Driving scientific discovery with AI

Nobel laureate and Cambridge alumnus Sir Demis Hassabis believes we are entering a new era of ‘digital biology’, where AI can help us reimagine the principles of drug discovery at ‘digital speed’.

Speaking at a special event in Cambridge, exploring how AI can accelerate scientific discovery, the Chief Executive Officer and Co-founder of Google DeepMind also said that despite the rise of quantum computing, classical computer systems still have the potential to advance knowledge using AI and could one day even help us uncover the true nature of reality.

Demis, who was last year jointly awarded the Nobel Prize in Chemistry with Google DeepMind colleague Dr John Jumper for their AI research contributions for protein structure prediction, told Cambridge students and alumni that AI was potentially the “perfect description language for biology”.

“Right now it takes an average of 10 years for a drug to be developed, and it’s extraordinarily expensive, billions and billions of dollars,” he said. “And so I’m thinking, why can’t we use these techniques to reduce that down from years to months? Maybe even, one day, weeks? Just like we reduced down the discovery of protein structures from potentially years down to minutes and seconds.”

During his talk at the Babbage Lecture Theatre – where he told guests he attended his first lecture as a student almost 30 years ago – Demis recounted his AI career and research up to now, and also provided fascinating glimpses of how the technology might evolve, including the development of Artificial General Intelligence, a theoretical AI system that can do the same kinds of cognitive tasks that a human can do.

He said: “Cambridge is an amazing place. It has inspired my whole career actually, and hopefully it is going to do the same for many of you students in the room.”

After graduating from Cambridge, where he studied Computer Science at Queens’ College as an undergraduate in the 1990s, he co-founded DeepMind in 2010, a company that developed masterful AI models for popular games. The company was sold to Google in 2014 and two years later, DeepMind came to global attention when the company achieved what many then believed to be the holy grail of AI: beating the champion player of one of the world’s oldest boardgames, Go.

“My journey on AI started with games, and specifically chess,” he said. “I was playing chess from the age of four and it got me thinking about thinking itself – how does our mind come up with these plans, with these ideas, how do we problem solve, and how can we improve? What was fascinating to me, perhaps more fascinating than even the games, was the actual mental processes behind it.”

This interest continued when he moved to playing computer chess. “I remember being fascinated by the fact that someone had programmed this lump of inanimate plastic to actually play chess really well against you. And I ended up experimenting myself in my early teenage years with an Amiga 500 computer, and building those kinds of AI programs to play games like Othello. And really, that was my first taste of AI, and I decided from very early on that I would spend my entire career trying to push the frontiers of this technology.”

Computer games were the “perfect proving ground” for AI systems, he said. And after creating learning systems – inspired by neuroscience – that mastered Atari’s catalogue of games, and developing the AlphaGo computer programme that defeated Go world champion Lee Sedol, he turned his attention to science.

“I felt that we were ready, we had the techniques that were mature enough and ready to now be applied outside of games and to try and tackle really meaningful problems.”

Protein folding – predicting the 3D structure of a protein from its amino acid sequence – was a prime example. Proteins are the building blocks of life, and the function of a protein is thought to be related to its structure. Knowing the structure of a protein could therefore help in drug discovery and disease understanding.

Scientists had been working on the challenge for at least 50 years when, in November 2020, DeepMind’s AlphaFold2 tool was declared to have solved it by the Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction, or CASP. DeepMind went on to use AlphaFold2 to fold all 200 million proteins known to science, and made the system and these structures openly and freely available for anyone to use.

“It’s kind of like a billion years of PhD time done in one year,” said Demis. “And it’s amazing to think about how much science could be accelerated. Two million researchers are using it from pretty much every country in the world. It’s been cited over 30,000 times now and it’s become a standard tool in biology research.”

Demis was jointly awarded the Nobel Prize in Chemistry last year in recognition of the major advances made possible by AlphaFold2.

And because these biological structures exist across much of life on Earth, he said new avenues of exploration had been opened up – in a wide range of fields –  including climate, agriculture, disease, and drug discovery.

“The mission of DeepMind mind from the beginning was about building AI responsibly to benefit humanity, but the way we used to articulate it when we started out was a two-step process, step 1 – solve Artificial Intelligence, step 2 – use it to solve everything else.

“If I look at all the work we've done in the last 15 years, first of all our games work, and then now with the scientific work that we’ve been working on, it’s all about making this search ‘tractable’. You have this incredibly complex problem, and there’s many possible solutions to the problem, and you've got to find the optimal solution –  kind of like a needle and a haystack. And you can't do it by brute force, so you have to learn this neural network model, so that you can efficiently guide the search and find the optimal solution.

“I think AI will be applicable to pretty much every field, and I think there are many, many advances to be made over the next 5-10 years by doing that.”

Discussing the path to Artificial General Intelligence (AGI) Demis said that Google DeepMind was making advances in all areas of AI’s understanding of the physics of the real world, and pointed to its new Veo 2 state-of-the-art video generation tool, which generates videos from a text description, and Genie 2, which can generate a computer game based on a single prompt.

And stressing the importance of AI safety, and the responsibility that came with building these kinds of transformative systems and technologies, he explained how Google DeepMind’s SynthID tool invisibly watermarked AI-generated content, which can then be detected as synthetically generated image, audio, text or video.

“AI has this incredible potential to help with our greatest challenges, from climate to health. But it is going to affect everyone, so I think it’s really important that we engage with a wide range of stakeholders from society. And I think that’s going to become increasingly important given the exponential improvement that we’re seeing with these technologies.”

Looking to the future, Demis said he was “very excited” about the next generation of virtual assistant technology, or ‘universal assistants’ as he described Google DeepMind’s work on a research prototype assistant that can understand the world around us.

“We call it ‘Project Astra’, where you have it on your phone or some other devices, maybe glasses. It’s an assistant you can take around with you in the real world and it helps you in everyday life.”

He said the next step in AI was building planning systems like we saw with AlphaGo, which can search and find good solutions to problems, ‘on top of’ world models like Google Gemini, which understand how the real world works. Combined, he said, they can plan and achieve things in the real world.

“That’s key to things like robotics working, which I think in the next two or three years is going to be a huge area that’s going to have massive advances.”

Bringing his talk to an end, Demis described himself as “Turing’s champion” – and posed the question “how far can these Turing machines and the idea of classical computing go?”

“There are a lot of things that are thought to require quantum computing to solve. My conjecture is that actually classical Turing machines that these types of AI systems are built on can do a lot more than we previously gave them credit for.

“If you think about AlphaFold and protein folding – proteins are quantum systems, they operate at the atomic scale and one might think you need quantum simulations to actually be able to find the structures of proteins. And yet we were able to approximate those solutions with our neural networks.

“And so one potential idea is that any pattern that can be generated or found in nature can be efficiently discovered and modelled by one of these classical learning algorithms. And if that turns out to be true, it has all sorts of implications for quantum mechanics and actually fundamental physics, which is something that I hope to explore. Maybe these classical systems will help us uncover what the true nature of reality might be.

“And that leads me back to the whole reason I started my path on AI many, many years ago. I always believed that AGI built in this way could be the ultimate general purpose tool to understand the universe around us and our place in it.”

Words: Stephen Bevan
Published: 24th March, 2025

The text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License