Instagram and WhatsApp – the new tools of social media propaganda

Shifting sands of disinformation: lessons from the EU elections

So how are the changes that Neudert identifies manifesting in reality? Earlier this year, Comprop turned its attention to the European parliamentary elections, examining social media trends in seven languages – English, French, German, Italian, Polish, Spanish and Swedish.

The team pored over more than two million tweets, and found that rates of disinformation from junk news were actually falling. Neudert explains: “In Germany and France, for example, we used to see about 20% of all information about politics being from junk news. This year it was about 10%. In other countries it was even lower – between 5% and 10% for all the different language spheres. That made us quite hopeful and positive about the state of junk news during the European elections.”

But the team then turned their attention to Facebook, where they examined the outlets classified as junk news and outlets classified as professional or mainstream media, and examined metrics that will be familiar to any social team – shares, likes and popularity.

“What we found there is that, on average, the stories that were shared on Facebook that were disinformation were way more popular than stories from professional outlets,” says Neudert. “The average interaction for junk news sites on Facebook were way higher by a factor of between three and six times per country. That made us less hopeful.”

Much of the junk news material consisted of scare stories linking immigrants to criminality – a racist trope exploited by the far right for years. Some of these stories were linked to the 15 April 2019 fire at the cathedral of Notre-Dame de Paris, which occurred a month before the EU election and was falsely linked to Islamist terrorism almost as soon as the first pictures of the blaze appeared online. “This really shows how flexible disinformation is, how quickly it develops and is shared on social,” says Neudert.

Photo and video content: a new disinformation vector

The visual component of propaganda, as exemplified by the spectacular photos of the Notre-Dame blaze or, more recently, the Amazonian wildfires in Brazil, highlights the growth of disinformation on platforms such as Instagram and TikTok.

“Disinformation is becoming more and more visual – it’s images and gifs – and often that kind of content is harder to detect and very shareable,” says Neudert.

In the case of the Amazon fires, the spread of disinformation was almost too subtle for most people to spot, including the likes of French president Emmanuel Macron, who shared a photo taken by someone who died in 2003, and actor, producer and environmentalist Leonardo DiCaprio, who shared a photo that wasn’t even of the Amazon.

Their high profiles – Macron has 4.2 million Twitter followers and 1.2 million on Instagram, DiCaprio 19.2 million on Twitter and 35.2 million on Instagram – meant that the lie spread, literally and figuratively, like wildfire.

“This was being shared by other influencers, celebrities and politicians, and it exemplifies how difficult it is to control and how easily it spreads,” says Neudert. “Often it’s not the Russian disinformation that we are concerned about when we think about elections, but disinformation in the public mainstream.”

Platforms like Instagram are still little understood in the context of online propaganda, but one important emergent factor is the use of hashtags – Instagram allows users to post up to 30 on one picture – and to follow them so that pictures with the hashtag “puppies” or “flowers“, or perhaps “news” or “politics” will show up in the user’s feed from time to time.

“That content is coming from influencers and pages with big followings, and also the likes of Russia Today [which has hundreds of thousands of followers across multiple targeted accounts],” says Neudert. “There is not a whole lot of content moderation going on on Instagram yet.”

This is also a source of concern on TikTok, the de facto replacement for the now-defunct Vine service beloved of the Generation Z demographic.

TikTok, which had more than a billion downloads as of February 2019, is owned by Beijing-based ByteDance, and has rapidly emerged as a key source of short-form videos. It is wildly popular not just in the West, but in markets such as India.

In its most benevolent form, TikTok content is very similar to Vine’s, with lots of quick-fire gags and memes, and the ubiquitous puppies and kittens, but, as Neudert notes, it has a very young audience that is often more susceptible to disinformation, and the people who spread it know this.

“In India, for example, we know there are conspiracy theorists and media personalities with TikTok channels using it to propagate misinformation,” she says. “In Hong Kong – where we know there is censorship going on – TikTok very prominently carries no content relating to the Hong Kong protests, so there is already censorship, filtering and disinformation from the Chinese government.”

