Demis Hassabis, CEO of Google DeepMind, believes that artificial intelligence capable of reasoning similarly to humans could emerge within the next 5–8 years. He made the statement during an AI Summit in New Delhi, according to Agerpres and reported by Economedia.
To measure progress toward that goal, Hassabis proposed a new benchmark called the “Einstein Test.” The idea involves training an AI system with all human knowledge available up to 1911 and then assessing whether it can independently develop the theory of general relativity, as Albert Einstein did in 1915.
“The concept consists of training an AI with all human information, but cutting off the data in 1911,” he explained during the summit.
Read also: Apple selects Google’s Gemini to power AI-based Siri launching this year
From digital encyclopedia to discovery engine
According to Hassabis, current systems act as encyclopedic experts but lack genuine creativity. “They only solve what already exists and do not generate new scientific hypotheses of that caliber,” the CEO emphasized.
The roadmap toward advanced reasoning involves combining the planning capabilities of systems such as AlphaGo with the large-scale processing power of modern foundation models. Models like Gemini are expected to play a key role, serving as structured representations of how the world works and enabling advanced learning tasks.
The long-term objective is to shift AI from a fast information processor to a discovery engine capable of producing original scientific breakthroughs and tackling problems beyond current human solutions.
Photo: freepik.com
UK
IT
ES
DE
FR
RO
PT
HU
