Last year, violence broke out in a north-western corner of China, when against the police. Dozens of people were reportedly killed, and social media lit up with posts about the riots. But as China鈥檚 many censors got to work, many posts also quickly disappeared. But how did the government decide which could stay and which had to go?
A group of US researchers can help answer that question. They have lifted the lid on Chinese online censorship 鈥� by pretending to be censors themselves. They found that the censors of China鈥檚 social media sites are only worried about posts that may incite mass protests, rather than ones that poke fun at individual politicians, for example.
Acting like a social media start-up, at Harvard University and his colleagues built a fake website, bought a domain name and server space inside China, and populated the site with their own posts. Then, they obtained a copy of the software that helps Chinese website administrators censor their content so they stay out of trouble with the authorities. When questions about how the software worked came up, answers were just a phone call to customer service away.
Advertisement
鈥淲e could call customer service and say, 鈥楬ow do we stay out of trouble with the Chinese government?鈥� They would say, 鈥楲et me tell you,'鈥� says King. 鈥淐ustomer service was pretty good at their job.鈥�
Reverse engineering
The software could automatically block posts that came from banned IP addresses or contained problematic keywords. But many of the censorship choices were still deferred to human judgment. Some sites held posts for manual review within 24 hours; others automatically published the post, and a person would check in later to decide if it was OK.
If in-house human censors, rather than software, were calling the shots, then perhaps the researchers could divine insight into the minds of both government and independent censors by examining their choices carefully. King鈥檚 team went back to real social media sites such as Sina Weibo. Of the 100 sites, 20 were run by central government, 25 were by overseen by local authorities, and 55 were privately owned.
The researchers wrote more than 1000 comments about controversial topics, such as artist Ai Weiwei鈥檚 music, a village protest in Panxu, and the Uighur riots in Xinjiang. The researchers tracked if and when the posts were taken down.
鈥淭he censorship programme is like an elephant tiptoeing around. It leaves big footprints,鈥� says King. 鈥淲e can look at those footprints and learn things about the intentions of the Chinese government that would be difficult to learn otherwise.鈥�
Collectivism a no-no
To their surprise, the posts that were most likely to be censored were those that mentioned collective action in the real world, like a protest or a boycott. Even posts that praised the government could be censored if they also referenced collective action.
Meanwhile, criticism and complaints often made it through to publication. A satirical post about the Uighur situation went viral, and only got censored about 10 per cent of the time.
King suspects this is because the government is more worried about regime stability than anything else, so they exert most of their pressure on posts that could jeopardise their power. 鈥淭hey don鈥檛 care what anybody says about them in and of itself. They care what people will do with that information,鈥� he says.
Degrees of criticism
Before this study, we only had 鈥渧ery impressionistic understandings鈥� of the Chinese government鈥檚 priorities, says at the University of California, San Diego. The findings suggest that censorship decisions are more nuanced than previously thought.
鈥淚t鈥檚 really an amazing piece of work, to be able to do that kind of intervention and look at the impact of the intervention. Nobody has ever done anything quite like that before,鈥� Shirk says. In some ways, the censorship system is tighter than we thought, she says, and in some ways looser.
is more sceptical of the results. She suspects there are still some kinds of complaints that the government would want to censor, like Western media reports that expose abuses of power. 鈥淐riticism has many different degrees,鈥� she notes.
Journal reference: Science, DOI: 10.1126/science.1251722
