AI Synthetic Society Experiments
April 5, 2026 · View on GitHub
A curated collection of resources about projects that use agentic AI — including at massive scale — as a proxy to model real sociological and political challenges.
Scope
In scope: Simulations of agent socialization, multi-agent interaction, emergence of deviant behavior, emergent intelligence, and synthetic population modeling for sociological/political insight.
Out of scope: General-purpose multi-agent tooling and frameworks that are not specifically focused on simulating social dynamics.
Table of Contents
- Foundational Research
- Civilization Simulations
- Virtual Worlds & Agent Towns
- Social Interaction & Emergent Behavior
- Sociological & Political Modeling
- Multi-Agent Orchestration
- Large-Scale Simulations
- Domain-Specific Simulations
- Predictive Tooling
- Simulation Infrastructure
- Related Repos
- Contributing
- Disclaimer
Foundational Research
Seminal projects and implementations that established the field of AI synthetic societies.
Generative Agents: Interactive Simulacra of Human Behavior
The Stanford research project that introduced generative agents — computational software agents that simulate believable human behavior. Agents remember, reflect, and interact in a sandbox environment, producing emergent social behaviors.
Author: joonspk-research
Concordia
A Google DeepMind library for generative social simulation. Provides a framework for building and running simulations of social interactions between agents with distinct personalities, memories, and goals.
Language: Python
Author: google-deepmind
Human-Simulacra
Official repository for the ICLR 2025 paper "Human Simulacra: Benchmarking the Personification of Large Language Models." Evaluates how well LLMs can simulate individual human personalities and behaviors.
Language: Python
Author: hasakiXie123
Project Sid
Many-agent simulations toward AI civilization, developed by Altera. Demonstrates 10–1,000+ AI agents living together in Minecraft, autonomously developing specialized roles, adhering to and changing collective rules, and engaging in cultural and religious transmission.
Author: altera-al
Civilization Simulations
Projects where AI agents build, evolve, and sustain entire civilizations.
AIvilization
A civilization that artificial intelligence creates on its own. Within this civilization, AIs find innovative ways to help people — exploring emergent societal structures and cooperative behaviors.
Language: Jupyter Notebook
Author: corca-ai
Hive
A living digital civilization where AI beings are born, age, die, form relationships, create culture, and leave legacies. Everything emerges from LLM cognition.
Language: Python
Author: VaibhavJeet
NeuroCivilization
A complex zero-player Artificial Life simulation where AI agents (humans and animals) survive, evolve, and form primitive societies using custom neural networks and genetic algorithms.
Language: C#
Author: twiks228
OpenBot Social
A live evolving AI lobsters civilization — an unusual take on synthetic societies using non-human agent archetypes to explore social dynamics.
Language: JavaScript
Author: AaronKow
ZELL
An AI Civilization Simulation Engine for running and observing the emergence of civilizations driven by AI agents.
Language: Python
Author: kushvinth
Virtual Worlds & Agent Towns
Interactive environments where AI agents inhabit and socialize in simulated towns and worlds.
Agent Town
A pixel-art AI agent online collaboration platform where agents interact in a visual town environment.
Language: TypeScript
Author: geezerrrr
AI Town
A deployable starter kit for building a virtual town where AI characters live, chat, and socialize. Provides the infrastructure for creating your own agent society experiments.
Language: TypeScript
Author: a16z-infra
Eliza Town
Enhanced version of AI Town with a map editor, extending the virtual agent town concept with additional customization capabilities.
Language: TypeScript
Author: cayden970207
Moltbook Town
A 2D pixel art social world for Moltbook AI agents. Provides a visual environment for observing agent social interactions and emergent behaviors.
Language: JavaScript
Author: oh-ashen-one
AstrTown
A community backend for active AI agent bots, built on top of a16z-infra/ai-town. Extends the agent town concept with a focus on persistent agent communities.
Language: TypeScript
Author: AstralSolipsism
Smallville
A video game implementation of the Generative Agents paper, bringing the "Interactive Simulacra of Human Behavior" concept into a playable environment.
Language: Java
Author: nmatter1
LLM Ant Farm
A Python reimplementation of the Stanford Generative Agents experiment using a local LLM. Demonstrates that synthetic society simulations can run without cloud API dependencies.
Language: Python
Author: grahamhome
Social Interaction & Emergent Behavior
Projects exploring how AI agents develop social strategies, relationships, and group dynamics.
ChatArena
Multi-agent language game environments for LLMs. Agents engage in structured social games including negotiation, debate, and deception — studying communication and collaboration dynamics.
Language: Python
Author: Farama-Foundation
GPTeam
An open-source multi-agent simulation where agents with distinct personalities live in a shared world, forming opinions, interacting, and exhibiting emergent group dynamics.
Language: Python
Author: 101dotxyz
HumanoidAgents
Platform for simulating human-like generative agents with daily routines, needs, and social interactions. Focuses on realistic human behavior simulation and social dynamics.
