DeepMind Chief Demis Hassabis Predicts AGI Could Arrive Within Three Years

Demis Hassabis, co-founder of Google DeepMind and a Nobel Prize-winning scientist, has reportedly revised his forecast for the arrival of artificial general intelligence (AGI), saying the technology could emerge within the next two to three years. In an interview with The Wall Street Journal, Hassabis indicated that his estimate had moved closer as artificial intelligence systems continued to improve. Earlier in 2026, he had generally discussed a longer timeline of approximately three to five years. His latest prediction reflects his view that advances in AI reasoning, coding and autonomous agents are bringing researchers closer to developing systems capable of performing a broad range of intellectual tasks at a human level. However, the timeline remains a personal forecast rather than an established scientific conclusion.

AGI refers to a hypothetical stage of AI development in which a system can perform a wide range of cognitive tasks at a level comparable to humans, rather than excelling only in specific areas. Current AI models can write code, analyse information, generate images and assist with scientific research, but they still have limitations in areas such as reliable long-term planning, continual learning and adapting to unfamiliar situations. Hassabis has argued that achieving AGI will require systems that demonstrate a much broader range of capabilities than those measured by conventional benchmarks. His definition emphasises general intelligence across different domains, making the question of whether AGI has arrived more complicated than simply comparing an AI model’s score with human performance on a test.

To make the concept more concrete, Hassabis has proposed what he calls an “Einstein test.” Under this idea, an AI system would receive only scientific knowledge available before the early twentieth century and then be challenged to independently develop insights comparable to Albert Einstein’s groundbreaking work in 1905. The proposal is intended to test whether AI can generate genuinely new scientific ideas rather than merely reproduce information contained in its training data. Such a demonstration would represent a significant milestone in machine intelligence. However, the test remains a proposed benchmark, and no publicly established result shows that a current AI system has passed it. Researchers also continue to debate how AGI should be defined and which experiments would provide convincing evidence that it has been achieved.

Hassabis believes the potential consequences of AGI extend far beyond the technology industry. He has repeatedly highlighted its possible contribution to scientific discovery, particularly medicine, biology and the development of new treatments. More capable AI systems could help researchers analyse complex datasets, identify promising drug candidates, formulate scientific hypotheses and accelerate experiments. His work at DeepMind has already demonstrated how AI can contribute to scientific research, most notably through AlphaFold, which transformed the prediction of protein structures. Nevertheless, the benefits of future systems will depend on their reliability, the quality of scientific validation and the ability of researchers to understand and verify their conclusions. Even if AI accelerates research, translating discoveries into approved treatments and practical technologies will still require further testing and human expertise.

Alongside his optimistic outlook, Hassabis has stressed the importance of AI safety and responsible governance. He has warned that increasingly capable systems could introduce serious risks if their behaviour is not adequately evaluated before deployment. His broader proposals have included stronger oversight, with governments and industry working together to establish proportionate safety assessments for advanced models. However, there is no universal agreement among researchers about when AGI will arrive, how it should be measured or what regulatory framework would be most effective. Some experts expect rapid progress, while others believe substantial technical breakthroughs are still required. Hassabis’s two-to-three-year prediction therefore signals his assessment of the current pace of development, not a guaranteed deadline. Whether his forecast proves accurate will depend on technological progress, scientific evidence and the emergence of systems that can demonstrate genuinely general capabilities.