The RSA awards Demis Hassabis the Albert Medal, and he argues society must intentionally shape AI.
Proteins are tiny machines that run the body, and what each one does depends on its shape, which you need to understand a disease or build a drug. Getting one protein's shape could take a scientist years, even a career. AlphaFold has now done 262 million of them and given every one away free to researchers.
He calls AI one of the most profound technologies humanity will ever invent, a revolution like electricity, the telephone, the internet and mobile, and perhaps bigger. He suggests maybe ten times the impact of the Industrial Revolution, happening ten times faster, unfolding over a decade instead of a century, with new institutions needed for its downsides.
Hassabis says something radical must happen and children should be trained to be native with AI. He recommends inverting the classroom: rote learning happens outside class with personalised AI tutors that give more practice where you struggle, while classroom time becomes Montessori-like, project-based and human-to-human, building creativity, entrepreneurialism and teamwork, like the Oxbridge supervision system.
He used games, especially his company Elixir Studios, as a Trojan horse to fund AI research, working with people like Dave Silver on AI to control characters and simulations. He also learned to manage startups and creative engineering teams, and took to DeepMind the skill of combining cutting-edge engineering with an open-ended creative objective.
He was building toward DeepMind and AGI and wanted to look at intelligence from every angle. Having pushed AI as far as known in the mid 90s, he sought an orthogonal source in the brain, using newly mainstream fMRI. He studied memory and imagination, because planning and imagination were central to chess and game design and no one knew how to do them in AI.
He says creative work has no linear notion of better, so it comes down to what is meaningful to other humans. Stories labelled 'based on a true story' matter because a real person lived them, and in a Van Gogh you see his pain in every brushstroke. An AI-created piece will probably never offer that human connection.
Unlike some competitors, they build tools in collaboration with top artists early on, so creators can say what helps versus gets in the way. Partners include Darren Aronofsky, top musicians, and the Icelandic company that builds EVE Online. He sees AI slop as one risk, but also great prototyping that lets artists test many more ideas cheaply.
Carefully stewarded AI could drive advances in science and medicine, curing terrible diseases, maybe all eventually, plus better materials, new energy sources, fusion and quantum computers. That would ease problems like environment and water access, create a world of abundance, and unlock proper space travel. The biggest remaining challenges would be economics, purpose and value.
Prince Albert ran the society for 18 years, and when he died it created a medal in his name. The first went in 1864 to Rowland Hill for the penny post. Later recipients include Marie Curie, Alexander Graham Bell, Winston Churchill, Stephen Hawking, Tim Berners-Lee and Sarah Gilbert for the Oxford vaccine.
During his postdoc at MIT, top AI professors told him in the cafeteria that learning approaches would never work, having failed at MIT in the 80s, and that he should use logic systems. He found their arguments boiled down to dogma, not first principles, which encouraged him: at worst he would fail in a new, interesting way.
“And he gave every single one away free to every researcher on earth.”
“He says, there is no 50% mode in Demis, right?”
“I don't think society is ready yet for what's coming.”
“I think the future is really not written.”
“Sir Demis Hassabis on the AI Revolution, the Future of AI and Human Creativity | Hannah Fry.” https://www.youtube.com/watch?v=rDteJAuKiBI. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/sir-demis-hassabis-on-the-ai-revolution-the-future-of-ai-and.html.