The bitterness and helplessness of the self-driving marker: 2 Mao Qian pulls a box with a monthly salary of 3000 yuan
CTOnews.com, May 15, Autopilot technology is one of the most important applications in the field of artificial intelligence, which requires a lot of data to train and optimize. And the source of these data is the autopilot data marking company. These companies employ thousands of people in third-and fourth-tier cities to provide Party A with pictures with data frames, according to Xuebao Finance and Economic Society. After these pictures are transmitted to Party A through the system, they are used to train autopilot skills to make it more like a natural person driver.
However, these companies and employees who provide "water" for the self-driving industry do not enjoy the dividends of the development of the industry. On the contrary, they are faced with the dilemma of low income, high pressure and no development. According to McKinsey, the total sales of self-driving cars in China will reach about $230 billion by 2030 and about $360 billion by 2040. According to the observation of one practitioner, the salary range of workers in this industry is generally between 3,000 and 6,000 yuan. "if you have just entered the function, you can get 3000, if you are proficient in the work, you can get 4000, and if you do a management post or quality inspection, you can get 6000."
Autopilot data labeling is a low threshold and repetitive work. Employees need to use the mouse to frame the picture and mark the type of car (SUV, car or truck). Each time it is finished, the income is about 20 cents. The labeled images can be divided into 2D and 3D, 2D is like ordinary photos, 3D is "spatial point cloud data". With the popularity of lidar, there will be more and more 3D pictures from 2021, according to an executive at an autopilot data tagging company.
Although the work is simple, it is very hard. In the work of autopilot data markers, overtime, staying up late and even staying up all night are common. An executive at an autopilot data tagging company says his company does not recruit employees over the age of 26.
CTOnews.com notes that suppliers of self-driving data tags also face price competition and cost pressures. It is reported that from 2016 to 2018, the unit price in this industry is quite high, and companies that do well can earn nine figures a year. However, with the influx of more teams, Party A has more and more choices, industry prices have become increasingly transparent, price comparison, mutual price reduction is a common thing. Now the unit price of pulling a box has dropped by 40% to 50% compared with 2016, and there will be a certain degree of decline every year. "Party A puts more and more emphasis on reducing cost and increasing efficiency, and wants to pull more boxes with less money."
In order to reduce costs, some practitioners have also tried to use intelligent tools to improve the efficiency and quality of data tagging, but this requires a lot of investment and research and development. Before AI completely replaced the manual landing, "now it is better to find employees with lower wages to reduce costs."
