Simulated Societies

Simulated Societies Satellite

September 3-4, 2025

Siena, Italy

Satellite Description

Online social media platforms are complex ecosystems where structural and social behaviors emerge in unpredictable ways. While data-driven social studies offer valuable insights, they face key limitations—short-term datasets, algorithmic opacity, and restricted data access from major platforms like X/Twitter, Meta, and Reddit, despite regulatory efforts (e.g., the EU’s Digital Services Act).

To bridge these gaps, researchers often turn to socio-physics models (e.g., opinion dynamics) to capture large-scale phenomena like echo chambers, network resilience, and information spread. These models provide theoretical clarity and predictive power, but they oversimplify real-world social media dynamics by overlooking algorithmic curation, personalized recommendations, and linguistic diversity.

A promising alternative lies in AI-driven social media simulations, particularly those powered by Large Language Models (LLMs). These simulations offer a controlled environment to experiment with variables often hidden in real-world data—such as algorithmic influence and language-based interactions. While they cannot replace empirical studies, they serve as in vitro laboratories, enabling researchers to test hypotheses on social media dynamics in ways traditional methods cannot.

The Simulating Societies satellites addresses this gap by providing the CCS community a space to discuss modeling and simulations of social media settings, either with opinion dynamic models and/or large language model (LLM)-based agent systems.

Topics of interest

Agent Based Modeling
Opinion Dynamics
Social Digital Twins
AI impact on social simulations
From Data to Models and back
Validation of social modeling
Replicating online users’ social/psychological behaviors in simulated settings

Objective

The satellite is specifically tailored to appeal to the diverse CCS audience, aiming to bring together disciplines such as computer science, physics, and the social sciences.

Researchers and practitioners in physics and data science, will learn how LLMs can power simulations that more closely resemble real-world social media environments. Social scientists and cognitive researchers can better explore online behavior, bias formation, and group dynamics through these methods.

We aim to: