Demis Hassabis believes AI could help scientists treat disease and give artists new ways to create. He also thinks society has barely begun discussing what people will value if machines can do much of the work that now takes years of human training.
At the Royal Society of Arts’ 2026 Albert Medal ceremony, he argued that those decisions need artists, philosophers and social scientists as well as technologists. Professor Hannah Fry presented the medal, then interviewed him about his career, creative work and the future he hopes AI will make possible.
Hassabis estimated that full artificial general intelligence, or AGI, was a few years away. By AGI, he meant AI with broadly general abilities rather than a system built for one narrow task. His timing and expectations were forecasts, and he repeatedly said the future was not settled.
Why the RSA awarded the medal
RSA chief executive David Joseph introduced the evening in the society’s Great Room. He recalled demonstrations of the telephone, electric telegraph and electric light there, and described the RSA as a place where people could encounter new ideas before knowing how they would change everyday life.
Joseph traced the Albert Medal to Prince Albert and the first recipient, Rowland Hill, in 1864. Hill’s penny post made sending a letter cheaper and simpler. Joseph placed Hassabis among recipients whose work had opened opportunities for others, including Marie Curie, Alexander Graham Bell, Stephen Hawking and Tim Berners-Lee.
He also recalled a Musicians’ Union branch voting to ban synthesisers in 1982 because of fears that they would replace orchestral players. In Joseph’s account, synthesisers helped musicians create new sounds, but some players still lost work. He used the story to distinguish the opportunities a technology creates from the losses particular people may experience.
Fry described the ambition behind Hassabis’s work. When she interviewed him in 2017, she jokingly asked whether someone who had already been a chess prodigy, games designer and neuroscientist could have tried harder. He responded with a childhood memory of his father’s advice to try his hardest, which he had taken as a serious question about the limits of effort.
Fry said his curiosity had repeatedly taken him beyond the immediate task: from playing chess to understanding what his brain was doing, and from building games to asking how much of reality could be simulated. She regarded his concern for humanity as part of that ambition. This was her assessment of him, rather than a guarantee about the effects of the technology.
How games led him towards AI
Hassabis said he had never drawn a firm boundary between creativity and science. His family encouraged artistic and unconventional interests. His formal training developed his mathematical and scientific skills, but he described his own motivation as creative and intuitive.
In the games industry, a creative goal could require new technology. Developers wanted to make something people enjoyed, and that pushed work on graphics, simulation and AI. The graphics processors later used to train AI models had originally been developed for graphics. Hassabis saw this as an example of a project producing benefits beyond its original purpose.
He said great scientists need creativity as well as technical ability. Technical knowledge makes the work possible; intuition helps a researcher decide which ideas to pursue.
In his own games company, Elixir Studios, he found ways to support AI research through game development. Researchers could work on the characters and simulations within games, and commercial success could help fund further ideas. He mentioned working with David Silver, who later played a central role in DeepMind’s research.
Designing games also taught him to manage projects whose success could not be scheduled precisely. A publisher might ask when a game would become fun, but the team could not guarantee when it would find the right design. He saw a similar uncertainty in research. Engineers could build the necessary tools while researchers explored an unanswered question, with each group’s work informing the other. He credited that approach with helping AlphaGo and AlphaFold succeed.
Why he studied memory and imagination
Hassabis described his move into neuroscience as another way to study intelligence. After exploring AI through computer science and games, he wanted evidence from the brain.
He chose memory and imagination partly because both mattered in his earlier work. A chess player imagines a future position and plans how to reach it. A games designer imagines a player using a system and anticipates where it will become confusing or dull. At the time, he said, researchers knew much less about how to give AI that kind of imagination.
His doctorate also taught him how research differs from engineering. A researcher cannot promise to produce a particular discovery on a timetable.
Games such as Theme Park had let him build small simulations by hand. That interest continued into questions about intelligence and the nature of reality. He said he did not subscribe to a simple simulation theory, the idea that our world is itself a computer simulation. He wanted to understand the physical and computational principles underlying the world.
His hopes for science and medicine
Hassabis compared AI’s possible impact with electricity and the Industrial Revolution. He suggested it might have ten times the impact of the latter while unfolding ten times faster, over a decade rather than a century. These were his broad estimates, not measured comparisons.
His main reason for working on AI, he said, was to advance science and medicine. He pointed to AlphaFold, which predicts protein structures, and Isomorphic Labs, which applies AI to drug discovery. He considers improving human health the most important potential use of AI.
Asked to describe the best possible future, he imagined new treatments for terrible diseases, perhaps eventually all diseases, along with better materials and energy sources. He mentioned fusion and quantum computing, and hoped progress in energy could help with water, environmental problems and space travel.
