ChinAI #111: The Human Cost of 30-min Food Deliveries

Renwu magazine's longform exposé on how drivers are trapped in a dangerous system

Greetings from a world where the price of punctuality is too high…

…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).

ChinAI Links (Four to Forward)

Switching up the order this week. There’s a feature translation of a really moving, special piece so I wanted to include the full text. Plugging a few things first, and then the last link will set the scene for this week’s 16page, 7000-word+ translation (and I only got halfway through this week).

First, GovAI is hiring for a project manager:

Should-read: Chinese Climate Narratives Newsletter

This summer I got to work with Mel Guo, a fellow at the Stanford Existential Risks Initiative. She is starting a newsletter titled “ To burn jade and common stone” (玉石俱焚), a Chinese idiom that means to destroy indiscriminately. It encapsulates the consequences without China’s buy-in on addressing climate change — the destruction of everything, precious or ordinary. At the intersection of Chinese civil society and climate change, the newsletter will unpack discussions of environmental organizers on Weibo, Chinese public opinion on climate change, etc. Check it out and subscribe!

Should-read: In China, Therapy is Going Mobile

RestofWorld, which reports on tech stories outside the Western bubble, has just published a story about the rise of online psychology apps in China, a country where psychotherapy was once condemned. Thanks to a thriving middle class looking to self-actualize, answers to questions like “are you happy?” are finally being addressed. For 100 to 1,200 TMB, a user can choose from hundreds of mental health professionals located anywhere from Beijing to Boston and book a consultation on such topics ranging from self-love to career pressure.

Must-reads: Some Food Delivery Background Reading before the Longform Translation

  • The thread that first linked me to the People (Renwu) magazine longform article:

  • An excellent article by Eliza Gkritsi, Nicole Jao, and Coco Gao for Technode that provides further insight into the labor status of drivers (most are independent contractors but they have made attempts at organizing).

  • After the People magazine exposé, China’s Meituan and Ele.me (the two giants food delivery platforms) have “announced tweaks to their algorithms as they try to head off a growing societal backlash against the stringent demands placed on their drivers.”

  • Your order, their labor: An Exploration of Algorithms and Laboring on Food Delivery platforms in China — an article in the Chinese Journal of Communication by Sun Ping, an assistant researcher at the Chinese Academy of Social Sciences who is cited throughout the People magazine article. She expands on the idea that delivery drivers have created their own “organic algorithms” or “inverse algorithms” to subvert the delivery platform system’s algorithm.

*By the way, if you’re a student and you need the full text to any of the academic article I link to, just hit me up and I’ll get you the pdf.

Feature Translation: Delivery Drivers, Trapped in the System

***I took an initial cut this week and got halfway through on the Google Doc. Can we crowdsource the rest of the translation? I think it’s a really worthwhile article to translate to get a sense of ordinary life in China. Just start writing suggestions in the Google doc so there’s no duplicate work! And then we’ll credit people when (hopefully) sharing the rest of the translation next week.

Context: Author: Lai Youxuan (赖祐萱); Editor: Jin Shi (金石); Source: 人物 (“People” - monthly magazine, first published in 1980, that started out focused on celebrity biographies. Original mandarin here. ChinAI previously translated a People article on a mother and her AI daughter.

No key takeaways this week, but I hope my informal translation can at least you give folks some sense of the type of journalism that can and does exist in China:


"Message Received"

Another two minutes disappeared from the system.

Ele.me driver Zhu Dahe clearly remembered that one day in October 2019. When he saw the system delivery time for an order, his hand holding the handlebar was sweaty: "2 kilometers, delivery within 30 minutes” — as someone who did food delivery in Beijing for two years, he knew that the previous shortest delivery time for the same distance was 32 minutes. But starting from that day, those two minutes were gone.

At about the same time, Meituan drivers also experienced similar “losing-time incidents." A Meituan driver who specializes in running long-distance deliveries in Chongqing found that the delivery time for orders within the same distance had changed from 50 minutes to 35 minutes; his roommate also did deliveries, and his delivery time limit for 3 kilometers had also been reduced to 30 minutes.

