An analysis of tweets about the 2016 election has revealed a crushing victory for the bots, with sock-puppets and bits of code having more online influence, on average, than real people. The extent to which these social media wars affect voting behavior remains in question, but in terms of social media influence, it seems we’re already having our opinions shaped by accounts pretending to be human.
Dr Timothy Graham and Dr Marian-Andrei Rizoiu, both of the Australian National University, collected 6.4 million tweets from 1.5 million accounts produced from 15 minutes before the first Presidential debate between Hilary Clinton and Donald Trump to 15 minutes after it finished. They used the established BotOrNot interface to assess each account’s humanity and looked at who it supported.
In the Proceedings of the 2018 International AAAI Conference on Web and Social Media (ICWSM) (preprint on arXiv), they report that only 4.8 percent of the accounts tweeting about the debate were clearly bots, much lower than some other estimates. BotOrNot provides a score of between zero and one for how likely an account is to be a bot, and Graham and Rizoiu eliminated all those in the gray zone, leaving bots outnumbered 20 to one by near-definite humans.
What bots lacked in numbers, they made up for in effectiveness. “We devised an influence measure which we could apply over all the millions and millions of possible unfoldings of retweet diffusions, as well as a separate measure of political polarisation and engagement which we used to determine the partisanship of a tweet,” Graham said in a statement. Bots, including “sock-puppets” amplifying the voices of individuals, and simple lines of code tweeting the same slogans repeatedly, averaged 2.5 times the retweet rate of humans.
This was not a contest of human versus machine. Bots fought on both sides politically, but those backing Trump were more numerous, more politically engaged, and more influential than those cheering Clinton on. Some of the more prominent pro-Trump bots have been shown to be controlled by Russia’s Internet Research Agency.
Twitter users may wonder how bots managed to have more influence than their human counterparts, despite the humans usually having more followers. Graham told IFLScience that bot strategy seemed to be to tweet at influential humans in the hope they would retweet the statement to their armies of followers. This sometimes worked, although it is unclear if there were other elements to bot success.
Graham acknowledged the study only captures a small moment in the long presidential election. He also told IFLScience it doesn’t address the much harder problem of determining how many people changed their votes on the basis of messages started by bots. However, in the context of recent evidence, Twitter bots are discouraging people from getting life-saving vaccinations, so there are definitely consequences.