We joke about our “robot overlord” all the time. It is like what Pete Davidson said once on SNL, it’s “a serious joke; you know like when you are joking, but you mean it”.
Well, joking or not, for some people, this has already become their reality. A friend of mine, who is a reporter for GQ China, recently published a report for the magazine about the workers and regular folks who are, wittingly or unwittingly, serving the future Skynet. With her permission, I am posting a synopsis of the story (badly put together by me, my apologies) here, as I think some of you might be interested.
For reference and those who could read Chinese, the original article is linked here: https://mp.weixin.qq.com/s/Orj8tNOS9W0mz5Vk6XVSMQ, with the author’s information available at the bottom, and some pictures taken during her interviews. Also, she recently told me that a German press institution had reached out to her for the right of full translation into German and English. If it becomes available I will link it here too. My injected commentary are put in italic to be separated from the original contents.
1. The story begins with the first main character, a 29-year-old girl named Ma. Everyday, she sits in front of her computer from 8am to 5pm, pictures with labeled annotations flash across the screen. Her job is to double check if all the objects in the pictures have been correctly labeled and make sure there is nothing missing.
First picture is an outdoor European cafe. She moves her mouse, and a green box appears on the picture with the word “Chair” in it. She drags another box over some flowers in a vase, and typed “Flowers (in a vase)”.
Next picture, a Japanese flower market. She examines every “Pot” labeled in the picture.
Next, an American style kid’s bedroom. “Chair”, “Desk”… she deletes a label “Ladder” — the client means those moveable ladders, not a stationary one on a bunk bed.
Behind her, a roomful of people in front hundreds of computer screens are doing the same thing. It almost feels like an old-fashion Internet cafe in the 90s. They are picking out all the “Garbage Can”s, “Coffee Table”s, “Carpet”s in numerous pictures grabbed from social network sites around the globe. Most of these pictures are blurry, taken from all sorts of angles, and do not have high resolution: some of them were clearly taken from random surveillance cameras on the streets.
Ma goes through 1000 pictures a day at least. The current order has been going on for two months, during which she and her co-workers must have identified at least tens of thousands of chairs and tables. As far as Ma knows, once they are done, all their work products will be sent back to a fancy unicorn company in Beijing, becoming learning data for AI.
What she doesn’t know, nor does she necessarily care, is that her work is an indispensable part of helping AI eventually recognize the “Moveable Ladders”. AI could be smart, but at first, they have to rely on the “artificial” labeling services performed by thousands of regular humans like Ma. In the past year, her responsibilities include labeling joints on human bodies, identifying different vehicles on the road, and recording words and phrases into a smartphone – the content of which are confidential.
Ma used to be one of the typical “factory girls” that you might have read about before. She did sewing jobs, worked for Foxconn, and worked assembly line. After moving back to her hometown, she got this job titled “Computer Operator”. She likes it: it pays less than Foxconn, but more than being a cashier; she gets to use computer while sitting in an air-conditioned room; the hours are good to take care of her kid; and she knows her bosses from her childhood.
2. Liu is one of her bosses, and the second main character. He started this company with two partners about a year ago in three days. One day they heard about a new business idea called “data labeling”, so he tried it out himself, and realized that the process was no more difficult than taking a screenshot. The next day, they bought some ethernet cables, and 20 office cubicles, $12 each. Third day, they sent out a hiring ad via a WeChat (think WhatsApp) group message, looking for 20-something workers, and immediately filled 20 seats.
His first employees included junior high graduates, stay-at-home moms, and former retail workers. Within a week, everyone has “mastered” the skills needed for the job. His “factory” soon expands to have 500 computers. Every worker is hired locally – Ma’s village alone has 7 people working here. Ma recently bought a SUV with a loan, and they carpool to work when it’s raining.
