by Kai-Fu Lee
But the California-based startup Traptic has created a robot that can handle the task. The device is mounted on the back of a small tractor (or, in the future, an autonomous vehicle) and uses advanced vision algorithms to find the strawberries amid a sea of foliage. Those same algorithms check the color of the fruit to judge ripeness, and a machine arm delicately plucks them without any damage to the berry.
Amazon’s warehouses give us an early glimpse of how transformative these technologies can be. Just five years ago, they looked like traditional warehouses: long aisles of sedentary shelves with humans walking or driving down the aisles to fetch inventory. Today, the humans stay put and the shelves come to them. Warehouses are covered with roving bands of autonomous beetle-like robots that scurry around with square-shaped towers of merchandise sitting on their backs. These beetles roam the factory floor, narrowly avoiding one another and bringing a handful of items to stationary humans when they need those goods. All the employees need to do is grab an item off that tower, scan it, and place it in a box. The humans stand in one place while the warehouse performs an elegantly choreographed autonomous ballet all around them.
All of these autonomous robots have one thing in common: they create direct economic value for their owners. As noted, autonomous AI will surface first in commercial settings because these robots create a tangible return on investment by doing the jobs of workers who are growing either more expensive or harder to find.
Domestic workers in the United States—cleaners, cooks, and caretakers—largely fit those criteria as well, but we’re unlikely to see autonomous AI in the home any time soon. Counter to what sci-fi films have conditioned us to believe, human-like robots for the home remain out of reach. Seemingly simple tasks like cleaning a room or babysitting a child are far beyond AI’s current capabilities, and our cluttered living environments constitute obstacle courses for clumsy robots.
SWARM INTELLIGENCE
But as autonomous technology becomes more agile and more intelligent, we will see some mind-bending and life-saving applications of the technology, particularly with drones. Swarms of autonomous drones will work together to paint the exterior of your house in just a few hours. Heat-resistant drone swarms will fight forest fires with hundreds of times the current efficiency of traditional fire crews. Other drones will perform search-and-rescue operations in the aftermath of hurricanes and earthquakes, bringing food and water to the stranded and teaming up with nearby drones to airlift people out.
Along these lines, China will almost certainly take the lead in autonomous drone technology. Shenzhen is home to DJI, the world’s premier drone maker and what renowned tech journalist Chris Anderson called “the best company I have ever encountered.” DJI is estimated to already own 50 percent of the North American drone market and even larger portions of the high-end segment. The company dedicates enormous resources to research and development, and is already deploying some autonomous drones for industrial and personal use. Swarm technologies are still in their infancy, but when hooked into Shenzhen’s unmatched hardware ecosystem, the results will be awe-inspiring.
As these swarms transform our skies, autonomous cars will transform our roads. That revolution will also go far beyond transportation, disrupting urban environments, labor markets, and how we organize our days. Companies like Google have clearly demonstrated that self-driving cars will be far safer and more efficient than human drivers. Right now, dozens of startups, technology juggernauts, legacy carmakers, and electric vehicle makers are in an all-out sprint to be the first to truly commercialize the technology. Google, Baidu, Uber, Didi, Tesla, and many more are building teams, testing technologies, and gathering data en route to taking human drivers entirely out of the equation.
The leaders in that race—Google, through its self-driving spinoff Waymo, and Tesla—represent two different philosophies for autonomous deployment, two approaches with eerie echoes in the policies of the two AI superpowers.
THE GOOGLE APPROACH VERSUS THE TESLA APPROACH
Google was the first company to develop autonomous driving technology, but it has been relatively slow to deploy that technology at scale. Behind that caution is an underlying philosophy: build the perfect product and then make the jump straight to full autonomy once the system is far safer than human drivers. It’s the approach of a perfectionist, one with a very low tolerance for risk to human lives or corporate reputation. It’s also a sign of how large a lead Google has on the competition due to its multiyear head start on research. Tesla has taken a more incremental approach in an attempt to make up ground. Elon Musk’s company has tacked on limited autonomous features to their cars as soon as they became available: autopilot for highways, autosteer for crash avoidance, and self-parking capabilities. It’s an approach that accelerates speed of deployment while also accepting a certain level of risk.
The two approaches are powered by the same thing that powers AI: data. Self-driving cars must be trained on millions, maybe billions, of miles of driving data so they can learn to identify objects and predict the movements of cars and pedestrians. That data draws from thousands of different vehicles on the road, and it all feeds into one central “brain,” the core collection of algorithms that powers decision-making across the fleet. It means that when any autonomous car encounters a new situation, all the cars running on those algorithms learn from it.
Google has taken a slow-and-steady approach to gathering that data, driving around its own small fleet of vehicles equipped with very expensive sensing technologies. Tesla instead began installing cheaper equipment on its commercial vehicles, letting Tesla owners gather the data for them when they use certain autonomous features. The different approaches have led to a massive data gap between the two companies. By 2016, Google had taken six years to accumulate 1.5 million miles of real-world driving data. In just six months, Tesla had accumulated 47 million miles.
