Chao News Written by Xie Danying.
In March, in the misty spring rain, the mountain roads on the high slopes of the loess were foggy and the visibility was extremely low. The local driver is accustomed to it, skillfully turning on the double jump lights, and still driving the car fast.
Looking out of the window, you can only see the red glow from the front taillights. Photo by Chao News reporter Xie Danying.
Different from the speeding car, the destination of the reporter's trip, Yonghe County, Linfen City, Shanxi Province, has been "walking" very slowly.
Here, the area is not large, since the West Hanzhi County, Sui changed its name to "Yonghe" and has a long history. According to public data, Yonghe has a registered population of less than 50,000. To date, there has been no train service. After the official opening of the expressway at the end of 2016, it will take at least two and a half hours to drive from Yonghe to Linfen.
It is hard to imagine that in this county with poor arable land, blocked transportation, and weak industry, AI (artificial intelligence) has opened a new window for the local area. Since August 3, 2020, more than 100 women in the county have joined Yonghe County Idol Technology *** to engage in the work of data annotators. A computer in hand becomes a part of the digital economy.
Data annotator is a new profession under the development of artificial intelligence. Intelligent training software is usually used to carry out database management, human-computer interaction design, performance test tracking, and other auxiliary operations in the actual use of AI products.
In the past four years, these female workers in remote counties who are engaged in data labeling have undergone earth-shaking changes in appearance, clothing, speech, thoughts, and even family status.
Group photo of data annotators at the 2023 World Artificial Intelligence Conference Courtesy of the interviewee.
Marking. In the county, the most stable structure is the home, and the most central part is also the home. In the past, the greatest expectation and praise for women in Yonghe was to have children and be a good wife and mother.
And now, at the end of the main street, in a low building marked "Yonghe Talent Base", it is full of women with all kinds of exquisite makeup, facing desktop computers, with straight backs and skillful hands. The large room was quiet, except for the sound of rapid keyboard taps and mouse clicks.
Yonghe County Idol Technology *** data annotators are working. Photo courtesy of the interviewee.
They are engaged in the profession of data annotators, also known as AI trainers, who annotate text, images, and voices by labeling, framing, sorting, etc., feeding AI and making AI smarter.
As for what exactly will be used for data labeling, even Wang Lina, the project manager, can't say, "Some of them have done too much, and they can guess a rough idea", such as making the map more accurate and the vending machine more accurate.
This company, which seems "abrupt" in the county, originated from the "AI Bean Plan" digital industry poverty alleviation project launched by Ant Group, Zhejiang Ant Public Welfare Association, and China Women's Foundation in 2019. Under the coordination of the National Health Commission, Yonghe became the second batch of pilots of the project in underdeveloped counties. In the second year of the project, the data annotation company was launched.
The main force is Bao Ma. Feng Qin, one of the first employees, was one of them. She is in her early 40s this year and has a good figure. When she used her hands to draw frames on the screen and explain the steps of labeling, her slender fingers and bright nails all attracted the attention of reporters.
*, automobiles, fashion. Feng Qin is like a treasure, and in every gesture, he does not have the slightest sense of isolation in the underdeveloped areas of the west. The temperament of a "working woman" and a "mother of two daughters" is skillfully integrated in her.
If you are willing to work, it is not a problem to earn five or six thousand a month, and it is no worse than working outside. Feng Qin didn't expect that in Yonghe, the salary could reach this number - there are few basic salaries in the data labeling industry, and piecework wages are the mainstay. She remembers that she was not proficient in business in the first month, and her income exceeded 3,000 yuan. ”
Early morning in Yonghe County. Photo by Chao News reporter Xie Danying.
The economic basis determines the status of the family. Wang Lina found that in the past four years of work, the status of their husband and wife has gradually equalized. Her husband, who has never touched her hands in housework, will now help take care of the baby and cook when she is busy with work. The three-year epidemic has made many female data annotators who can work from home become the breadwinners of small families for the first time. "There are fewer quarrels, there is no idle time, and it is important to make money. ”
In their free time, they spend "getting beautiful" – tattooing their eyebrows, dyeing their hair, getting their nails done, and buying new clothes. There is no commercial complex in the county, so I made an appointment with three or five little sisters and drove to Linfen on weekends or holidays.
Sometimes a full load returns, and a month's salary is cleared. Feng Qin said frankly that having this job made them dare to spend money, "at most a few days of overtime to earn the money back." ”
Unconsciously, all kinds of data annotation projects have built up their broad imagination of the world - Wang Lanlan, who used to do tourism-related annotation business, has searched for a lot of information about West Lake, Wuzhen and other scenic spots. "It's beautiful. Sitting in front of the computer screen, she seemed to see the appearance of Zhejiang 1,400 kilometers away, "I want to go there when I have the opportunity in the future!" ”
Transfiguration. Because of this job, in April 2023, Wang Lina and Feng Qin, as representatives of the Yonghe project of the "AI Bean Project", were invited to participate in Ant Group's first "Digital Mulan" Women's Development Annual Conference and flew to Hangzhou, where they "learned a lot of new things, met a lot of new friends, and played a lot of places".
It is precisely because of this job that they usually mark data behind the scenes, and in July 2023, they rushed to the Shanghai World Expo Exhibition and Convention Center to have a glimpse of the grand occasion of the World Artificial Intelligence Conference. "Before, I only knew I was working for AI. This time, I finally got to know what the latest big models are doing. The opportunity was not easy to come by, at that time, Wang Lina was 7 months pregnant, with a big belly, and she was not willing to be absent.
At the 2023 World Artificial Intelligence Conference, Wang Lina (first from left) and Feng Qin (middle) are listening to the introduction of the relevant person in charge Photo provided by the interviewee.