Liu was born in 1987, and has been quite a drifter. He used to study computer in a vocational school before dropping out. Then he became an excavator operator while studying computer and programming by himself. Over the years, he became a sales representative for beverage bottles, sold imported fertilizer, even went to Chile for a month. His boss at the time wanted to sell off-brand Chinese cell phones to South America, and made him learn Spanish for a month. It didn’t last, and the only thing he remembered was how to say “Hola” and “Gracias”. Before eventually returning home, he also spent some time in Zhuhai, a city in Guangdong Province. While there, he helped design a small microcontroller that automatically performs assembling tasks for phones – what Ma used to do on an assembly line.
As we hear all the time, automation is replacing labor jobs in an unprecedented scale. Experts in the field have expressed concerns over the fast developing of AI and automation, and the social impact the new technologies might have. However, for people like Liu, this is not something that is going to keep him up at night. In this booming industry where new startups are gobbling up billions of dollars of investment, companies like Liu’s are the most insignificant Davids surviving in the world of Goliaths. His employees currently make about 1 cent for each labeling done, and that’s twice higher than it was a year ago. They often have to tolerate the arrogance of the contractees. Most of the times that means steady orders, but occasionally, the bosses have to cover their employees’ salaries while some big companies delay the payments for months.
3. By this point, you have probably gathered that Liu and Ma do not live near metropolitans such as Beijing and Shanghai. Liu’s company slash factory is located in Jia County, a rural county of Henan Province with a population less than 600k. Translating to American terms, this is quite similar to the fly-over/rustle belt regions. In fact, Jia County is literally a wheat growing coal mining county, and it might just be the third character in this story.
Being here means Liu has much cheaper labor cost than his competitors in cities like Beijing, and Liu does not require his workers to have advanced degrees. None of the three founders has a college degree anyway. In Liu’s words, when it comes to dragging boxes, “all men are created equal”.
The benefits of small town also manifest in unexpected this year. Human facial recognition has become the newest trend in the Chinese AI industry recently, which means there is a sudden increase of demands in relevant raw data: videos of Chinese adults under various lightings. Liu’s company is not labeling videos yet, but gathering raw data itself has become a brand new business opportunity. The contractees pay about $15 for each video sample, which takes about 45 minutes to collect. Considering the running cost, there simply is not enough money left to attract volunteers in any bigger cities.
In the small county, things are quite different. Liu has a subsidiary company in a small town. Boxes and boxes of rice, soybean oil and toilet paper pile at the front door: those are the prizes being given away “for free” to any town folks who stop by to be recorded. Liu and his partners used to do something similar in villages all over the country, offering gifts such as off-brand laundry detergents (think “T1de” instead of “Tide”) in exchange of registering new accounts on a start-up “pay day loan app” (they are causing all kinds of crisis in China these days, but that’s a different topic altogether). The small town farmers don’t use the app at all, neither do they care about the prizes being off-brand. Back then, Liu could make $10 kickback for every new account. On a good day, he could get hundreds of them and earned over $1000 easily. Recording “head-shot” videos is an easier task now, but much less profitable: about $100 per day at most.
While the company is doing pretty well right now, ironically, the three founders think that they could probably last for 5 years tops, because one day, the AI they help create might eventually replace themselves. These data labeling companies do not have access to any actual AI technologies: no computers in their office even has a hard drive as everything is done through the cloud; they are unable to store samples, nor do they have the permission to insert new ones directly. Nevertheless, in the short term, their laborious work is not going to be replaced in Jia County. Some of Liu’s biggest contracts these days are servicing self-driving cars now. They are labeling human faces, vehicles, 3D atlas, and audios, as well as handling different weather conditions.
The last day of the reporter’s visit, they have collected 37 videos: 20 females and 17 males, not a particularly good day. An employee asks one of the last volunteers, a middle-aged local man to post an ad on his WeChat page to get more people to come. The man stops, but he has no idea how to do it.
The employee grabs his phone, adds the two as friends so he could send over the text of the ad, and then posts the message on the man’s page – his first post ever.
The man takes his phone back, glances at the post, doesn’t pay much attention to what it says, and goes home with his bottle of soybean oil.
That’s most of the article. Thanks for reading! The original article is much better written of course, and I don’t take credit for the story itself at all. In any case, if humans still exist in the self-driving car future, maybe we’ll remember how much humans have to do with it.