especially in France, it is that of the responsible entrepreneur who develops technologies with a positive impact.

The agricultural sector is a key domino in the climate transition. What role can AI play here?

Parametric agriculture powered by AI is a green revolution. It requires radically changing practices. But it makes it possible to significantly increase productivity, while massively reducing the need for inputs. [NDLR : engrais, pesticides, etc.], which significantly increase the sector’s greenhouse gas footprint. We often think, wrongly, that agriculture has reached its optimum. But if we look abroad, we see that we are far from having such high productivity. The Netherlands is the second largest agricultural exporter in the world. Certainly, many crops there are in greenhouses, but when we see the size of the country, compared to that of France, this should concern us.

And in global supply chains?

THE supply chains are at the heart of the productive system of tomorrow and its transformation towards a more resilient model. Covid and the war in Ukraine have underlined their importance. The beauty of AI is that it brilliantly manages contradictory injunctions: I don’t want to pay a lot, I want it to go quickly, I want buffer stocks but not too much, I want to reduce my carbon footprint… Almost insoluble equations. But AI figures out how to help the shareholder and the planet simultaneously. It reveals the narrow path that serves both parts at once.

Is the carbon footprint of digital and AI a problem, or is it likely to become one?

There is a lot of misinformation about the carbon footprint of AI and digital technology. AI will be everywhere tomorrow. We therefore obviously need to control its environmental impact. But for it to spread, it will have no other choice than to become very efficient. What happened with computers will happen in AI: their operation has been radically optimized, and their consumption has plummeted. That of major language models is already falling precipitously. Anthropic’s latest LLM, Claude 3, has reduced its consumption by a factor of 50. I am convinced that eventually large language models will run on smartphones, with 3 watts of energy. Which means that the problem of their consumption will have been resolved.

READ ALSO: Joëlle Barral (Google DeepMind): “In AI, France has many advantages”

AI on mobile opens up vast perspectives. Who are these “AI companions” that we will all have tomorrow?

This is the most spectacular transformation that AI will bring: in the medium term, we will be able to ask almost anything and everything from our smartphone: you want to make an appointment with an ENT, he will take care of it -even find a specialist of this type with a slot at times when you are free, in a perimeter that suits you; find a better route, it will suggest a route allowing you to be on time for your meeting while avoiding crowded trains. The next phase is when the system will become transactional and will be directly responsible for reserving train tickets or a hotel room, for modifying our meetings… But for this we will have to control the risks of AI hallucinations.

Methods, such as augmented recovery generation (“RAG” in English) seem to be an effective medicine against hallucinations. But will we ever be able to completely cure AI of this disease?

AIs are a bit mythomaniac, but various methods can reduce these hallucinations. The real question is the degree of acceptance. For critical systems, error is not allowed. But for more mundane actions, like planning meetings, this will happen because the risks will be moderate and the benefits large.