Will China Become Global AI Trainer?
China is becoming a global factory for artificial intelligence (AI). The country has proposed becoming a world leader in this field by 2030. With the introduction of robots and machine learning into production processes, China will be able to advance in the global value chain.
However, to adjust the automatic training, the data must be entered manually and it’s a monotonous process that requires low-cost labor.
Christine Lagarde, Managing Director of the IMF, has made a frightening forecast at the Bloomberg Global Business forum in New York. An IMF study says that in the near future 26 million jobs will be replaced by artificial intelligence. According to Lagarde, the most vulnerable group is women as they are mainly engaged in areas requiring monotonous and precise work.
However, the head of China’s largest technology company, Baidu, which is involved in AI development, questioned the IMF director’s remarks. He believes that the development of artificial intelligence won’t lead to unemployment. According to him, it doesn’t matter that certain jobs may disappear since there will be new ones to which people will be able to adapt quickly.
One new profession that’s already being actively learned around the world, and especially in China, is that of data labeling. AI development is based on three main elements: hardware, software, and data. So, we need advanced chips and software that supports the work of artificial intelligence. AI needs data to learn: the more information is added to the system, the more “intelligent” it’s going to be.
However, you shouldn’t train the system uncontrollably and feed it with a large amount of data. This has been proved by the sad experience of Facebook and Microsoft, which were forced to turn off their AI systems due to the failure of chatbots.
In the case of Facebook, the bots began talking to each other in an unknown language. As for Microsoft, their bots turned into Nazi supporters and started using foul language. It’s all because the machines were in a way trained “in the combat conditions” of Twitter, where the subjects under discussion were often of a xenophobic and racist character.
It is obvious that at the data entry stage, a person must strictly control the content. In addition, the data for AI training must be labeled. For example, to teach a machine to recognize people in photos you not only need to upload a variety of photos to the system but also mark the people in each photo. It is not a difficult process, but monotonous and slow one. And the Chinese have started it.
The factories that 20 years ago were used to manufacture clothes are now being used for data labeling. For the most part, the workers are women who for low wages spend 12 hours a day in front of the computer screen and tag huge amounts of data for AI training.
According to the South China Morning Post, the quality supervisor of the Chinese Basic Finder company earns between 3,000 and 4,000 yuan per month which is less than $450. Labelers earn even less, but there are lots of people who want to work there because it’s much easier than sewing or working in the field.
As reported by Liu Xingliang, a Chinese expert in Internet technology, it seems that China is creating a new niche by becoming a global factory for the 21st century, laying all the groundwork to create artificial intelligence.
China also has an advantage in labor that can process this data. Therefore, while China may not be the leader in the field of advanced innovations, it may easily become a global AI trainer.
The services of local data labeling companies are in demand not only in China. One-third of Basic Finder’s orders are from foreign companies, including the University of California-Berkeley, California’s Auto X which is engaged in the production of unmanned vehicles, as well China’s SenseTime and iFlyteK. In addition, Beijing’s Mada Code, which employs about 10,000 freelancers, processes data for Microsoft and Carnegie Mellon University.