The online petition, which appeared on the British government site petition.parliament.uk in May, earned more than 3.7 million new signatures over the weekend — a surge widely interpreted as a sign of voters’ remorse. But even as pundits passed around the petition, inconsistencies surfaced: By noon on Sunday, archived records show, the petition had 42,000 signatures from Vatican City (population 800) and almost 25,000 from North Korea (where Internet access is extremely limited).
On 4chan’s /pol/ board, trolls and pranksters posted screencasts suggesting they were automating signatures from exactly those places — a claim that’s impossible to verify, given the nature of 4chan, but in keeping with the community’s ethos and modus operandi. The parliamentary committee that oversees the petition site has said it’s investigating allegations of fraud, and that it’s already removed 77,000 bad signatures.
As of this writing, 96 percent of the petition’s signees claimed to live in the U.K., and the remaining four percent come from places with large expat populations, like France, Australia, the U.S. and Spain. (That doesn’t necessarily mean those signatures are legit, mind you — just that they aren’t quite so obviously fraudulent as the 4chan pranks.)
Meanwhile, even before outcry unfolded over the potentially fraudulent referendum petition, researchers at Oxford’s Computational Propaganda project quietly released a preliminary paper on the role of bots in the Twitter conversation around Brexit. They found that, of the almost 314,000 accounts that tweeted about the vote, pro or con, in the week of June 5 to 12, 15 percent were heavily or entirely automated. While pro-E.U. tweets saw a higher rate of automation than pro-Brexit tweets did, the two most prolific accounts on both sides of the debate are believed to be entirely automated.
Did anyone actually base her Brexit vote off the RTs of some Twitter hashtag drone? And did the surge of robotic signatures prompt more people to sign the referendum-redo petition? We don’t know the answers to either of those questions.
Still, these revelations come at a time when political influence bots are becoming both more sophisticated and more prevalent, raising questions about how they could impact things like, say, the approaching presidential election in the United States. Already in this election, observers have accused bot-makers of faking Latino support for Trump and rallying robot opposition against Ted Cruz. Accusations of astroturfing and bot-networking fly in the Democratic camp, as well, particularly since the launch of an “anti-trolling,” pro-Clinton task force called Barrier Breakers.
“Some parties are using them quite extensively,” said V.S. Subrahmanian, a computer scientist whose team recently won a Defense Department contest to improve Twitter bot-detection technology. “They’re subverting the free flow of opinion and influencing elections in a covert way. It’s worth knowing when something is actually a person’s opinion, and when it’s part of a more orchestrated campaign.”
And yet, as the recent online drama over Brexit suggests, we’re still a pretty long way from being able to consistently distinguish between the two. Dozens of news outlets reported credulously on the petition signature numbers over the weekend; in that Defense Department contest, even Subrahmanian’s specialist team fell for one bot that claimed to be human. (“They’ve gotten very, very, very good,” he said.)
Until we’re as good as the bot-makers, we might want to give a little less authority to things like Twitter trends and online polls. You never know when your big Brexit petition is possibly just a bored American, running a program while he showers.