This is not the first instance of when time disappeared from the system.

Jin Zhuangzhuang has been the leader of the Meituan distribution station for three years. He clearly remembers that from 2016 to 2019, he received three notifications from the Meituan platform to "accelerate.” In 2016, the longest time for delivery within was 1 hour. In 2017, it became 45 minutes. In 2018, it was shortened by 7 minutes and fixed at 38 minutes. According to relevant data, in 2019, the average delivery time of delivery orders for the entire industry was 10 minutes faster than three years earlier.

The system has the ability to continuously "swallow" time. For the creators, this is commendable progress and a manifestation of the deep learning capabilities of AI’s intelligent algorithms. In Meituan, this "real-time intelligent distribution system" is called the "Super Brain.” Ele.me, named its system "Ark." In November 2016, Wang Xing, the founder of Meituan said in a media interview, "Our slogan is ‘Meituan waimai, song sha dou kuai (Meituan delivery, we deliver anything quickly.’ Deliveries arrive in 28 minutes on average." He said, "This is the embodiment of very good technology.

As for the delivery people who implement "technical progress," this can be "crazy" and "deadly."

In the setting of the system, delivery time is the most important indicator, and going over time is not allowed. Once it happens, it means bad reviews, reduced income, or even removal. In a Baidu Tieba where delivery drivers gather, one driver wrote, "Delivering food is a race against death, a competition with traffic police, and a friendship with red lights."

In order to keep himself alert, a Jiangsu driver changed his social account nickname to: going over time is 狗头. A Shanghai driver who lives in Songjiang said that he would drive in the wrong direction on basically every delivery. He calculated that he could save 5 minutes each time. Another Ele.me driver from Shanghai did a rough calculation. If he did not violate the rules, the number of deliveries he could run in a day would be reduced by half.

"Drivers will never be able to rely on their personal strength to fight the time allocated by the system. We can only use speeding to retrieve some time." One Meituan driver told People that the "craziest delivery" he has experienced is 20 minutes for one kilometer. Although the distance is not far, he needs to go get the food, wait for the food to be made, and deliver the food within 20 minutes. On that day, his speed was so fast that his butt bounced off his seat several times.

Speeding, running red lights, going in the wrong direction... In the opinion of Sun Ping, an assistant researcher at the Chinese Academy of Social Sciences, these delivery drivers’ actions to challenge traffic rules are a kind of "inverse algorithm," which is a last-resort work practice of drivers who have been under the control and discipline of system algorithms for a long time. The direct consequence of this "inverse algorithm" is that the number of traffic accidents encountered by delivery workers has risen sharply.

Sun Ping began to study the digital labor relationship between delivery system algorithms and drivers in 2017. In the exchange with "People” about the relationship between "shorter and shorter delivery times" and "more and more traffic accidents," she said, "It must be (the most important reason)." 

Actual data strongly supports this judgment. In the first half of 2017, data from the Traffic Police Corps of the Shanghai Public Security Bureau showed that in Shanghai, there would be a delivery driver casualty every 2.5 days on average. That same year, there were 12 casualties among Shenzhen delivery drivers within 3 months. In 2018, the Chengdu traffic police investigated and dealt with nearly 10,000 illegal drivers in 7 months, with 196 accidents and 155 casualties. On average, one driver was injured or killed every day. In September 2018, the Guangzhou traffic police investigated and dealt with nearly 2,000 delivery drivers who violated the law. Meituan accounted for half of them, and Ele.me ranked second.

The hashtag --- #外卖骑手,已经成为最危险的职业之一# (delivery driver has already become one of the most dangerous occupations) has become a Weibo hot search term more than once.

Specific cases from public reports are far more affecting than data——

In February 2018, an Ele.me driver was speeding on a lane closed to motorized vehicles and knocked down Li Mouqiu, one of the founders of the Shanghai Emergency Department, Ruijin Hospital, and Huashan Hospital emergency department. Li Mouqiu died one month later. In May 2019, a delivery driver in Jiangxi was in a hurry to make a delivery and hit a passerby who entered a vegetative state. A month later, a Chengdu driver ran into a Porsche while running through a red light, and his right leg flew off on the spot. In the same month, a delivery driver in Xuchang, Henan, went the wrong way on the motorway, was hit and flew in the air and rotated 2 times to the ground, causing multiple fractures all over his body...

