For the French researcher, head of artificial intelligence at Meta, however, we should not see AI as a threat.
It is one of the figures of artificial intelligence. Yann Le Cun, head of AI at Meta, made a detour to Paris this Wednesday for a round table at the Vivatech show. The opportunity for him to put a little serenity on this subject which impresses as much as it worries.
And clearly, the French researcher is one of the enthusiasts. “AI is intrinsically good because it makes people smarter, it’s an intelligence amplifier,” he says. “We shouldn’t see it as a threat, they will be like assistants in your daily life.”
However, he assures us that the potential of AI is enormous: “There is no doubt that AIs smarter than humans will eventually arrive,” explains Yann Le Cun. “Currently, AIs can surpass humans in some areas, but those are narrow areas.”
Best example according to him: some programs can certainly pass the lawyer’s exam but autonomous driving is still far from being a reality.
“Obeying Humans”
“We are therefore missing something very important,” he continues, recalling that the language models were ultimately quite limited. “Most human knowledge does not pass through language. Otherwise one could become a surgeon or a mechanic just by reading books.”
It remains to be seen what will become of the AIs when they have finally passed this stage of learning.
“The ‘purpose in life’ of every AI is to satisfy a ‘super-goal’,” he continues. “It could be anything, generate text, real, non-toxic… but it could also be ‘obey humans’”.
With an implacable argument against the “intelligence” of future AIs: “there is no correlation between “being intelligent” and “taking over the world”, we see it in politics, they are not necessarily the most smart people who take control”.
