Traditional social media sites tend to prioritize showing their users what will achieve the most “engagement”, thereby exacerbating echo chambers and affective polarization. Drawing inspiration from cultural and political sociological theory, this project imagines a new, more prosocial way of engaging on social media, in news comment sections, and on public fora. User interfaces shape user behavior, and an alternative user interface can support more productive, enlightening discussions where new ideas can come to the forefront. We will use large language models as well as traditional NLP/ML methods to visually supplement the current conversation with selected supplemental posts from other conversations, and to give users a preview of how their messages may be received by others before they send them.

I aim to develop technology that serves the public interest. This SuperUROP will equip me with a deeper understanding of both the technically rigorous methods of designing and evaluating complex AI systems currently being used in academia and industry and the opportunities that exist to apply those methods in unexpected and beneficial ways in the real world. My background lies in web design and development and in ML model uncertainty and bias mitigation, and I’m excited to synthesize these skills and supplement them with NLP to build my researcher and developer toolkit.