New study shows how hard it is to spot bot accounts on Twitter

Fake news is one of the biggest topics of the day, but fakery is nothing new online.  Several years ago the issue was less about fake news as it was about fake social media followers.  It was brought to a head after presidential candidate Mitt Romney gained huge numbers of followers overnight, and a number of tools were developed to gauge the authenticity of ones social following.

This has become especially key as research has shown that such bots can be rather effective at persuading us to follow their points of view, and hence they have become a weapon of choice in anything that requires persuasion to occur.

Not only are such fake activities effective, they’re also very hard to spot.  A recent study from Northeastern University highlights the difficulty many people have in identifying bot accounts on Twitter.

The author argues that most of us believe ourselves to be good at spotting the bots in our midst, but whilst that may be true for simpler bots that struggle to behave in a ‘natural’ way, bot owners are increasingly adept at appearing natural online.

Appearing natural

For instance, whilst starting conversations is relatively easy for bots, engaging in an active one is much harder, so bot owners are working hard to develop bots that can do just that.  What’s more, with the scale of many ‘bot armies’ it can create an illusion of reality that should not exist.

“If something is fake but tweeted by lots of accounts, it creates an illusion of reality,” the author says. “We might think, ‘maybe this is true because so many people are talking about it.'”

This fundamental difficulty in identifying bot accounts is one reason why Twitter themselves have not managed to crack down on them more.  As such, it has often required the crowd themselves to come together to root out fake accounts, as they did when identifying roughly 20,000 bot accounts working on behalf of Daesh last year.

Such vigilante action isn’t ideal however, and the authors estimate that roughly 15% of Twitter accounts could be fake.

As part of their work, the team developed a Botometer tool, which has been used over 30 million times since its creation.  It provides users with a potential bot rating for any account on Twitter.  It put my own odds at roughly 50/50, but whether that signifies a flaw in the tool or my own activity on Twitter remains open for debate.

It’s a constant arms race that shows little signs of abating.