On the Loess Plateau, women who have not gone to college are feeding AI

Author: Nanfengchuang Senior Reporter Zhu Qiuyu

Source: Nanfengchuang

When I saw Cao Yali, she was staring at the computer screen without distraction. The three fingers of her left hand covered the four keys of A, W, S, and D, and her right hand was placed on the up and down keys of the keyboard.

Her hands are well coordinated, and her back is straight, as if a pianist is flexibly pressing the keys.

Every two seconds, Cao Yali moved her finger, which was the time for her to make a judgment in front of a picture.

Cao Yali's profession - AI trainer (artificial intelligence trainer) - will be officially listed as an emerging profession by the country in 2020. Every day, she and her colleagues annotate text, images, and voices and feed AI by labeling, framing, sorting, and finding differences. As for what the data will be used for after labeling, Cao Yali can't tell.

Shaanxi Yulin Aidou Technology Co., Ltd. (hereinafter referred to as "Qingjian Aidou"), where Cao Yali works, was established in 2019 and currently has more than 180 data labelers. According to Yu Tao, the person in charge, 66% of them are women.

Shaanxi Yulin Aidou Technology Co., Ltd. / Photo by Zhu Qiuyu

A similar scene also happened in the Yonghe Digital Employment Center in Linfen City, Shanxi Province in Luliang Mountain. The latter is also a company that does AI data labeling. Among the 133 employees, 128 are women.

Among them, 80% of them have another identity: treasure mother.

Many of these women have high school or technical secondary school education, and have never gone to college. After staying in their hometown to take root, they became wives and mothers in other people's mouths.

When we met in June, several employees recalled a scene to me - as a full-time mother in the past, there would always be times when they would "reach out for money" from their husbands. Such a life is not shameful, but it makes people feel shameless value.

Today, the new digital industry built on the Loess Plateau has given the left-behind women dignity and hope. In the interaction with artificial intelligence, science and technology endowed them with magic that can eliminate the innate gap, and also radiated new vitality to the people on the Loess Plateau.

mouse point, all money

Among the three brothers and sisters, Cao Yali thinks that "I am the worst". She is the second child in the family, her elder brother works in a state-owned enterprise in Inner Mongolia, and her younger sister is a teacher. There is only her, who did not go to college, and has been doing post-production in a photo studio for many years.

In 2019, due to family changes, she returned to her parents' home in Qingjian County from Xi'an.

Cao Yali was born and raised in Qingjian County. This county, which was lifted out of poverty in 2020, is the hometown of writer Lu Yao. The ravines made of loess surround the surrounding area, and only the long river valley impacted by the river gives people room to live. In people's memories, there is a large piece of wasteland in the county seat, "there are no buildings, but mountains".

Accompanying it is the increasingly hollowed out local industry.

Qingjian County, Shaanxi Province/Photo by Zhu Qiuyu

Qingjian County is the "Hometown of Red Dates in China" and is famous throughout the country for planting the big dog's head dates. But in recent years, with climate change, it rains in the Loess Plateau in autumn. Mature jujubes are most afraid of rain, farmers often fail to harvest, and more and more farmland is abandoned.

As a result, the young people left their hometowns and went to Yulin, Xi'an and other places to make a living and look for hope.

Cao Yali, 30, is a retrograde.

After returning to her hometown, she couldn't find a job, and she was anxious about it for a year. In 2020, she became an AI data labeler in an office. This is a rare new job in a small county.

After only one month of training, Cao Yali will officially start working. For more than two years, she has been doing similar things: marking a certain map navigation.

She showed me a dashcam image with two timestamps above and below the screen. The data labeler needs to manually subtract the two times to judge the difference. "If it is less than 2 seconds, it is qualified, press A; if it is more than 2 seconds, it is unqualified, choose S; if you can't see clearly, choose D, and there is no time stamp W."

On February 16, 2023, in Qingjian County, Yulin City, Shaanxi Province, the staff were marking data. Data annotation is a part of developing machine learning models in the artificial intelligence industry. The Qingjian County Digital Employment Center has been established for more than three years. By training local women and returning young people as data labelers, it has helped rural women find employment locally/Photographed by China News Agency reporter Zhang Yuan

Cao Yali said that the difficulty lies not in subtraction, but in the face of massive pictures, the annotator must maintain focus and accuracy.

