April 4 2018 photo of a train station in Chicago featuring a Facebook ad warning against fake accounts
One of the biggest surprises from Friday’s tornado, besides the tornado itself, was a related thread on the Facebook page “Impressions of Luxembourg”.
The page usually posts stunning photos of the grand duchy. This time the administrator posted a stock photo of a tornado, which had reportedly been circulated with the claim it was the tornado that struck Luxembourg. The post was shared 162 times and received 47 comments. Some asked why anyone would mislead in such a way. I can only speculate. Sometimes news organisations or bloggers purchase stock photos to illustrate a story where they don’t have a photo from the scene. It could be because it is an incident that has just occurred, or the site lacks the resources to send a reporter. In the majority of instances, the goal is not to mislead. Simply, most legitimate news sites are structured in a way they have to carry a photo so the editors are trying to be resourceful, not malicious.
I was positively surprised by the scepticism of the thread. One account owner even wrote they had recognised the photo as a stock picture. It’s reassuring for me because I am noticing more and more suspected fake Facebook accounts accessing Facebook groups in Luxembourg. There are (in my mind) justifiable reasons why some people might create fake accounts: perhaps they risk imprisonment for criticizing an oppressive government, or they wish to ask a question anonymously. I’m not talking about these kinds of fakes. I’m talking about the more malicious ones.
In the first quarter of 2019, Facebook disabled 2.2 billion fake and compromised accounts, many of them hiding behind stolen photos. Millions more have cropped up since. I’m talking about the accounts that are fronts for bots (a software application performing tasks automatically), cyborgs (automatic with some human curation or posting) or trolls (fully human). According to Nato Stratcom report “The Black Market for Social Media Manipulation”, these accounts can be purchased on the open internet, with prices depending on the time and effort needed for a manipulation.
What do they do?
They do many things that fall under the umbrella term “coordinated inauthentic behaviour”. These accounts could be used to “like” certain posts, to post fake news stories and make them go viral, or they could be used to troll journalists and activists who challenge a narrative that the manipulator wants to spread. This kind of activity hasn’t been identified in Luxembourg but that doesn’t mean the bots aren’t here, in our favourite expat groups.
Fadi Safieddine, associate professor in management information systems at Queen Mary University of London will next year publish a book on the subject. He believes that these kinds of groups are an ideal cover for the fake accounts, helping them to avoid detection by the Facebook algorithms created to identify fakes. “We believe they may be preparing a trojan horse or sleeper cells where they look genuine and have a history,” he explained.
Account history is very important for fooling the Facebook algorithms, Safieddine says. He tested them by creating four fake profiles. Facebook shut down the first within three weeks, the second in a week, the third in 24 hours and the fourth in 30 minutes. Each time, the platform conducted an ID check, which successfully called the new accounts out as fake. “If you’re trying to create an adult account joining Facebook, they look at you with suspicion. Because they close it so quickly, it means those who are determined, will try to find a way to bypass the algorithm,” he said. In a lot of cases, this means fake accounts blend in until the moment when an election or referendum looms (anywhere in the world). That’s when we see their behaviour change--they will post or share fake content that pushes an agenda--be it to create chaos, discredit a candidate or otherwise. “By the time the fact-checkers come along, the average is 1.6 days, by then the post has gone viral,” said Safieddine.
How to spot fake accounts
The worrying thing is it’s quite possible we’ve already interacted with a fake account already in a Facebook group. Identifying them is tricky. Just over a year ago, the advice was to do a reverse image search of the photos they use. However, the searches by platforms like Google and tineye which offer this service cannot filter through all Facebook photos. Facebook restricted such searches to avoid people tracking down ex-partners in bad faith. There was a time when poor grammar was a sign of coordinated inauthentic behaviour. But language tools have been integrated into bot codes. Plus, in a multicultural country like Luxembourg, who can expect everyone to write English fluently?
Some group moderators have introduced a short questionnaire for new joiners to sort out the fakes. Others exclude accounts which do not say they are based in Luxembourg. But, semi-human run accounts with enough guile will always find a way through. I’ve spent hours trawling through some suspected fake accounts, searching for “evidence”. But, the inconsistencies are sometimes subtle. I don’t have the same resources as the Facebook information operations disruption unit and nor does the average Facebook group moderator. “What if there was a tool built into the browser in which you could just write a comment at meta level and say “I believe it’s a fake”? That crowdsourcing element could give you the result in maybe 30 seconds,” Safieddine said.
Critical thinking and scepticism are almost certainly going to be essential going forward. The tornado story reassures me, Luxembourg already has a healthy dose of both.