by Xiaowei Wang
For a long time, China used its own version of community policing: the local neighborhood association—volunteers from the neighborhood, nosy neighbors and gray-haired grannies—was the eyes and ears for police officers. In recent years, the Chinese police have been relying on outside expertise to help modernize training and policing. There have been numerous “learning exchanges” between China and Europe, as well as the United States, where police officers from China visit police departments abroad to see how modern policing is done. Like capitalism and the free market, China’s models of policing have been based on U.S. models. Chinese police academy exchange students have ended up at Sam Houston State University in Texas, and the Los Angeles police even proudly tout “Police Diplomacy,” citing the number of police exchange visits between China and L.A.
In places that are experiencing rapid development, like Guiyang, the policing focus is on addressing the surge in crime that comes from economic disparity. Swindles and other schemes to make money, robbery, and muggings are the target. Despite official statistics of a “low crime rate,” there’s also a proliferation of “criminal villages” in China, where residents of an entire village will engage in some kind of criminal activity. Such villages are like Taobao villages (rural villages that produce merchandise for the e-commerce platform Taobao), but instead of households achieving the Chinese Dream through legal means, they draw on illicit tactics ranging from financial scams involving cake delivery to running kidnapping businesses.1 China’s low crime rate is also due to skewed statistics that don’t include urban migrant workers, who are the victims and perpetrators of up to 80 percent of urban crime.
Xiaoli pulls up the Real Population Platform and the first page is a gorgeously shot aerial image. It’s not a satellite image, but the result of drone photography, put together by the Beijing geographic information systems company SuperMap.
Each house in the image has been assigned an arbitrary number starting from 1. I ask Xiaoli if it’s the actual address and he explains, “So, the thing with urban villages in Guiyang is that there are no actual addresses. All of the construction is very informal. A few of the buildings were here to begin with, village buildings on farmland. Those old buildings are still around, and because there’s money to be made renting rooms out, landlords are always adding to their buildings. Or constructing new buildings in between existing buildings.”
Xiaoli continues, “The whole reason for this platform is because right now, Guiyang is developing fast, with so many migrants. And 80 percent of Guiyang’s migrants live in urban villages. And 70 to 80 percent of all Guiyang crime occurs in the urban villages. So what are we supposed to do? That’s why we have this platform, we have to register and track everyone. It’s for public safety.” The migrants he’s talking about are the same people I have met in the countryside throughout my travels: young men and women who have left their rural homes in search of economic opportunity. As migrants they are called liudongrenkou (流动人口) in Chinese, the “floating population”—a floating population in a floating world.
Xiaoli’s office is small and messy. There’s a couch, which I am sitting on, and a blanket and pillow in one corner for a quick nap on long shifts. In another corner, folded clean police uniforms are stacked on top of a mini-fridge, next to a teapot and tins of tea. He clicks around some more on the map. Each house is perfectly numbered, some with hyphens like 1-1 or 684-1. When he clicks on the house, a list of residents pops up in a small window. I ask him how the numbers are so precise, in the absence of formal addresses, and how they get the information about the residents.
In my mind, I imagine some sophisticated computer vision tool that looks at the aerial image, calculates the boundary of the house, and then assigns it a number. I imagine that the city has sensors and surveillance cameras to capture how many people leave the house. I also imagine that the surveillance cameras would know the face and personal ID number of each resident, perhaps tracked all the way from their tiny rural village through the numerous cameras I see everywhere—in train stations, at vending machines, on the street.
Instead, this system relies not on automation but on people. Xiaoli clicks around the map some more. “Every police station has numerous police assistants that live in the urban village. They are our eyes and ears. They are embedded in the community and they’re the ones who ground-truth the existence of every single house on this aerial image, giving each house a number. And they are the ones who help landlords register on the platform. People register through WeChat using the mini program.” Xiaoli pulls the WeChat app up on his screen, and taps into the mini program section. After tapping one of the registration buttons, a form comes up. The police assistants Xiaoli refers to are ordinary citizens who have deep ties in the urban village and have lived in the community for a long time. They also alert the police officers at this station of incidents that happen in the village.
“But registration is voluntary, there’s no way logistically we can force landlords to register everyone. Plus there’s also the problem…” He pauses and reaches for a thin stapled book that’s hidden under a mess of papers on his desk. On the cover are three passport photos: two men and a woman. “There’s the problem that some landlords are old, illiterate, or both. They have no clue how to use WeChat to scan and register their tenants. So the police assistants just give them this booklet to fill out and we put all their information in manually.”
He glances at another stack of papers near the window. “It takes a really long time. So right now we only have about sixty thousand people registered on this platform, even after a year or so.”
All this information sits in a database, a hulking engineering marvel that underpins so much of our modern world. Databases allow people to read, write, update, and destroy data in a fairly dependable way. They also require the people who build databases to form strong opinions about the world and the way it’s structured. For example, the attributes of a user on a platform are dictated by columns an engineer defines in the database. Different databases have different logics for the way data must be formatted, which in turn shapes the way we have come to encode the world. In the case of Real Population Platform, Xiaoli tells me the hardest part is data compatibility.