Zhu Dahe, the driver with sweaty palms who was "terrified" by the delivery time, also once had an accident. In order to avoid a bicycle, he fell off his electric bike as he was speeding on a lane off-limits to motorized vehicles. The mala hot pot that he was delivering also flew off. At that time, what ran through his mind first before the physical pain was, "Oops, that will mean I went over time."

In order to avoid going over time and bad reviews, he called the customer and asked the other party to cancel the order, and bought that hotpot using his own money. "It's too expensive, more than 80 RMB," he said, "but it tastes good. "He still feels bad about it, because he had just entered the industry at the time and had insufficient experience. A more reasonable approach would be to pay the customer the money for the spicy pot and have them place another order. This way, “At least I could get the delivery fee for this trip," he said. "6.5 RMB, I remember it very well."

"Car crashes are all too common. As long as you don’t spill your meal, it’s not a big deal for a person to fall." Zhu Dahe said. When running deliveries, he has seen too many colleagues who had traffic accidents. “Normally I won’t stop,” because “I don’t have time for my own delivery.”

The experience of the Meituan driver Wei Lai confirms this statement.

At noon this spring, Wei Lai and a driver in uniform of the same color waited for a red light at an intersection. It was only a few seconds before the other driver got anxious and rushed past just as a car came at high speed. “The person and the car both flew in the air, and he died on the spot." Wei Lai said that when he saw his colleague lying in the middle of the road he did not stop. "The delivery in his hand would go over time." And just then, a new order came, and a familiar female voice sounded out —"Delivery. From XX to XX, please reply after the beep, and receive it."

Heavy Rain

According to the settings of the system, after the drivers reply "received,” it will start to operate.

In 2019, at the ArchSummit global architects summit, Wang Shengyao, a senior algorithm expert on the Meituan distribution technology team, introduced the basic operation of this intelligent system-

From the moment the customer places the order, the system starts to decide which driver to send to take the order based on the driver’s route, location, and direction. The order is usually dispatched in the batches of three or five orders. An order has two task points: fetching and delivering food. If a driver is doing  5 orders and has 10 task points, the system will complete the "ten thousand orders to ten thousand second-level solutions" in the 110,000 route planning possibilities. Plan out the optimal distribution plan.Two task points and food delivery. If a driver carries 5 orders and 10 task points, the system will map out the optimal delivery plan based on its "10,000 orders for 10,000 people at the speed of seconds” solution.

But in reality, if you want to crush this “optimum," a heavy rain is enough.

The drivers' attitudes towards rain vacillate greatly. They like rain because there will be more orders on rainy days, but if the rain falls too much, the system will easily "burst with orders,” and they are more likely to "get into trouble."

Gengzi, a Meituan driver rin Hunan, had a terrible rainy night. The rainstorm kept pouring down for a whole day, orders poured in frantically, and the system exploded. Every driver in the site carried more than a dozen orders at the same time, the boxes were full, and bags crowded the handlebar space. Gengzi remembered that he could only lightly lean his feet on the edge of the pedal, as he was keeping an eye on the delivery boxes wedged in between his thighs.

The road was too slippery. He fell down several times, then quickly got up and continued to deliver. It was not until 2:30 in the morning that he delivered all the orders on his hand. A few days later, he received the salary slip for the month, and the figure was actually much lower than usual. The reason is simple. On the day of the heavy rain, many of his orders were delivered late, so his salary was cut.

It is not only Gengzi who has been deducted from wages, but also the stationmaster of the distribution station.

"I’m someone who eats data." Meituan's distribution station director Jin Zhuangzhuang defined himself in this way. For a distribution station, the most important data include: orders received, late delivery rate, bad review rate, and complaint rate. Among them, the late delivery rate is the most important because it is the source of many bad reviews and complaints.

