Can AI Ever Match Humans? DeepMind’s Demis Hassabis, Yann LeCun and Elon Musk Spark Fresh AGI Debate
The question of whether artificial intelligence can ever truly rival human intelligence is once again at the centre of a heated debate—this time involving some of the most influential minds in the field. Google DeepMind CEO Demis Hassabis and renowned AI researcher Yann LeCun are publicly disagreeing over what “general intelligence” really means and whether machines can achieve it. Adding fuel to the discussion, Tesla and X owner Elon Musk has openly sided with Hassabis.
At the heart of the disagreement is the idea of Artificial General Intelligence (AGI). Unlike today’s AI tools such as ChatGPT or Gemini, which excel at specific tasks, AGI is imagined as a system that can learn, reason, and adapt across a wide range of problems—much like humans do. Proponents believe such systems could handle unfamiliar situations by learning on the fly, instead of relying purely on pre-trained data.
Despite impressive achievements—solving complex exams or winning math competitions—current AI systems remain far from this vision. Their abilities still fall short of basic human reasoning in everyday, real-world situations. This gap has led Yann LeCun to question not just the feasibility of AGI, but the very concept of “general intelligence” itself.
LeCun argues that intelligence, even in humans, is not general at all. Instead, he says it is deeply specialised and shaped by biology and environment. Humans, he notes, evolved to solve human-specific problems, and even among people, intelligence varies widely. To illustrate this, he points to chess: while machines can calculate millions of possibilities instantly, even elite players like Magnus Carlsen can only evaluate a limited number. For LeCun, this difference highlights why comparing machine and human intelligence using a single “general” label is misleading.
Demis Hassabis strongly disagrees. Responding to LeCun on X, he said LeCun was “confusing general intelligence with universal intelligence.” According to Hassabis, human brains are among the most flexible and general learning systems known. While they operate under biological limits, they are still capable of learning an enormous variety of skills over a lifetime.
Hassabis further supports his argument by referencing the theoretical Turing Machine—an abstract model capable of performing any computation with enough memory and resources. He suggests that human brains are approximate biological versions of such machines, and modern AI foundation models are increasingly moving in that direction. Neither humans nor machines are perfect, he admits, but both are general enough to learn a vast range of tasks.
Elon Musk quickly weighed in on the exchange, backing Hassabis with a succinct endorsement: “Demis is right.” Musk has long maintained that superintelligent AI is inevitable, often warning that it represents both a major risk and an unprecedented opportunity for humanity.
LeCun, however, has not softened his stance. In a follow-up, he clarified that his objection is mainly about terminology. “I object to the use of "general" to designate "human level" because humans are extremely specialised,” he said. The debate, it seems, is far from settled—but it underscores how uncertain and contested the future of AI intelligence remains.