Wraiths in the Machine: How Bots Artificially Amplify Online Hate#

In 2018, I was beginning my journey as a researcher in social sciences. I was also the only person in my lab who could write a roughly usable Twitter bot. So my first field work was an attempt at curbing hate speech in online settings. The plan was to find people who spread antisemitic hate speech on Twitter, send them a message based on theoretical considerations (e.g. calling upon authority – “the Polish Pope would not condone such hateful speech”), and then come again in two weeks to see whether that mitigated their tendency to spew hate. What I did not expect was that most of the subjects would be bots. Not real hate-driven people, but computer code oblivious to the consequences of its actions.

Hate speech has been notoriously evading attempts at drawing a precise definition, so we took the conservative approach, and only targeted tweets which were undeniably antisemitic. It is not an elegant solution to the problem, but it’s workable something you can work with. Then we found eight hundred people users, who posted this kind of speech, and we sent out our minions – an algorithm which replied to those 800’s antisemitic rants with replies randomly picked from a prepared set, then we got kicked out of Twitter for using bots after roughly 200 replies were posted, and came back after two weeks to see whether anything can be salvaged. We gathered the second set of data and checked whether our interventions had any effect. They had none. So we did what should be done, and took it all to one conference, and then to another. Graphs were shown, ideas exchanged… And many of those ideas were good and worth employing. As we soon found out, they were all missing the point, and by far. With no hope left for this particular study, the data were archived, put aside to a deep cellar of a hard disk, and all but forgotten. We moved on, leaving the idea of making the internet a better place behind. Some time later, I stumbled upon tools (https://tweetbotornot.mikewk.com/) which allow you to assess, with good confidence, whether a particular user is human or a bot. It was roughly 80%. Eighty percent of the users spewing antisemitic hate speech on Polish Twitter were not actual human beings holding such opinions, but scripts designed by someone to do so. And of course, they were not readily distinguishable from real human beings. There is a good side to it, and a bad side. The good one is simple – the more percent of those writing such nasty stuff are not human, the fewer are. There is something comforting in the knowledge that actually the cesspool of social media is artificially made to be more stinky, than the actual sea of human opinions and beliefs. We might not be so bad after all, when you filter out the artificially amplified voices of hate and contempt. But that leads us to the bad thing. As humans we are quite smart. But we are also super lazy, when it comes to cognition. When we encounter a new and sophisticated problem, and intergroup relations tend to fall into that category, we do possess the tools to figure out what is going on, individually or collectively. But when someone else had the job done for us? And if their opinion is popular among people, then it must be quite valid! From the evolutionary point of view, this apparent naïveté is not as wrong as it seems. Not wasting resources on solving problems already dealt with by someone else allows for dividing more complicated tasks among many people, and is the foundation of civilization. And this adaptation mechanism is exploited here – a false consensus is fabricated, a false common – and hateful – belief. And so it seeps from ones who, for their own reasons (I assume political, or money. Or most likely – abusing the political system for personal gain) propose hatred, to our minds – those of genuine users, who want to know what’s going on in the big and complex world.

There is a concept called the Dead Internet Hypothesis. It says that the internet is less and less genuine, with artificial agents creating more and more content – and often it’s bots who read that content, too. Search engines, like Google or Bing, send out bots (spiders) to check out websites, assess their relevance to search results etc. Automated spam factories send out terabytes of emails advertising penis-enlarging pills, online casinos, dubious dating websites and other bad ideas, daily. According to several independent estimates, roughly half of the traffic is generated by various kinds of bots. Does that mean that one cannot fulfill the promise of the internet, and engage into a meaningful discussion with a stranger from the world’s other end online? Not necessarily. The bots are not evenly distributed, not even randomly. Compare the 50% mentioned above and the 80% from our study. Some places are relatively safe, from bots at least, and some are swarming with bots.Keeping to the grim analogy of dead internet filled with zombies and wraiths, we can say that is not only haunted, but it is haunted by vile ghosts, who want to turn you into hateful ghouls.