Sir Demis Hassabis discovered this week that he had won the Nobel Prize in Chemistry when his wife – also a scientific researcher – received several calls via Skype to urgently request his telephone number.
“My mind was completely exhausted, which almost never happens. It was. . . almost like an out-of-body experience,” says Hassabis, co-founder and CEO of Google DeepMind, the artificial intelligence arm of the Silicon Valley search giant.
The Nobel Prize in Chemistry, which Hassabis shared with his colleague John Jumper and the American biochemist David Baker, was won for unlocking an impossible problem in biology that had remained unsolved for fifty years: predicting the structure of every protein known to humanity knows, using AI software known as AlphaFold.
Now that he’s solved this long-standing challenge, with widespread implications for science and medicine, Hassabis has set his sights on climate change and healthcare. “I want us to help solve some diseases,” he told the Financial Times.
His team is working with drugmakers Eli Lilly and Novartis on six drug development programs, targeting disease areas such as cancer and Alzheimer’s disease. Hassabis said he expects to have a drug candidate in clinical trials within two years.
His other major areas of interest include using AI to more accurately model the climate and pushing the ultimate frontier in AI research: inventing machine intelligence on par with human intelligence.
“I hope when we look back in ten years’ time [AI] will have ushered in a new golden age of scientific discovery in all these different domains,” said Hassabis, a former neuroscientist and video game designer. “That’s what brought me to AI in the first place. I see it as the ultimate instrument to accelerate scientific research.”
The DeepMind duo was recognized on Wednesday, a day after former Google colleague and veteran AI scientist Geoffrey Hinton won the physics prize alongside physicist John Hopfield for their work on neural networks, the foundational technology for modern AI systems that underpin healthcare, social media, self-reliance. cars – and AlphaFold itself.
The recognition of AI breakthroughs marks a new era in research, highlighting the importance of computational tools and data science in solving complex scientific problems on much shorter time scales, in everything from physics to mathematics, chemistry and biology.
“It is of course interesting that the [Nobel] The committee has decided to make such a statement by bringing the two together,” Hassabis said.
The awards also include the promises and potential pitfalls of AI.
Hopfield and Hinton were pioneers in this field in the early 1980s. Hinton, who is 76 and left Google last year, said he had no plans to investigate further. Instead, he plans to advocate for work on the safety of AI systems, and for governments to facilitate it.
The DeepMind pair, on the other hand, won for work revealed mainly in the past five years, and remain extremely optimistic about its social impact.
“The impact of [AI] in particular in the field of science, but also in the modern world more broadly is now very, very clear,” says Maneesh Sahani, director of the Gatsby Unit at University College London, a research institute focused on machine learning and theoretical neuroscience. Hinton was the founder and director of Gatsby in 1998, while Hassabis worked there as a postdoctoral researcher in 2009 and eventually spawned DeepMind from the UCL institute in 2010.
“Machine learning is popping up everywhere, from people analyzing ancient texts in forgotten languages, to x-rays and other medical imaging. There is a toolkit that we have now that will advance science and academic disciplines in many different directions,” says Sahani, who is also a professor of neuroscience.
The recent iterations of AlphaFold have “ramifications throughout medicine, biology and many other fields” because they are so fundamental to living organisms, says Charlotte Deane, professor of structural bioinformatics at the University of Oxford.
“Many were skeptical when they started, but soon their program outperformed all other programs in predicting protein structures,” says Venki Ramakrishnan, a biologist who won the 2009 Nobel Prize in Chemistry for his work on protein synthesis. “It really changed the field dramatically.”
AlphaFold has been used by more than two million scientists to, among other things, analyze the malaria parasite, develop a vaccine, improve plant resistance to climate change and study the structure of the nuclear pore – one of the largest protein complexes in humans. body.
Rosalyn Moran, professor of neuroscience at King’s College London, and CEO of AI start-up Stanhope AI, said: “Building tools is scientific work by hand. . . they are often the unsung heroes of science. For me that was the most exciting part of the prize.”
AlphaFold still has flaws, as reported by its makers earlier this year, including “hallucinations” of “spurious structural order” in cell areas that are in fact disordered. Another challenge in using AI for scientific research is that some important research areas may be less rich than protein analysis in experimental data.
In physics, Nobel, Hinton and Hopfield used fundamental concepts from physics and neuroscience to develop AI tools that can process patterns in large information networks.
The Boltzmann machine, which Hinton invented, could learn from specific examples rather than from instructions. The machine could then recognize new examples of categories it had been trained on, such as images of cats.
This type of learning software, known as neural networks, now forms the basis of most AI applications, such as facial recognition software and large language models, the technology that underpins ChatGPT and Google’s Gemini. One of Hinton’s former students, Ilya Sutskever, was co-founder and chief scientist of ChatGPT maker OpenAI.
“I would say I’m someone who doesn’t really know what field he’s in, but would like to understand how the brain works,” Hinton, a computer scientist and cognitive psychologist, said at a press conference this week. “And in my efforts to understand how the brain works, I have helped create a technology that works surprisingly well.”
The AI awards have also highlighted the interconnected nature of scientific discoveries, and the need for sharing data and expertise – an increasingly rare phenomenon in AI research occurring within commercial organizations such as OpenAI and Google.
Neuroscientific and physics principles were used to develop today’s AI models, while data generated by biologists helped invent the AlphaFold software.
“Scientists like me have traditionally solved protein shapes using laborious experimental methods that can take years,” says Rivka Isaacson, professor of molecular biophysics at King’s College London, who was an early beta tester of AlphaFold. “However, it was these solved structures, which the experimental world deposited for public use, that were used to train AlphaFold.”
She added that the AI technique had allowed scientists like her to “dive deeper into protein function and dynamics, ask different questions and potentially open up whole new areas of research.”
Ultimately, AI – like electron microscopy or x-ray crystallography – remains an analytical tool, not an independent agent conducting original research. Hassabis emphasizes that technology cannot replace the work of scientists.
“Human ingenuity is involved – asking the question, the hunch, the hypothesis, our systems can’t do any of that,” he said. “[AI] now just analyzes data.”