Generally, the late delivery rate of drivers should not be higher than 3%. If this is not reached, the rating of the distribution site will be lowered, and the unit price of the entire site will also drop. Everyone including the site manager, personnel, quality control, etc., or even managers connected to the site will be affected.

At the end of each year, the site will also face assessments by the Meituan and Ele.me platforms. The bottom 10% of the distribution stations in each region will face the risk of being eliminated.

Under this systematic evaluation system, the "late deliveries" bring not only the loss of income to the drivers, but also a second harm in the form of mental hurt.

"He will become a thorn on the team’s side." Sun Ping said, "Late deliveries are serious. It not only deducts a large amount of money, but it’s also closely tied to the issue of group honor. Since he is dragging everyone back, the distribution manager will find him, and then the section manager will seek him out, and then the district manager will seek him out. And then everyone will dislike him.

This will bring great mental pressure to the driver. Zhu Dahe, who fell on the road with the mala hotpot, told "People" that in his first few months as a driver, he spent every day in depression.

He came from a small place and was not familiar with the roads of Beijing, not to mention the huge amount of traffic and people on the road. He tremblingly abided by the rules, and was deducted for late deliveries every day. This makes him feel incompetent. "Isn’t it said that delivery staff can earn more than 10,000 RMB (a month)? Why am I so bad at it?" He said, "I thought I wasn’t cut out for being a delivery driver."

Later, as the electric bike rode more and more smoothly, and the road became more and more familiar, he transformed from a novice to a master on the road, and this sense of incompetence gradually disappeared. "Compared with delivering an order late, driving on the wrong side of the road is nothing." He said that he can even experience a sense of "flow" when driving on the wrong side of the road with his colleagues.

Nowadays, under normal circumstances, Zhu Dahe rarely has late deliveries, but extremely bad weather is still a spell he cannot escape. At this time, the out-of-control system will also rope him in --- carrying an excessive amount of deliveries, completely losing control of the delivery times, facing late delivery penalties, and plus he cannot ask for leave

In August 2019, Typhoon Lekima struck Shanghai. An Ele.me driver accidentally got electrocuted while delivering goods in the rain. Then, a screenshot of the WeChat group chat of the delivery station was uploaded to social networks. In the screenshot, the distribution manager wrote @人民 (all): "No leave for the next three days... If I can’t find you in the next three days then you will face double the penalty for absenteeism. Reply when you receive this message." Under the manager’s message, a long list of drivers responded “1” -- which represents they received the message.

This screenshot caused a huge public controversy. Some netizens asked why could Hema, KFC, and McDonald's all suspend delivery during the typhoon, but the delivery platform could not?

In this regard, Meituan station manager Jin Zhuangzhuang can only express helplessness. Every time it rained heavily, the drivers would come to him for leave, a flat tire, a fall, and family trouble, for various reasons. But in the face of a large influx of orders, for the sake of the site’s performance metrics, he had to forcefully stipulate, "Except for birth, old age, sickness, or death, you cannot ask for leave in bad weather, and you will be fined if you ask for leave.

In heavy rain, when Jin Zhuangzhuang was the most tired, he had to sit in front of the computer on the site and constantly monitor the position of each driver, the amount of orders they carried, and the time of delivery. For his site, Meituan stipulates that each driver can only receive 12 orders at a time. If the number exceeds 12, the system will stop dispatching orders. However, in severe weather or major holidays, this limit makes it impossible for drivers to carry the huge orders that flood in. At this time, the system is most likely to collapse: some drivers carry double the orders, and some drivers get almost no orders; some drivers get orders in the complete opposite direction; the delivery time of the closer order is longer than the one far away...

At this time, Jin Zhuangzhuang needs to play another role --- "manual scheduler." Under this status, he can enter the system and transfer driver A's order to driver B in order to balance the capacity. Although the system is capped at 12 orders, manual rescheduling is not restricted. As long as someone is manually controlling the system, the number of orders in the hands of the driver can become "a very scary number" -- the most was when a driver carried 26 orders at the same time; one delivery station with a little more than 30 drivers once completed 1,000 orders within 3 hours; another driver was allocated 16 orders  at once when running orders in a county with a population of 500,000 during peak hours.

