Join the Cyber Policy Center October 15th, from 1 PM-2 PM Pacific, for the next seminar in our Fall Seminar Series. Sessions will run October through December and will be moderated by Nate Persily, co-director of the CPC. Tiziano Piccardi, postdoctoral scholar in the Stanford Human-Computer Interaction Group (HCI), joins us for his talk Tuning Our Algorithmic Amplifiers: Designing Pro-Social Values Into Social Media Algorithms.
Can we design feed algorithms that are mindful of their societal impact? AI-driven feed algorithms typically optimize for user engagement and retention but may inadvertently contribute to harmful outcomes, such as affective polarization. Recent interventions aimed at mitigating these effects have produced mixed and sometimes contradictory results. However, advancements in AI now offer an opportunity to rethink how algorithmic curation is done by embedding important societal values—such as supporting democratic principles—directly into the objectives of feed algorithms. In this presentation, we showcase the benefits of this approach through a field experiment on X (formerly Twitter). Using a browser extension powered by large language models (LLMs), we re-rank users' feeds in real time, controlling exposure to posts that express political distrust and animosity. Our findings demonstrate that this intervention influences users' feelings toward individuals from opposing political parties, highlighting a potential pathway for designing feed algorithms that explicitly promote societal values like democracy.