LLMs will need to store memories similar to human brain functions, such as working memory and long-term memory, which current AI lacks. Another essential step, according to LeCun, is enabling these systems to plan and reason.
Artificial intelligence may need more than a decade to reach human-like intelligence, suggests Yann LeCun, Meta’s Chief AI Scientist. Speaking at Meta’s Build with AI Summit in Bengaluru, LeCun indicated that achieving human-level intelligence, or what Meta calls “advanced machine intelligence” (AMI), is still quite a way off.
Speaking to the Economic Times, LeCun explained that building AI capable of understanding the world like humans will require significant progress.
He emphasised that the journey towards advanced AI will involve systems that are not only aware of the physical world but also possess persistent memory and reasoning abilities. Even after six to ten years of development, he believes, the intelligence of such systems might only rival that of a cat.
Learning from videos, memory, and reasoning are key
LeCun described the steps Meta is taking to bridge the intelligence gap. He shared that the company is developing systems that learn from observing videos, enabling AI to better understand the physical world.
Additionally, large language models (LLMs) will need to store memories similar to human brain functions, such as working memory and long-term memory, which current AI lacks. Another essential step, according to LeCun, is enabling these systems to plan and reason.
Although progress may take several years, LeCun doesn’t expect the first versions to achieve full human-level intelligence. He likened the early systems to having the cognitive capacity of a cat.
Regulation could slow AI innovation
LeCun also expressed concerns about global regulations, which he feels are often based on unfounded fears around AI. He mentioned that these regulatory uncertainties could hinder innovation, citing the example of Meta’s smart glasses, developed in partnership with Ray-Ban. The glasses can translate text in real-time, but they are not available in Europe due to unclear legal frameworks.
Content retrieved from: https://www.firstpost.com/tech/ai-will-need-at-least-a-decade-to-reach-human-like-intelligence-claims-metas-chief-ai-scientist-yann-lecun-13828929.html.
