‘Building AI responsibly, to benefit humanity’: 2023 UCL Prize Lecture in Life and Medical Sciences

Last November, Sarah Jilani attended the UCL Prize Lecture to hear world-renowned researcher and founder Demis Hassabis CBE speak about his remarkable career and the rapid development of AI.

“We really can change the world in an incredible way, we just have to try.”

This is the advice that Demis Hassabis CBE, CEO and co-founder of Google DeepMind, would have given to his younger self.

UCL’s Annual Prize Lecture in Life and Medical Sciences took place in November and saw a theatre filled with 900 in-person attendees joined by more than 450 online viewers. This year’s lecture was given by Hassabis, whose leadership has driven the development of Google DeepMind into one of the world’s leading AI research groups.

The Prize Lecture, established in 1997, aims to celebrate significant scientific advancements while also creating a space to debate and discuss their implications. With AI at the forefront of conversations across many sectors – especially academia and security – Hassabis’ lecture addressed some of these topical discussions, acknowledging that we must not overlook concerns but ensure we take time to carefully consider them.

Hassabis’ lecture began by charting the trajectory of his early work, starting with his days as a chess prodigy (he continues to play the game daily), an undergraduate degree in computer science, and early career in programming games, during which period he founded the independent games developer Elixir Studios.

He would then return to education and complete a PhD in Cognitive Neuroscience at UCL, which equipped him with a deep understanding of the human brain and its cognitive processes that would prove highly beneficial to the development of AI learning systems.

After founding Google DeepMind in 2010, the company’s breakthrough success occurred in 2015, when their AI program AlphaGo succeeded in winning a five-game Go match against leading professional Lee Sedol. With 10¹⁷⁰ possible combinations (more atoms than there are in the universe), Go is a highly complex game. Experts in the field predicted it would be many years before AI could succeed in competing against top human players, yet through machine learning, DeepMind were able to achieve the feat.

The implications of the victory were interesting. While it was undoubtedly a huge breakthrough in the field of AI, one might pity a skilled and practised player defeated by a computer. Yet Hassabis did not see it in this way: he discussed how Go players now study such moves in a sort of ‘role reversal.’ Through this, he claimed, players are encouraged to be more creative, harness their intuition and to change the ways they think about the game.

These games, however, were ultimately a demonstration of ‘the means to an end’, and ensuing developments have proven just how crucial those are. Immediately after AlphaGo’s victory, the team began working on AlphaFold, which had the ability to predict the 3D shape of a protein, almost instantly, with atomic level accuracy.

The knowledge of the 3D structure of a protein is essential to understanding its function, yet prior to AlphaFold we were aware of only about 17% of the roughly 20,000 proteins in the human body. AlphaFold increased this to 98.5%, and the accuracy of its predictions rendered it the ‘solution’ to what was known as the ‘protein folding problem.’

Since then, there has been widespread use of related data for medical research, from targeting early onset Parkinson’s to tackling antibiotic resistance. Hassabis revealed that he was particularly proud of the advances made in the Drugs for Neglected Diseases Institute, where AlphaFold’s predictions accelerated research into curing diseases that are major public health concerns in less economically developed regions.

He also revealed that he ‘couldn’t have dreamed’ of the positive implications and new breakthroughs that AlphaFold would lead to. The relevant research has been cited more than 150,000 times, placing it in the 100 most-cited papers of the last decade, and in the 900 most-cited papers of all time.

This extraordinary impact was made possible by his company’s decision to make AlphaFold’s calculations freely available to anyone in the scientific community. Prior to release, 30 experts in biosecurity were consulted to ensure that the benefits outweighed the risks and unanimously agreed that they did.

Today, Google DeepMind maintains a strong commitment to ethical use of AI. The company’s mission statement is ‘building AI responsibly to benefit humanity’, and we can see the multitudinous benefits their research is bringing. A commitment to harnessing AI for good is also reflected in the company’s ethical implementations, such as their Responsibility and Security Council, which works to monitor projects and mitigate any risks that may arise.

To move forward with AI, Hassabis says, we must not to be fearful and halt it completely, as this would deprive us of so many positive possibilities. But at the same time, such research cannot be in any way rushed, and we cannot ‘move fast and break things’ in a style historically favoured by Silicone Valley.

Instead, Hassabis advised his audience to strive towards a middle ground, in which we go slowly, and ‘move with foresight, rather than hindsight.’

In all, the audience were left in no doubt that AI should be treated with exceptional care and reverence. If that is the case, it will be able to fulfil its incredible potential without causing harm, and bring about the ultimate aim of benefiting humanity.

About the author

Sarah Jilani is a second-year undergraduate at UCL, studying English Language and Literature. As someone who enjoys learning widely, she aspires to write articles which improve understanding of topics and issues as well as broaden her own knowledge. To contact Sarah, please email sarah.jilani.22@ucl.ac.uk