PoliSim@CHI 2026

Overview

Large Language Models are rapidly evolving from text generators into reasoning systems that can act as autonomous agents. When placed in social contexts, these agents display emergent behaviors such as forming coalitions, spreading information, and making collective decisions.

Policymaking is fundamentally collective, high-stakes, and uncertain. Unlike laboratory science, it rarely allows for controlled experimentation. LLM agent simulations offer a new kind of in silico testbed, enabling policymakers to explore interventions, stress-test communication strategies, and surface unintended consequences across large, diverse populations before acting in the real world.

Drawing on HCI traditions such as participatory and user-centered design, we argue that the value of these simulations does not come "out of the box." Instead, it emerges through iterative, stakeholder-engaged design—where policymakers build trust, probe system boundaries, and continuously recalibrate expectations.

Guiding Questions

Usefulness

How can LLM agent simulations move beyond technical demonstrations to become practical tools for policymaking?

Responsible Use

How can simulations be designed and interpreted responsibly, ensuring appropriate reliance, transparency, and fairness?

Co-evolution

How can simulations and policy processes be developed simultaneously, so that each informs and adapts to the other?

Call for Participation

OpenReview Submission Portal

We invite position papers (2–4 pages) or short reports describing case studies, design explorations, methodological insights, or reflections on using LLM agent simulation for policy. Encore submissions of relevant published work are welcome.

Word LaTeX Overleaf

What Makes LLM Agent Simulations Useful for Policy? Insights from an Iterative Design Engagement in Emergency Preparedness Can a Society of Generative Agents Simulate Human Behavior and Inform Public Health Policy? — A Case Study on Vaccine Hesitancy Sotopia: Interactive Evaluation for Social Intelligence in Language Agents Sotopia-S⁴: A User-Friendly System for Flexible, Customizable, and Large-Scale Social Simulation The Future of HCI-Policy Collaboration Generative Agents: Interactive Simulacra of Human Behavior Agentic AI for Sustainable Development: Leveraging Large Language Model-Enhanced Agent-Based Modeling for Complex Policy Strategies Validation Is the Central Challenge for Generative Social Simulation: A Critical Review of LLMs in Agent-Based Modeling Large Language Models That Replace Human Participants Can Harmfully Misportray and Flatten Identity Groups Position: LLM Social Simulations Are a Promising Research Method

Topics of Interest

LLM agent simulations for public health, climate adaptation, education reform, and other policy domains
Institutional and organizational policy applications (emergency preparedness, online community management)
Methodological innovations in memory architectures, environment embedding, and agent heterogeneity
Responsible AI considerations for policy simulations
Participatory and user-centered design approaches for simulation tools
Case studies of policy-simulation co-development
Evaluation frameworks for policy-relevant simulations

Submission Format

2–4 page position papers or short reports following the ACM template. Papers will be reviewed for relevance and diversity of perspectives.

Publication

All accepted papers will be published on the workshop website and in the proceedings with CEUR-WS. Selected authors will be invited to extend their work for established venues.

Attendance

At least one author of each accepted paper must register for and attend the workshop.

Accepted Papers

Keynote Speaker

Amanda Swearngin

Microsoft Research AI Frontiers

Magentic Marketplace: Testing Societies of Agents At Scale

As AI agents move from personal assistants to participants in shared digital marketplaces, questions of safety become questions of social reasoning: when should an agent act, when should it pause, and whose interests are affected. This talk centers on Magentic Marketplace, a simulation environment for studying how agents behave when they interact under shared constraints, incentives, and competition in marketplaces. I contextualize this work relative to prior research on personal computer‑use agents, including Magentic‑UI and From Interaction to Impact, which shows the importance of reasoning about context, consent, and irreversible actions. Magentic Marketplace extends these concerns to societies of agents, revealing how incentives and coordination can amplify risk and produce failure modes that do not appear at the individual level. Together, this work shifts attention from task success to socially aware agent behavior, framing responsible AI deployment as a problem of social reasoning rather than capability alone.

Amanda (she/her) is a researcher at Microsoft Research AI Frontiers where she conducts research at the intersection of AI and Human-Computer Interaction, building agentic workflows and experiences. Prior to Microsoft, Amanda researched and productionized technologies for computational UI understanding, deploying these technologies into multiple widely used Accessibility features. Amanda earned her Ph.D. from the University of Washington, where she was advised by Amy Ko and James Fogarty. Her research focused on developing novel interfaces for UX/UI designers that leveraged AI and program analysis. Amanda also spent three years prior to grad school as a Software Engineer in the Microsoft Dynamics group.