Meta’s chief scientist Yann LeCun says a “new paradigm of AI architectures” will emerge in the next three to five years, going far beyond the capabilities of existing AI systems.
LeCun also predicted that the next few years could be the “decade of robotics,” where advances in AI and robotics combine to unlock a new class of intelligent applications.
Speaking at a session called “Debating Technology” in Davos on Thursday, LeCun said the “flavor of AI” we have at the moment — namely, generative AI and large language models (LLM) — isn’t for everyone. so much. It’s useful, sure, but it fails on many fronts.
“I think the shelf life of the current paradigm (LLM) is pretty short, maybe three to five years,” LeCun said. “I think within five years, nobody in their right mind would be using them anymore, at least not as the central component of an AI system. I think (….) we will see the emergence of a new paradigm for AI architectures that may not have the limitations of current AI systems.”
These “constraints” prevent truly intelligent behavior in cars, LeCun says. This depends on four main reasons: lack of understanding of the physical world; lack of continuous memory; lack of reasoning; and lack of complex planning skills.
“LLMs really aren’t capable of any of that,” LeCun said. “So there will be another AI revolution over the next few years. We may have to change its name, because it probably won’t be generative in the sense that we understand it today.”
“World Models”
This echoes sentiments LeCun has espoused in the past. At the heart of this is what will be known as “world models” that promise to help machines understand real-world dynamics. This includes having memory, common sense, intuition, reasoning skills – traits far beyond those of current systems, which are primarily concerned with pattern recognition.
Previously, LeCun has said that could still be about 10 years away, but today’s estimate brings things closer to the horizon. Although to what extent it will reach in that time frame is not exactly clear.
“LLMs are good at manipulating language, but not at thinking,” LeCun said. “So that’s what we’re working on — having systems that build mental models of the world. If the plan we’re working on succeeds, on the schedule we hope, within three to five years we’ll have systems that are a completely different paradigm. They may have a common sense level. They may be able to learn how the world works by observing the world and perhaps interacting with it.”
“Decade of Robotics”
As impressive as generative AI is, capable of passing the bar exam or discovering new drugs, LeCun thinks robotics could be a central component of the next wave of AI applications in such real-world scenarios.
Meta himself is doing some research work in the field of robotics, but so is the AI darling of the moment, the creator of ChatGPT OpenAI. Earlier this month, new job listings appeared detailing a new OpenAI robotics team focused on “general-purpose,” “adaptive,” and “versatile” robots capable of human-like intelligence in environments the real world.
“We don’t have robots that can do what a cat can do – a cat’s understanding of the physical world is far superior to anything we can do with AI,” he said. “Maybe the next decade will be the decade of robotics, maybe we’ll have AI systems that are smart enough to understand how the real world works.”
