Chinese internet censoring

Hi, Habr! I present to your attention the translation of an article with nytimes.

Thousands of low-paid workers in the so-called “censorship factories” are combing the online world for prohibited content, where even a photograph of an empty chair can cause big problems.

Li Chengzhi had a lot to learn when he first got a job as a professional censor. Like many young people in China , the 24-year-old college graduate knew little about the events on Tiananmen Square in 1989 . He had never heard of the most famous Chinese dissident, Liu Xiaobo , the Nobel Peace Prize winner, who died in custody two years ago.

Now, after a certain period of training, he knows what to look for and what to block. He spends hours browsing online content for Chinese media companies looking for something that could incite government anger. He knows how to identify code words that indirectly relate to Chinese leaders and scandals, or memes that touch on topics that the Chinese government would not mind keeping silent about.

Mr. Lee, who still has acne marks on his face, is serious about his work. “I help clean the online space,” he says.

For Chinese companies, it is very important to “be on a short foot” with government censors. In addition, the authorities demand that companies self-censor themselves, thus encouraging them to hire thousands of people to control content.

This, in turn, created a new growing and profitable industry: censorship factories.

Mr. Lee works for Beyondsoft , a Beijing-based technical services company, which, among other things, takes on the burden of censoring other companies. He works in an office in the city of Chengdu (Chengdu), located in the heart of a high-tech industrial area, in a fairly new and bright office that resembles the offices of well-funded start-ups in technology centers like Beijing and Shenzhen. They recently moved here because customers complained that their previous office was too small for their employees to do a good job.

"Just one small omission can lead to a serious mistake," said Yang Xiao, head of the Internet services and content analysis department at Beyondsoft. (Beyondsoft refuses to advertise which Chinese media or online companies work, citing a privacy policy).

China has created the most extensive and complex online censorship system in the world. This system has become even stronger under President Xi Jinping, who wants the Internet to play a more important role in strengthening the influence of the Communist Party. More and more content falls into the category of dubious , and the punishments are becoming more severe .

After a thorough study of the issues of control, China is now preaching the policies of the Internet controlled by the government. Even traditional strongholds of freedom of speech, like Western Europe and the United States , consider their own measures to limit the digital space. Platforms like Facebook and YouTube have said they will hire thousands of people to better control content. Workers such as Mr. Li demonstrate the extremes of this approach, which controls what more than 800 million Chinese Internet users see every day.

In Beyondsoft, content analysis factories employ more than 4,000 people (compared to about 200 in 2016). They browse and check content day and night.

“We are Foxconn in the data industry,” says Mr. Yang, comparing his firm with the largest contract manufacturer that stamps iPhones and other products for Apple.

Many online media companies have their own internal content analysis teams, the number of which sometimes amounts to thousands. They are learning how to introduce artificial intelligence to automate and speed up work. The head of the AI lab at a major online media company who chose to remain anonymous said the company had developed about 120 machine learning models.

But success is not constant. Users can easily fool algorithms. “The algorithms of artificial intelligence are smart enough, but not to the extent of the human brain,” said Lee. "They miss a lot of things when viewing content."

Beyondsoft has a team of 160 people in Chengdu, working four shifts a day, viewing potentially unacceptable content in a news aggregation application.

For the same application, Beyondsoft has another team in the western city of Xi'an (Xi'an), which filters potentially offensive and vulgar content. As in the rest of the world, the Internet in China is replete with pornography and other materials that many users would find inappropriate.

At the Chengdu office, workers must leave their smartphones in special lockers in the corridor. They are also forbidden to take screenshots or send any information from their computers.

Almost all employees are college graduates aged 20 and over. For the most part, they are not interested in politics, or simply indifferent to it. In China, many parents and teachers tell young people that worrying about politics leads to one trouble.

To address this issue, Mr. Yang and his colleagues have developed a comprehensive training system. New employees start with a weekly “theoretical” training, during which senior staff teach them how to work with confidential information that they could not have known before.

“My office is next to a large training room,” says Mr. Yang. "I often hear exclamations of surprise coming from there." “They sometimes don’t even know things like June 4,” he added, referring to the 1989 Tiananmen event.

Beyondsoft has developed an extensive database based on such information, which Mr. Yang calls one of his “core competencies”. The company also uses anti-censored software to regularly visit banned websites blocked by the Chinese government. After that, the database is updated. New employees study this database as thoroughly as they do when preparing for the college entrance exam. After two weeks of study, they pass the test.

The screensaver on each computer is the same: photos and names of current and former members of the Standing Committee of the Politburo , the top leadership of the Communist Party. Employees must remember these faces: only government websites and specially approved political blogs - the so-called “white list” - can post photos of top executives.

At the beginning of the shift, employees are informed about the latest censoring instructions sent by customers, who in turn receive them from government censors. Workers must then answer 10 questions designed to test their memory. The results of such checks affect their salary.

Here is an example of one of these questions: Which of the following names is the name of Li Peng’s daughter, Li Peng , the former Prime Minister of China? The correct answer is Li Xiaolin (Li Xiaolin) - an old target of online ridicule for her preference for expensive fashion items and for being one of many high-ranking children of high-ranking officials .

This is a relatively easy question. More difficult questions are the analysis of devious ways in which Chinese Internet users avoid harsh censorship, just to chat about everyday affairs.

Take, for example, the commentary of the Hong Kong 2017 news site, which compared six Chinese leaders from the time of Mao Zedong and emperors from the time of the Han dynasty. Some Chinese users began to use the names of emperors, speaking of political leaders. Employees of Beyondsoft should have known exactly which name of the emperor is associated with a certain name of the leader.

There were even pictures of an empty chair. These are references to Mr. Liu (Liu) , a Nobel Prize winner , who was not allowed to leave China to attend the awards ceremony, as a result of which he was represented by an empty chair. Links to George Orwell’s 1984 novel are also banned.

The Beyondsoft software scans the web pages and marks the potentially offensive words with different colors. Thus, if the page is full of color-labeled words, then you need to study it more carefully. Only one or a few tagged words are allowed.

According to the Beyondsoft website , the Rainbow Shield Content Monitoring Service has collected over 100,000 key sensitive words and more than three million of their derivatives. Politically incorrect words make up one third of the total, followed by words related to pornography, prostitution, gambling and knives.

Workers, like Mr. Lee, earn between $ 350 and $ 500 per month, which is the average wage in Chengdu. It is assumed that each employee views 1000 to 2000 articles during a shift. Articles in news viewing applications must be approved or rejected within an hour. Unlike Foxconn workers, they do not work overtime, because recycling can damage the accuracy of verification, says Mr. Yang, the executive head.

Making a mistake is pretty easy. An article about Peng Liyuan, the first lady of China, mistakenly used a photograph of a famous singer who was rumored to have been associated with another political leader. According to Mr. Yang, the error was found by someone else before the article was published.

Mr. Lee said that almost all of the worst mistakes were related to information about top managers. Once he missed a tiny photo of Mr. Xi on a non-whitelisted website, because very tired. He still blames himself for this misstep.

When asked if he had shared with family and friends what he had learned, for example, news about Tiananmen Square, Mr. Lee violently replied with a categorical "no." “This information is not intended for outsiders,” he says. “It can give rise to rumors as soon as people find out about it.”

But cruel persecutions are historical facts. It was not a rumor. How does he reconcile all this in his head? “For certain things,” he says, “you just need to follow the rules.”