A Cross-National Study to Promote Mutual Understanding via Social Media Using Generative Artificial Intelligence

Duke University, Durham NC

Investigators

Abstract

There is broad concern that social media inhibits productive discussion. Yet, few scholars have explored how social media platforms might be redesigned to counter such trends. This project employs generative artificial intelligence, social simulation models, and online experiments to identify how algorithms that shape the information users see on social media could promote mutual understanding, increase trust, and affect polarization. The project will also develop and, in a laboratory setting, deploy and investigate the consequences of alternative algorithms. Through these comparisons, we hope to reveal how digital platforms can promote mutual understanding and trust, informing future business leaders and policy makers within the industry. This project employs generative artificial intelligence, agent-based models, and online laboratory experiments to identify how algorithms that shape information users see online influence social norms, trust, and polarization. The first phase of the project trains Large Language Models to simulate social media users by calibrating them with empirical data derived from nationally representative surveys. Preliminary results indicate this research design can reproduce large-scale behaviors on social media platforms, and enable scholars to prototype alternative newsfeed algorithms that lead to different outcomes. The second phase of analyzes the impact of such alternative newsfeed algorithms via randomized controlled trials with human respondents via a tool that enables scholars to build social media platforms for the purposes of scientific research, and allows users to interact with such environments via mobile apps or the web. Embedded surveys will allow the researchers to assess multiple indicators, as well as behavioral data generated by research participants. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.