Britain’s new £225m national artificial intelligence supercomputer will be used to spot sick dairy cows in Somerset, improve the detection of skin cancer on brown skin and help create wearable AI assistants that could help riot police anticipate danger.
Scientists hope Isambard-AI – named after the 19th-century engineer of groundbreaking bridges and railways, Isambard Kingdom Brunel – will unleash a wave of AI-powered technological, medical and social breakthroughs by allowing academics and public bodies access to the kind of vast computing power previously the preserve of private tech companies.
The supercomputer was formally switched on in Bristol on Thursday by the secretary of state for science and technology, Peter Kyle, who said it gave the UK “the raw computational horsepower that will save lives, create jobs, and help us reach net zero-ambitions faster”.
The machine is fitted with 5,400 Nvidia “superchips” and sits inside a black metal cage topped with razor wire north of the city. It will consume almost £1m a month of mostly nuclear-powered electricity and will run 100,000 times faster than an average laptop.
Amid fierce international competition for computing power, it is the largest publicly acknowledged facility in the UK but will be the 11th fastest in the world behind those in the US, Japan, Germany, Italy, Finland and Switzerland. Elon Musk’s new xAI supercomputer in Tennessee already has 20 times its processing power, while Meta’s chief executive, Mark Zuckerberg, is planning a datacentre that “covers a significant part of the footprint of Manhattan”.
The investment is part of the government’s £2bn push to attain “AI sovereignty” so Britain does not have to rely on foreign processing chips to make AI-enabled research progress. But the switch-on could trigger new ethical dilemmas about how far AI should be allowed to steer policy on anything from the control of public protests to the breeding of animals.
One AI model under development by academics at the University of Bristol is an algorithm that learns from thousands of hours of footage on human motion, captured using wearable cameras. The idea is to try to predict how humans could move next. It could be applied to a wide range of scenarios, including enabling police to predict how crowds of protesters may behave, or predict accidents in an industrial setting such as a construction site.
Dima Damen, a professor of computer vision at the university, said based on patterns in the human behaviours a wearable camera was capturing in real time, the algorithm, trained by Isambard-AI, could even “give an early warning that in the next two minutes, something is likely to happen here”.
Damen added there were “huge ethical implications of AI” and it would be important to always know why a system made a decision.
“One of the fears of AI is that some people will own the technology and the knowhow and others won’t,” she said. “It’s our biggest duty as researchers to make sure that the data and the knowledge is available for everyone.”
Another AI model under development could detect early infections in cows. A herd in Somerset is being filmed around the clock to train a model to predict if an animal is in the early stages of mastitis, which affects milk production and is an animal welfare problem. The scientists at Bristol believe this could be possible based on detecting subtle shifts in cows’ social behaviour.
“The farmer obviously takes a great interest in their herd, but they don’t necessarily have the time to look at all of the cows in their herd continuously day in, day out, so the AI will be there to provide that view,” said Andrew Dowsey, a professor of health data science at the University of Bristol.
A third group of researchers are using the supercomputer to detect bias in the detection of skin cancer. James Pope, a senior lecturer in data science at the University of Bristol, has already run “quadrillions if not quintillions of computations” on Isambard to find that current phone apps to check moles and lesions for signs of cancer are performing better on lighter coloured skin. If confirmed with further testing, apps could be retuned to avoid bias.
“It would be quite difficult, and frankly impossible to do it with a traditional computer,” he said.
