The high-risk agency dreamt up by Dominic Cummings – and its seven big ideas to save the world

The internet, satellite navigation and Covid vaccines have one thing in common. They all owe their existence to a secretive US research agency set up in one of the tensest moments of the Cold War. 

In 1957, the ghostly beeps of the Soviet Union’s Sputnik satellite orbiting the world shocked America into an explosion of scientific research. The following year, the US government set up the Advanced Research Projects Agency, a lavishly-funded skunkworks initiative designed to ensure Uncle Sam had a technological lead on its superpower rival.

“Arpa” led to an explosion in invention. Self-driving cars, the computer mouse and drones also owe their existence in some way to the lab and its defence-focussed successor, Darpa.

Now the Cold War is over, researchers worry that the rate of scientific progress has stalled. We are simply not inventing any more.

But in a nondescript office tucked in a corner of the British Library, there is a bold project to change that. This is the headquarters of the Advanced Research and Invention Agency, otherwise known as Aria. The agency, the UK’s answer to Arpa, is an £800 million effort to beat our scientific stagnation and get Britain inventing again.

“Over the last five, six years there’s a growing momentum around the idea that good ideas seem to be getting harder to find,” says Matt Clifford, Aria’s chairman. “There’s this sense that, to maintain a constant output of great science, we’re having to spend more and more.”

Nobody could accuse Aria of a lack of ambition. The projects listed on its website include engineering the planet to reverse the extreme effects of climate change, replacing physical labour with robots and merging human brains with computers to turn us into cyborgs. 

If any of these projects succeed, the man to thank is Dominic Cummings. The architect of Vote Leave and Boris Johnson’s former chief adviser may have left government in 2020 after a spectacular falling out with the then-prime minister, but by then, his push for a “high risk, high reward” science agency was already well under way. 

Cummings’s mantra for government was “Get Brexit done, then Arpa”. The phrase was even used as his WhatsApp profile. He had blogged about the US agency as early as 2014, before Vote Leave had been founded; and wrote a 47-page paper on how to fund “crazy” ideas before he entered Number 10. Among the ideas he proposed was that Brexit Britain could have a manned base on the moon.

Not many people noticed at the time, but Aria’s Clifford did. The Bradford-born technology investor, who ran the venture capital firm Entrepreneur First, regularly wrote about Cummings’s ideas in his newsletter, recommending the blog in his online reading club, Organising Genius.

In 2022, well after Cummings had left Downing Street, Clifford was appointed Aria’s chair (he was later thrust into the political spotlight as Rishi Sunak’s representative for last year’s Artificial Intelligence Safety Summit, which saw nations and tech companies make commitments on preventing AI disasters). Ilan Gur, an American scientist, was made Aria’s chief executive.

Gur has said he and Cummings have never met. Clifford says the two have only crossed paths a couple of times, over issues such as support for tech companies during the pandemic. “There are a lot of people that deserve a lot of credit,” he says. “I think it’s sort of a bit unhelpful to characterise it as one person’s thing”

Within parts of the government, however, Aria was seen as Cummings’s pet project. “It was developed and pushed through with very little civil servant involvement,” says one source. “It would’ve been completely frustrated by the machine if they were involved as it’s a radically different model of organisation than Whitehall likes.”

Crucially, Aria had the support of Sir Patrick Vallance, the chief scientific adviser who gained more influence during the pandemic. Aria was pushed through by Cummings’s band of self-labelled “weirdos and misfits”. 

What Aria is supposed to do differently to the thousands of academics and inventors across Britain, however, is be prepared to fail. 

“There’s definitely a question about, ‘Do we incentivise enough risk-taking in academic careers?,’” says Clifford. “It’s very hard to make progress in a scientific career if you can’t show a lot of publications that have positive results. At the margins that’s going to push people towards things that they know will work, whereas presumably most breakthroughs will come from things that we don’t know will work.” 

He draws parallels with the highwire world of venture capital investing, which operates on a so-called power law – even if nine in 10 of the start-ups you back crash and burn, the one survivor will change the world.

“If you want things that 10 years from now, can be placed alongside the Darpa-type achievements – the internet, GPS, mRNA vaccines – then you’ve got to find ways culturally to take a lot of risk, but also institutionally take out a lot of the veto points that would get rid of things. You’re going to fail more than you’re going to succeed, and most work is not like that. If we ran schools like that it would be a disaster.

“But it’s an awkward thing, because you tend to see the failures before you see the successes, and it’s hard for any public sector organisation to do that,” says Clifford.

Aria’s supporters believe the key is to borrow the Darpa model – each project has an individual programme director tasked with bringing their crazy ideas to life, and with the control that allows them to avoid getting bogged down in bureaucracy and endless meetings.

James Phillips, who was Boris Johnson’s technology adviser during Aria’s founding, says this work hierarchy is crucial. “The American programme manager model didn’t exist in the UK science system until Aria did. That’s what’s so important for me about that model, it empowers an individual deep technical expert to coordinate a risky bet about the future, like the internet or satellites were, that are too early for commercial investment.”

