Chinese social media platforms have been cracking down on content in the lead up to the 240+ pages of censorship guidance. This documentation is a deeply collaborative effort to ensure that any reference to sensitive or unknown incidents are removed. The Tiananmen Square massacre remains one of the greatest worries for the Chinese government. ByteDance, the parent company of TikTok and Douyin, has produced a highly-specialized army of censors. Their human experts do a beautiful job finding the best content, complemented with systems oriented around advanced artificial intelligence.
Difficult as it may be for censors, it’s even harder during the run-up to the anniversary of the Tiananmen Square massacre. They do this under extremely tight productivity quotas, required to review hundreds of posts a minute. It’s the collaboration between human labor and AI technology that makes this a truly sophisticated system. This system acts as a filter for any content that would incite dissent or otherwise disrupt the government’s story.
The Role of AI in Modern Censorship
AI systems have transformed the way content is surveilled and censored on Chinese social media. These technologies automatically sweep through every single post and flag anything that mentions the Tiananmen anniversary, including the most innocuous images. An algorithm specifically developed for detecting visual metaphors has been able to do just that. Most importantly, it can understand the iconic scene of one man standing up to four tanks.
This capability extends to even benign images. For example, just a straightforward arrangement of a single banana and four apples can trigger censorship filters. As the 2022 Douyin censors training manual makes clear with precise, illustrative examples. This is an indication of how far the Chinese government will go to suppress even the slightest mention of what happened in 1989.
“Even if you replace the tank man image with bananas and apples, the algorithm has learned the pattern.” – Dr. Lennon Chang, cyber risk expert
The consequences of these AI-driven censorship measures are deep. As Dr. Chang warns, “If misleading data continues to flow outward, it could influence the AI models the rest of the world relies on.” This has rightfully called into question the integrity of information coming from outside of China. It illustrates the global stakes at play in insidious, localized practices of censorship.
Human Moderation and the Censorship Season
We’ve seen in China too, despite all the technological AI advances, the human censors are introductory key Where else content moderation in China. Workers on platforms such as Douyin undergo rigorous training. They are taught how to identify “subversive pictures,” especially ones related to the legendary Tank Man photo. They refer to the time between China’s cultural purge and the Tiananmen anniversary as a “censorship season.” During this short window, all-encompassing pressure builds not to get anything wrong.
It’s in this moment that censors need very clear parameters of what is and isn’t allowed. An internal memo advises them to “delete first. Ask questions second,” adding to the high pressure situation of their job. This is the season, kinder critics agree, where there is no room for error. Any mistake, big or small, will directly impact the censors’ reputations and judgment on these platforms.
A current worker at ByteDance described their experience, saying, “It’s like reliving the darkest pages of history every day while being watched by software that records every keystroke.” This creates a culture of fear where those in the community are highly aware that they are being surveilled.
“It’s the most important event in the whole censorship system. Nothing is as significant.” – Liu Lipeng, former censor
Liu Lipeng, a veteran of several years as a censor, describes what moderation looks like today. He notes that as important as AI will be to this effort, human creativity has a long history of sidestepping automated detection. “After working as a censor for years, I found human creativity can still crush AI censors many times over,” he remarked.
The Impact of Censorship on Society
The effects of this intense censorship regime go farther than just the experience of single users. They influence the attitudes and conversation inside of China itself. The Chinese government has a tight control over all information, particularly around events such as Tiananmen Square that are in the realm of sensitive historical political occurrences. Remarkably, even 36 years later, officials have still not released an estimated official death toll from the massacre.
This lack of transparency is a very difficult situation for citizens to be placed into. They need to cut through the thicket of unsubstantiated claims and alternative facts. The government’s fear of social upheaval is what motivates its harsh censorship campaign. Content moderators for social media platforms such as Weibo know that their work occurs under an acute state of alert at this time.
“They use computer vision, natural language processing and real-time filtering. It doesn’t change the nature of censorship, but it makes it more powerful.” – Dr. Chang
China is getting better at censorship with technology and human moderation. Experts are adamant about the importance of keeping these databases neutral and factually accurate. Dr. Chang advises that “we need to think hard about how to maintain databases that are neutral, uncensored and accurate because if the data is fake, the future will be fake too.”