Everything is something that the previous Yonghe women couldn't think about.
According to the locals, the main coal-producing areas, iron ore resource areas and wheat-producing areas in Linfen, Shanxi Province, all "pass by" Yonghe. It wasn't until 2015 that natural gas was detected in Yonghe, and several gas stations became the main employers of the local male labor force.
Walking on Yonghe Street, in addition to the elderly, most of the figures in the county town are women and children. "Yonghe has the most common family division of labor: men go out to drive big trucks and earn seven or eight thousand a month; The woman stays behind to take care of the children. Li Linfeng, the person in charge of Yonghe County Idol Technology, said.
Leaving work and becoming a housewife is the epitome of most of Yonghe's female roles. Many female data annotators told reporters that even if there are elderly people at home to help take care of the children, and women have time, the county itself is too small and it is not easy to find a job; There are a few job opportunities such as supermarket clerks and restaurant waiters, and the monthly income does not exceed 3,000 yuan.
used to rely on her husband to work outside to earn money, and after sending her children to school, her life was either playing cards or shortening**. Wang Lina said frankly that in those years, some chess and card rooms in the county even opened directly next to the school, targeting these mother groups.
Yonghe girl ** Dan, talking about her life before becoming a data annotator, especially after giving birth, her eyes are wet, "She doesn't do anything all day long!" A sentence from her husband's friend made her feel like a fish in her throat, cooking, washing clothes, and breastfeeding her children every day, "Why do you say I didn't do anything?" ”
It is still that small county town, which is said to be much more "prosperous" than it was four or five years ago. On the main street, there is a Honey Snow Ice City, and every weekend, the nearby elementary and middle school students can buy all the ice cream in the store. When the reporter arrived, the red lanterns wrapped around the street trees on both sides of the main street had not yet been taken off, and the stars were red, revealing the joy and vigor of the small town.
The street trees have not yet grown green shoots, and the red lanterns of the Chinese New Year are hung high on the trees. Photo by Chao News reporter Xie Danying.
For migrant county women, they are even more difficult in terms of identity, facing the embarrassment of 'not being able to enter the city and not being able to return to their hometown'. Through digital employment, they have achieved a new type of urbanization that combines work and life. The relevant person in charge of Ant Group said.
It is reported that Yonghe County Idol Technology is a county-owned state-owned enterprise, jointly supported by the National Health Commission, Yonghe County, Human Resources and Social Security Bureau, and Ant Group, with 110 employees, and is currently the largest employment enterprise in Yonghe County, with an average monthly income of more than 4,000 yuan. The company's employees are 90% female, and more than 60% are returnees from other places in the past two years.
Ahead. Since 2023, ChatGPT has been born, AI has received unprecedented attention, and the entire industry has begun to speed up. But the female workers on the Loess Plateau are still living in their own rhythm, nervous and comfortable.
Data labeling has not been an industry for a long time, with its beginnings dating back to 2012. It took 8 years from a new concept to be officially listed as an emerging profession by the state in 2020.
At the beginning, the quality of data labeling work is not high, and the project needs can be realized by repeated framing.
In recent years, the development of autonomous driving has driven the market for data labeling. According to Deloitte's report, the demand for labeling in the field of autonomous driving accounted for 38% of the entire AI downstream application in 2022, and it is expected to account for 52% by 2027.
Intelligent driving Source: Visual China.
Autonomous driving requires a high level of data annotation because it requires nearly 100% accuracy. Industry insiders admit that most artificial intelligence products have requirements for model accuracy higher than 90%, but when the accuracy wants to be increased from 90% to 95%, or from 95% to a little higher, the requirements for the amount of data behind it may be millions or even tens of millions. "The higher the accuracy requirements, the more data is needed, which means that the number of data annotations is also doubled. ”
The rise of large models this year has added another fire to the data labeling industry. A large number of orders based on large model training scenarios flew towards data labeling companies, injecting vitality into the boring business of data labeling again.
As a result, some technology companies have gone to the forefront and tried to use AI to automatically synthesize data for AI training. Synthetic data is based on a small amount of real data, which is infinitely generated by AI and does not need to be labeled, and it no longer relies on manual annotation.
In their vision, synthetic data will replace manual annotation in the future.
Labeling companies that do not have the technology and rely only on manpower will be phased out. According to a data, 70% of the basic data used for artificial intelligence abroad is synthetic data, and this path is being verified.
*Dan is doing annotation work Courtesy of the interviewee.
Li Linfeng told reporters that the company has not been affected too much at present, because the company itself originated from poverty alleviation projects and has a public welfare nature. Over the past four years, the business orders received have been relatively stable, mostly from within Ant Group, or the group acts as a hub to bring in business from other companies for these small county-level enterprises.
Overall, our business volume is increasing, but at the same time, the difficulty of labeling business is also increasing. In Li Linfeng's view, the 200-person labeling company is the ceiling of Yonghe County. At present, there has begun to be a demand for highly educated or professional employees, such as young people majoring in finance and medical care, except for a very few, not all female workers have the ability and desire for self-improvement.
Why can't I find a direct flight from Linfen to Hangzhou on some ticketing software? The reporter's journey has aroused heated discussions among several data annotators. They have recently been in charge of a cultural tourism project, which makes navigation smarter, maps more accurate, and recommendations smarter through annotation. After repeated searches, we found that we need to accurately search for "Yaodu" in order to book a direct flight once a day, "It seems that the accuracy of map AI recommendations needs to be improved, and the projects we are doing need to be further promoted!" ”
The words on the computer screen, ** flashed one by one, and the interval was calculated in seconds, which dazzled others.
*Please indicate the source".