Back to search

FRIHUMSAM-Fri prosj.st. hum og sam

The Double-edged Sword of News Recommenders Impact on Democracy (NEWSREC)

Alternative title: Anbefalingsalgoritmers konsekvenser for nyhetsmedienes demokratiske rolle

Awarded: NOK 8.0 mill.

The NEWSREC project investigates the promise and perils of recommender algorithms that are designed to influence people’s news selection. We focus on one of the most heavily debated consequences of recommender algorithms: the extent to which recommender algorithms lead people into “filter bubbles” where they are separated from information that disagrees with their viewpoints. Because of the “black box” nature and poor transparency of algorithmic technology, the precise conditions under which recommender algorithms are a threat to or an opportunity for democracy remain a puzzle. For instance, we lack knowledge on how to make informed choices about designing recommender algorithms such that they do not, as a biproduct, filter out information that disagrees with people's viewpoints. The project addresses this puzzle by studying how recommender algorithms can be designed to either create or counter the formation of online filter bubbles. To achieve that, we aim to shift the scholarly attention from the dominant perspective of uncovering whether the current recommender algorithms (for instance Facebook’s or YouTube’s algorithms) threatens democracy to understanding the conditions under which recommender algorithms are for the better or the worse for democracy, given that they are designed for that purpose. This new perspective shifts the responsibility for the democratic implications of recommender algorithms from the technology itself to those who make the decisions on the implementation and design of the technology. The project conduct experiments in surveys where we study how people navigate on news sites where algorithms influence the presentation of news. The project also develops the first news recommender algorithm that is tailor-made to pioneer research on the democratic implications of recommender algorithms. We also test our algorithm on authentic replicas of Norwegian news sites in a large field experiment.

NEWSREC deals directly with one of the most pressing questions facing the news media today: What are the precise conditions under which news recommender technology are for the better or the worse for the democratic role of the news media? Evidence of news recommenders’ dystopic democratic threats (e.g., Filter Bubbles) and of their opportunities to counter such threats remain largely anecdotal. Despite an increasing scholarly attention to recommenders, the precise conditions under which they are a threat to or an opportunity for democracy remain a puzzle. We will address this puzzle head-on by offering a radically new perspective: We aim to shift the scholarly attention from the dominant perspective of uncovering and describing whether the current news recommenders amplify or reduce selective exposure and sharing to understanding the conditions under which recommenders do so, given that they are designed for that purpose. By focusing on this counterfactual (i.e., what has not happened but could or might under differing conditions), we radically shift the responsibility for the democratic implications of recommenders from the technology itself to the decisions surrounding the implementation and design of the technology. We mobilize this novel perspective by developing the first news recommender that is tailor-made to pioneer research on the conditions under which news recommenders amplify or reduce selective exposure and sharing. We will: (a) develop a framework for understanding when and how news recommenders can increase or decrease selective exposure and sharing, and delineate the ethical considerations pertaining to designing recommenders to do so; (b) develop the first news recommender equipped with factors that increase or decrease selective exposure and sharing; (c) use a randomized field experiment to test this recommender to gain a precise understanding of when and how news recommenders increase or decrease selective exposure and sharing.

Publications from Cristin

No publications found

No publications found

Funding scheme:

FRIHUMSAM-Fri prosj.st. hum og sam