Communicating with Algorithms: A Transfer Entropy Analysis of Emotions-based Escapes from Online Echo Chambers

Abstract

Online algorithms have received much blame for polarizing emotions during the 2016 U.S. Presidential election. We use transfer entropy to measure directed information flows from human emotions to YouTube's video recommendation engine, and back, from recommended videos to users' emotions. We find that algorithmic recommendations communicate a statistically significant amount of positive and negative affect to humans. Joy is prevalent in emotional polarization, while sadness and fear play significant roles in emotional convergence. These findings can help to design more socially responsible algorithms by starting to focus on the emotional content of algorithmic recommendations. Employing a computational-experimental mixed method approach, the study serves as a demonstration of how the mathematical theory of communication can be used both to quantify human-machine communication, and to test hypotheses in the social sciences. Communicating with algorithms. 2 Communicating with algorithms: A transfer entropy analysis of emotions-based escapes from online echo chambers Algorithms intermediate almost all of the roughly 3.5 hours per day that each American communicates online (Center for the Digital Future, 2016). Algorithms decide how fast information is transferred and what information is presented (Hannak et al., 2013; Lazer, 2015). The proactive role of algorithms in communication processes can be conceptualized as an active dialogue between users and algorithms. Users communicate preferences (advertently or not), which are interpreted by algorithms (e.g., by recommender engines or bots), which then send responses back to users, who receive, interpret, and react to the algorithmic reply. In this study, we quantify the flow of information in the conceptual communication channel between users and socially responsive algorithms. Particularly, we first demonstrate that human emotions (joy, fear, sadness, etc.) are communicated to the selection mechanism of videos recommended by YouTube (via an individuals' choice of search terms), and second that the video's emotions influence how the viewer feels after being exposed to the videos. We study emotions related to Presidential candidates from the 2016 U.S. election. Rather than making assumptions about how recommender systems work, we systematically manipulate YouTube's real-world recommender system to create stimuli in a mixed-method computational-experimental approach. Algorithmic Filter Bubbles and Echo Chambers In the two-step communication between algorithms and humans it is common (but by no means obligatory) that the algorithm takes on the role of a confirmatory communication partner that reassures and often reinforces received information through positive feedback. This leads to the notorious "filter bubbles" (Pariser, 2011). The separation of users from contradictory information also gathers likeminded people in similar communication spaces, which then creates reinforcing echo chambers (

Machine behaviour

“…Although it can be methodologically convenient to separate studies into the ways that humans shape machines and vice versa, most AI systems function in domains where they co-exist with humans in complex hybrid systems 39,67,125,128 . Questions of importance to the study of these systems include those that examine the behaviours that characterize human-machine interactions including cooperation, competition and coordination-for example, how human biases combine with AI to alter human emotions or beliefs 14,55,56,129,130 , how human tendencies couple with algorithms to facilitate the spread of information 55 , how traffic patterns can be altered in streets populated by large numbers of both driverless and human-driven cars and how trading patterns can be altered by interactions between humans and algorithmic trading agents 29 as well as which factors can facilitate trust and cooperation between humans and machines 88,131 .…”

Propaganda Model in the Age of Social Media

“…Similar symptoms also occurred among Twitter The 10th IGSSCI users in the momentum of the Presidential Elections in the United States in 2016 whichshowed a level of significant political homogeneity and opinion leadership which created homogeneous communities(Guo, Rohde, and Wu 2018). In this political event, it can also be seen how YouTube users have also experienced a split, namely the feeling of joy in relation to emotional variation and sadness and fear to play important roles in emotional convergence(Hilbert, Ahmed, Cho, Liu, and Luu 2018). This shows, again, that social media plays a more important role in emphasizing emotional strength than rational power in decisive political momentum.…”

Behavioral Experiments With Social Algorithms: An Information Theoretic Approach to Input–Output Conversions

“…It is also important to start watching the video, as we found that YouTube's recommendation algorithm works on basis of the watch history, not on basis of the search history. We speculate that the reason is that the final consumption of online content is a mix of own search results and of input from their online friends (Gottfried & Shearer, 2016;Hilbert, Ahmed, Cho, Liu, & Luu, 2018).…”

“…Such interpretive power has given rise to omnipresent online recommender algorithms, which have become crucial gatekeepers in the management of today's communication landscape (Ricci, Rokach, Shapira, & Kantor, 2011). Their critical role has then again received much blame recently for creating filter bubbles and echo chambers that clearly restructure our communicational landscape (Bakshy, Messing, & Adamic, 2015;Colleoni, Rozza, & Arvidsson, 2014;Hilbert, Ahmed, Cho, Liu, & Luu, 2018;Pariser, 2011).…”