Decoding Algorithms: Exploring End-Users' Mental Models of the Inner Workings of Algorithmic News Recommenders

Abstract

Algorithmic recommenders are omnipresent in our daily lives. While a multitude of studies focus on how people use algorithmic recommenders, far too little attention has been devoted to how they perceive and understand these complex systems. In this study we focus on Algorithmic News Recommenders (ANR). Drawing on 26 semi-structured interviews, we investigated how laypeople decode Google News and Facebook News. In our method we employ the scroll-back method, make use of visualizations and a double interview design. Our results differentiate between those with a high and low level of understanding. Those with a high level of understanding acknowledged the role of companies and developers in the workings of ANR. Others, who were less cognizant had a more instrumental view and mostly focused on the relation between their individual data disclosed and the ANR. More importantly, in both groups, their feelings (ranging from admiration to frustration) about and everyday interactions (both dominant and deviating) with ANR shape their general understanding. In the discussion we argue how it's necessary for future research endeavors and algorithmic literacy initiatives to be mindful of the interconnection between knowledge, feelings, and interactions to understand layman's perspectives.