Inside X, Google's top-secret moonshot factory

X – formerly Google X – aims to pursue technological breakthroughs by taking crazy ideas seriously. When will its bets pay off?

Gandalf arrives on rollerblades. It’s morning in the cafeteria at X – formerly Google X – and Astro Teller, X’s Captain of Moonshots, glides over dressed in coarse grey robes and a pointed hat, carrying oatmeal. Jedi stroll past to their desks, gripping coffee. Star Fleet officers queue for breakfast. This, it should be said, is unusual – it’s Halloween. But X is a surreal place. Outside, self-driving cars loop around the block. Sections of stratospheric balloons, designed to broadcast internet to remote places, hang in the lobby. Robots wheel around, sorting the recycling. Teller likens X to Willy Wonka’s chocolate factory; it seems only fitting that there be costumes.

Even standing inside X – a cavernous former mall in Mountain View, California – it’s hard to articulate exactly what X is. Within Alphabet, Google’s parent company, it is grouped alongside Deepmind in "Other Bets", although in that metaphor, X is more like the gambler. Its stated aim is to pursue what it calls “moonshots” – to try to solve humanity’s great problems by inventing radical new technologies. To that end, besides the self-driving cars (now a standalone company, Waymo) and internet balloons (Loon), X has built delivery drones (Wing), contact lenses that measure glucose in the tears of diabetics (Verily) and technology to store electricity using molten salt (Malta). It has pursued, but ultimately abandoned, attempts to create carbon neutral fuel from seawater, and replace ocean freight with cargo blimps. It once earnestly debated laying a giant copper ring around the North Pole to generate electricity from the Earth’s magnetic field.

That might sound fantastical or even absurd, but every day you almost certainly use something developed at X. Google Brain, the deep-learning division that now informs everything from Google Search to Translate, began at X. So did camera software GCam, used in Google Pixel phones; indoor mapping in Google Maps; and Wear OS, Android’s operating system for wearable devices.

But those are beside the point. “Google Brain, the cars, Verily, everything else – those are symptoms. Side effects of trying weird things, things that are unlikely to work,” Teller says. “We are a creativity organisation, not a technology organisation.” The rollerblades, which he wears every day, are tucked neatly under the table. (They save him eight minutes a day between meetings.) X, he explains, is not so much a company as a radical way of thinking, a method of pursuing technological breakthroughs by taking crazy ideas seriously. X’s job is not to invent new Google products, but to produce the inventions that might form the next Google.

X was once seen as a punchline in Silicon Valley (and on Silicon Valley). Today, its self-driving cars have logged 10 million miles on public roads, and operate an autonomous ride-sharing service in Arizona. Loon’s balloons provide internet access to communities in rural Peru and Kenya. Wing, X’s drone delivery effort, is carrying food and medicines to customers in Australia. Still, as Alphabet continues to be buffeted by employee protests and leadership changes – in December 2019, founders Larry Page and Sergey Brin stepped down, handing the company to Google CEO Sundar Pichai – X is facing renewed scrutiny to prove that its moonshots are more than just an indulgence, or expensive PR stunts. X celebrates its tenth anniversary in 2020. When will its bets pay off?

Alphabet is not the first company to set up a laboratory for chasing moonshot ideas. In 1925, AT&T and Western Electric founded Bell Labs, which assembled scientists and engineers from different disciplines to advance the field of telecommunications. Bell Labs invented the transistor, the first lasers and photovoltaic cells, winning nine Nobel prizes in the process. Ever since, corporate research labs, from Xerox PARC to Lockheed Martin's Skunk Works and DuPont’s Experimental Station, have played a central role in producing breakthrough inventions. Apple, Facebook, Microsoft and Amazon all have corporate research labs. Google has several, including Google AI (formerly Google Research), Robotics at Google, and Advanced Technologies and Projects, which works on things like AR and smart fabrics.

But corporate research labs are flawed. Big companies, chasing quarterly results, often ignore transformative ideas even from within their own organisations. Xerox PARC invented the graphical user interface, but we don’t sit at Xerox laptops. As startups grow into corporations, bureaucracy can take over, and their capacity to think creatively wanes. “Over a 20 to 30 year period, companies tend to move from experimentation to process,” Teller explains. “Process is an attempt to get surprise all the way down to zero. Experimentation is the complete, constant bathing in surprise. You can’t have both.”

