SocioSim: A Framework for Rapid, Policy-Relevant Audience Simulation

Eugene Lensky

Keywords: Large Language Models, Social simulation, Computational social science, Survey methodology, Policy research

TL;DR: SocioSim uses LLMs to simulate survey responses from diverse audiences. 3-stage process generates personas then simulates answers. AI companion case study revealed stark generational divide. Enables rapid hypothesis testing for researchers.

Abstract: Recent advancements in Large Language Models (LLMs) have enabled complex social simulations featuring persistent, interactive agents. While these models offer high-fidelity insights into emergent behaviors, they are often computationally intensive and ill-suited for rapid exploration of public attitudes or policy reception. This paper introduces SocioSim, a complementary framework designed for the rapid simulation of large, demographically-nuanced audiences. Instead of modeling longitudinal agent interactions, SocioSim employs a multi-stage LLM pipeline to generate cross-sectional insights. The paper details the SocioSim methodology, which consists of three core stages: Persona Blueprint Definition, Iterative Persona Instantiation, and Contextualized Response Generation. It presents a case study demonstrating how the framework was used to model societal attitudes toward AI companionship, revealing a non-obvious generational divide in public sentiment. SocioSim provides a novel, structured method for researchers and policymakers to generate hypotheses, de-risk research, and anticipate public reaction to emerging social and technological phenomena.

Submission Number: 7