When Cao Yali first joined the company, she had to look at 13,000 pictures every day, which she couldn't finish within 8 hours of work. But now, she is proficient enough to complete the same workload in more than 6 hours; if she works hard to work overtime, "I can read 30,000 pages a day."

The above tasks are related to the popular "deep learning" in the artificial intelligence community. Since 2014, one of the mainstream development paths of deep learning is supervised learning. That is to say, if you want to make the machine intelligent, you must first feed a large amount of labeled data, so that it can automatically learn, make decisions, and complete specific tasks.

When AI performs algorithm iterations, it also needs to process and label a large amount of data. According to the statistics of Cognilytica, an AI analysis company, in AI projects, data-related processing accounts for more than 80% of the time. As a result, there is a widespread rumor in the industry that "how powerful artificial intelligence is, how powerful intelligence is."

The business of artificial intelligence has fallen from the tall Grade A office buildings to the Loess Plateau. Talk to the women in the county for a while, and you will feel their desire for this new business.

Working environment of Shaanxi Yulin Aidou Technology Co., Ltd./photo by Zhu Qiuyu

Cao Yali said that because she is paid on a piece-by-piece basis, she would get up at 5:30 every morning and go straight to the office without looking at her mobile phone. There is no lunch break at noon, and I go to work until 10:00 or 11:00 pm after eating.

Cao Yali is not the only one who loves her work. Even many female employees of Aidou Technology Co., Ltd. said that they wanted to work overtime. Sitting in front of me in a flowing pink dress, a mother of two looked me in the eyes and said firmly, "I just want to make more money."

After working for two years, she became a project manager of a certain business line, working overtime every day until the children's bedtime at night. When the monthly salary was the highest, she took more than 7,000 yuan.

"A click of the mouse is money," she said.

She repeated: "The mouse is money at all."

When she spoke, her eyes were bright.

Bao Ma's Rose

The wind of artificial intelligence has blown from Qingjian County to surrounding counties and cities. In August 2020, a new digital employment center was established in Yonghe County, Linfen, Shanxi Province, which is close to Qingjian, and artificial intelligence trainers are also recruited.

For this county with long-term traffic congestion, only two main roads, and most buildings with no more than 4 floors, artificial intelligence is a new thing that is out of place.

Li Linfeng, general manager of Yonghe County Aidou Technology Co., Ltd., told me about his unforgettable scene. In 2012, he went to university from Yonghe County to Linfen City, and the first highway leading to the county had just opened.

He excitedly shared with his classmates that there is a road in his hometown. The classmate was very surprised: "Highway, isn't this something that (we) have had since birth?"

"Others can't imagine that some places don't even have expressways." Li Linfeng said.

Street View of Qingjian County/Photo by Zhu Qiuyu

At the same time, Linfen City's main coal-producing areas, iron ore resource areas, and wheat-producing areas all pass by Yonghe County. Until 2015, natural gas was detected in Yonghe County, and several gas stations became labor companies that absorbed local male labor.

The backwardness of the economy makes people unstoppable outflow. Except for the elderly, most of the figures in the county are women and children. So Li Linfeng introduced: "We in Yonghe have the most common family division of labor: the men go out to drive large trucks and earn seven to eight thousand a month; the women stay behind and take care of the children."

This explains why when Yonghe County Aidou Technology Co., Ltd. first recruited in July 2020, most of the applicants were women.

Liu Xia, from Yonghe County, will become an artificial intelligence trainer in 2021 when her second baby just turned 1 year old. In the previous seven years, because of the birth of her child, Liu Xia experienced resignation, work, and resignation.

Her last job was at a gas station, where she worked two shifts. After working night shifts for more than a year, Liu Xia found that not only was she unable to take care of her children, but staying up late also caused wrinkles around her eyes. She, who loves beauty, felt the changes in herself very sensitively, so she simply resigned. "I even despise myself", but she also believes that this kind of life is like the doomed fate of women in the county.

Liu Huanmei, 34 years old, Qingjian, Shaanxi, she is an artificial intelligence trainer. This trendy-sounding job was brought to the local area by cooperation between Qingjian County and Alipay

Before getting married and having children in Yonghe County, Liu Xia sold clothes in a clothing store in Linfen City for six years. When she was a sales girl, she had just graduated from high school and her academic performance was at the top of her class. But among the four brothers and sisters, none of them went to college. "At that time, I thought it would be better to learn a technology and earn money early when I went to college." Her parents didn't care about her, so she set foot on the road of hard work in a foreign land alone.