“To be honest, many of the recent upgrades in Guiyang have been a headache.” Xiaoli looks at me, and then suddenly asks, “It is true that Americans each have a number that allows them to be tracked? But that there is only one database that has that number? The social benefits number?” It takes me a second to realize that he means social security numbers. After all, it’s not immediately obvious to me that a social security number tracks us. But it does, as any American can attest to: the social security number and credit score follows us, it dictates if we get loans, if we can access credit, and if we can access housing. And while we give our social security number out somewhat casually, research has shown the ways credit scores, attached to our social security numbers, exacerbate deeply entrenched inequality in the United States. For an individual, it’s an innocuous number, but on a large scale, it forms a hulking system.
I nod and tell Xiaoli that it is just one number, the social security number, and some private credit-scoring bureaus. Xiaoli looks quizzical. “Well, it’s been extra difficult because in China, we have multiple databases. You can use your ID number and one address for work papers, while using a different address for your electricity bill. Especially for migrant workers, it’s hard for anyone to tell where you actually live. No one has a permanent address. Before the Real Population Platform, we used a database where the data format was very strict. Each person could only have one address. Duplicated names were not allowed. The Real Population Platform works a lot better, and we have a database where one person can have multiple addresses. Still, trying to reconcile all the entries is difficult. A lot of people have the same names, you know?”
On his screen, he continues to click around. Each entry for an individual has their address, date of birth, and national ID nu
mber. Some of the entries have photos, others do not. I imagine a horrible scenario where a case of mistaken identity arises from a database error, a typo, or a dropped row.
For Xiaoli, the difficulties of creating data emphasize not only how important data is but how it doesn’t just appear out of thin air, extracted by technology. Data needs to be collected, chased down, massaged, tidied. Which means that the collection of data and the design of a database are intentional: the dimensions of data that need to be collected, defined by a database, already imbue a vision of how a world should be turned granular. Like the translating of a human life into columns and rows, any image of an object is simply a representation of the object. No matter what you use to photograph an object, the image remains a grainy approximation.
I ask Xiaoli when they will be done collecting data on everyone. Xiaoli responds that they are trying to work as fast as they can, but they will never be sure if they have everyone in the database. “We rely on the landlords, because we think a landlord would only like to rent to dependable people. But you never know. We have community police assistants that visit apartments during the day and at night. Many of the migrant workers do jobs that have night shifts. It’s hard to tell how many people live in a place. There’s some places crowded with bunk beds that allow double the capacity. Night workers sleep in the beds during the day, day workers sleep in the same beds at night.”
To collect more data, the local government has been partnering with companies like China Unicom, the mobile carrier, to advertise the advantages of registering on the Real Population Platform. When I ask why China Unicom would push such a campaign, Xiaoli remarks in a cynical, disaffected tone, “Because the more people you have using the platform, the more data people burn. Unicom can make money.”
After an hour of me questioning Xiaoli, and several cups of tea, we start talking about his family life and his time working at the police station. He’s optimistic about the state of policing, and feels strongly that the community in the urban village is best suited to guide policing strategies. We talk about how he’s a young police officer, which he admits hasn’t been easy. Many of the older police officers have their own views about how things have traditionally been done and how things should continue to be done. He makes a joke about these older police officers being ancient rocks. “You kick them a little bit, and they move an inch. Not a foot, but an inch. I guess you just keep kicking!” Xiaoli adds that older police officers tend to view the technology as an annoyance, but since they have to use it, they will resort to a kind of magical thinking about it, convinced that technology can do more than it actually does, or they eschew deeper understanding for lazy, one-click solutions.
As for the future of Real Population Platform, I wonder if everyone will be as realistic and understanding of what technology can and cannot do as Xiaoli. He has brought up issues before with the platform builders about the user interface, adding certain columns into the database, but has been met with pushback by the engineers. So Xiaoli comes up with his own work-arounds.
Finally, I blurt out, “What about predictive policing?”
Xiaoli handles this question smoothly, laughing at me. “Listen, if we could actually prevent crime, that would mean I found a way to predict the future. None of us can predict the future. If someone wants to commit a crime, they will commit it. But with this platform, we will try to collect the best data sources, and then, with all that data, we can check up on people and know existing areas where crime has happened, who has committed it, and who we should watch out for. For example, we want to figure out who is making homemade drugs in this urban village. We are trying to get electricity usage per household, and sudden spikes in the electricity meter would likely indicate illicit activity. But there are strict data-sharing laws across organizations, so it’s not so easy. Even then, there’s the question of reconciling all that data, making it useful.”
