Demis Hassabis moves AGI timeline to 2-3 years, with an "Einstein test" as the bar

Demis Hassabis moves AGI timeline to 2-3 years, with an "Einstein test" as the bar

N
News Editor
2026-10-23 03:25:56
Google DeepMind chair Demis Hassabis has shortened his expected timeline for artificial general intelligence, saying in an interview cited by The Wall Street Journal that AGI could arrive within two to three years. Earlier this year, he had placed the window at three to five years. In Hassabis’ framing, AGI is not a system that handles only part of human capability, but one that covers the full cognitive range of the human brain, remains stable, and can produce original thought. He also described a benchmark he calls the "Einstein test." The idea is to train a model only on human knowledge available before 1901, then see whether it can independently reach the breakthroughs Albert Einstein published in his 1905 "miracle year," including work on the photoelectric effect, Brownian motion, special relativity, and mass-energy equivalence. Hassabis said AI that can invent modern physics is still years away, but closer than he had expected when he received the Nobel Prize. He added that one of the first things he would want to do with AGI is use it to understand the nature of reality and derive the next layer of physics.

Google DeepMind chair Demis Hassabis has again pulled forward his timeline for artificial general intelligence, or AGI.

According to an interview by Wall Street Journal reporter Cohen, Hassabis now expects AGI to arrive within two to three years. Earlier this year, he had said the technology was still three to five years away.

How Hassabis describes AGI

In Cohen’s account, Hassabis sees AGI as a model with the full cognitive range of the human brain, not just a subset of human abilities. He said general intelligence should be stable and capable of original thinking. For him, the "G" in AGI matters as much as the AI itself.

The "Einstein test"

Hassabis uses what he calls the "Einstein test" as a threshold for AGI. The idea is to train a model only on human knowledge available before 1901 and then ask whether it can produce the results Einstein published in his 1905 "miracle year."

That year, Einstein published papers on the photoelectric effect, Brownian motion, special relativity, and mass-energy equivalence. Put simply, the test asks whether AI, without access to later developments in physics, can re-derive modern physics on its own.

As relayed in the article, Hassabis said AI that can invent modern physics is still years away, but that point will come sooner than he expected when he won the Nobel Prize.

Asked what he most wants to do with AGI, he said, "I want to use AGI to understand the nature of reality ... to try to derive the next layer of physics. That’s probably one of the first things to do."

How this compares with his earlier public comments

The new two-to-three-year estimate can be compared with Hassabis’ public remarks from earlier this year.

On May 19, at the end of Google I/O’s keynote, he said, "Looking back on this period, I think we’ll realize that we were standing at the foothills of the singularity."

According to a May 28 Computerworld report citing an Axios interview after I/O, Hassabis said AGI would likely arrive around 2030, possibly in 2029 or even earlier. From a May vantage point, that put the timeline roughly three to four years out, broadly in line with the three-to-five-year range Cohen said he had given earlier in the year.

This latest view pushes the lower end forward again, pointing to roughly 2028 to 2029. If his comments cited by BlockTempo in March 2025 are also included, he had then said AGI was still five to 10 years away. On that comparison, the timeline has been cut by more than half in a little over a year and a half.

The debate is still about definition

One unresolved issue is the definition of AGI itself.

The article notes that some people see Hassabis’ definition as one of the strictest among major lab leaders. Others argue that by the time AI has all human capabilities, it may already be superhuman across dozens of fields, which makes AGI a poor finish line in the first place.

Under that view, the same year can mean very different things depending on how AGI is defined.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
7800

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.

Demis Hassabis moves AGI timeline to 2-3 years, with an "Einstein test" as the bar | Bit.Fan