Blab AI
DeepMind CEO Demis Hassabis Discusses AIs Past, Present, and Future at Stanford GSB
On June 18, 2026, the Stanford Graduate School of Business staged a live dialogue between DeepMind’s co‑founder and chief executive, Demis Hassabis, and Stanford University President Jonathan Levin. Filmed for the GSB’s Applied AI initiative, the conversation traced DeepMind’s evolution—from its earliest experiments to its most recent breakthroughs—and explored the company’s stance on artificial general intelligence, regulatory hurdles, and the broader impact of AI on society.
Hassabis opened by charting his own journey. He described how a childhood obsession with chess, a stint in video‑game development, and a fascination with neuroscience all converged on a single ambition: to build an AI that learns and reasons like a human. “I think there’s several through lines actually with what seems maybe somewhat unconnected subjects,” he said, noting that his chess training taught him to decompose ambitious goals into manageable steps and that game‑development provided a sandbox for testing reinforcement‑learning algorithms.
The discussion then pivoted to AlphaGo, DeepMind’s 2016 triumph over world champion Lee Sedol. Hassabis recounted the early days of training a model on Atari games, where the system initially struggled to score a single point in Pong. He said the breakthrough arrived when the model “started winning a lot of points” and “then it started winning the games.” The success of AlphaGo paved the way for AlphaZero, MuZero, and other systems that generalized the same approach to chess, shogi, and beyond.
Next came AlphaFold, the protein‑folding AI that addressed a decades‑old scientific challenge. Hassabis explained that the project was driven by the need to predict three‑dimensional protein structures from amino‑acid sequences. “If you could unlock that and find the structures of proteins, that would unlock whole new avenues of research,” he said. AlphaFold’s 2020 release achieved unprecedented accuracy, and by 2025 DeepMind had made predictions for more than 200 million proteins publicly available. The decision to release the model openly, Hassabis noted, stemmed from a belief that its benefits would be far greater if it were freely accessible.
A central theme of the talk was the prospect of artificial general intelligence and the so‑called “singularity.” Hassabis warned that AGI could have an impact “10 times the Industrial Revolution… 10 times faster.” He characterized the current AI landscape as “the most ferocious competitive environment” and cautioned against a “race to the bottom” among leading laboratories.
Regulation surfaced as a key concern. Hassabis argued that the pace of AI development outstrips traditional regulatory mechanisms. He said, “The hard part there, of course, is that anything to do with regulation, it’s too slow.” He called for a “fleet‑footed” approach that could adapt quickly to new developments and involve leading labs in designing safety checks. The conversation also touched on public perception, noting that many Americans view AI with skepticism. Hassabis suggested that demonstrating tangible benefits—such as accelerated drug discovery—could help shift public opinion.
In the final segment, Hassabis highlighted DeepMind’s global reach. He cited collaborations with the Drugs for Neglected Diseases initiative and plant‑protein research to illustrate how AlphaFold can accelerate solutions in low‑resource settings. He also mentioned Isomorphic Labs, an Alphabet spin‑off working on a drug‑discovery platform that aims to reduce development time from years to months.
The session concluded with a brief Q&A. Students asked about the future of AI, the ethical limits of the technology, and how to prepare for a world where AGI may become mainstream. Hassabis reiterated that the field must balance rapid progress with careful consideration of societal impacts.
Overall, the Stanford GSB event showcased DeepMind’s trajectory from game‑playing AI to life‑science breakthroughs, underscored the urgency of responsible governance, and highlighted the potential for AI to transform science, medicine, and society at large. The conversation left open questions about how best to regulate AGI, how to ensure equitable access to AI benefits, and how to prepare the next generation of professionals for an AI‑augmented world.
Hassabis opened by charting his own journey. He described how a childhood obsession with chess, a stint in video‑game development, and a fascination with neuroscience all converged on a single ambition: to build an AI that learns and reasons like a human. “I think there’s several through lines actually with what seems maybe somewhat unconnected subjects,” he said, noting that his chess training taught him to decompose ambitious goals into manageable steps and that game‑development provided a sandbox for testing reinforcement‑learning algorithms.
The discussion then pivoted to AlphaGo, DeepMind’s 2016 triumph over world champion Lee Sedol. Hassabis recounted the early days of training a model on Atari games, where the system initially struggled to score a single point in Pong. He said the breakthrough arrived when the model “started winning a lot of points” and “then it started winning the games.” The success of AlphaGo paved the way for AlphaZero, MuZero, and other systems that generalized the same approach to chess, shogi, and beyond.
Next came AlphaFold, the protein‑folding AI that addressed a decades‑old scientific challenge. Hassabis explained that the project was driven by the need to predict three‑dimensional protein structures from amino‑acid sequences. “If you could unlock that and find the structures of proteins, that would unlock whole new avenues of research,” he said. AlphaFold’s 2020 release achieved unprecedented accuracy, and by 2025 DeepMind had made predictions for more than 200 million proteins publicly available. The decision to release the model openly, Hassabis noted, stemmed from a belief that its benefits would be far greater if it were freely accessible.
A central theme of the talk was the prospect of artificial general intelligence and the so‑called “singularity.” Hassabis warned that AGI could have an impact “10 times the Industrial Revolution… 10 times faster.” He characterized the current AI landscape as “the most ferocious competitive environment” and cautioned against a “race to the bottom” among leading laboratories.
Regulation surfaced as a key concern. Hassabis argued that the pace of AI development outstrips traditional regulatory mechanisms. He said, “The hard part there, of course, is that anything to do with regulation, it’s too slow.” He called for a “fleet‑footed” approach that could adapt quickly to new developments and involve leading labs in designing safety checks. The conversation also touched on public perception, noting that many Americans view AI with skepticism. Hassabis suggested that demonstrating tangible benefits—such as accelerated drug discovery—could help shift public opinion.
In the final segment, Hassabis highlighted DeepMind’s global reach. He cited collaborations with the Drugs for Neglected Diseases initiative and plant‑protein research to illustrate how AlphaFold can accelerate solutions in low‑resource settings. He also mentioned Isomorphic Labs, an Alphabet spin‑off working on a drug‑discovery platform that aims to reduce development time from years to months.
The session concluded with a brief Q&A. Students asked about the future of AI, the ethical limits of the technology, and how to prepare for a world where AGI may become mainstream. Hassabis reiterated that the field must balance rapid progress with careful consideration of societal impacts.
Overall, the Stanford GSB event showcased DeepMind’s trajectory from game‑playing AI to life‑science breakthroughs, underscored the urgency of responsible governance, and highlighted the potential for AI to transform science, medicine, and society at large. The conversation left open questions about how best to regulate AGI, how to ensure equitable access to AI benefits, and how to prepare the next generation of professionals for an AI‑augmented world.
