Meta’s AI Boss: Human Level AI Not Just Around the Corner

According to Yann LeCun, Meta’s chief AI scientist, large language models (LLMs) like ChatGPT and Claude cannot lead to human-level artificial intelligence (AI) in the near future. LeCun recently discussed artificial general intelligence (AGI), which refers to a theoretical AI system that can perform any task given the right resources, with Time Magazine. While there is no consensus on what an AI system needs to do to be classified as AGI, Meta CEO Mark Zuckerberg announced a shift towards AGI development. LeCun disagrees with Zuckerberg’s terminology and prefers to call it “human-level AI,” as humans are not general intelligences either.

LeCun believes that LLMs, including Meta’s LLama-2, OpenAI’s ChatGPT, and Google’s Gemini, are far from possessing the intelligence of a cat, let alone human intelligence. He emphasizes that computer systems struggle to reproduce tasks that humans easily accomplish, suggesting that achieving AGI or human-level AI requires significant perceptual changes.

LeCun also discussed the ongoing debate surrounding the potential threat posed by open source AI systems like Meta’s LLama-2. He dismissed the idea that AI poses an exaggerated danger to humanity. When questioned about the possibility of a human programming a goal of domination into an AI, LeCun argued that if such a “bad AI” existed, smarter and benevolent AIs would counteract it.