Yann LeCun, a distinguished professor at New York University and former chief AI scientist at Meta, has delivered a sobering critique of the artificial intelligence industry's current trajectory. Often referred to as one of the 'godfathers of AI', LeCun has declared artificial general intelligence (AGI) to be the most overrated concept in the field today.
The Fundamental Flaw in Current AI Systems
Speaking recently at Davos, LeCun presented a compelling argument that should concern everyone involved in AI development. He emphasized that the AI systems driving today's most significant breakthroughs are fundamentally flawed in their approach. According to LeCun, the industry's enthusiastic push toward 'agentic AI' – systems designed to autonomously take actions and complete tasks – represents a potential recipe for disaster.
Why Scaling Won't Solve the Problem
The pioneering computer scientist did not hold back when discussing the limitations of current large language models like ChatGPT and similar systems. His central claim challenges the entire direction of the AI industry. "We're not going to attain human-level intelligence by simply making these systems bigger or better," LeCun stated unequivocally. "We're not going to get to human-level intelligence or superintelligence by scaling up or refining the current paradigm. There is a need for a paradigm change."
LeCun's most alarming criticism targets the industry's ongoing obsession with 'agentic systems'. He argues that these systems are being built on large language models that lack a crucial capability: the ability to predict the consequences of their actions. "How can a system possibly plan a sequence of actions if it can't predict the consequences of its actions?" LeCun questioned. "If you want intelligent behavior, you need a system to anticipate what's going to happen in the world and predict the consequences of its actions."
The Real-World Understanding Gap
To illustrate this fundamental gap, LeCun provided a striking comparison. "The first time you ask a 10-year-old to solve a simple task, they will do it without necessarily being trained. Within the first 10 hours a 17-year-old drives a car; they can drive. We had millions of hours of training data to train autonomous cars, and we still don't have level-five autonomous driving. That tells you the basic architecture is not there."
According to the former Meta executive, the core problem lies in language models operating in a simplified universe. "The real world is way more complicated than the world of language," he explained. While many consider language to represent the pinnacle of human intelligence, LeCun insists that "predicting the next word in text is not that complicated."
True intelligence, he argued, requires understanding the physical world. "Sensory data is high-dimensional, continuous, and noisy, and generative architectures do not work with this kind of data," LeCun noted. "The type of architecture we use for LLM generative AI does not apply to the real world."
The Dangers of AI Consolidation
Beyond technical limitations, LeCun addressed what he considers the most pressing danger facing society: the consolidation of AI control among a handful of powerful companies. "Capture and centralized control of AI is the biggest danger," he warned, "because it will mediate all of our information diet."
Unlike those who fear killer robots or AI takeovers, LeCun's concerns focus on a future where our entire digital experience will be mediated by AI systems controlled by just a few proprietary companies based primarily on the West Coast of the United States or in China. "We're in big trouble for the health of democracy, cultural diversity, linguistic diversity, and value systems," LeCun stated. "We need a highly diverse population of AI assistants for the same reason we need diversity in the press, and that can only happen with open source."
The Shift Away from Open Research
The AI pioneer also expressed his dismay at the industry's gradual abandonment of open research principles. "The biggest factor in progress was not any particular contribution, it's the fact that AI research was open," he shared, describing how researchers traditionally published papers, shared code, and accelerated collective progress through collaboration.
"What's been happening the last few years, to my despair, is that increasingly more industry research labs have been closing up," he said, specifically pointing to OpenAI and Anthropic as organizations that have "never been open, in fact very closed." Even formerly open organizations like Google's and Meta's FAIR have become more restrictive in their approach.
Meanwhile, LeCun noted that "the best open-source models at the moment come from China, they're really good. Everybody in the research community is using Chinese models." This significant shift, he argues, is slowing Western progress at a critical moment in AI development when open collaboration is most needed.
LeCun's comprehensive critique presents a challenging perspective for an industry currently obsessed with scaling existing models and developing increasingly autonomous AI agents. His arguments suggest that without fundamental architectural changes and a return to open research principles, the AI industry may be heading toward significant limitations and potential societal risks.
