Pro-Social Design with Pluralistic and Narrative AI Systems

Abstract

In this talk, I will present two complementary lines of work that explore how AI systems can be designed to serve pro-social outcomes not by collapsing social diversity into a single voice, but instead surfacing pluralism and by making abstract societal issues more personally meaningful.

First, I will discuss Plurals, a system that uses multi-agent deliberation to simulate socially diverse ensembles rather than a single neutral model output. Plurals provides a flexible framework for configuring agents, interaction structures, and moderation strategies inspired by deliberative democracy. Across multiple case studies and experiments, we show that simulated social ensembles can produce outputs that better resonate with real audiences than standard single-model generation.

Second, I will introduce an ongoing line of work that explores AI-assisted narrative autocompletion as a tool for reducing psychological distance to complex societal issues. This work uses interactive, personalized narratives co-written with users to help people imagine how distant or abstract events could plausibly unfold in their own lives. 

Taken together, these projects illustrate two complementary strategies for AI in society: one that emphasizes pluralistic deliberation across perspectives, and another that leverages narrative imagination to connect individual experience with collective outcomes.