We may question the extent of its scope, but generative AI is good at at least one thing: opening doors. At the last CES in Las Vegas, the big global high-tech gathering that took place in early January, it was in front of everyone’s eyes and at the front of every booth. Around the world, funding sources have dried up for a year, but those that manage to slip AI into a PowerPoint presentation still stand a chance of being picked up by investors.
Never before has a technological innovation spread so rapidly in use and yet its implementation remains largely dysfunctional. This is the topic of a new episode of DeepTechs, a podcast Challenges produced in collaboration with Delight. Our guest is artificial intelligence researcher Laurence Devillers, author of the book Robots and men (Plon publishing house).
There are still many misconceptions about these generative AIs, especially when it comes to their true potential. They are given the power to think intelligently or to instigate the great replacement of humans by machines. In Silicon Valley in particular, the Cassandras predict a Terminator-style future world where machines replace humans, adding to the buzz around the technology.
But the biggest specialists on the subject put the most imaginative minds at ease about the current capabilities of AI. Artificial intelligence does not exist, says Renault’s scientific director Luc Julia. Meta’s chief scientist, Yann Le Cun, one day sees that she is able to experience feelings, but not immediately. Very positively, when it comes to the use of these technologies, he thinks that “these machines must learn, interacting with the world around them, with emotions.”
In his latest book Artificial mind (Observatory), Raphaël Enthoven considers her incapable of philosophizing: “When we ask her to understand a problem, she finds herself like a chicken before a knife. » Laurence Devillers draws his revolver when he hears that AI-doped machines would be intelligent. We must remember, she says, how these intelligences were constructed: there is no intention, no emotion, no consciousness in the machine. “They play ratatouille with different information,” assures the researcher, “when they have lost all sources, they reconstruct improbable bits in the middle of plausible bits. »
They only have forms they have seen elsewhere and put them together like a jigsaw puzzle of billions of pieces. And finally, they can produce sensible things or fake things. They may also “hallucinate” when they absolutely want to answer an abnormal question. In order for us to answer “I don’t know”, they must have been programmed that way.
Ultimately, these AIs must be seen and used for what they are: disruptive technologies that now permeate our lifestyles, but are only there to assist, support, and empower our actions.