With Microsoft and OpenAI building a $100?billion (?77.5?billion) supercomputer and technology giants set to invest more than $1?trillion in artificial intelligence, it is difficult to see how publicly funded bodies can keep up in this highly expensive field of research.
That is, however, the challenge that Charlotte Deane, who took over as executive chair of the UK’s Engineering and Physical Sciences Research Council (ESPRC) in January, faces – with the added pressure of a new Labour government with potential uses in science, business and beyond that may be enormous but are still unclear.
That Professor Deane is firmly rooted in the world of AI research may certainly help with this task. Prior to her arrival at the ESPRC – whose annual budget, including PhD fellowships, exceeds ?1 billion – the University of Oxford professor of structural bioinformatics was chief scientist of biological AI at Exscientia, a leading British AI-led drug discovery firm.
Unusually, she has been engaged in this world since her student days at Oxford in the late 1990s.
“My friends would probably start my ‘Résumé for Researchers’ as ‘too clumsy to be a wet lab chemist’ – which is a factually correct statement – so I chose to do a research project in the computational area, as I’d always liked working with computers,” reflected Professor Deane.
On how AI has changed science in recent years, she reflected how computers have long been “a toy and tool to help us get better understanding or change the experiments we do” but “we now have some new toys like machine learning…which allow us to do things that five or six years ago we thought were impossible. It’s really exciting because you can see how it will change how we do science – it allows us to think about things in a different way and do experiments differently.”
Even as a relatively old hand in AI research, Professor Deane admitted she has been surprised by the rapid progress made in recent years, such as DeepMind’s , which used AI to solve the 50-year-old grand challenge of predicting the structure of proteins. “I’d worked on this problem and if you’d asked me if this problem would have been solved in my lifetime, I’d have said definitively yes. But was I surprised by the date it was solved? I was,” said Professor Deane.
While DeepMind can, however, spend tens of millions of?dollars on experiments, that is not an option for the ESPRC. So how can university-based researchers keep up?
“It’s never about competition. Yes, we don’t have computers as big as they have but that’s OK, because there are different things that we’ll do in the public sector versus the private sector,” she said.
“In academia, we can play at the edge of what’s possible – I’m not going to try to build the perfect large language model…but we can think about things like uncertainty, explicability, decision-making…which are less important to DeepMind because they just want their model to get it right. I want to know why it got it right because that will change my way of doing science.”
Academics can also investigate areas that might prove highly valuable within a decade or so, continued Professor Deane. “These areas might not seem cool now but might be some day…quantum [technology] is a really good example. It’s worth about $9 billion and will be worth $90 billion in 10 years’ time. The reason this exists, and is a strength for the UK, is because – a long time ago – investment was put into something that still doesn’t sound that cool – which is photonics. AI too – it wasn’t that long ago that the idea you’d have companies in this area wasn’t such a good idea.”
There are, of course, challenges – not least attracting top AI talent when big tech is offering high six-figure salaries to AI engineers. “There is no point in pretending that, as the academic sector, we can compete on salaries for some of these things, but I have brilliant PhD students in my research group, and I see people make active choices on where they want to be,” she said.
Yet even with UKRI’s recent , Professor Deane feels it is important that AI researchers can move more easily between academia and industry.
“It’s one of the things that I want to drive in UKRI – I’ve been able to work across academia and industry at the same time. Many companies want to have people with that freedom to do other things.”
后记
Print headline: ‘Never about competition’: all AI research valid