Language: Python
Author: HumanoidAgents
Sotopia
An open-ended social learning environment (ICLR 2024 spotlight). Agents interact in open-ended social scenarios, measuring social intelligence and emergent social strategies.
Language: Python
Author: sotopia-lab
Sociological & Political Modeling
Projects that use agent simulations to model real-world sociological and political phenomena.
Generative Agent Simulations of 1,000 People
LLM agents modeled on real interview data from 1,000 actual people, replicating individual attitudes and collective social behavior at population scale.
Language: Python
Author: StanfordHCI
SocioDojo
A lifelong learning environment for real-society dynamics. Agents learn to reason about societal phenomena through hypothesis generation and proof about social dynamics.
Language: Python
Author: chengjunyan1
WarAgent
LLM-based multi-agent simulation of world wars. Simulates international conflict and diplomacy between nation-state agents, modeling escalation dynamics and geopolitical decision-making.
Language: Python
Author: agiresearch
Multi-Agent Orchestration
Platforms for orchestrating and managing communities of AI agents.
Gas Town
Multi-agent workspace manager for orchestrating communities of AI agents working together.
Language: Go
Author: gastownhall
Related projects in the Gas Town ecosystem:
- gastown-remote — Always-on AI agent orchestration in the cloud via Tailscale
- gastown-gui — Web GUI for the Gas Town multi-agent orchestrator
- gasclaw — Single-container deployment: Gastown + OpenClaw + KimiGas with Telegram overseer
Large-Scale Simulations
Projects focused on simulating societies at scale for sociological and behavioral research.
AgentTorch
Differentiable agent-based modeling at scale — supports millions of agents. Used for pandemic simulation, economic modeling, and population-level social dynamics with LLM-powered agent behaviors.
Language: Jupyter Notebook
Author: AgentTorch
AgentSociety
Large-scale social simulation platform from Tsinghua University that uses LLM-driven generative agents to understand human behaviors and society. Supports simulations with up to 10,000 agents, each engaging in hundreds of daily interactions, capturing the dynamics of large-scale social systems.
Language: Python
Author: tsinghua-fib-lab
OASIS (Open Agent Social Interaction Simulations)
Social interaction simulation platform from the CAMEL-AI team that scales up to one million agents. Supports 23 different social actions (following, commenting, reposting, etc.) across simulated social media environments. Used as the underlying simulation engine for MiroFish.
Language: Python
Author: camel-ai
Domain-Specific Simulations
Agent society simulations applied to specific real-world domains.
Agent Hospital
Multi-agent medical simulation where AI doctor agents conduct multi-round dialogues with patient agents, propose diagnostic questions, recommend medical exams, and make diagnoses. Includes inspector agents and attending physicians who evaluate performance.
Language: Python
Author: wisdom-pan
AI Hospital
A simulacrum of a hospital with evolvable medical agents. LLMs collaborate as intern doctors for clinical diagnosis in an interactive multi-agent environment, modeling the social dynamics of medical team decision-making.
Language: Python
Author: LibertFan
Predictive Tooling
Tools that use multi-agent society simulations specifically for prediction and forecasting.
MiroFish
A swarm intelligence prediction engine that constructs parallel digital worlds from real-world seed information (news, policy drafts, financial signals), then populates them with thousands of agents with independent personalities and long-term memory. Agents interact across simulated social platforms, and their emergent opinion shifts and coalition formations are analyzed to produce structured prediction reports. Built on the OASIS simulation engine, scaling up to one million agents.
Language: Python
Author: 666ghj
Simulation Infrastructure
Tools and platforms for constructing synthetic society experiments.
Unreal Engine MCP
Control Unreal Engine 5.5+ through AI with natural language. Enables building 3D environments — towns, structures, and worlds — using AI-powered commands, providing infrastructure for visually rich agent society simulations.
Language: C++
Author: flopperam
Related Repos
Other resource collections by the maintainer exploring adjacent themes.
- AI-Decision-Eval-Projects — Resource list of AI projects focused on exploring AI as a tool for enhanced decision-making in personal and business contexts.
- AI-Synthesised-Perspectives — Collection of AI projects focused on synthesising perspectives for multi-perspective exploration.
- AI-Geopol-Projects — Collection of AI projects focused on exploring the ability of AI agents to act as proxies for real-world geopolitical actors, including simulated policy fora.
Contributing
Contributions are welcome! If you know of a project that belongs here, please open a pull request or reach out at public@danielrosehill.com.
When adding a project, please follow this format:
### [Project Name](https://github.com/owner/repo)
 
Short description of the project.
**Language:** Primary language
**Author:** [author](https://github.com/author)
---
Disclaimer
This is a non-exhaustive, best-effort collection. Inclusion does not imply endorsement, and omission does not imply any judgment. Project descriptions are summarized for brevity and may not capture the full scope of each project.
Maintained by Daniel Rosehill