Reddit’s subforum ChangeMyView is a subreddit where users post about a specific view they hold and challenge other users to present arguments to change their view. Here, we investigated what distinguishes the most convincing participants on Reddit ChangeMyView, as measured by the all-time deltaboard, from equally active users who are nonetheless less likely to persuade people on the forum. We investigated the most common features of top persuaders and compared them to their similarly active Reddit counterparts. We found that top persuaders were more likely to provide external evidence for their claims and use morality-based reasoning, made longer comments and engaged in more back-and-forth argumentation, were less likely to use informal language in their arguments, and were more likely to show some overlapping semantic content with an original poster’s comment. These findings replicate and expand on previously established effects and may highlight ways to develop more effective educational interventions.

Humans are resistant to changing their beliefs even in the face of disconfirming evidence. The Bayesian brain theory suggests that we should update our beliefs optimally in light of new evidence, but recent research indicates that belief formation is far from the Bayesian ideal. Individuals can exhibit "stronger-than-rational" updating or be resistant to revising their beliefs. The present study proposes a novel paradigm to explore perceptions and preferences for belief updating patterns in stable and dynamic stochastic environments, using an advice-taking paradigm. In an experiment (N=567) based on a fishing task, we introduce three advisor characters representing formal updating models: Bayesian, Volatile and Rigid. We find that participants exhibit higher trust for the Bayesian advisor than the Rigid advisor, in the stable but not changeable environment conditions. In the changeable environment, participants exhibit higher trust for the Volatile advisor, compared to both the Bayesian and Rigid advisors. The findings also suggest that participants' own learning closely mimics the pattern of the Volatile model. This study illustrates that people can differentiate between Bayesian updating, and its "stronger-than" and "weaker-than" variations, and exhibit preferences for these updating patterns, in different environment structures.

Although research in the area of belief updating has flourished in the last two decades, most studies do not treat beliefs as part of a complex and interactive network. In this study, we investigate humans' use of auxiliary hypotheses as a mechanism to avoid belief updating in light of conflicting information. In Experiment 1, we replicate an unpublished study by Kahneman and Tversky, introducing two additional domain conditions (N=119). Participants construct an initial model, express a prior belief, and face conflicting information. They are then prompted to provide an explanation. Across three domains, only 37% of responses demonstrate belief updating, by attributing the information conflict to the original report being unreliable or invalid. In Experiment 2 (N=29), a within-participants manipulation of credibility shows no effect on generating auxiliary hypotheses. Even in the presence of credibility cues to explain away information conflicts by invoking the reliability of either source, participants instead generated auxiliary hypotheses to resolve them in 27% of the cases.

We introduce a novel methodology to scrutinize blame attributions in 'Tweets', focusing on Artificial Intelligence (AI) incidents - a contemporary issue that provokes regular discourse. The method identifies the agents that get blamed and the factors that are associated with blame attributions. The proposed methodology replicates and contextualizes findings from experimental settings, revealing AI entities are often held accountable for adverse outcomes, while human agents are judged based on intentions. It also identifies unexplored factors, such as blaming data for perceived biases or AI for replacing humans. This method offers a robust tool for mitigating measurement bias in specific fields, enabling the continual rejuvenation of theoretical frameworks with emerging variables.