An Ele.me station manager told People that this kind of manual intervention is not for rescuing the drivers, but to "excavate the potential and speed of each driver to the maximum."

The driver’s potential has been tapped to the extreme. If it still doesn’t work, Jin Zhuangzhuang will go out and deliver the orders himself, the most he carried was 15 once. "(When the orders explode), I first let the drivers carry on for a while. If they can’t go on any more, I appeal to Meituan to reduce the scope of deliveries. After 2018, our site no longer allows these appeals. No matter how many orders there are, we will have to deliver them." He said that when one finally finishes deliveries during a period when orders explode, your entire body feels numb, as you’ve been completely running on instinct -- “there is no human emotional response."

Last year, Jin Zhuangzhuang left the business because his family was sick. He said he would not come back again. Recently, a friend wanted to take over a distribution site, but Jin Zhuangzhuang dissuaded him, "This industry gives people a sense of time pressure and pressure to meet metrics, which you can't imagine." This summer there were heavy rains in southern China, and while Jin Zhuangzhuang was thankful that he had escaped, he was also worried. He didn't know how many sites had explosions of orders, or how many drivers had to desperately rescue their performance metrics.

Navigation

In order to complete her subject research, Sun Ping contacted nearly one hundred delivery drivers in the past four years, many of whom have complained about the delivery route given by the system.

In order to allow drivers to focus more on food delivery, this intelligent system will replace the human brain as much as possible — helping drivers plan the order of picking up and delivering food for multiple orders, and provide food delivery route navigation for each order, so drivers do not need to rack their own brains and can follow the prompts of the system to complete orders, and at the same time bear the risk of being led the "wrong way."

Sometimes, the navigation will show a straight line. A driver once said angrily to Sun Ping: “It (the algorithm) predicts the length of time based on the straight-line distance. But this is not the case for our food delivery, as we need to detour, and we have to wait for traffic lights... Yesterday, I delivered an order, and the system displayed five kilometers, but I ended up driving seven kilometers. The system treats us like helicopters, but we are not.”

Sometimes, the navigation will also include sections where drivers drive the wrong direction on the road.

In October 2019, Guizhou driver Xiaodao posted on Zhihu that Meituan had guided drivers to go on the wrong side of the street. In his communications with "People," he said that he had just been a driver for half a year and had encountered several navigations that guided him to go the wrong direction on the road. One of them was to deliver a meal to a hospital. Normal driving required a U-turn, while the route on Meituan’s navigation system was to cross over and drive on the other side of the road. According to the screenshots he provided, this segment extended close to 2 kilometers.

"Some are even more intimidating," said Xiaodao. "Some places are not convenient for going against traffic flow. If there are overpass bridges, the system navigation will have you drive over the overpasses, including those that do not allow electric vehicles to go up. It will also have you go straight through a wall."

In Beijing, a short video content creator, Cao Dao also encountered the same situation. To gain experience in the profession, she worked as a Meituan driver for less than a week. What surprised her was that when she took the order, the system navigation actually displayed as if it were a walking route -- there was no difference between walking forward or backward on a particular route, but the delivery system calculates time based on the shortest route, which contains a large number of segments which involve going against the traffic.

From Xiaodao's point of view, whether it is the straight-line distance or the wrong-direction distance the purpose of the system has been achieved — the system will pay delivery fees according to the distance and time calculated by the navigation model. With shorter distances and quicker delivery times, the delivery platform can keep more users, and also hold down delivery costs.

At the end of 2017, the Meituan technical team also mentioned "costs" in an article introducing the optimization and upgrading of the intelligent distribution system. The article pointed out that the optimization algorithm has reduced the platform's capacity loss by 19%. The meal that required 5 drivers to deliver in the past can now be delivered by 4 drivers. Finally, the word "costs" appeared in the conclusion of the article: "Efficiency, experience, and cost will become the core indicators pursued by the platform."

In fact, Meituan has also gained tremendous benefits.

According to data released by Meituan, in the third quarter of 2019, Meituan’s delivery orders reached 2.5 billion, and revenue per order increased by RMB 0.04 compared to the same period in 2018. At the same time, cost per order was reduced by 0.12 RMB compared to the same period --- This also helped Meituan earn a full 400 million RMB more in Q3 2019 (than it did in Q3 2018).