X does not call itself a corporate research lab (it uses the term “Moonshot Factory”), but when it was founded in 2010, its remit wasn’t entirely clear. X originally grew out of Chauffeur, Google’s self-driving car project, then spearheaded by the Stanford roboticist Sebastian Thrun. Page and Brin admired Thrun for his work on Streetview and turn-by-turn directions in Google Maps, and at X they offered him free reign to pursue similarly offbeat ideas. “Initially, the title was called ‘Director of Other’,” Thrun says. “We wanted to push technologies in many different directions, including self-driving cars.”

For at least a year, X’s existence was a closely guarded secret. Other Google employees were denied keycard access. Even within Google, where bottom-up management is a founding principle and employees are allowed to spend 20 per cent of their time working on their own ideas, X had a free-wheeling, intellectually anarchic style. Engineers from Project Chauffeur worked alongside those from Google Brain, Loon and a handful of other equally audacious projects. “I wanted to get no bureaucracy, no PowerPoints, no financial reporting, no oversight, so that the people in charge could focus entirely on the challenge,” Thrun says. Most of the early project ideas came from Page and Brin themselves, who took a close interest, and eventually moved into X’s building. (Teller once described X as Brin’s “batcave”.)

When Thrun left X in 2012 for Udacity, his online education company, Teller took over. He was, in many ways, the natural choice. His paternal grandfather is Edward Teller, known as the father of the hydrogen bomb, and the co-founder of the US’s Lawrence Livermore National Laboratory. His maternal grandfather was a Nobel prize-winning economist. “I grew up as the dumb one in my family,” says Teller. “My family believed that being smart was the only thing that mattered. I wasn’t going to win on those terms. As a result, it forced me to explore other ways to be successful.” Before joining X, Astro (a nickname; his real name is Eric) founded an AI hedge fund, and sold a wearable sensors company. He has finished two novels and co-written a book of relationship advice. At school, to compensate for mild dyslexia, he would do every problem twice, using different methods. “Then if I came up with the same answer, it was the right answer,” he says. The experience taught him early the value of experimental thinking – “to just try things fast, to approximate at the beginning, and come at the problem from different angles.”

When Teller took over at X, it had little structure. “I would describe it as like the Wild West. We just started projects when we were interested in things. There was almost no process whatsoever,” says Obi Felten, who joined X from Google in 2012. Teller hired Felten, who then worked in Google’s product marketing department, to formalise the moonshot process. While the engineers pushed the frontiers of artificial neural networks and high-altitude ballooning, Felten says she dealt with “literally everything that wasn’t the tech. Legal policy, marketing, PR, partnerships. They’d never had a business plan for any of the projects before.” Her job title was "Head of Getting Moonshots Ready for Contact With the Real World".

Not all X projects survived first contact. One of the first was Google Glass, a wearable computer inside a pair of spectacles. Brin loved the idea, and pushed X hard to turn the early prototypes into a consumer product. When Glass eventually launched in 2013, Google created huge fanfare. Skydivers wearing Glass parachuted on to the roof of its annual developers' conference. Models wore them on the runway at New York Fashion Week. They were featured on The Simpsons and in Vogue.

But in the real world, Glass faced poor reviews, mockery (“Glassholes”) and outrage at potential invasions of privacy. “The real failure that we had with Glass was when we were trying to talk about it as a learning platform, the public started responding to it like a product,” Teller says. “What was worse is we fell into the trap of talking about it that way ourselves. And that was terrible, because it was not a finished product.”

Glass was discontinued as a consumer product in 2015. It still exists, but as an enterprise tool mostly used in manufacturing and other manual industries. “Sometimes it just doesn’t work, the technology’s not ready, and we have to stop doing it, pause it, slow it down,” Teller says. He still believes that a Glass-like device will catch on eventually. (Apple is reportedly working on AR glasses for debut in 2022.) “There's no way to take moonshots and never be too early. By the definition of what we’re doing, we’re erring on the side of being too early, rather than being too late.”

Once a week or so, X’s smartest minds gather in a conference room and set about systematically killing each other’s craziest ideas. To be considered a moonshot, an idea must fulfil three criteria: it must address a significant global problem, involve inventing breakthrough technology, and result in a radical outcome – at least “10X” better than what exists today. Jetpacks and hoverboards might be fun, but don’t serve a common good.

Similarly, distributing vaccinations is a worthy goal, but isn’t a moonshot. “If most or all of the audacity is just in the scale of the thing, that is not what we’re interested in,” says Teller.

