A Silent Silencing Industry: Unveiling the Labor Market of Chinese Censorship

• In Event: Political Regimes and Economic Policies

Abstract

The academic discussions about content moderation and censorship, in both democratic and non-democratic countries, often focus on the technical aspects – algorithms, effectiveness, and implications. However, this attention overlooks the significant role of numerous human labors and related industries in these information control practices, which had been occasionally highlighted in media coverages. Particularly in China, a country with the most sophisticated information control system, censorship work is delegated to private sectors which hired a large number of humans, and the state-owned companies are also known to have provided the censorship service to internet platforms. Our work aims to redirect the focus towards this critical yet understudied aspect and provide valuable insights about human-powered censorship industry and the market dynamics of Chinese censorship.

To fill this knowledge gap about human elements in the censorship, we started from nearly the universe of all job ads posted on several major job seeking websites in China from 2015 to 2022. We constructed a dataset of 10,000 job ads by random oversampling for human coding, where each job ad could be labeled for containing full-time, part-time or no censorship responsibility. Then, we developed a neural network classification model based on these human-labeled training dataset after manual cross-checks and identified around 560,000 censorship-related job ads with high accuracy. We obtained all public information of these job ads, including job titles, descriptions, expected salary ranges, companies, locations, and so on. In the meantime, we also leveraged human coding and GPT-4 collaboratively to identify the target social media platforms in the censorship job descriptions, which enabled us to extract the potential outsourcing relations. In addition, we collected other datasets like all job ads from censorship-hiring companies, official company registration data and the announcements from Cyberspace Administration of China.

This work presents several new insights about censorship in China. First, there is a wide range of censorship-related works in different types of business, beyond just social media companies. Between 2015 and 2022, more than 1,730,000 censorship-related workers were demanded in China, with the peak in 2019. Second, there was a specialization transition documented in the early years of our data. Censorship was mostly taken as a part-time task for occupations like editors in the beginning, but dramatically shifted to numerous full-time jobs in around 2018. Third, there was a geographical redistribution of censorship works from major cities to less developed inland regions, creating several censorship hubs for tech companies. Fourth, most censorship works are low-skill and labor-intensive jobs, and not more special, or being paid more or less, than other similar non-censorship jobs, but some special talents like language skills could be rewarded. Fifth, it is ubiquitous for tech companies with strong censorship demands to outsource their censorship works to human resource companies, while the rationale behind this choice remains uncertain. Sixth, the recruitment of censorship-related works on company levels is not a significant reaction to the political sanctions from governmental agencies, but more of a reaction to the market and product needs.

This paper explores the labor market of human censors and the dynamics between different entities in this game. We argue that the specialization of censorship works not only strengthen the capacity of state-driven information controls but also reshapes the landscape of information controls in the social media age.

Authors

• Andy Zhao, Cornell University

• David Y. Yang, Harvard University

• Leo Yang, University of California San Diego

• Eddie Yang, University of California, San Diego