Reasoning, autonomous creativity, and the "Einstein Test": Google DeepMind's roadmap to AGI
The arrival of a general artificial intelligence (AGI) capable of reasoning similarly to humans will occur within a timeframe of 5 to 8 years. This was stated by Demis Hassabis, CEO of Google DeepMind, during his speech at the AI Impact Summit 2026 held in New Delhi, India.
The "Einstein Test" and the limits of encyclopedic AI
When assessing the current state of large language models, Hassabis warned that contemporary tools function as encyclopedic experts that solve problems based on existing information, but still lack a real creative dimension to generate new scientific hypotheses.
To measure the true leap towards human capability, the executive proposed a new evaluation standard called "Einstein Test":
- Methodology: It consists of training an AI model with all accumulated human knowledge, applying a strict cutoff of data in the year 1911.
- The challenge: To observe if the algorithm is capable of independently deducing and inventing the General Theory of Relativity, just as Albert Einstein did in 1915.
The architecture of the future: Planning and scale
To move beyond the phase of data imitation and turn AI into a true engine of scientific discovery, Hassabis outlined a technical roadmap:
- Combination of architectures: Integrate the planning and strategic searching capabilities of systems like AlphaGo with the processing scale of modern foundational models.
- The role of Gemini: Models like Gemini will act as a map of how the world operates, on which processes of learning and advanced reasoning can be executed.
The meeting in the Indian capital brought together over 20 heads of state and 500 leaders from the tech industry to discuss the rules and security of the digital economy for the next decade.
From the editorial perspective of next+, Demis Hassabis's statements mark a turning point in the narrative around enterprise artificial intelligence. The competitive frontier will no longer be about building models capable of summarizing documents or programming repetitive code, but about developing systems equipped with planning and autonomous scientific inference capabilities. For directors of innovation, strategy, and technology, the transition from "encyclopedic" AI to "discovery" AI will dramatically change corporate R&D models, requiring governance frameworks ready to validate hypotheses and discoveries generated autonomously by machines.