How Social Networks Give Rise to Extreme Beliefs

Jason W. Burton, Momme Von Sydow, Christoph Merdes, Ulrike Hahn

Research output: Contribution to conferencePaperResearchpeer-review

Abstract

Much of what we think we know we know from the testimony of others. Other people tell us what the world is like: whether in personal conversation, via traditional media outlets, or, increasingly, social media. This raises the question as to how we might maintain the accuracy of our beliefs given that the reliability of testimonial sources is often unknown, and even the most trustworthy, expert sources make errors. In light of this question, past research has shown that people naturally engage in an “expectation-based” (or “belief-based”) updating strategy, whereby an agent simultaneously revises its belief in a communicated claim and its estimate of the source’s reliability based on the expectedness of said claim (i.e., the congruence with the agent’s prior) [Hahn, Merdes, & von Sydow, 2018; Collins et al., 2018]. While this strategy is an intuitive, normatively appropriate solution to situations in which one possesses no or limited knowledge of a source’s past judgmental performance, it runs the risk of establishing a confirmation bias among agents. Moreover, this process of communication and expectation-based updating occurs not in just isolated dyads, but across wider social networks. This reality of testimony means there may be dependencies among sources, such that the same piece of evidence could reach an agent via multiple paths leading to a kind of “double counting.” Consequently, agents may over-estimate the amount of evidence there is and be led toward more extreme, overconfident beliefs than objectively justified.
In this work, we demonstrate how these two fundamental aspects of testimony — expectation-based updating with respect to unknown source reliabilities and the fact that our sources are themselves parts of wider social networks — give rise to overconfidence [in terms of collective calibration, i.e., the extent to which subjective degrees of belief (probabilities) match objective probabilities]. First, we present agent-based simulations showing how these two aspects are causally sufficient for overconfidence to emerge, and then we re-analyze a past behavioral experiment that provides further support for our conclusion.
Original languageEnglish
Publication date2022
Number of pages4
Publication statusPublished - 2022
EventACM Collective Intelligence Conference 2022 - Virtual event, WWW
Duration: 20 Oct 202221 Oct 2022
https://web.cvent.com/event/32341921-13bf-442f-9f91-f8ceef42e5c9/summary

Conference

ConferenceACM Collective Intelligence Conference 2022
LocationVirtual event
Country/TerritoryWWW
Period20/10/202221/10/2022
Internet address

Cite this