Demis Hassabis: The Future of AI Is Far Bigger Than We Think

The strangest detail in this recording is what a Nobel laureate is doing in a London arts venue: not lecturing, but folding mathematics, cinema and neuroscience into a single conversation. Demis Hassabis is accepting the RSA's Albert Medal here, and his very first question is the right one. What has the relationship between creativity and AI actually become over the past year?

Tools designed with artists, not for them

Hassabis describes an approach that deliberately differed from his competitors: the tools were designed from the start with a select group of leading creators. The names he offers are telling, including the director Darren Aronofsky and the Icelandic company behind EVE Online. The idea is not to ship a product and then measure artist reaction, but to ask creators what a tool should actually do before building it. The same instinct runs through DeepMind's science work, where they sat down with leading domain authorities to argue about the protein-folding problem rather than simply declaring an answer.

The concrete payoff of that collaboration is that prototyping with generative tools has become far cheaper. Making a short teaser instead of a pitch deck is an early validation step for a film. What Hassabis emphasises is that the biggest names do not fear these tools; they like them precisely because they can test ideas before committing serious money. For emerging creators, this is a route to professional-quality proof of concept without a financier behind them.

What he rejects just as clearly is bland AI output built to hold attention for a few seconds. In his phrasing, that is weak creative work, maybe acceptable as light entertainment but not worth a serious investment. The test DeepMind applies is blunt: does the tool multiply the output of the best scientists and creators tenfold, or does it merely put on a show.

A two-stage forecast

The analytical core of the talk is a two-stage prediction about creativity. Stage one is the next ten years: extraordinary tools that let the best scientists and artists multiply their output tenfold. Stage two is those tools becoming independent systems, at which point the real question changes. In that second era, Hassabis argues, human connection, human creativity and craft will matter more than ever.

The worry behind that forecast is concrete: if craft, meaning years of practice at something, stops being the measure of creativity, what happens? We can copy a masterpiece today, yet we do not value the copy the same way. Does a shortcut still produce a concert pianist or a great mathematician? His answer is not a definition but a summons: what we value is a collective decision we have to make deliberately.

Nobody knows the future

This is the hinge of the whole argument. Hassabis pushes back on any technologist who claims to know what is coming, calling it either a lie or a hidden motive. Immediately after, he balances it: the future has not been written, which is the good news. Given that this technology can make almost anything easier, society has to know what it actually wants from it. Left to itself, technology follows the path of capital, producing the loneliness and dependence we already saw with social media.

He applies the same treatment to the physical world. Data centres are the most visible infrastructure of the AI revolution, yet how they sit in their surroundings is a matter of health and inspiration. He argues we have lost that sensitivity in large cities since the second world war, and that the elegance of Art Deco across London eventually stalled for no good reason. An age of abundance, he suggests, is a chance to stop treating cost as the only measure of what we build and start counting its effect on people.

Education: turning the classroom upside down

Education closes the event. The question is how AI gets into the school system without damaging how children learn, and how to make that integration a positive one. Hassabis's answer compresses into two claims: education needs a radical change, and children must be trained to use these tools. His analogy is plain, and it is to how computers and the internet were taught in the nineties. Children who never learn that technology simply fall behind the world.

His concrete proposal is to invert the classroom model. Instead of memorisation inside the period, learning happens outside it with AI assistance, so a student who finds algebra hard simply gets more questions on it. He describes it as a personal tutor for everyone, built from the recorded lectures of the best teachers available. What is then left for the classroom is not preparing the same material again, but project-based work: creativity, enterprise, teamwork and leadership, the skills he thinks actually matter for the future.

Behind the proposal sits an uncomfortable observation. The current system is fifty years old, a product of the post-war period, and it was built for one type of pupil. Gifted students are not challenged enough and fall out of interest, while late bloomers do not get enough help. Meanwhile teachers keep rebuilding identical materials, which is neither meaningful nor efficient. Turning the model upside down is, on his reading, the only way to fix all three problems at once.