Editorial illustration for AI Pioneers LeCun and Hassabis Clash Over Definition of General Intelligence
AI Leaders Clash Over General Intelligence Definition
LeCun and Hassabis dispute meaning of ‘general intelligence’
The debate over artificial general intelligence often feels like a battle over a mirage. Yann LeCun, Meta’s chief AI scientist, says the mirage is us. Humans, he argues, are not broadly general, we are exquisitely specialized creatures, trapped inside the narrow sliver of problems we can even conceive.
“It’s simply an illusion,” he insists. Demis Hassabis, the co-founder of DeepMind, sees it differently. He shoots back that LeCun is mixing up general intelligence with universal intelligence.
Brains, he writes, are “the most exquisite and complex phenomena we know of in the universe (so far),” and they are in fact extremely general. Yes, the no free lunch theorem applies, no system can be optimal for all tasks. But a genuinely general system, he argues, can still learn any computable function in principle, given enough time and memory.
Human brains, like today’s AI foundation models, are “approximate Turing machines.” The clash is not just semantic. It cuts to the core of what intelligence even is, and whether our own minds are as open-ended as we like to believe.
This, he said, shows that humans are not broadly general but highly specialised. "We think of ourselves as being general, but it's simply an illusion because all of the problems that we can apprehend are the ones that we can think of," LeCun said. Hassabis responded that LeCun was conflating general intelligence with universal intelligence.
"Brains are the most exquisite and complex phenomena we know of in the universe (so far), and they are in fact extremely general," he wrote in his post on X. He argued that while no system can escape the no free lunch theorem, a general system can still learn any computable function in principle. "In the Turing machine sense, the architecture of such a general system is capable of learning anything computable given enough time and memory," he said, adding that human brains and AI foundation models are "approximate Turing machines".
This dispute is not a debate over definition. It is a referendum on what we believe intelligence actually is, and whether we are brave enough to see our own limits. LeCun’s point is uncomfortable: we mistake our narrowness for breadth because the universe only shows us problems we are built to solve.
Hassabis counters with a deeper claim: brains may be imperfect, but their architecture is general in the only sense that matters, given time, they could learn anything. He is betting that foundation models are the same kind of machine. Both men are right about something critical.
LeCun reminds us that generality is always constrained by hardware and evolution. Hassabis insists that constraint is not the same as specialisation. The real question is whether those constraints can be stretched far enough to match the wild, messy, open-ended world we actually inhabit.
The answer will not come from a single tweet or a laboratory showdown. It will emerge from the systems we build next, and from the uncomfortable truth that even a general intelligence must live inside a finite skull.
Related Reading
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
From Archives Qwen3-4B-Instruct-2507: 4B-parameter model boosts Raspberry Pi AI
From Archives Dell and NVIDIA host AI developer meetup in Bengaluru on deployment trade-offs
Common Questions Answered
How do LeCun and Hassabis differ in their views on human and machine general intelligence?
LeCun argues that human intelligence is actually highly specialized and limited by our cognitive frameworks, challenging the notion of true generality. Hassabis counters this by emphasizing that brains are extremely complex and general, representing the most sophisticated cognitive systems known in the universe.
What philosophical tension exists between Meta's and DeepMind's perspectives on artificial intelligence?
The debate centers on defining what constitutes 'general intelligence' and whether humans or AI systems can truly be considered broadly adaptive. LeCun and Hassabis represent different interpretations of cognitive capability, with LeCun suggesting human intelligence is more constrained than commonly believed.
Why does LeCun claim that human perceived generality is an 'illusion'?
LeCun argues that humans only perceive problems they can conceptualize, which inherently limits their cognitive range and specialization. This perspective suggests that our understanding of intelligence is fundamentally restricted by our existing mental frameworks and ability to comprehend complexity.
Further Reading
- Papers with Code Benchmarks — Papers with Code
- Chatbot Arena Leaderboard — LMSYS
- Google DeepMind CEO Demis Hassabis says AI scaling 'must be pushed to the maximum' — Business Insider
