AI is currently as smart as a precocious adolescent nerd – plenty of book learning, but not very street-wise.
Yann LeCun, VP and chief AI scientist at Meta, however, believes a new model he’s developed called JEPA – Joint Embedding Predictive Architecture – can give AI the blueprints of human common sense in five to six years.
Before anyone gets excited or fearful about dealing with a truly intuitive AI, at the CES 2025 Era of AI is Here conference moderated by Wing Venture Capital head of research, Rajeev Chand, LeCun warned “It’s going to take some more time to actually make this work. It’s always harder than we think. People have been sort of overly enthusiastic about the capabilities of the new techniques we just came up with and turned out to be disappointed.”
To achieve AI common sense, LeCun argued for the abandonment of current generative, probabilistic, contrastive, and reinforcement learning AI approaches since they haven’t really achieved much beyond book learning. “We’re not even close to matching the understanding of the physical world of any animal, a cat or a dog. They have a much better understanding of the physical world than any system we have today,” LeCun opined.
Instead, LeCun is advocating for the more object-driven JEPA. LeCun believes current AI models are useful and should continue to be developed and deployed. “But as a path towards human-level AI,” LeCun offered, “they are kind of a distraction, a bit of an off-ramp. What you want is a system capable of doing reasoning and planning and has some understanding of the physical world.”
The first JEPA model, Image JEPA (I-JEPA), was announced by LeCun and Meta in June 2023, and LeCun believes JEPA could be part of Meta’s release 5 of Llama (Large Language Model Meta AI); Llama 4 could be released in the next few months.
While many in the AI world aren’t as bullish on JEPA as LeCun, he is bullish on the eventual development of intuitive AI agents with which humans would coexist in both the virtual and physical worlds.
“Every one of us would be running around with AI assistants within 10-15 years,” LeCun predicted. “It might live in our smart glasses or things like that, and they will be with us at all times. All of us would be like an executive in a large company with a staff of people following them, but they would be virtual people.”
See also: CES Spotlight: John Riddle, Head Of NATM, On Innovation, Collaboration, And The Future of Retail