It’s just that in the background of the platform’s huge profits is the decrease on drivers’ personal income. Xiaodao said that whenever there is a wrong-side segment in the system navigation, he will face a dilemma of no choice. Either give up on wrong-side driving and face the risk of delay by taking a longer path, or risk safety to follow the navigation, but matter which choice, "There is really less money."

"Every driver must balance safety and income." As an "outsider" who temporarily entered into this system, Cao pointed out the plight of the drivers. "All delivery platforms are chasing profit maximization. In the end, they have passed on all the risks to the drivers who have the least bargaining power."

In the exchange with "People," several drivers all expressed the same sentiment, "They don't worry that no one will do delivery. If you don't do it, others will."

Before becoming a Meituan driver, A’Fei was a KFC delivery worker. "At most, one person would deliver 600 or 700 orders a month. Because of store limits, KFC can give the delivery company a high unit price of 12 or 13 RMB (order). Therefore, the delivery fee staff held at 9 RMB and has not changed." He used the phrase "the most standardized" to describe the job, but the income is not high, "At most you can earn a little over 5,000 RMB a month." In the end, inspired by the "over 10,000 RMB income of delivery drivers" (narrative), he decided to leave KFC to do food delivery.

In Meituan and Ele.me, drivers are divided into two categories-special delivery and crowdsourcing.

"Professional Deliverer" (专送) is a full-time driver attached to the distribution station, with a basic salary, prescribed work hours, and receive dispatch orders from the system, which also evaluates their on-time rate. "Crowdsourced deliverers" (众包) are part-time drivers with extremely low barriers to entry. One person, one car, and one app can be employed immediately after registration. They have no basic salary, can freely take orders, and can refuse system dispatches, but if they refuse too many items they can be restricted from taking orders. Crowdsourced drivers are not affected by bad reviews and complaints, but will face heavier penalties for delays. One second of going over time will directly deduct half the delivery fee. Regardless of whether you are a professional deliverer or a crowdsourced deliverer, no one has a labor employment relationship with the delivery platform.

A Fei finally chose to join Meituan and became a crowdsourced driver. That was around 2017. He worked about 9 hours a day, specializing in running long-distance delivery, and earned about 10,000 per month, and earned 15,000 in the best month — low threshold (for entry), high incomes — This is considered an important reason why delivery platforms "are not afraid of nobody coming to work".

In the eyes of sociology scholars, "over 10,000 in delivery income" is nothing more than an “exceptional situation" in the initial stage of the platform. After a long-term investigation of the labor process of couriers and deliverers in Wuhan, Zheng Guanghuai’s team at the Central China Normal University’s school of sociology found that with the end of platform subsidies and more and more drivers joining, "over 10,000 in delivery income"  is becoming an illusory dream.

The research report released by the team showed that: only 2.15% of delivery drivers had a monthly income of more than 10,000 RMB, while 53.18% of the interviewees reported that their current income could not meet family expenses.

A’Fei told "People" that after delivering orders in Beijing for a period of time, due to personal reasons, he went to Chongqing and his income dropped. Especially after the epidemic, more and more people joined, and it was even difficult for him to get orders. Sometimes the monthly income is less than 7000 RMB.

According to the "Report on the Employment of Meituan drivers during Epidemic Periods in 2019 and 2020" issued by the Meituan Research Institute, during the epidemic, the number of newly registered delivery drivers on the Meituan platform reached 336,000. Among the sources of new drivers, factory workers ranked first, followed by salespeople.

As for "when do you make the most money now", A Fei's answer is, "Only when it is very cold and very hot." Because, at such times, "most people are not willing to go out."

****You can read the rest of the sections (that I’ve translated so far) in the Google Doc with some annotations: Semi-Finished Translation

Thank you for reading and engaging.

These are Jeff Ding's (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is a PhD candidate in International Relations at the University of Oxford and a researcher at the Center for the Governance of AI at Oxford’s Future of Humanity Institute.

Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).

Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99