Google and Tesla are now inching toward one another in terms of approach. Google—perhaps feeling the heat from Tesla and other rivals—accelerated deployment of fully autonomous vehicles, piloting a program with taxi-like vehicles in the Phoenix metropolitan area. Meanwhile, Tesla appears to have pumped the brakes on its rapid rollout of fully autonomous vehicles, a deceleration that followed a May 2016 crash that killed a Tesla owner who was using autopilot.
But the fundamental difference in approach remains, and it presents a real tradeoff. Google is aiming for impeccable safety, but in the process it has delayed deployment of systems that could likely already save lives. Tesla takes a more techno-utilitarian approach, pushing their cars to market once they are an improvement over human drivers, hoping that the faster rates of data accumulation will train the systems earlier and save lives overall.
CHINA’S “TESLA” APPROACH
When managing a country of 1.39 billion people—one in which 260,000 people die in car accidents each year—the Chinese mentality is that you can’t let the perfect be the enemy of the good. That is, rather than wait for flawless self-driving cars to arrive, Chinese leaders will likely look for ways to deploy more limited autonomous vehicles in controlled settings. That deployment will have the side effect of leading to more exponential growth in the accumulation of data and a corresponding advance in the power of the AI behind it.
Key to that incremental deployment will be the construction of new infrastructure specifically made to accommodate autonomous vehicles. In the United States, in contrast, we build self-driving cars to adapt to our existing roads because we assume the roads can’t change. In China, there’s a sense that everything can change—including current roads. Indeed, local officials are already modifying existing highways, reorganizing freight patterns, and building cities that will be tailor-made for driverless cars.
Highway regulators in the Chinese province of Zhejiang have already announced plans to build the country’s first intelligent superhighway, infrastructure outfitted from the start for autonomous and electric vehicles. The plan calls for integrating se
nsors and wireless communication among the road, cars, and drivers to increase speeds by 20 to 30 percent and dramatically reduce fatalities. The superhighway will have photovoltaic solar panels built into the road surface, energy that feeds into charging stations for electric vehicles. In the long term, the goal is to be able to continuously charge electric vehicles while they drive. If successful, the project will accelerate deployment of autonomous and electric vehicles, leveraging the fact that long before autonomous AI can handle the chaos of urban driving, it can easily deal with highways—and gather more data in the process.
But Chinese officials aren’t just adapting existing roads to autonomous vehicles. They’re building entirely new cities around the technology. Sixty miles south of Beijing sits the Xiong’an New Area, a collection of sleepy villages where the central government has ordered the construction of a showcase city for technological progress and environmental sustainability. The city is projected to take in $583 billion worth of infrastructure spending and reach a population of 2.5 million, nearly as many people as Chicago. The idea of building a new Chicago from the ground up is fairly unthinkable in the United States, but in China it’s just one piece of the government’s urban planning toolkit.
Xiong’an is poised to be the world’s first city built specifically to accommodate autonomous vehicles. Baidu has signed agreements with the local government to build an “AI City” with a focus on traffic management, autonomous vehicles, and environmental protection. Adaptations could include sensors in the cement, traffic lights equipped with computer vision, intersections that know the age of pedestrians crossing them, and dramatic reductions in space needed for parked cars. When everyone is hailing his or her own autonomous taxi, why not turn those parking lots into urban parks?
Taking things a step further, brand-new developments like Xiong’an could even route the traffic in their city centers underground, reserving the heart of town for pedestrians and bicyclists. It’s a system that would be difficult, if not impossible, to implement in a world of human drivers prone to human errors that clog up tunnels. But by combining augmented roads, controlled lighting, and autonomous vehicles, an entire underground traffic grid could be running at the speed of highways while life aboveground moves at a more human pace.
There’s no guarantee that all of these high-flying AI amenities will be rolled out smoothly—some of China’s technologically themed developments have flopped, and some brand-new cities have struggled to attract residents. But the central government has placed a high priority on the project, and if successful, cities like Xiong’an will grow up together with autonomous AI. They will benefit from the efficiencies AI brings and will feed ever more data back into the algorithms. America’s current infrastructure means that autonomous AI must adapt to and conquer the cities around it. In China, the government’s proactive approach is to transform that conquest into coevolution.
THE AUTONOMOUS BALANCE OF POWER
While all of this may sound exciting and innovative to the Chinese landscape, the hard truth is that no amount of government support can guarantee that China will lead in autonomous AI. When it comes to the core technology needed for self-driving cars, American companies remain two to three years ahead of China. In technology timelines, that’s light-years of distance. Part of that stems from the relative importance of elite expertise in fourth-wave AI: safety issues and sheer complexity make autonomous vehicles a much tougher engineering nut to crack. It’s a problem that requires a core team of world-class engineers rather than just a broad base of good ones. This tilts the playing field back toward the United States, where the best engineers from around the globe still cluster at companies like Google.
