DeepMind CEO: 'AGI' Is Coming Soon, But Here's the Test It Must Pass First

Google’s DeepMind CEO Demis Hassabis claimed artificial general intelligence (AGI) will emerge within the next six years, estimating its arrival around 2030. Speaking at Google I/O’s opening keynote, Hassabis dismissed AGI as a single disruptive event, instead describing it as a series of incremental upgrades that will unlock transformative innovation. He framed the process as a two-step approach: first solving intelligence itself, then applying it to broader challenges. To determine AGI’s achievement, Hassabis proposed an 'Einstein Test': training an AI system on physics knowledge up to 1901 and assessing whether it could independently replicate Einstein’s groundbreaking 1905 discoveries. He acknowledged current AI systems fail this test but expressed confidence they will eventually pass it. When asked about Google’s competitive edge in AI, Hassabis rejected the idea of a single winner, calling the current landscape the most intense tech competition in history. He noted Google’s strength lies in its broad research capabilities and ability to integrate professional-grade AI into consumer-facing services, such as its newly launched AI Mode in web search, which he described as 'unbelievably popular.' The discussion touched on growing public skepticism toward AI, highlighted by a recent *Wall Street Journal* report and an incident where former Google CEO Eric Schmidt was booed at a University of Arizona commencement for advocating AI’s potential. Hassabis acknowledged the backlash but reiterated Google’s commitment to developing AI for broad societal benefit. Hassabis emphasized that AGI’s development will not be dominated by one company, as top-tier AI research is distributed across multiple organizations. He stressed the collaborative nature of the field, where advancements depend on collective progress rather than individual competition.