Show Hide image

Improbable: the billion-dollar startup that aims to reprogramme government itself

Co-founders Herman Narula and Rob Whitehead explain why virtual worlds are technology's next big earthquake

We stand on the wooden deck of a ship, staring out at a sea of clouds. A school of flying manta-ray creatures drifts past. In the distance, islands hang in the sky.

“That isn’t background,” says Herman Narula, gesturing to the distant horizon in the video game, Worlds Adrift, which we’re looking at. “Those rocks in the distance are places you can go to. You can go to all of these places - there is no painted background.”

This, says Narula, is the fundamental difference between this new game and every other video game ever made. Every game made prior to Worlds Adrift offered “the illusion of a world; this is an actual world.”

This is what makes Improbable, the London-based startup company founded by Narula five years ago from his parents’ house while he was still at university, now worth over a billion dollars: it offers the means to make actual worlds.


The million-player game


The technology developed by Narula, his co-founder Rob Whitehead and his team at Improbable has been widely misrepresented in the rush to explain why a small startup was able to secure second-round funding of $502m, without handing over control of the company. The BBC, the Guardian and others have described Improbable as a ‘virtual reality’ company, but this is simplistic and misleading. Improbable’s product is not a game or a headset, but an operating system. In the same way that Microsoft makes Windows and software developers write games and applications that run on Windows, Improbable makes a platform called SpatialOS, upon which new kinds of games are being built. The difference is that each installation of Windows – or any other operating system, such as the iOS software that runs on every iPhone – runs on a single computer. SpatialOS runs on thousands of computers, all at the same time. Improbable’s technology is distributed computing, and Narula believes it will be even more influential in our future than the Windows PC or the iPhone have been in our past.

People describe Improbable as a virtual reality company because this is what SpatialOS enables people to make. By bringing together thousands of machines into a distributed supercomputer, it provides software developers with the means to create huge virtual worlds that can be inhabited by many thousands of people at the same time.

The first industry that Improbable aims to transform is the gaming industry, the biggest and fastest-growing of the entertainment media. The biggest event in online gaming so far took place in a game called Eve Online. Over 21 hours on a Monday in 2014, a conflict over a space station escalated into an all-or-nothing showdown between the game’s warring factions. At its peak, more than 2,500 players joined a single map. The game had to run ten times slower than usual to accommodate the number of players present. For SpatialOS, this number of simultaneous interactions would represent a normal day in a small virtual world. Despite the large number of players currently enjoying the early preview of Worlds Adrift – Narula says there are “thousands” – the developers are currently limiting the size of the world they’re playing in, to give them any chance of bumping into one another.

“When you play World of Warcraft,” explains Narula, “you’re not really in a single world. You’re in thousands of copies of the same world, and those copies are all static. They never change, nothing you do in the world ever creates a lasting impact. So the kind of engagement you can have involves going through a linear story, like in a single-player game, but you have other people around that can help you with it.” Worlds Adrift, on the other hand, “is a single world the size of Wales, with millions of simulated entities inside it. And every action any player takes in the world permanently impacts the world in a way that then creates effects that other people encounter.”

“It has cartographers,” adds Rob Whitehead.  There is a cadence to the way Improbable’s two co-founders talk. Narula’s speech is very fast but also very detailed and composed, as if he is reading as quickly as possible from an internal textbook. Whitehead says less, dropping in observations that Narula quickly elaborates upon, but in speaking to either there is the same sense of chatty yet formidable intelligence. At one point in our conversation Whitehead makes a comment about the Mandelbrot set, and from his expression it is clear that he assumes everyone in the room knows what he is talking about.

“Yeah, cartographers,” continues Narula. “There are people on the internet, flying around this world with compasses and wind speed, to try to create maps of how big this world is.”

Narula is eager to stress that this is not an incremental development in games programming. He describes the difference as “binary”; “you’re either running on one server or on a massive distributed cluster. If you’re running on a massive distributed cluster, thousands of new things are possible. The endgame for this company, what we’re trying to build and to make happen, is literally to create new realities.”

It is at this point that Narula’s mountainous ambitions loom into view. He talks about upscaling “from thousands of machines to millions of machines, to datacentres that are located near population centres, to the point where we can have low-latency, massive interactions that are happening with millions of people.”

