Demis Hassabis moves up AGI timeline, says it could arrive within two to three years
Demis Hassabis, the Google DeepMind chief executive and Nobel laureate, has sharply shortened his timeline for artificial general intelligence, saying in a Wall Street Journal interview that AGI could arrive within two to three years. The remark stands out because Hassabis has long been seen as one of the more cautious voices in AI. The report traces a series of revisions in his public forecasts, from saying in 2023 that AGI might take years or even a decade, to five to 10 years in January 2025, five years in February, and 2030 in April, before his latest two-to-three-year estimate. The interview also lays out Hassabis’ strict definition of AGI, including an "Einstein test" that asks whether an AI system given all human knowledge up to 1901 could independently derive the breakthroughs Einstein published in 1905. Hassabis said AI is still years away from independently inventing modern physics. He also discussed AI safety, warning at the Dublin international AI summit that the industry could face a serious safety incident in about two years if governments and companies do not raise standards. Hassabis said he drafted a proposal this summer for an industry-funded, federally supervised body to test frontier models before release, and argued that bad regulation could be worse than no regulation at all.
