The Empty Chair: Using LLMs to Raise Missing Perspectives in Policy Deliberations

TL;DR

This paper investigates using large language model-generated personas to surface missing perspectives in policy deliberation, addressing representational gaps and polarization. It implements a real-time transcription-based tool that creates dynamic stakeholder personas and extracts points of disagreement to seed discussion in a campus sustainability assembly with 19 undergraduates. The results suggest that AI personas can spark new discussions and increase empathy toward differing viewpoints, but they also raise significant concerns about misrepresentation and the need for careful framing so that AI complements rather than substitutes genuine participation. The work provides design guidance and underscores the importance of grounding AI-generated perspectives in authentic data to ensure responsible deployment in deliberative contexts.

Abstract

Deliberation is essential to well-functioning democracies, yet physical, economic, and social barriers often exclude certain groups, reducing representativeness and contributing to issues like group polarization. In this work, we explore the use of large language model (LLM) personas to introduce missing perspectives in policy deliberations. We develop and evaluate a tool that transcribes conversations in real-time and simulates input from relevant but absent stakeholders. We deploy this tool in a 19-person student citizens' assembly on campus sustainability. Participants and facilitators found that the tool was useful to spark new discussions and surfaced valuable perspectives they had not previously considered. However, they also raised skepticism about the ability of LLMs to accurately characterize the perspectives of different groups, especially ones that are already underrepresented. Overall, this case study highlights that while AI personas can usefully surface new perspectives and prompt discussion in deliberative settings, their successful deployment depends on clarifying their limitations and emphasizing that they complement rather than replace genuine participation.

Figures (6)

• Figure 1: An example persona generated from the conversation transcript and deliberation context. We have redacted any specific references to locations or universities.
• Figure 2: The flow for generating missing stakeholders, stakeholder reflections, and finally stakeholder questions based on the assembly context and a live transcription of the conversation. The solid lines capture how the assembly context and the transcript helps generate missing stakeholder personas, which then is used to generate stakeholder reflections, and then finally the stakeholder questions.
• Figure 3: An example of points of disagreement and missing perspectives from a persona.
• Figure 4: An example question generated by a persona.
• Figure 5: Changes in understanding towards different points between pre-survey (prior to first day) and post-survey (after first day). Red lines capture pre/post means, and blue lines and dots represent individuals.