If the idea of a million people playing a game at once sounds unlikely, consider that this would amount to 0.05 per cent of Facebook’s current membership.           

New nations

On 27 June, Facebook announced that its membership has now exceeded two billion people. If Facebook was a country it would be by far the most populous nation on Earth, with a population more than six times that of the United States (assuming every account is a real person). In 2015, Facebook claimed it had indirectly created 4.5 million jobs worldwide. 

For people who work in traditional industries, the idea that a computer game could hold its own internal economy and create large numbers of jobs may sound flimsy. For Rob Whitehead, it’s as normal as a paper round. Before he went to Cambridge, Whitehead “built virtual goods - weapons, gadgets, that kind of thing” in Second Life, the online virtual world that has been running since 2003. “That was my job. I went into uni with a couple of tens of thousands of dollars’ worth of virtual money, made from this virtual thing. Of course from my side, I saw it as just content creation, but people within that world derived real meaning and had real experiences because of the things I made.”

Narula says most of the value now applied to goods is no more or less real than the things Whitehead built in Second Life. Design, scarcity, social value and skill dictate whether a piece of leather becomes a £1,000 handbag or a £3.99 windscreen cleaner. “So much of what we already do in the industries in which we create value is constructing realities on top of the base value of whatever it is you’re dealing with.”

A game with millions of players would have, Narula says, “its own massive, internally consistent economy”.

With millions – or hundreds of millions, or billions – of people joining a virtual world, the current thinking about what constitutes a video game would be left a very long way in the past. In fact, says Narula, “That stops being a game, and it starts being a country.”

If the games industry can produce new worlds of this size, the effects will be felt far beyond the community of players and PCs. “This is a fundamental thing,” says Narula, “equivalent to AI in terms of its potential importance to society. This comes under the category of those enabling technologies, industrial processes that no-one thinks of, but that happen to enable really important things that happen in the world. Distributed computing, and in particular this type of distributed computing, is that important, that fundamental – and that hard. It requires you to rethink the foundations of how applications are written.”

If Narula is right, it will require people to rethink more than just programming. The simulations that can be created in the hive-mind of Improbable’s distributed supercomputer are so complex that they may offer the chance to rethink how cities and transport systems and economic markets are run, and how governments create and justify policy. It may be the most disruptive technology in government since the ballot box.

The what-if machine

For policymakers the most important aspect of Improbable’s work is not the transformation of the $100bn gaming industry but the disruption of how policy is written and tested. “One of the most important uses of virtual worlds,” Narula explains, “is answering questions about the real world – a world of complex adaptive systems, which can probably not be understood unless we can recreate their behaviour at very large scale. The weather is a complex adaptive system, and the way we predict the weather is through massive-scale simulation. The same is true of so much of what we do.

“Most public policy is built on shaky assumptions on top of these complex adaptive systems, and being able to properly model these systems is going to make a big difference. Virtual worlds will become our collective “what-if” machine, before we try things in the  real world.”

Improbable is already using its technology in this way, with surprising results. One of the first projects was a simulation of the city of Cambridge and everyone in it: “the population, the transport network, the sewage network, the mobile phone network, electricity, gas, water, and also the spread of panic in certain situations. One of the things we found that was quite counter-intuitive was that in some situations, when something bad happens, damage can be limited by actually turning off communication, because it prevented panic spreading, which prevented traffic building up. That was startling. The idea that making people communicate less is going to help people in a disaster isn’t something we’re proposing as a policy, but these counterintuitive behaviours look to be interesting.”

In the planning of infrastructure, rational but counterintuitive arguments can often lose to more easily grasped assumptions. “The classic challenge,” says Narula, “is that a lot of the obvious thinking - building a road to reduce congestion, for example - turns out not to be true when you look at the real behaviour of these systems.” Infrastructure simulation could provide a cheaper, more effective way to determine the efficacy of contentious projects such as HS2 and allow for better planning of everything from telecoms networks to healthcare spending.

What Narula and Whitehead assume, of course, is that people will listen to reason. Brexit’s predicted impact on the economy, for example, actually became part of the anti-intellectual rhetoric of the Leave campaign. No prediction about the future of the country can be made apolitically, and it is more likely that virtualisation will be used more by businesses to pitch their case for policy changes than a newly enlightened government. 