Whatever the problem, no solution is too outlandish. “Everything has to be on the table,” says Teller. “So someone said in a brainstorming session: what if guns actually shot some kind of lethal poison, but there was an antidote in all the jails in the country?” Teller grins. “First of all, that’s a gorgeous idea – I mean, it’s a terrible idea. But in terms of the creativity of the idea, that is good, and that person will think of other really weird ideas that are awesome for society.”

After a moonshot is proposed, X’s Rapid Evaluation team – a rotating cast of Xers with expertise from materials science to artificial intelligence – begin what is known as a pre-mortem. “Let's imagine everything fell apart. What are the things that caused that?” says Phil Watson, who heads the Rapid Evaluation department. Ninety per cent of ideas fail at this stage. Some fail for obvious reasons: too expensive, too difficult. Others break the laws of physics. If an idea can’t be killed easily, then it becomes an investigation, and is assigned a small team to interrogate it further. “We start looking into it more systematically,” Watson explains. “What would it take for this to succeed? What are the skill sets we would need to be able to advance it to the next level? What's the thing most likely to make this fail?” Successful investigations become projects, with a name, budget and full-time staff.

One of the foundational tenets at X is "Monkey First". That is, if you were asked to teach a monkey to stand on a pedestal and recite Shakespeare, you should resist starting with the simplest task (building the pedestal) and start with the hardest (teaching a monkey to speak). Teams are encouraged to set both performance milestones and "kill targets" – thresholds that, if missed, will automatically end the project. For example, Project Foghorn, X’s attempt to turn seawater into fuel, succeeded in producing fuel but couldn’t do so cheaply enough. X killed the project, published its findings as a scientific paper, and gave the team a bonus.

One afternoon I follow Kathryn Zealand, of the Rapid Evaluation team, to see an investigation at work. The idea is to build a pair of assistive trousers that might help the elderly and immobile walk independently. “The space of ageing in general hasn’t had enough investment,” says Zealand, who is Australian. “The demographics mean it’s going to be a big thing in the future.”

The trousers, code-named Smarty Pants, are inspired both by recent advancements in soft robotics and Zealand’s own experience with her 92-year-old grandmother, who has Alzheimer’s. At that age, even simple movements like standing become difficult. “If you help them do that one thing, they take 30 per cent more steps over the course of the day. And the longer people stay walking, the few other health issues they’ll have,” Zealand says. She is wearing what looks like 3D-printed armour on one leg, wired with sensors, which collect data on her gait as we walk.

Early on in any investigation, X begins prototyping. X’s Design Kitchen houses virtually anything required to conduct experiments in numerous fields: a wet lab, milling machines, laser scanners, 3D printers. “We say, OK, what is the quickest experiment we can do that will get us to a yes/no?” Zealand says.

We arrive in a large, airy atrium. Zealand has enrolled her mother, who happens to be visiting, as a test subject. “She struggles with stairs,” Zealand explains. The trousers prototype is rough: actuators on each knee joint, connected to fabric panels around the wearer’s legs. The outer seams are laced, corset-style, giving them a Victorian steampunk vibe. The motors are controlled by a Raspberry Pi in a pearlescent bumbag. Zealand’s team – including a deep learning specialist, a clothing designer and a world expert in biomechanical exoskeletons – fit her mother into the trousers, then monitor her as she climbs several flights of stairs. “It’s amazing,” she says, stepping down, delighted. “Normally I’d be out of breath by the time I got up there.”

Zealand asks me if I want to try them. After a brief fitting, I tread gingerly on the first step, and immediately feel myself being pulled upwards, as if by an additional set of muscles. Climbing is noticeably easier. By using the sensor data and machine learning, Zealand explains, the trousers are learning to "see" the stairs, knowing exactly when to apply force. She hopes that, eventually, soft robotics and material advances might enable a lightweight product a fraction of the weight, with a flexible frame, that could aid a range of mobility issues. “That’s probably 10 years out,” she says. Still, it’s early days. Fewer than half of X’s investigations become Projects. By the time this story is published it will probably have been killed.

The ability to work on such long-term problems is X’s great advantage: the patience of research, without the financial pressures of a startup. “There are some technologies that, because of safety, you have to have several ‘nines’ of reliability before you can even get started,” says Teller. “There’s a really big difference between a 1 per cent error rate and 0.001 per cent error rate.” A software glitch in a mobile app is unlikely to be fatal, but one in a self-driving car might be.

