AI pioneer Yann LeCun doesn’t think artificial intelligence is on the verge of becoming truly intelligent.
LeCun, a professor at New York University, senior fellow at Meta, and winner of the prestigious AM Turning Award, has been open about his skepticism before. For example, before we started worrying about controlling superintelligent AI, we tweeted this: “This is the beginning of a design hint for a system smarter than a house cat.”
In an interview with the Wall Street Journal, he elaborated on his opinion when asked whether AI has become smart enough to pose a threat to humanity, replying, “You’ll have to forgive my French, but that’s complete BS.” ”
LeCun argued that today’s large-scale language models lack key cat-level features such as persistent memory, reasoning, planning, and understanding of the physical world. In his view, LLMs only show that “you can manipulate language, but you’re not smart” and will never lead to true artificial general intelligence (AGI).
He’s not a complete AGI skeptic. But he said a new approach would be needed. For example, he pointed out that Meta’s basic AI research team is working on digesting real-world videos.
