Research: Social Identity Theory and Algorithmic BiasResearch

Title: Social Identity Theory and Algorithmic Bias: Ingroup and Outgroup Acrophily in Recommender Systems

Paper signed by: Carlos Carrasco-Farré, Didier Grimaldi, Marc Torrens & Enzo Longobuco

published in the Journal of Management Information Systems (FT50), 

Summary

“Most debates about social media algorithms focus on echo chambers: platforms show you more of what you already like. Our new paper, published in the Journal of Management Information Systems (FT50), shows that this is only half of the story. We study YouTube’s recommendation system using 300k+ political videos and 1.7M+ recommendation links. We combine Social Identity Theory with network models to examine how recommender systems shape not only what we see from “our side,” but also what we see from “the other side.”

We focus on a phenomenon called acrophily: the tendency for users to be nudged toward more extreme content. First, we document ingroup acrophily: if you start from a moderate video aligned with your own political views, the algorithm is disproportionately likely to recommend more extreme content on your own side. For example, someone watching a center-left video will often be guided toward increasingly radical left-leaning videos rather than toward more moderate or balanced content.

More surprisingly, we introduce and test a new concept: outgroup acrophily. Here, users are not only exposed to their own extremes, but also to the most extreme content from the opposing camp. Imagine watching mainstream left-wing commentary and being recommended highly provocative far-right videos that caricature your views, or vice versa. This “hate-watch” dynamic means that users may come to know the outgroup primarily through its loudest and most inflammatory voices, amplifying perceptions of threat and moral outrage.

In other words, we show how recommender systems can simultaneously intensify identification with one’s ingroup and intensify hostility toward the outgroup. The result is a structural, algorithmically mediated push toward societal polarization: more extreme “us,” more demonized “them.”

Carlos Carrasco-Farré