What is certain is that if Narula and Whitehead succeed in providing the means to create new worlds for millions of people and to change the way governments and businesses plan for the future, their platform will become, like Windows and iOS and Facebook, a very powerful force in the world. Mark Zuckerberg once thought of himself as mere platform provider, outside politics, but recent events have forced him to confront the very real influence his company wields. Do the Improbable founders have a simulation for that?

“There are maybe six Olympian businesses on this planet,” Narula agrees, “and the fact is that no-one can argue that they don’t have a special and unassailable status of almost undisruptable, hegemonic control over certain aspects of life. I don’t know if that’s a situation that can continue without some form of rethink.

“Nobody wants to see another group of tech people get rich in a non-socially conscious way. We like to think that what we’re doing here will actually improve people’s lives in tangible and direct ways. That’s one of the reasons that public policy and government work were really fundamental in the early stages of the company. We didn’t go there because we thought it was where the most revenue was – we felt it was a way for us to understand what real problems we can impact.”

“But,” he adds, “video games are cool, too.” 

Will Dunn is the New Statesman's Special Projects Editor. 

Show Hide image

Why the government refused to invest in the most successful computer Britain has ever made

Eben Upton, whose Raspberry Pi computer was denied government support – went on to become the most successful British computer ever made. Here, he tells us how

On 25 June 2009, Eben Upton received a reply from the East of England Development Agency. He and his colleagues had recently applied for a loan guarantee to start a new company making a very small, very cheap computer that would help people learn programming. The response was not encouraging.

“Thank you for your application,” it read. “Given the widespread availability of ‘proper’ computers, the rationale for . . . a device that reverts to the early days of computing is not persuasive. The panel also noted that the demand for application programmers has fallen since the last decades of the 20th century, and it was felt that the number . . . is unlikely to rise to previous levels.” The email then added that “the National Curriculum does not specifically mandate programming skills”.

This letter is tech’s equivalent of Decca Records’ declaration, when declining to sign The Beatles in 1962 that “guitar groups are on the way out”. Despite the doomsaying of the East of England Development Agency, Upton’s Raspberry Pi computer quickly became the biggest selling computer Britain has ever produced, with more than 10 million units sold in less than five years.

Better still, it is – unlike almost every other computer on the market – made in Britain, and its popularity in schools has led to a demonstrable increase in the number of people applying to study computer science. Why does its creator think the government is poor at spotting promising emerging technologies?

“Large organisations struggle to allocate money to the right things,” Upton says. “We know that central planning doesn’t work, for a variety of reasons. There’s an information problem – it’s hard to get information into the centre about what’s going on at the edges – and it’s also difficult because of politics.

“Resources tend to flow towards people who are good at playing the game. If you’re good at playing the game of Get Resources, you get more resources and more power and more experience of getting resources. The wrong things get funded. That’s true of government, and it can also be true of large companies. It can be true of any large, centrally planned organisation – corporations are little mini Soviet Unions, they’re miniature centrally planned economies. But corporations have a way out – they can buy other corporations.

“Over the last 50 years, we’ve seen a gradual de-emphasising of the idea of research and development, or at least internal R&D, towards – I’ve heard it called research and acquisition. Rather than trying to develop new technologies or products yourself, you as a corporation focus on trying to do the things you’re already doing really well. And you rely on the little nimble guys to run around discovering new things to do, then make sure you’ve got a good enough radar to spot them while they’re still small, and you snaffle them up and apply your big-company economies of scale to make those things work. That’s the established private-sector solution to the problem.

“On the startup side, you see a lot of companies that are starting up really with the intention of being acquired. So both sides co-evolve towards each other – and you see this in pharmaceuticals as much as in the tech sector. It leads to a profusion of small companies that might never turn into large, self-sustaining businesses, but that’s not the point. It works, it delivers growth. The annoying thing is, I’m not sure anyone has a plausible story as to how the public sector can learn from this experience.”

