Computational Multi-Agents Society Experiments: Social Modeling Framework Based on Generative Agents

This paper introduces CMASE, a novel framework that integrates generative agent-based modeling with virtual ethnography to transform researchers into embedded participants, enabling real-time social intervention, causal reconstruction of social phenomena, and predictive modeling with high empirical accuracy.

Hanzhong Zhang, Muhua Huang, Jindong Wang

Published Wed, 11 Ma
📖 4 min read☕ Coffee break read

Imagine you want to understand how a city works. Traditionally, sociologists have two main ways to do this:

  1. The Survey Method: They send out questionnaires or interview people. This is like asking people, "How do you feel about your neighbors?" It's accurate but slow, expensive, and you can't force people to act out specific scenarios without ethical issues.
  2. The Robot Method (Old School): They build computer simulations where "agents" (digital people) follow strict, pre-written rules like, "If it rains, go inside." This is fast, but the robots are too simple. They don't have personalities, they don't get angry, and they can't surprise you.

Enter CMASE: The "Virtual Reality" for Sociologists.

This paper introduces CMASE (Computational Multi-Agent Society Experiments). Think of CMASE not as a spreadsheet or a rigid robot factory, but as a massive, living, breathing video game where the researcher isn't just the player holding the controller—they are actually inside the game world.

Here is how it works, broken down with simple analogies:

1. The "Dungeon Master" Approach

In old simulations, the researcher was like a programmer writing code from the outside. In CMASE, the researcher is like a Dungeon Master (DM) in a tabletop role-playing game (like Dungeons & Dragons).

  • The World: You build a map with houses, parks, and shops (the "Environment Maker").
  • The People: You create digital characters with names, jobs, ages, and personalities (the "Agents").
  • The Twist: Unlike a video game where you just press buttons, these characters are powered by Large Language Models (AI). They can think, feel, and chat just like real humans. They don't just follow a script; they react to the world based on their "personalities."

2. The "Time Travel" Experiment

The coolest part of CMASE is that it lets researchers step inside the simulation.

  • Old Way: You set up a simulation, run it for a week, and look at the results at the end. It's like baking a cake, waiting for it to cool, and then tasting it.
  • CMASE Way: You are in the kitchen while the cake is baking. You can taste the batter, add more sugar, or even take a character aside for a private chat to ask, "Why are you so angry right now?"
  • The Goal: This allows researchers to test "What if?" scenarios in real-time. What if we add a new park? What if we change the tax laws? They can watch how the digital society reacts instantly, rather than waiting for the simulation to finish.

3. The "Emotion Engine"

To make the characters feel real, CMASE gives them an emotional dashboard.

  • Instead of just saying "Happy" or "Sad," the AI calculates three hidden numbers for every character: Valence (good vs. bad), Arousal (calm vs. excited), and Dominance (in control vs. submissive).
  • If a character is "High Arousal" and "Low Valence," they might be furious. The system uses this to stop them from making calm, rational decisions and instead makes them act impulsively. This mimics how real humans behave when they are stressed or angry.

4. The Real-World Test: The "Park Experiment"

To prove it works, the researchers recreated a famous real-world study about green spaces and social trust.

  • The Setup: They built two digital neighborhoods. One was a concrete jungle (low trees), and the other was a lush park (high trees). They put 10 AI characters in each.
  • The Result: Just like in the real world, the characters in the "green" neighborhood became more trusting and less suspicious of each other. The characters in the "concrete" neighborhood became more isolated and indifferent.
  • The Bonus: Because the researchers were "inside" the game, they could interview the AI characters. They found that creative characters (like a fashion designer) handled isolation differently than analytical characters (like a lawyer), revealing why the green space helped some people more than others.

Why Does This Matter?

Think of CMASE as a flight simulator for society.

  • Pilots don't learn to fly by crashing real planes; they use simulators to test dangerous situations safely.
  • Similarly, policymakers and sociologists can use CMASE to test new laws, community programs, or crisis responses in a safe, virtual world before trying them on real people.

In a nutshell: CMASE turns the dry, boring world of computer simulations into a living, interactive storybook where researchers can walk among the characters, ask them questions, and watch how their digital society evolves in real-time. It bridges the gap between cold data and the messy, emotional reality of being human.