Demis Hassabis, founder of Google DeepMind and a Nobel laureate in chemistry, has moved his estimate for artificial general intelligence closer. He now believes AGI could arrive within two to three years, after previously giving a range of three to five years.
What Hassabis means by AGI
Hassabis defines AGI as a system that matches the full cognitive range of the human brain. That definition sets a broad bar: the idea is not simply a system that performs well at one task, but one that can cover the range of human cognitive abilities.
The estimate is Hassabis’s judgment about when such a system might appear. His changing forecast signals that he believes the possibility is drawing nearer, but a timeline is not itself evidence that AGI has arrived. The definition matters because it clarifies the ambition being discussed, while leaving open how to determine whether a system has met it.
An Einstein-inspired test
Hassabis has described an “Einstein test” as a way to think about that question. The test asks whether a system trained only on knowledge from before 1901 could reproduce Einstein’s 1905 papers.
The example frames AGI as more than recalling information or repeating familiar answers. It asks whether a system could produce work associated with a major intellectual advance, using only knowledge available before that work appeared. The test is a proposed litmus test, rather than a reported result showing that a system has passed it.
That distinction helps keep two questions separate: what capabilities would count as evidence of AGI, and when those capabilities might emerge. Hassabis’s two-to-three-year estimate addresses the second question; the Einstein test illustrates one way he thinks about the first.
A forecast with wider implications
Hassabis has compared AGI’s potential impact to “10 times the industrial revolution at 10 times the speed.” The comparison conveys his view that the effects could be both very large and rapid. It also makes the forecast relevant beyond the technical definition: if the technology develops on the timeline he expects, questions about preparation and oversight become pressing.
His comments do not specify exactly what changes would follow or how quickly particular effects would be felt. The comparison is a description of potential scale and speed, not a detailed prediction of outcomes. Still, it helps explain why a shorter AGI timeline is accompanied by discussion of safety and regulation.
Oversight and Hassabis’s current focus
Since stepping down from Google DeepMind, Hassabis has focused on Isomorphic Labs, Google’s drug discovery AI spin-off. He has also proposed an industry-funded oversight board under government control to vet models before release.
The proposal points to a role for review before powerful models reach the public. Hassabis has warned that bad regulation is worse than none, putting the quality of oversight at the center of the debate. His remarks therefore link the prospect of AGI to two practical concerns: how to assess what systems can do, and how to examine them before release.
For now, the two-to-three-year timeframe remains Hassabis’s estimate. His definition and proposed test show what kind of capability he has in mind, while his comments on oversight suggest that he sees preparation as part of the same conversation. Whether the forecast proves accurate is an open question, but his revised timeline makes the debate over AGI’s meaning and governance more immediate.