Today, the only memory of working in a clothing store is "standing for a day, my legs are numb". But Liu Xia has a stubborn personality. Even though she was tired, thinking "Why can't I do it when others can", she continued to work in the clothing store.

Until she was 25 years old, the age that the county considered that women should get married. Under the urging of her relatives, Liu Xia felt that she had not "made a name for herself" in Linfen, so she returned to her hometown and entered into marriage.

After a year of marriage, like other women in her hometown, she quit her job because of her pregnancy and was out of work for three years. But she knew it in her heart: "I like to go out."

Since childhood, like her mother, she would rather go to the countryside to do farm work than do housework. She told me that in Yonghe County, there are many full-time housewives who think like her, and everyone hopes to have a career of their own. However, the county has long only recruited people from the service industry. It is unrealistic to work 7 days a week and take care of the baby.

People of Qingjian County/Photo by Zhu Qiuyu

In 2020, the AI company opened in the county will clearly write "weekends and weekends" in the recruitment advertisement, which makes Liu Xia, who is confinement, brewing to go out to work.

What made her even more irresistible was that among her friends of the same age, several mothers shared their working status in AI companies in their circle of friends-"March 8th Women's Day, each of them has a rose."

This made Liu Xia very envious. In seven years of marriage, she has not received a gift from her husband.

In this way, the work of artificial intelligence that is friendly to women spreads in the county, attracting a large number of "baby mothers" to sign up.

Li Linfeng was also surprised to find that although gender was not deliberately emphasized, more than 95% of the idol companies in Yonghe County are women.

"We do what others give us, we don't choose." A Yonghe mother said. What they are more concerned about is that the per capita salary of 3000 to 4000 to be an annotator exceeds the level of the county's career editor.

One of the largest employers

The AI labeling industry makes people in the Loess Plateau feel like a spring breeze. But in the outside world, it is a highly competitive market.

According to the statistics of Qianzhan Industry Research Institute, Chinese data labeling companies emerged from 2014 and reached the peak of development in 2017. In that year, there were 9 related financing incidents. This number is also the peak for the next few years. Since 2018, the financing volume of companies related to AI data has begun to decline. By 2021, there will only be two related financings a year.

According to an industry insider, a major feature of the AI data labeling industry is labor-intensive. Therefore, there are a large number of small workshops and studios for labeling in China. They recruit people with low salaries and fight "price wars" with each other. For a long time, the data labeling industry was in a state of extensive volume price.

Comparison of production capacity between different labels/36 krypton

In contrast, Cao Yali, Liu Xia, etc. are not deeply aware of the competition in data labeling. When asked if they were worried about the company going bankrupt and losing their jobs, they shook their heads.

Previously, Qingjian County had tried many industrial poverty alleviation projects, such as pig raising, fruit industry, and animal husbandry. Yu Tao, general manager of Qingjian County Aidou Technology Co., Ltd., recalled: "Everything that people can think of in rural areas, our county has done it, and almost all of them have failed."

This statement was confirmed by a person close to the government. He recalled that Qingjian County has introduced fruit companies many times to process red dates, but without exception, the enterprises that came to Qingjian all ended in bankruptcy.

In 2019, it was Liu Qinghai, the deputy county magistrate of the National Health and Medical Commission in Qingjian County, who first proposed the development of artificial intelligence in small counties. In fact, in 2019, no one in Qingjian County understood what artificial intelligence was except for Liu Qinghai.

Yu Tao recalled that that year, he was still the vice president of the Qingjian County Urban Investment Company, and received a call from the county asking him to be the person in charge of the newly established company.

He asked the other end of the phone: "What does this (company) do?"

The other end of the phone replied: "We don't know either, only Secretary Liu knows."

Qingjian County, Shaanxi Province/Photo by Zhu Qiuyu

After contacting Liu Qinghai, Yu Tao understood the idea of developing artificial intelligence. "This is at least an asset-light project." In case of poor development, it will not be as tragic as the previous bankrupt enterprises. "It's easy to turn around when the boat is small."

In 2019, under the coordination and introduction of the National Health and Medical Commission, the digital industry incubation project "AI Dou Project" initiated by the China Women's Development Foundation, Ant Group, Ant Charity Foundation, etc., landed in Qingjian.