The thought lingers that afternoon, as a driverless Waymo pulls up outside the X campus. Since its start at X in 2009, Waymo has now logged more than 10 million autonomous miles on public roads. For the last year, it has operated Waymo as a small scale ride-hailing service in Phoenix, Arizona, and is currently working with Jaguar on its next generation of vehicles. Morgan Stanley recently valued the company at $105 billion (£80 billion). “Waymo’s goal is to build the world’s most experienced driver,” says Andrew Chatham, a Waymo software engineer. “It is not to build a car. Other people are quite good at that.”

We pull away. A safety driver, Rick, is sitting in the front seat, but the wheel turns itself. Inside the car, a white Chrysler Pacifica, headrest displays show a live view of what the roof-mounted sensors ‘see’: pedestrians in yellow wireframes, the purple outlines of other vehicles. The ride is surprisingly quiet, and, other than some slight hesitations – even now the cars struggle to predict drivers’ intentions at junctions – entirely uneventful.

Even so, mass adoption of self-driving cars is still a long way off. “If I look at where we were in the first six months [back in 2009], we had some really cool videos. And here we are more than a decade later, and those videos… it was too easy to get there,” Chatham says. “Actually getting to the level of deployment is a whole other ball game.”

During the early years, X employees could happily work on technologies that might be decades away, knowing that all the while advertising revenue was flooding into Google. Thrun recalls asking former Google CEO and Alphabet executive chairman Eric Schmidt for $30 million to fund a project. Schmidt gave him $150 million. “Eric said to me, ‘If I give you $30million, you're going to come back next month and ask for another $30million.’”

Then, one morning in 2015, Brin and Page announced that Google was restructuring, to become Alphabet. The news came as a shock inside the company. There were widespread reports of budgets being tightened. But at X, being spun out clarified the team’s essential purpose: “[It] became even more clear that the goal of X is to produce new Alphabet companies,” Felten says.

When projects reach a certain scale, they “graduate” from X to become standalone companies. Most, like Waymo, join Alphabet’s Other Bets. A few have been acquired by Google, or spun out independently, such as the renewable energy startups Dandelion and Malta. Upon graduation, project leaders become executives, and employees are given a stake in the company. “When projects leave here, they're not done,” says Teller. “There's plenty of learning still to be done.”

The transition is not always easy. Since Alphabetisation, the original leaders of several X projects, including Waymo, Loon and Wing, have either left or been replaced. “If you’re trying to accelerate something to grow ambitious and large, the chances that the original person that invented it can take [it through] every stage is pretty hard,” Wendy Tan White, a vice-president at X who oversees growth-stage projects, says. “They would have to grow very fast themselves.”

Alphabet as an organisation has also faced various controversies in its first five years. In 2018, the company was rocked by allegations of sexual misconduct by senior executives; 20,000 employees, including many at X, staged a global walkout in protest. One of the executives named by the New York Times was Richard DeVaul, then head of Rapid Evaluation at X, and one of the original creators of Loon. (DeVaul resigned, reportedly without an exit package.)

Teller has publicly expressed his regret at the episode, and his admiration for those involved in the walkout. “It made me believe in Google and Alphabet more,” he tells me. “Awesome that employees should say, 'This is our company too and that needs to reflect us.’”

There has also been widespread upset at Google’s involvement in Project Maven, a Pentagon artificial intelligence project, and Project Dragonfly, a reported plan to launch a censored search engine in China. (Both have reportedly been shelved.) These events have reignited debates about the responsibility that Alphabet has, in Google’s once famous phrase, to not be evil.

Although these projects did not come under X, Teller says he thinks deeply about the ethics of his team’s work. His grandfather, after all, worked on the Manhattan Project. “There definitely have been things people brought up [at X] that are like, 'Nope, that's evil'. We're just not doing that,” he says. But other issues are less clear-cut – for example, projects that may cause jobs to be lost to automation. One current X project is to create all-purpose “everyday robots” that might automate menial tasks. “New technologies tend to create concentrated harm and diffuse benefits,” Teller says. “If the benefits of automation are 100 times the downsides, that leaves us with 99 per cent positive. But we owe it to those people who have experienced those concentrated problems – and it is a public policy matter – to make sure we take care of them.”

