Now Reading: Demis Hassabis and Yann LeCun Clash Over Whether AI Can Achieve Human-Level Intelligence
-
01
Demis Hassabis and Yann LeCun Clash Over Whether AI Can Achieve Human-Level Intelligence
Demis Hassabis and Yann LeCun Clash Over Whether AI Can Achieve Human-Level Intelligence
Estimated reading time: 3 minutes
- Google DeepMind CEO Demis Hassabis and AI researcher Yann LeCun publicly disagreed on whether artificial intelligence can achieve human-level general intelligence.
- LeCun argued that intelligence, including human intelligence, inherently specializes, rejecting the idea that “general intelligence” exists at all.
- Hassabis countered that humans are broadly capable learning systems and that AI models can, in principle, reach similar levels of generality.
- Elon Musk supported Hassabis’s position, saying that AI achieving human-level intelligence is a matter of when, not if.
- The exchange highlights ongoing uncertainty and disagreement among AI leaders about the future limits and direction of artificial intelligence.
A renewed debate over artificial general intelligence has emerged among some of the most influential figures in artificial intelligence (AI). Google DeepMind CEO Demis Hassabis and prominent AI researcher and VP & Chief AI Scientist at Meta, Yann LeCun, are taking opposing views on whether machines can ever match human-level intelligence.
The disagreement centres on artificial general intelligence (AGI), a term widely used by companies such as Google and OpenAI to describe AI systems that could learn, reason, and adapt across a wide range of tasks in a human-like manner. Proponents argue AGI would allow machines to solve unfamiliar problems without relying solely on pre-training.
While today’s AI models, such as ChatGPT and Gemini, can outperform humans on specific benchmarks, including exams and advanced mathematics, they still struggle with everyday reasoning tasks. This gap prompted LeCun to question not only the feasibility of AGI, but the concept itself.
LeCun argues that there is no such thing as general intelligence, even in humans. He maintains that biology and evolution inherently specialize intelligence, shaping it to address human-specific challenges. Individuals excel in different areas, he says, and human capability varies widely across domains.
He has pointed to chess as an example. He noted that while machines can compute millions of possibilities per second, even elite players like Magnus Carlsen can only evaluate a limited number of moves.
LeCun has also criticised the idea that simply scaling data and computation will lead to human-like intelligence. Instead, he has argued for systems with longer memory and richer sensory understanding, cautioning against equating data ingestion with cognition.
Hassabis pushed back strongly against this view, saying LeCun was “confusing general intelligence with universal intelligence.”
According to Hassabis, the human brain remains one of the most broadly capable learning systems known. While constrained by biological and physical limits, he argues humans do possess general intelligence, and that AI systems can, in principle, reach a similar level of generality.
Hassabis pointed to the theoretical Turing Machine, a model capable of performing any computation given sufficient resources. He argued that human brains are approximate biological versions of such systems. He said modern AI foundation models are increasingly approaching this form of general learning capacity.
Musk briefly weighed in on the debate, responding on X with a short endorsement of Hassabis’s position. He stated, “Demis is right.”
Musk has repeatedly argued that the emergence of superintelligent AI is inevitable, while also warning about the risks it may pose.
Despite the exchange, LeCun has not retreated from his position. In a follow-up clarification, he said, “I object to the use of ‘general’ to mean ‘human level’, because humans are extremely specialised.”
The disagreement reveals a deeper divide within the AI research community. And this is not just over technical pathways, but over how researchers should define intelligence itself. As investment and expectations around AI continue to rise, there’s uncertainty and contestation surrounding the road to human-level machine intelligence.
Editorial Note: This news article has been written with assistance from AI. Edited & fact-checked by the Editorial Team.
Interested in advertising with CIM? Talk to us!