Examples of the misguided attempts by government to participate in the private sector’s ecosystem of research and acquisition are dismayingly easy to find, but the most resounding clanger of recent years must be, the “gift economy” website that was awarded £200,000 of government funds at around the same time as Upton was launching the Raspberry Pi. Described by The Spectator as “disastrously vacuous” and The Register as “a website that replicates the ‘Help Needed’ pages of Craigslist and Freecycle”, is reported to have lost around £250,000 per year since it launched. The crucial difference in securing funding was not the validity of its premise, but the fact that it is the brainchild of the supermodel Lily Cole, who at the time had a personal wealth estimated at £7m.  

Upton says a more effective way for government to stimulate innovation can be to invest at scale, over time, and with the goal of creating regional centres of excellence. “In 1978, the government, through the National Enterprise Board, invested £50m in a company called Inmos, based in Bristol. They built memory and processors, including the famed transputer, which was innovative but never became a commercial success. The business never became profitable, and after consuming more than £200m of government funding was sold to the French-Italian company SGS-Thomson in 1989; it was a failure. But if you go to Bristol now, the place is crawling with chip companies, and in fact the chip we use in the Raspberry Pi 3 was largely designed in Bristol. When industrial policy investments work, it is often in this sort of non-linear way.”

One of the key differences in the way the public and private sectors successfully invest in technology is duration. The staying power of public-sector investment is its unique selling point. “Inmos was there for more than a decade. Throwing some money in, letting it burn up in two years and then it dries up and blows away – that’s very different from putting money in for something that’s going to exist for 10 years.”

Magnify this kind of investment, says Upton, and you can explain the success of the biggest and most lucrative centre of innovation in the world: Silicon Valley. “People draw completely the wrong conclusion about Silicon Valley. Why is Silicon Valley where it is today? The answer is, about a trillion dollars in defence spending. The federal government spent an enormous amount of money over a long period with giant aerospace companies such as Lockheed and Westinghouse: that’s where a lot of the smart people came from. And often, those people had kids, and their kids didn’t go to work for Lockheed. The kids went to Stanford, and then founded tech companies, and those companies were able to hire from a pool of talent that was already in place.

“The mistake that governments make, a lot, though, is to try to build copies of Silicon Valley as it is now, rather than looking at where it came from. A long-term investment in something, even if it fails, can have the side effect of pulling lots of clever people into one place.” Counterintuitively, then, it may be that the most successful policy for fostering innovation is to look for the things that seem likely to fail, and which have no obvious commercial application.

“They have the Defense Advanced Research Projects Agency [DARPA] model in the US, where they spend money on ‘moonshot’ ideas that might turn into something, with one or two very notable successes – one of which was the internet.

“DARPA will pretty much fund ideas in proportion to their outlandishness – the internet is a very weird idea – where the East of England Development Agency was only able to fund things in direct proportion to their conformity to existing ideas.”

Should the government, then, be in the business of funding “moonshot” ideas? “In my view, the right thing to do is actually to fund basic university research better. We already have a system for doing moonshots: university researchers. If you work in the government and you’re tempted to be a venture capitalist, quit and try to get a job as a venture capitalist. We know that the government is good at funding universities, and academics are good at reviewing each other’s grant proposals to allocate that funding effectively. So, we have a working system there.

“It’s worth bearing in mind that the government activities that have really helped Raspberry Pi have been very traditional ones. The provision of free education, which created our workforce, the promotion of foreign direct investment, which led to the construction of the Sony factory in South Wales where we build the Raspberry Pi, and then reform of the computing curriculum, which was also to our massive advantage – this is pretty conventional stuff. It’s always been this way. Look at the BBC Computer Literacy Project, which gave us the BBC Micro: the government simply gave grants to schools to buy computers, which again is very conventional, but the long-term impact on UK industry was transformative – much more so than any speculative investment in an individual start-up.”

For Upton, while there are many questions to be asked about funding at the surface level, the broader perspective is that he and many other successful technologists have been aided over the decades by the education, the infrastructure and the economic environment that the UK government has provided for them.  

“I think our government’s quite good at what it does, actually. It’s been there for hundreds of years, and has an enormous depth of experience. We’ve got this very capable machine for doing traditional government things, and as long as we can keep it focused on those then it has the potential to be an incredible source of competitive advantage to our country.”

Will Dunn is the New Statesman's Special Projects Editor.