Meta’s AI chief says world models are key to ‘human-level AI’ — but it might be 10 years out

Oct 17, 2024

Yann LeCun believes world models could enable human-level AI in a decade.

Meta’s AI chief says world models are key to ‘human-level AI’ — but it might be 10 years out

: Yann LeCun, Meta's chief AI scientist, suggests that achieving human-level AI may take a decade, requiring the development of 'world models.' Unlike current AI models, which predict tokens or pixels, 'world models' would allow machines to understand and plan within actual three-dimensional spaces. While companies like OpenAI explore this approach, it remains computationally demanding, with uncertain progress. Meta's research lab, FAIR, is dedicated to this long-term endeavor.

Yann LeCun, Meta's chief AI scientist, argues that current AI models are incapable of true human-like reasoning and planning. He believes that the key to achieving human-level AI lies in 'world models,' a concept that could take a decade to realize and involves creating AI systems capable of understanding and interacting with the three-dimensional world.

World models, unlike current language and image models, aim to enable machines to perform complex tasks by forming mental models of their environment. These models are computationally intensive because they can process more data, making them attractive to venture capitalists and cloud providers racing to partner with AI companies.

Several AI labs, including Meta's FAIR, are pursuing world model development. LeCun acknowledges the considerable challenges ahead, suggesting that while Meta's FAIR lab is focused purely on such long-term AI research, progress remains slow due to the tasks' complexity.