Xiaoli’s job is simultaneously chaotic and mundane. Over the course of our entire conversation, phone calls come in, and he scribbles down numbers and addresses. Much of modernizing the police department, becoming more “United States–like,” is the emphasis on reporting, tallying, measuring the impact of policing itself. The reasoning is that left without performance indicators and statistics, policing becomes a haphazard endeavor. So to modernize, you get the Real Population Platform and the database, every movement and interaction between the police and a suspect tracked.
Through the computer screen, Xiaoli types in ID numbers, finding people. On-screen, everyone is just another entry in the database. It’s when he gets called into the neighborhood that these abstract numbers become animated with emotion: a domestic disturbance between a man and a woman, a belligerent man threatening his cousin. Xiaoli says he tries his best to stay calm. But it’s in the gulf between that number in the database and the visceral, adrenaline rush of responding to a call that fear comes in, a gulf created by the abstraction of numbers.
2.
The Megvii (Face++) office in Beijing is boring, banal. It reminds me of many offices in the Bay Area. A beverage fridge sits in the corner, filled with tea instead of LaCroix cans.
After the B or C stage of funding in the startup world, life in the office is marked by a level of nervous comfort. Making software is expensive. It requires engineers, whose market-rate salary is high, whether it’s New York, Beijing, or Shanghai. On top of labor costs is the careful massaging of “office vibes,” creating a space that’s productive but casual, designed to attract talent. There are also constant server costs—computation time can become expensive. On top of that, building certain products involves a long research-and-development phase; it may be years before any customers actually appear. The strangeness of the market and the cost of building software makes it an endeavor bankrolled by a slew of VCs who bring buckets of money, then show up in the office for quarterly meetings, for assurances that their wealth is being well spent.
As an engineer, or “individual contributor,” your days are filled with caffeine and snacks, computer screens, and work that alternates between a technical challenge that makes your soul tremble, and unsatisfying procedural tasks. Whatever task you are working on is just a small piece of a bigger picture. It can be difficult to understand the full technical scope of any project, untangling years of code written by someone else who has long since left the company. During my stint as an engineer, working in a San Francisco office, I would sporadically change my text editor to have a custom set of colors, or create new aliases for bash commands—as if variegating my visual landscape would knock me out of a daze. I felt a visceral pleasure in building and optimizing systems. Untangling bugs gave me a surge of adrenaline, like I was a private investigator. Yet coding for work meant placing an emphasis on details. Minutiae can take on significance—maybe a respite, or a natural reaction to inject coding with a touch of sentiment.
At Megvii, engineers sit on one side of the room, some at standing desks, pointing to lines of code on each other’s screens. A few have two screens set up, typing into a text editor while idly browsing a shopping site on another screen. Most of the engineers are men, somewhere between the ages of twenty and thirty-five, bespectacled and wearing untucked T-shirts. Sitting among them are some designers, who stare at mock-ups of interfaces. At the other end of the office are what I presume to be the sales and marketing teams. A bored man at the front desk collects packages from the couriers who walk through a set of jumpy automatic glass doors.
If the military science lab was seen as the birthplace of twentieth-century nuclear annihilation, the twenty-first century’s death by ecological destruction and unfettered capitalism is symbolized by a glass-cube conference room with a whiteboard. Down the hall, a large product showroom proudly heralds Face++’s achievements. This is a tech company that could be any tech company in the world. This generic geography allays my apprehensions about a Chinese surveillance state. It’s instead overshadowed by worry over the making of a global surveillance industry,
by people who stand to profit heavily from it.
Face++ is powering many platforms with its facial- and image-recognition algorithms. To be clear, it stores no data on any of its servers—it simply provides the mechanism to recognize a face. First, the algorithm has to recognize that there is a face within an image, and perhaps a primary face within an image of multiple faces. It can distinguish eyes from a nose, which is handy for many of China’s Meitu photo-beautification apps that allow you to edit someone’s image beyond recognition: changing the shape of your lips, adding lipstick and eyeshadow. In a world inundated with social media images, your selfies need to always look good.
Beyond recognizing parts of faces, the algorithm can start to discern characteristics about you, characteristics to classify you—the distance between your eyes, or the width of your chin. These distances are compared to the average set of collected measurements in order to make categorical assumptions about you—whether you’re looking happy or sad that day, or even more troubling judgments.
One step after gathering characteristics is facial recognition: taking the image of your face, distilling it down to measurements, and being able to search a database of faces. Face recognition is a system with numerous parts, and each part is the domain of a private company—whether the one that owns the surveillance cameras used, the algorithm, or the computational power rented out on a server.
The Face++ showroom has plush white carpeting and shiny white walls with inset screens. One wall features real-time camera footage from outside the showroom, in the office and outside the building. The display showcases how fast and precise Face++ computer vision algorithms are—as someone walks by the building, the algorithm detects their blue pants and umbrella. There’s also a hidden camera that you can stand in front of and the algorithm instantly classifies your age and gender. I am for some reason characterized by the Face++ algorithm as a twenty-seven-year-old male. The algorithm does not compute my thirty-four-year-old non-binary existence.
