The Great Firewall: China looks to AI to censor online material

The Great Firewall: China looks to AI to censor online material

Much is made of the ability of artificial intelligence to automate monotonous tasks. Powered by machine learning algorithms, bots can churn through huge amounts of data. Aside from the practical and productivity benefits, the notion of freeing up human minds for more creative and fulfilling tasks is an appealing one.

However, there can be negative applications of the technology. In China, artificial intelligence and machine learning algorithms are increasingly being deployed to stifle free speech and shut down material deemed objectionable by a government that is introducing mandatory social profiling and active citizen surveillance, and which employs teams of social influencers (aka ‘the 50 cent party’) to spread positive messages about the government.

One of the country’s newest industries, the mass enforcement of internet censorship, is well on the way to becoming automated.

Access to global internet giants from across the Pacific, including Google, Facebook, and Twitter is completely restricted. Yet the number of people in China with internet access through mobile devices has grown dramatically in recent years: from 420 million in 2012 to 753 million in 2017, according to the government’s own statistics. This has created an enormous challenge for state censors.

Sifting through millions of images and videos every day is no longer feasible without the help of AI.

The dawn of machine censors in China

Widely used platforms, such as video streaming service iQiyi, are looking to machine learning algorithms to enforce censorship directives from Bejing. Just as with many western internet platforms, sensitive material such as pornography and violence is rooted out and removed.

“Machines are very good at recognising sexual content, so our human staff now spend very little time on that,” Kent Liu, iQiyi’s chief technology officer told the Financial Times.

Beyond that, internet companies in China are also required to remove content with themes that may be politically sensitive. Particularly those that are ‘developing’: a word vague enough to include everything from references to homosexuality and political ridicule to veiled criticism of government policy.

These developing themes are ones that AI often has trouble recognising: metaphors and subtle references that no algorithm can yet detect.

For example, only human censors could have enforced the removal of Winnie the Pooh from blogging platform Sina Weibo last year. The cartoon bear was found by the platform’s users to share a likeness with President Xi Jinping, so he disappeared – despite being first published 27 years before the president was born.

Facebook parallels

US internet companies such as Facebook are also looking to AI to help ensure their content is in line with company policy. In May, the social media giant revealed it was using publicly available Instagram photos to better train its image recognition software.

Facebook already uses AI to spot nudity and violence in images posted on its platform, though it draws the line at deploying the same technology to shut down criticism of elected officials and government policy – as far as we know.

But there is a wider similarity. Internet companies in both countries are being pressured by their respective governments to better police their platforms, eradicate extremism, and remove sensitive content – through armies of human censors, AI, or other techniques.

It’s in the process of defining what constitutes ‘sensitive’ and ‘extremism’ that problems arise.

Internet of Business says

Indeed. Hard-right eurosceptic MEP Nigel Farage used Facebook CEO Mark Zuckerberg’s appearance before the European Parliament yesterday to, first, credit social platforms with Brexit and the election of Donald Trump as US president, and, second, to simultaneously accuse Zuckerberg of clamping down on right-of-centre views. He claimed that an algorithm change had caused a fall-off in engagement of 25 percent on his own and Trump’s pages.

Inconveniently for Farage – who appeared to call for a social media bill of rights – Crowdtangle engagement data (see below) for those pages reveals no meaningful change at all in his engagement statistics since October last year.

It does show a downward trend in Trump’s statistics, but – post Cambridge Analytica and evidence last year of external political interference on social platforms – Facebook is known to have been clamping down on fake accounts, troll farms, and Russian-financed advertising that favours Trump, so the cause may be a rebalancing, and not a matter of alleged ‘censorship’. It may also simply mean that people are disengaging with Trump of their own volition.

After all, can censorship be said to exist when people are still able to express their own opinions, and others are still able to read and respond to them on a social platform? Beware straw men seeking bills of rights when no rights have been infringed to begin with.

Either way, Zuckerberg’s own lack of engagement with MEPs’ questions did him few favours with just two days to go before GDPR.