The effect of algorithmic bias and network structure on coexistence, consensus, and polarization of opinions

Abstract

Individuals of modern societies share ideas and participate in collective processes within a pervasive, variable, and mostly hidden ecosystem of content filtering technologies that determine what information we see online. Despite the impact of these algorithms on daily life and society, little is known about their effect on information transfer and opinion formation. It is thus unclear to what extent algorithmic bias has a harmful influence on collective decision-making, such as a tendency to polarize debate. Here we introduce a general theoretical framework to systematically link models of opinion dynamics, social network structure, and content filtering. We showcase the flexibility of our framework by exploring a family of binary-state opinion dynamics models where information exchange lies in a spectrum from pairwise to group interactions. All models show an opinion polarization regime driven by algorithmic bias and modular network structure. The role of content filtering is, however, surprisingly nuanced; for pairwise interactions it leads to polarization, while for group interactions it promotes coexistence of opinions. This allows us to pinpoint which social interactions are robust against algorithmic bias, and which ones are susceptible to bias-enhanced opinion polarization. Our framework gives theoretical ground for the development of heuristics to tackle harmful effects of online bias, such as information bottlenecks, echo chambers, and opinion radicalization.

Opinion formation on social networks with algorithmic bias: Dynamics and bias imbalance

“…A simple implementation of algorithmic bias has been proposed by us in a previous study [28]. Here we generalize the definition of that paper by introducing two parameters characterizing the bias, instead of one.…”

“…the following regions can be distinguished (i) low 0 < α < 2 (pairwise interactions); (ii) high 2 < α < 5, (group interactions); and (iii) very high α > 5. Each of these cases displays a distinct phenomenology [28,33] and represents a different archetypal way for humans to influence each other (either in pairs or in groups).…”

“…In the present work, similarly to the approach of [23], we consider that a fraction of the neighbors of an individual holding disagreeing opinions are filtered, and thus interactions with those neighbors are not possible. Recently we have proposed a general formalism within the binary-state approach that includes this implementation of algorithmic bias [28]. We have extended previous theoretical tools to describe the macroscopic dynamics on networks, including mean-field and pair approximations.…”

“…We have studied the static, asymptotic behavior of archetypal models of opinion formation in the presence of algorithmic bias and concluded that, systematically, pairwise interactions lead to polarization, while group interactions promote coexistence of opinions. Here we use the same formalism of [28], but focus on dynamical aspects of the opinion formation process. We also extend the algorithmic bias mechanism to consider situations in which the online social platform promotes one opinion over the other, thus unbalancing the dynamics in favor of the preferred opinion by the platform, in what we call bias asymmetry.…”

“…This general framework enables us to consider several interaction mechanisms and leads to various opinion formation scenarios. In [28], we have considered other archetypal models of opinion formation (including voter-like and majority-vote models), and realized that the language model essentially interpolates between dynamics with either pairwise or group interactions depending on the value of α, and is thus a good candidate to explore the effects of bias asymmetry within a single model.…”