Tolerance in Polarized Societies: A Game Theoretic Agent-Based Model

TLDR

• Built an agent-based game-theoretic simulation in Python modeling ideological polarization and opinion dynamics.
• Implemented the Multiplicative Weights no-regret learning algorithm for enabling agents to adapt their strategies throughout the simulation.
• Designed utility functions balancing ideological distance and tolerance.
• Simulated various initial conditions to observe their impact on polarization dynamics.

Project Overview

This project simulates ideological polarization in a population of agents, each with a fixed belief and an adaptive strategy for tolerating disagreement. At each round, agents are randomly paired and decide how tolerant to be, earning utility based on both ideological distance and mutual tolerance.

We observed that despite initial biases, many agents converge toward higher tolerance over time. This emergence of cooperation highlights how simple rules and repeated interactions can produce complex societal dynamics.

Simulation Details

The simulation models 1,000 agents. Each agent:

• Holds a fixed ideology (from -1 to 1).
• Chooses a tolerance level (low, medium, or high) each round.
• Receives utility based on ideological distance and the joint tolerance decision.

Tolerance strategies evolve over time using no-regret learning — specifically, the Multiplicative Weights (MW) algorithm. Agents shift probability mass toward strategies that perform better in interactions.

Full Report and Code

What I Learned