After its flurry of early successes, X’s moonshots in recent years have struggled to strike the popular imagination – or financial success. Of its energy startups, only Malta and Dandelion have to date built a commercial product. Chronicle, its cybersecurity moonshot to create an immune system for the internet, was recently folded back into Google. And while Wing’s drones may end up transforming the logistics industry, it’s hard to paint delivering burritos as a genuine moonshot.

Recently, X has increased its effort on tackling problems that might threaten humanity, like climate change. “Climate change is by any reasonable standard the single biggest problem that humanity has,” Teller says. There are several climate-focused projects in development, including research into ocean health. One of the most advanced, currently untitled, is focused on agriculture. “It’s one of our basic needs. It’s one of the largest industries in the world. It has the largest carbon footprint of any major industry,” Teller says.

In an X workshop on the second floor, engineers are working on several boxy blue vehicles with stilt-like legs and off-road tyres. They are farming drones, designed to comb a field in groups, taking hyperspectral images of the crops and topsoil. The drones are already in testing on some farms in California, “collecting millions of images of plants, where every strawberry has a unique ID”, explains Benoit Schillings, an ebullient Belgian who oversees several moonshots at X. “Agriculture is a huge, complex optimisation problem. Right now the way it’s done is by simplifying the problem: we’re going to put hybrid corn over 10,000 acres,” Schillings says. By analysing the data and making suggestions, X hopes to improve crop yields and soil health.

The agriculture project is a typical X moonshot: take a huge problem and apply Alphabet’s massive advantage in computing power, intellectual expertise and financial resources to try and solve it – creating a global business in the process. “Going after ‘Let's solve agriculture’ – this is pretty hubristic,” Schilling laughs. “We take the challenges that I think very few other players would have the courage to take.”

But it’s also possible to see moonshots as something more cynical: attempts to dominate industries that don’t yet exist. Agriculture, worldwide, is a trillion-dollar market. It’s hard not to picture Waymo as the operating system inside every car, Wing as the air traffic control for every airborne package. Google itself, after all, started as a kind of moonshot, to systematically map all of human knowledge. For all of X’s rhetoric about changing the world, it ultimately exists to produce new enterprises – and profit – for Alphabet.

Teller, however, seems unphased by the notion that creating wildly profitable companies and solving issues like climate change are in opposition. To X, and Alphabet, creating the next Google and saving the world are essentially the same thing. “Things that lose money tend to get smaller over time, and things that make money tend to get bigger over time,” he says. “I see purpose and profit not in opposition. I see them as intensely synergistic.”

X recently celebrated its tenth birthday. The day I visit, the senior executives have been in meetings, trying to map out what the next decade might look like. “The world is changing,” Tan White says. Fields that X once pioneered, such as self-driving cars, are now cottage industries. The concept of moonshots is now used widely by both startups and governments. Massive venture capital funds, such as SoftBank’s Vision Fund, are enabling startups to take moonshot-sized risks of their own.

X’s true impact may not be clear for another decade, or more. While it has created sizeable returns for Alphabet – Teller has said the value of Google Brain alone paid for several years of X’s entire budget – there’s still no telling if its graduated companies will survive, let alone become the next Google. In 2018, Alphabet’s "Other Bets" lost $3.36 billion (£2.57 billion). “We have to accept that some of these businesses still won’t make it,” Teller says. But then, almost all attempts at invention end in failure. Genuine breakthroughs require both immense capital, creativity and, perhaps most important, patience.

“If we define moonshots as trying to create something really radical… that’s really hard. The really big things, it’s tough to see who’s going to do it,” Nathan Myhrvold, the former director of Microsoft Research and founder of Intellectual Ventures, told me. “But the flipside is if you’ve got those resources and you don’t do it, then we’ll never know if there was some fabulous technology that took that kind of effort.”

Sitting in front of Teller-as-Gandalf, it’s hard not to think of an even earlier moonshot factory: the labs of Thomas Edison, known as the Wizard Of Menlo Park. It may be that the self-driving car, or one of X’s many other moonshots, will end up transforming society in ways we can’t yet foresee. It might save the world, or it might just help Alphabet grow ever richer and more powerful.

“The real test is 15 to 20 years from now, when the dust is settled and we look backwards. Then how are we doing?” Teller says. Until then, there will always be more crazy ideas worth chasing. “The world’s got more than enough problems, sadly.”

Updated 17.02.20, 11:50 GMT: This article originally stated that of X’s energy startups, only Malta had built a commercial product. Dandelion has also built a commercial product. Sebastian Thrun also created Udacity, not Coursera

This article was originally published by WIRED UK