Sir Demis Hassabis on the AI Revolution, the Future of AI and Human Creativity | Hannah Fry

Demis Hassabis argues that AI could expand human creativity, but better tools will not decide what society should value. In his RSA Albert Medal remarks and conversation with Hannah Fry, he connects that argument to creative work, storytelling, architecture, and education.

Evidence basis: Based on the recording's English auto-generated captions, captured September 13, 2026; not checked against the original audio. These are Hassabis' views and first-party accounts, not independent evaluations of AI's benefits. His predictions and proposals remain uncertain.

Main ideas

• Creative research needs engineering that can change with the question. Hassabis compares making a game fun with achieving a scientific breakthrough: neither outcome has a guaranteed schedule. Engineers still need to build supporting technology, but research changes what they need to build next. He credits the feedback between engineering and research—not technical skill alone—for projects such as AlphaFold and AlphaGo. 26:42–30:40(opens in a new tab)

• Cheaper prototypes can change which creative ideas get pursued. He says artists working with his teams can test more ideas before committing substantial resources. An emerging filmmaker, for example, could pitch a prototype trailer instead of only a slide deck. The proposed benefit is earlier experimentation and access to projects previously too expensive to attempt. He contrasts that use with attention-grabbing but creatively uninteresting “AI slop”; he does not suggest that generating more material automatically makes it worthwhile. 35:38–38:43(opens in a new tab)

• Being computable is not the same as being meaningful to another person. Asked whether aspects of storytelling might be beyond computers, Hassabis calls this an open scientific question. He leans toward the view that computers will eventually be able to reproduce most human capabilities. But unlike chess, art has no single ranking of “better.” His example is the phrase “based on a true story”: knowing that another person actually experienced the events can add meaning for an audience. His point is about human connection, not proof that machines cannot create. 46:41–49:44(opens in a new tab)

• Technical progress leaves questions about safety, distribution, and purpose unresolved. Hassabis predicts roughly a decade of greatly amplified human creativity, followed potentially by more autonomous tools. He also insists that nobody knows how the future will unfold. He contrasts experiences that connect people with addictive, isolated experiences: market incentives alone need not produce what society wants. Even technically safe AI would leave open who shares its benefits and what people choose to value. 38:43–42:41(opens in a new tab)

• His optimistic future depends on breakthroughs, not just more AI output. He imagines advances in medicine, materials, and energy reducing disease and resource constraints. In this vision, new energy sources could ease resource constraints, including water access, and help produce fuel for space travel. This is a conditional vision of abundance, not a demonstrated outcome; he says economic arrangements and questions of human purpose would remain major challenges. 42:41–45:53(opens in a new tab)

• Architecture makes the question of value concrete. Responding to Thomas Heatherwick, he argues that cost should not be the only criterion for buildings: health and inspiration should count too. He suggests spending more to make new AI infrastructure, including data centers, beautiful and considering its effect on the physical environment. 51:49–53:56(opens in a new tab)

• His education proposal moves practice outside class and collaboration into it. Hassabis criticizes a system aimed at the middle of the class, leaving some pupils insufficiently challenged and others insufficiently supported. He proposes personalized AI tutoring outside class—for example, extra algebra questions where a learner struggles. He also suggests reusing recorded lessons and teaching materials so teachers do not each recreate the same material. Classroom time would instead emphasize projects, creativity, teamwork, and leadership. He presents this as an approach to explore, not an evaluated intervention; the exchange does not establish whether it would resolve the questioner's concern about children's cognitive development. 57:41–1:02:54(opens in a new tab)

For the full ceremony and conversation, watch from the beginning(opens in a new tab).