Three leading AI researchers offered sharply different views on the state of artificial intelligence this week, ranging from imminent technological singularity to outright dismissal of current models as truly intelligent.
Hassabis predicts AGI within five years
Demis Hassabis, co-founder of DeepMind, said humanity is now "standing in the foothills of the singularity." He made the remark at the close of his Google I/O 2026 keynote, around the 1:50:17 mark.
Hassabis expects artificial general intelligence (AGI) to become possible within the next five years. When it arrives, he said, it will be "10 times the industrial revolution at 10 times the speed." He called it a "profound moment for humanity."
The singularity concept, popularized by futurist Ray Kurzweil, refers to a point where AI surpasses human intelligence and triggers runaway technological growth. Hassabis, known for cautious public statements, surprised many with his optimistic timeline.
LeCun says intelligence is about doing, not knowing
Yann LeCun, AI researcher at AMI Labs, took the opposite stance. He argued that current large language models (LLMs) are not intelligent because real intelligence does not show up in accumulated knowledge or learned skills.
"Intelligence is not what you know, it's what you do when you don't know," LeCun wrote on LinkedIn, paraphrasing psychologist Jean Piaget.
LeCun has long advocated for AI systems that go beyond Transformer-based LLMs. In the past, he has debated even a DeepMind researcher about whether LLMs can achieve child-like learning, which he sees as a necessary precursor to true intelligence. His current work focuses on AI architectures that break free from the limitations of today's language models.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Vinyals splits the difference
Oriol Vinyals, co-lead of Google's Gemini program, offered a more measured view. He noted that today's models are strong at coding and mathematics, and their reasoning capabilities are becoming more general.
If someone had shown him these models seven years ago, Vinyals said, he probably would have called them AGI. But he added that the ability to learn from experience and produce real breakthroughs is still missing.
Vinyals' comment highlights the rapid pace of progress while acknowledging that current systems remain narrow compared to human adaptability.
Context and significance
The debate reflects a fundamental divide in the AI community. One camp, represented by Hassabis, sees exponential progress hurtling toward superhuman intelligence. Another, led by LeCun, argues that today's statistical patterns do not constitute understanding or reasoning. Vinyals occupies a middle position, impressed by performance but aware of limitations.
All three researchers agree that AI is advancing quickly, but they disagree on what counts as true intelligence and how close the field is to achieving it. The discussion has practical implications for investment, regulation, and public expectations.
Hassabis's singularity prediction is among the boldest publicly stated timelines from a major AI leader. LeCun's critique challenges the entire premise of scaling up language models. Vinyals' perspective grounds the conversation in measurable progress and remaining gaps.
Related on Neura Market
- AI Models Directory, Browse the latest AI models and benchmarks
- AI Research News, Stay updated on breakthroughs and debates
- AGI and Singularity Coverage, Track developments in AGI timelines