Silicon Valley companies also have a substantial head start on research and development, a product of the valley’s proclivity for moonshot projects. Google began testing its self-driving cars as early as 2009, and many of its engineers went on to found early self-driving startups. China’s boom in such startups really didn’t begin until around 2016. Chinese giants like Baidu and autonomous-vehicle startups like Momenta, JingChi, and Pony.ai, however, are rapidly catching up in technology and data. Baidu’s Apollo project—an open-source partnership and data-sharing arrangement among fifty autonomous-vehicle players, including chipmakers like Nvidia and automakers like Ford and Daimler—also presents an ambitious alternative to Waymo’s closed, in-house approach. But even with that rapid catch-up by Chinese players, there’s no question that as of this writing, the most experienced self-driving technologists still call America home.
Predicting which country takes the lead in autonomous AI largely comes down to one main question: will the primary bottleneck to full deployment be one of technology or policy? If the most intractable problems for deployment are merely technical ones, Google’s Waymo has the best shot at solving them years ahead of the nearest competitor. But if new advances in fields like computer vision quickly disseminate throughout the industry—essentially, a rising technical tide lifting all boats—then Silicon Valley’s head start on core technology may prove irrelevant. Many companies will become capable of building safe autonomous vehicles, and deployment will then become a matter of policy adaptation. In that universe, China’s Tesla-esque policymaking will give its companies the edge.
At this point, we just don’t yet know where that bottleneck will be, and fourth-wave AI remains anyone’s game. While today the United States enjoys a commanding lead (90–10), in five years’ time I give the United States and China even odds of leading the world in self-driving cars, with China having the edge in hardware-intensive applications such as autonomous drones. In the table below, I summarize my assessment of U.S. and Chinese capabilities across all four waves of AI, both in the present day and with my best estimate for how that balance will have evolved five years in the future.
The balance of capabilities between the United States and China across the four waves of AI, currently and estimated for five years in the future
CONQUERING MARKETS AND ARMING INSURGENTS
What happens when you try to take these game-changing AI products global? Thus far, much of the work done in AI has been contained within the Chinese and U.S. markets, with companies largely avoiding direct competition on the home turf of the other nation. But despite the fact that the United States and China are the two largest economies in the world, the vast majority of AI’s future users still live in other countries, many of them in the developing world. Any company that wants to be the Facebook or Google of the AI age needs a strategy for reaching those users and winning those markets.
Not surprisingly, Chinese and American tech companies are taking very different approaches to global markets: while America’s global juggernauts seek to conquer these markets for themselves, China is instead arming the local startup insurgents.
In other words, Silicon Valley giants like Google, Facebook, and Uber want to directly introduce their products to these markets. They’ll make limited efforts at localization but will largely stick to the traditional playbook. They will build one global product and push it out on billions of different users around the globe. It’s an all-or-nothing approach with a huge potential upside if the conquest succeeds, but it also has a high chance of leaving empty-handed.
Chinese companies are instead steering clear of direct competition and investing in the scrappy local startups that Silicon Valley looks to wipe out. For example, in India and Southeast Asia, Alibaba and Tencent are pouring money and resources into homegrown startups that are fighting tooth and nail against juggernauts like Amazon. It’s an approach rooted in the country’s own native experience. People like Alibaba founder Jack Ma know how dangerous a ragtag bunch of insurgents can be when battling a monolithic foreign giant. So instead of seeking to both squash those startups and outcompete Silicon Valley, they’re throwing their lot in with the locals.
RIDE-HAILING RUMBLE
There are already some precedents for the Chinese approach. Ever since Didi drove Uber out of China,
it has invested in and partnered with local startups fighting to do the same thing in other countries: Lyft in the United States, Ola in India, Grab in Singapore, Taxify in Estonia, and Careem in the Middle East. After investing in Brazil’s 99 Taxi in 2017, Didi outright acquired the company in early 2018. Together these startups have formed a global anti-Uber alliance, one that runs on Chinese money and benefits from Chinese know-how. After taking on Didi’s investments, some of the startups have even rebuilt their apps in Didi’s image, and others are planning to tap into Didi’s strength in AI: optimizing driver matching, automatically adjudicating rider-driver disputes, and eventually rolling out autonomous vehicles.
We don’t know the current depth of these technical exchanges, but they could serve as an alternate model of AI globalization: empower homegrown startups by marrying worldwide AI expertise to local data. It’s a model built more on cooperation than conquest, and it may prove better suited to globalizing a technology that requires both top-quality engineers and ground-up data collection.
AI has a much higher localization quotient than earlier internet services. Self-driving cars in India need to learn the way pedestrians navigate the streets of Bangalore, and micro-lending apps in Brazil need to absorb the spending habits of millennials in Rio de Janeiro. Some algorithmic training can be transferred between different user bases, but there’s no substitute for actual, real-world data.