Meta AI chief Yann LeCun discussed the limitations of current generative AI tools (via CNBC) Wednesday, suggesting that man’s best friend still has a leg up.
“What it tells you we are missing something really big … to reach not just human level intelligence, but even dog intelligence,” LeCun said at the Viva Tech conference in Paris, after noting that AI can pass the Bar exam to become a U.S. attorney but cannot learn to load a dishwasher.
“Most of human knowledge has nothing to do with language,” he said, “so that part of the human experience is not captured by AI.”
He said that because many systems are trained purely on text, they lack an understanding of “the underlying reality of the real world.” This circles back to a point LeCun has previously addressed in his emphasis on the value of teaching AI with videos.
“By training itself to predict what will happen in the video, it will produce hierarchical representations of the world,” LeCun wrote in a Meta AI post discussing systems to help AI learn and reason like animals do.
This may be why OpenAI reportedly used YouTube videos as training data for a while, despite such activity appearing to be a violation of YouTube’s rules regarding using its content for commercial purposes.
In any event, AI’s current limitations aren’t enough to stop companies from being cautious with the tech. Samsung has banned third-party generative AI use on its company devices, and Google parent company Alphabet recently took to warning its employees about the dangers of chatbots including its own Bard (a ChatGPT competitor).
And even if the tech isn’t as smart as a human yet, it’s still potentially capable of performing human jobs better than real people, as evidenced by customer satisfaction rankings referenced by the CEO of UK energy supplier Octopus Energy.