Whether human intelligence is broadly general or highly specialized has emerged as a major issue within the AI community, and two leading AI researchers, Yann LeCun and Demis Hassabis, have differing views.
LeCun, who is retiring as head of AI at Meta, said the concept of general intelligence does not exist, as it refers to human-level intelligence that is hyperspecialized rather than general. Although humans navigate real-world environments and deal well with other people, LeCun argues that humans perform poorly on structured tasks like chess.
“We think we're common, but that's just an illusion, because all the problems we can understand are the problems we can think about,” the Turing Award winner said in a recent podcast appearance.
Reacting to his remarks, Google DeepMind CEO Demis Hassabis said LeCun is “clearly wrong” because he “confuses general intelligence with universal intelligence.” “The brain is (so far) the most elaborate and complex phenomenon we know of in the universe, and it is actually quite common,” Hassabis wrote in a post to X.
“Obviously, you can't get around the no-free-lunch theorem, so in any practical, finite system, you always need some degree of specialization in terms of the target distribution you want to learn,” he said.
In machine learning (ML), the no free lunch theorem means that no single machine learning algorithm is the universally best performing algorithm for all problems. For example, an AI/ML algorithm can analyze X-rays with maximum accuracy, but it cannot be good at predicting the stock market.
“But the point about generality is that in theory, in the sense of a Turing machine, the architecture of such a general system can learn anything that can be computed, given enough time and memory (and data), and the human brain (and the underlying AI model) is an approximate Turing machine,” the Nobel Prize-winning AI researcher argued.
Yang is clearly wrong here. He confuses general intelligence with universal intelligence.
The brain is (so far) the most elaborate and complex phenomenon we know of in the universe, and it is actually quite common.
Obviously, there's no getting around free lunch… https://t.co/RjeqlaP7GO
— Demis Hassabis (@demishassabis) December 22, 2025
A Turing machine is a virtual machine that can be used to simulate any computer algorithm and is used to explore the limits of what is computable.
Story continues below this ad
Hassabis pushed back on LeCun's chess analogy, dismissing the idea that human performance in a narrow domain undermines generality. “It's amazing that humans were able to invent chess in the first place (and every other aspect of modern civilization, from science to the 747!), much less invent it with the genius of someone like Magnus (Carlsen),” he said.
“He (Carlsen) may not be exactly optimal (after all, his memory is limited and he has limited time to make decisions), but given the brain that evolved for hunter-gatherers, what he and our brains can do is incredible,” the DeepMind co-founder added.
I think this disagreement is mainly a matter of vocabulary.
I object to using “general” to refer to “human level” because humans are highly specialized.You may disagree that the human mind is specialized, but it is. It's not just a matter of theoretical ability…
— Yan Lecun (@ylecun) December 22, 2025
The exchange between LeCun and Hassabis highlighted wide differences in their views on the path to achieving artificial general intelligence (AGI), a virtual level of intelligence in which AI models perform tasks better than or comparable to human performance.
Demis Hassabis believes that scaling existing large-scale language models (LLMs) is not enough to achieve AGI; one or two more breakthroughs are needed.
LeCun, on the other hand, contends that an LLM is a dead end because it does not allow for continuous learning. Instead, he supported the development “World Model”It serves as an internal representation of the real world, incorporating physics, causality, and temporal dynamics. He has also previously said that he prefers the term “advanced machine intelligence” to AGI.