From deep fakes to shallow fakes: the tale of Jim Acosta

The other topic of interest for Comprop when it comes to visual content, and one that is gaining traction in the mainstream media, is the rise of deep fakes – photos and videos that are generated by artificial intelligence (AI) and purport to show real situations, but are, in reality, completely fake.

“This could be, for example, a speech given by president Trump that is not actually him – but it will look like him, sound like him, and will have his demeanour,” says Neudert.

“This kind of content is being produced by a technique called generative methods, so you have an AI that is basically watching tonnes and tonnes of video of Trump, and it learns from that kind of content how to generate something that will look and sound just like him.”

To the relief of everyone, no fake videos of Trump ordering a nuclear strike have yet emerged, as the AI techniques used to create them are still very expensive and at the bleeding edge of technology. The more relevant concern, says Neudert, is cheaper, shallow fakes.

Back in 2018, for example, CNN’s chief White House correspondent, Jim Acosta, found himself making the headlines rather than reporting them, when a testy exchange with Trump at a press conference culminated with a White House worker trying to grab the microphone from Acosta.

In the wake of this incident, the White House shared video footage, allegedly created by Infowars, that was cut in such a way that it seemed to show Acosta aggressively pushing the staffer away, when actually he was trying to pull the microphone back.

“It was shared by Trump himself and by all sorts of policy-makers and high-up people in the Republican Party,” says Neudert. “It was a very simple fake and it was debunked very quickly, but it still made it to the top trending topic on all sorts of channels and it led to Acosta being suspended from the White House.”

Encrypted channels: a great idea, but causing complications

Online disinformation is also spreading into private and encrypted forums – as the bright light of media scrutiny, and the threat of regulation, shines down on Facebook and Twitter – and it is only natural that producers of disinformation are starting to take advantage of the built-in encryption features of Telegram, Signal and WhatsApp.

After all, their reasoning is exactly the same as the day-to-day user – privacy and protection from government snooping. Objectively, such people have more reason to be concerned about that than most.

“Messenger apps are, in many countries, the most important form of social media,” says Neudert. “In Germany we have 82 million people, and 43 million people active every month on WhatsApp. Messengers are also becoming more and more important in countries like India, where the mobile internet really is the internet.”

These apps are now a huge source of news for millions of people, and because they are encrypted, it is much harder to know what is actually going on there, what content is being shared most widely, by whom it is being shared, and who is viewing it. But, anecdotally, threat researchers have described seeing closed groups that were cesspools of disinformation, extremist content and other forms of hate speech.

Reframing the debate: whose responsibility is it anyway?

Neudert’s work highlights the importance of education in countering evolving propaganda techniques. Everybody needs to take responsibility for training themselves to spot disinformation, but crucially, parents, educators and those in the media need to shoulder more of this burden.

“One thing that I hope to see from tech reporting is not to fall for the story the social media companies are telling you, that they just need guidance,” she says. “Yes, of course they do, but it’s important to be critical about that, to ask if guidance has already been provided, and no action has followed. It is clear by now that there are examples where policy-makers have provided guidance, and companies have had difficulty following that.

“That is something I hope journalists can be critical of, and ask questions that I as a researcher cannot.”

Neudert agrees with the widespread view that government regulation of social media is desirable, but unfortunately, she says, we are now past the point where social media companies can be trusted to self-regulate, particularly in the light of the Cambridge Analytica revelations. The landscape is complex, cultural and highly politicised, so regulation needs to be finely balanced.

There is one thing that may help, she concludes, and that is reframing the debate over what social media companies actually exist for.

“Back in 2016, there was debate over whether or not social media companies should be considered publishers,” she says. “I think that is now outdated.

“Acknowledging that those platforms are now information intermediaries, commodities, that they are also networks so broad that we can’t avoid them, that can maybe help change some of our thinking.”