Exploring Algorithmic Contestation: User Preferences for Non-Profiling Recommender Systems on Social Media Platforms

Social media platforms (SMP) have fundamentally transformed people’s daily lives. People use them to communicate with family and friends, receive and share news, or connect with like-minded people worldwide. However, SMPs also have a dark side. Much of the scrutiny focuses on the companies’ AI-powered algorithms. Personalized recommendation algorithms are designed to capture users’ attention and maximize their time on the platform. As a result, automated recommendations are biased toward extreme and polarizing content leading to a surging spread of false information (Vosoughi et al., 2018) and inappropriate content recommendations for children (Papadamou et al., 2020). The harmful effects of attention-extractive algorithms on democracy, social cohesion, and health are well-documented (LorenzSpreen et al., 2022; Meier & Reinecke, 2021). Policy makers call for stricter regulations to counter the negative externalities of SMPs. For instance, article 38 of the Digital Service Act (DSA) by the European Union (EU) states that SMPs “shall provide at least one option for each of their recommender systems which is not based on profiling“. Thus, under the DSA, users can opt out of personalized content recommendations, thereby contesting attention-extractive algorithms. Due to the so-called ‘Brussels effect’ (Bradford, 2020), this legislation may also affect people outside the EU jurisdiction in the future. Our pre-registered study (AsPredicted #109476) explores users’ preferences for nonprofiling recommender algorithms. We further test individual drivers of algorithmic contestation and investigate differences both on country-level and SMP-level. We conducted a large (n= 6300) cross-national survey in six countries to answer our research questions. Preliminary findings suggest that 19.3% of users prefer a non-profiling content recommendation algorithm for their SMP feeds, with substantial differences between platforms. Results of logistic regression show that attitudes towards recommender systems, trust in SMPs, and privacy concerns have the most notable influence on users’ preferences for profiling vs. non-profiling recommender algorithms.