Aria has started to outline seven of the ideas it will be working on. One concerns intervening in the climate to the extent that we can control weather, almost in the way that we adjust our thermostats. Another suggests that we could make brain implants to manage brain disorders and make them as commonplace as pacemakers. A third proposes developing artificial intelligence computers to be one thousand times more efficient than today’s systems – an attempt to counter concerns that AI will require a third of the world’s electricity within a few years.

Disappointingly, given these ambitions, Aria’s headquarters look more like David Brent’s Wernham Hogg offices than Q’s laboratory. But that is by design. The hard work here is choosing what researchers and experiments to spend money on. 

The total budget – £200 million a year – is small, and will consist of small grants of as little as £10,000.

Phillips says the hope is that there is a wider impact, encouraging others to take more risk. “We spend substantial sums on R&D in the public sector, but in my view, it is generally too risk averse to pursue unusual ideas. Aria can help shift the culture for the rest of it. It shows you a different way of doing things.”

Aria does at least have heavyweight leadership. Gur, its chief executive, had previously worked at Arpa-E, an Arpa spin-off dedicated to new energy. Clifford is widely seen as one of Britain’s best-networked techies, having secured investments from the likes of LinkedIn founder Reid Hoffman and the Collison brothers, the billionaire Irish founders of payments company Stripe. Both are crucially young: Clifford is 38, Gur is 43, and come from outside the fusty world of academia.

Aria’s board includes Sir Patrick Vallance, as well as Dame Kate Bingham, who led the vaccines taskforce, and Sarah Hunter, a former executive at Google’s “moonshots” laboratory X.

How should we judge success? Clifford says Aria should be judged on a decade-long horizon. That could be seen as a way of evading scrutiny and lead to accusations of being an expensive vanity project. The agency has already come under fire for being exempt from Freedom of Information laws.

But patience will likely be needed. Arpa began work on the “Arpanet” – the technology that would underpin the internet – in 1966. It would be three decades before the Information superhighway made it into people’s homes. 

If Aria works, we might be enjoying its inventions a century on from now – long after anyone will remember Dominic Cummings.

Aria’s seven big ideas

1. Better robot bodies

While robot brains are improving dramatically as artificial intelligence gets more powerful, their bodies are largely made up of metal and plastic, which is heavy and unsubtle compared to our dexterous fingers.

Making more supple or flexible materials to build robots with could mean they are less likely to crush items they pick up and could manipulate objects more easily, a potential route to a luxurious life of robot butlers and cleaners.

2. Proving that AI is safe

The rise of systems such as ChatGPT has sparked new warnings of killer artificial intelligence that could inadvertently cause catastrophes or help terrorists. At the moment, ensuring that such systems are safe is more art than science – to make them safer, chatbots are stress-tested by experts or tweaked when people discover ways to make them misbehave.

One project at Aria is devoted to developing mathematical algorithms that could prove that AI will not do harm, or to ensure that it would follow certain requirements. This would allow AI systems to have “safety guarantees” so they could be used more widely.

3. More accurate weather measurement

Accurately predicting weather and measuring climate change is likely to become more crucial over the coming decades. Aria researchers are looking at whether optical devices such as infrared sensors could measure things such as methane being released by permafrost ground or to better study cloud patterns.

The sensors could be placed on satellites and drones to map much of the world and understand its composition, allowing better prediction of weather patterns.

4. Making computers work like brains

Artificial intelligence uses huge amounts of computing power and energy, with data centres expected to account for up to a fifth of global electricity demand by the end of the decade. In comparison, biological brains compute information extremely efficiently.

By reducing electricity demand and following an emerging field called “nature-inspired computing”, computers could be modelled on how information is processed by human brains.

5. Brain implants/technologies

Tens of thousands of pacemakers are fitted each year to manage how hearts beat. What if brain implants could manage disorders such as anxiety, addiction and Alzheimer’s? One Aria project is seeking to develop “non-invasive” technologies that could manage these conditions.

It comes amid rising interest in brain implants including from companies such as Elon Musk’s Neuralink that have shown early promise.

6. Engineering plants to feed the world

Genetically modified crops are widely used to grow food, but researchers believe this could be taken a step further with “programmable plants” whose genes are edited to make them grow faster or to be more resilient to climate change.

Gene editing technologies such as CRISPR could be used to edit plants’ genetic codes more quickly than existing techniques which would typically take eight years to develop a new crop. 

7. Managing the weather 

Scientists are looking at how they could counteract the growing number of extreme weather events such as storms and heatwaves, as well as rising global temperatures. One solution would be tweaking where and when weather events take place.

Powerful supercomputer climate modelling could release elements into the atmosphere to cool the planet. And drones could make sure that rain storms happen at sea rather than over flood-prone areas.