He harbors no sympathy for billionaire Elon Musk and doubts that six months of « pause » in the development of artificial intelligence (AI) are enough to stem the risks to humanity. So why did Raja Chatila sign, on Wednesday March 29, alongside hundreds of world experts in the sector, an open letter calling “all AI labs to immediately suspend for at least 6 months training on AI systems more powerful than ChatGPT”? “Something had to be done to make the public, the media and policy makers realize there is a problem”, simply answers this 71-year-old French roboticist, emeritus since 2020 from Sorbonne University. A ” Red card “in short.
The problem in question is these so-called generative AIs, capable of creating original content (text or image) at the simple written request of a user. Or rather, the very wide distribution of these technologies since the release in November of a conversational agent now known around the world. “The texts produced by ChatGPT are so similar to human speech that we tend to treat them as such”deplores this CNRS researcher. “Except that the true and the false mingle in an imperceptible way… The risk is a complete relativism with regard to the truth. »
Atomic bomb
However the truth, for the one who directed two important laboratories of the CNRS (1), has nothing relative. By June, this native of Damascus, Syria, will have co-published with several colleagues an ethical opinion on generative AI, at the request of the Minister Delegate for Digital Affairs, Jean-Noël Barrot. This will be the fifth opinion of the National Digital Ethics Pilot Committee (CNPEN), of which Raja Chatila has been a member since its creation in 2019. In the meantime, whoever pleads for rapid regulation – “and coercive! » – of these AI systems will undoubtedly have the opportunity to discuss the subject with the engineering students to whom he has been teaching the ethics of technologies for five years.
Alerting the general public to the risks raised by techniques to which we ourselves have devoted our lives… Paradox? “I have famous predecessors”, smiles the interested party, before quoting the American physicist Robert Oppenheimer. The one that history has remembered as the ” father “ of the atomic bomb had indeed opposed military nuclear research after Hiroshima, of which he carried the guilt all the end of his life.
The mobile robots on which Raja Chatila has worked for four decades, since a pioneering thesis in 1981, are certainly much less offensive. And the academic does not question the relevance of his work, which has paved the way for today’s autonomous cars. But this word “autonomous”, precisely, he handles with the greatest precaution; to believe that he is trying to avoid it. As well as the now hackneyed term “intelligent”, which the industry uses at will to designate all kinds of connected objects.
Humanoid robots
Raja Chatila, who initially considered a career in astrophysics, began his shift towards ethics – a subject of which he deplores the lack of success with his scientific colleagues – in the 1990s, after several “shocks” successive: the first American “battlefield robotization” program, then the first Japanese humanoid robot. “The machines I was working on looked like carts… All of a sudden, we discovered bipedal robots – which of course had a feminine appearance – with some even a silicone skin. The resemblance to the human was extremely disturbing. » But such a resemblance, the researcher is convinced, brings nothing but confusion.
This confusion between what is human and what is not echoes the tragic news of his country of origin, Syria, where his sister still lives. “The dehumanization of the enemy is the condition for being able to kill him”he believes, before confiding that having always considered the Syrian people to be peaceful, he did not expect the deployment of“such savagery” in this war.
———
His compass: “What is intelligence? »
“As a scientist, this question is fundamental for me. However, for the first time in the history of humanity, we have a tool, the computer, capable of things that resemble what the brain does. This “universal computing machine” manipulates data to produce new ones. Can it be developed in such a way as to simulate human intelligence? This question, which presided over the birth of artificial intelligence in the 1950s, I cannot avoid asking myself. Developing robots was above all a way for me to respond to this. But I also took part in industrial applications with a more obvious “usefulness”. In particular, I worked on a robot which was to be sent to Mars in 2022. still been the case. »