Idol’s first recruitment was in the summer of 2019, and the recruitment publicity was vigorous. The local Women’s Federation, Youth League Committee, Human Resources and Social Security Bureau, etc. all helped to post recruitment notices, and even the TV station advertised, but still failed to recruit people.

This profession is too unfamiliar to the people of the county. "In addition, there were some industries that went bankrupt before, and people are not willing to believe it." Yu Tao said.

Until 2020, affected by the epidemic, most people in the county will stay at home. At this time, Yu Tao and several management members sent computers to the employees' homes so that they could work from home.

The advantages of the digital economy are reflected at this time.

One month, an employee received a salary of more than 10,000 yuan, far exceeding the salary of about 2,000 yuan in the local service industry.

This quickly became explosive news in the small county. Since then, Qingjian Idol Company's reputation has grown, and there is an endless stream of applicants, and it soon becomes the largest employment company in the area.

Qingjian became a model. Liu Qinghai concluded in an interview with the media recently that the experience of Qingjian County shows that there is room for the digital economy and artificial intelligence to be transferred to the western region.

In the next three years, the digital employment centers of the "AI Dou Project" in Yonghe, Zizhou, Suide, and Yanchuan counties in the surrounding counties were successively launched, all with Qingjian Aidou as the parent company. According to statistics, these centers have absorbed more than 600 local women and youths who have returned to work.

There is no ceiling in the county seat

Digitalization has not stopped the innovation of small counties.

Li Aqian, an idol employee in Qingjian County, mentioned to me that during an interview, a reporter asked her: "Do you know ChatGPT?"

Li Aqian heard this word for the first time, "At that time, I felt that I was backward, and there was still a big gap between here and outside."

Jia Yuhang, general manager of Cloud Measurement Data, also told Nanfengchuang that with the development of ChatGPT and large models, the types and contents of data annotation are becoming more and more complex. "In the earliest face recognition, you only need to make a frame mark on the face to complete the corresponding training." Jia Yuhang gave an example: "Now, it is also required to recognize the key points of the face, expressions or some people. The attribute or posture of the face, such as when half of the face is blocked, is marked."

Early face recognition is different from the content that needs to be trained for current face recognition

Yu Tao also found that the large-scale model frenzy triggered by ChatGPT and others has had a great impact on the data labeling industry, and the difficulty of labeling is also increasing step by step. "Some of our junior high school graduates have been replaced by AI. We have no tasks for them."

What we are facing now is still the problem of four years ago-difficulty in recruiting people.

Yu Tao said: "A labeling company with 200 people is already the ceiling of our county." At present, what is needed are people with high education or professional background, such as young people majoring in finance and medical care. Most of these talents go out to work, or go back to the county for examinations.

Compared to Yu Tao, Li Linfeng was less anxious.

In his view, data labeling companies rooted in the Loess Plateau are a more special and rare practice. Previously, the Yonghe County Government had tried to support industries such as e-commerce live broadcasting, but failed to achieve similar results. The reason is that "e-commerce has not provided so many jobs."

In Qingjian County, new buildings are rising from the ground/photo by Zhu Qiuyu

He believes that the advantage of the data labeling industry in the development of the Loess Plateau is precisely that it is a labor-intensive industry, which helps to absorb women's employment.

According to reports, the idol company will use more than 90% of its revenue for employee salaries and benefits. 59% of the employees are former key monitoring populations for poverty alleviation, subsistence allowances, and disabilities.

Only when there are industries and jobs can it be possible to stimulate the return of talents and give full play to the power of women. Taking Yonghe as an example, when the AI company started in 2020, there was only one college student among its employees. Since its establishment more than two years ago, the proportion of employees with a college degree or above has exceeded 30%.

Nowadays, mothers are not only seen at home, vegetable markets, school gates and mahjong parlors. At work, at nail salons and at gyms, they wear makeup and keep their backs straight.

Yu Tao, who made the "Qingjian Model" widely known, has even greater ambitions.

If it is a ceiling for a county to recruit 200 people, then he hopes to "bloom more"-in the future, each surrounding county will set up a digital employment center. "In this way, we will be stronger. Whoever has a lot of data (needs to be labeled), we will swallow it in one gulp."

The pictures in this article are partly from Visual China, and partly from the Internet