It's worth pointing out that this research was conducted by arguing with people on twitter via a bunch of automated bots. It's not really that surprising that people will [be slightly more likely to] double down if they notice that the person trying to talk them out of their beliefs is part of a legion of literal robots.
