Who Do We Blame for the Filter Bubble? On the Roles of Math, Data, and People in Algorithmic Social Systems

Book Chapter August 2020

Filter bubblesRecommender systemsSimulation modeling

Overview

This chapter introduces a three-factor responsibility framework—Data × Algorithms × People—for analyzing who (or what) drives online “filter bubbles.” Rather than blaming math alone, we show how outcomes emerge from the interactions of training data, algorithmic logic, and human behavior. A custom simulation of 500 iterations across 2×2 experimental conditions (overlapping vs. polarized preferences × deterministic vs. stochastic collaborative filtering) reveals that the same tweak can either widen or shrink ideological overlap depending on context. The chapter situates these results in current U.S. legislation debates, including the Algorithmic Accountability Act (AAA) and Filter Bubble Transparency Act (FBTA).

Read the Book Chapter (Cambridge UP)
Invited chapter in an edited volume on Networks, Algorithms, & Humanity (2020)

What the Chapter Argues

Taken as a whole, the scientific literature does not support treating the filter bubble as a simple problem of algorithmic bias. Media environments emerge from the interaction of data, algorithmic choices, and human behavior, which means responsibility is necessarily shared across all three.

Our 2×2 simulation makes that interdependence concrete by comparing polarized and overlapping preference distributions under classic and stochastic recommenders. The same algorithmic change can either widen or shrink ideological overlap depending on the data and the behavior of the people using the system.

This context dependence complicates policy proposals built around broad appeals to “transparency,” “bias,” or “accountability.” The chapter argues instead for a more empirical understanding of how particular data, algorithms, and behavioral patterns combine before deciding which intervention is likely to help.