He presented this as a future to work towards. Even if technical progress delivered greater material abundance, people would still have to decide how to share it and what gave their lives purpose.
Hassabis also wanted artists to describe plausible futures worth pursuing. Isaac Asimov’s Foundation books and Iain M. Banks’s Culture series had helped shape his own hopes. He asked for more science fiction grounded in science that could offer people constructive possibilities.
What AI could change for artists
Hassabis said DeepMind was developing creative tools with artists involved early enough to influence their design. He mentioned filmmaker Darren Aronofsky, musicians and the Icelandic games company behind EVE Online.
He sees benefits in testing ideas before committing a large budget. A filmmaker could try more approaches or make a short prototype trailer instead of pitching with slides alone. Established artists might undertake projects they previously could not afford, while newcomers might find it easier to demonstrate what they wanted to make.
He also criticised the growth of disposable AI-generated content made mainly to hold attention for a few seconds. Although it might serve as light entertainment, he did not want that to become the main use of the tools.
Hassabis expected roughly a decade in which strong artists and creative scientists could greatly increase what they produced with AI. He used a tenfold increase as his forecast. After that, he thought more autonomous systems might make questions about human authorship and craft harder to answer.
He suspected that people would place greater value on human connection, much as they value an original artwork differently from a copy. But he acknowledged a difficulty: people often admire mastery because of the effort and discipline behind it. If technology provides a shortcut, effort alone may no longer explain what they value.
He wanted a public discussion about what counted as craft, which achievements people would celebrate and what they would encourage children to create.
Whether a human story has something a machine cannot reproduce
Kate Varah, the National Theatre’s co-chief executive, asked whether any part of human storytelling was non-computable, meaning beyond what a computer could reproduce, and how such qualities might be protected.
Hassabis called this an open scientific question. He thought building AI and comparing it with the human mind could help researchers understand what, if anything, was uniquely human.
He referred to Roger Penrose’s ideas about quantum processes and consciousness, which might imply limits for conventional computers. Hassabis said he had not seen evidence that convinced him of this view. He expected most human abilities would eventually be shown to be computable, while acknowledging that researchers did not yet know how many of them worked.
He did not think this meant an AI artwork would necessarily be better. Games such as chess have an agreed way to measure winning; art does not have one equivalent scale.
His example was a story described as being based on real events. Knowing that another person lived through it can add meaning for the audience. He similarly described a connection with Van Gogh’s experience when looking at his paintings. He doubted that an AI producing the same image would create that same connection, but treated the question as something people needed to examine rather than a settled rule.
Buildings, health and London
Designer Thomas Heatherwick asked how technology could help improve the physical places people inhabit. He argued that buildings should be considered partly in terms of public health, and raised concerns about loneliness and disconnection in the digital era.
Hassabis agreed that surroundings could encourage creativity and affect wellbeing. He wanted the design of housing and other buildings to account for health and inspiration as well as cost. He suggested that new AI infrastructure, including data centres, could be made more beautiful, and mentioned possible collaboration with Heatherwick.
He also explained why he valued London. Finance, government, technology and the arts were concentrated in one city, making it easier for people from different fields to work together. He saw institutions such as the RSA, Royal Society and Royal Academy of Engineering as places that could bring them together.
Hassabis wanted Europe to combine its concern for social consequences with more ambition and optimism. He recalled conversations at MIT in which senior researchers argued that learning-based AI would fail because earlier approaches had failed. He did not find their arguments convincing. He said he had not known his approach would work, but thought even an unsuccessful attempt could teach researchers something new.
His proposal for education
An audience member who introduced herself as Usha, an independent member of the House of Lords, asked how education could help children use AI’s potential while addressing concerns about their cognitive development.
Hassabis thought education needed substantial changes. In his view, a system aimed at the average pupil could leave both those ahead and those behind without enough support. He wanted children to learn how to use AI, as earlier generations had needed to learn about computers and the internet, but with learning as the purpose.
He proposed moving some routine practice outside the classroom, where AI could provide personalised support. A pupil struggling with algebra, for example, could receive more questions and help with that topic. Teachers could also reuse strong teaching materials instead of each creating similar lessons from scratch.
He wanted classroom time to focus more on projects, creativity and working with other people. He compared this with aspects of Montessori education and small-group university teaching, while saying he had not attended a Montessori school himself. It was a proposal he wanted explored, not evidence that the approach had already improved learning.
Throughout the discussion, Hassabis argued that technical success would leave important choices unresolved. He warned that market incentives alone could favour addictive experiences, as he believed had happened with social media. Making AI safe, sharing its economic benefits and deciding what people valued would require people from many fields to take part.