Skip to content

README.md

Latest commit

 

History

History
76 lines (54 loc) · 3.35 KB

README.md

File metadata and controls

76 lines (54 loc) · 3.35 KB

Epistemic Communities under Active Inference

This repository contains the code and simulations for the paper "Epistemic Communities under Active Inference" (Albarracin et al., 2022).

Paper Link: https://pubmed.ncbi.nlm.nih.gov/35455140/

Overview

This codebase implements a multi-agent simulation of opinion dynamics using the Active Inference framework. The model investigates the cognitive and social mechanisms behind confirmation bias and echo-chamber formation.

In this model:

  • Agents are active inference systems that sample information to minimize free energy (uncertainty).
  • Agents generate observable social claims (e.g., "tweets") supporting one of two mutually exclusive ideas.
  • Agents choose which neighbors to "listen" to based on the expected epistemic value (information gain) of their observations.
  • Confirmation Bias emerges as agents preferentially sample information that validates their existing model of the world, leading to self-reinforcing belief structures and the consolidation of epistemic communities.

Key features of the simulation include:

  • Generative Models: Agents maintain beliefs about an abstract "Idea" and the beliefs of their neighbors.
  • Active Sampling: Agents select actions (who to attend to) to resolve uncertainty.
  • Network Dynamics: The structure of social attention evolves as agents change who they sample from over time.

Setup

  1. Install dependencies: 
    pip install -r requirements.txt

Reproducing Experiments

To run the main experiments and generate plots, run the reproduction script:

python3 reproduce_experiments.py

The results (plots) will be saved in the results/ directory.

Individual Demos

You can also run individual demos as modules:

  • Active Inference Demo (3 agents):

    python3 -m demos.active_inference_demo

    Generates plots showing belief updates and sampling probabilities for a simple 3-agent scenario. This demonstrates the basic dyadic interaction and sampling logic.

  • Inference Demo:

    python3 -m demos.inference_demo

    Runs inference simulations and plots belief switching rates across different parameters (e.g., Epistemic Confirmation Bias vs. Environmental Determinism).

  • Network Demo:

    python3 -m demos.network_demo

    Simulates a larger network of agents (e.g., N=10) and plots the evolution of beliefs over time, showing how communities of belief emerge or fracture.

Directory Structure

  • model/: Core agent and generative model implementation.
    • agent.py: The Agent class implementing active inference.
    • genmodel.py: The generative model (matrices A, B, C, D, E).
    • pymdp/: Python library for MDPs and active inference math.
  • demos/: Executable scripts for running simulations.
  • analysis/: Tools for analyzing simulation results.
  • simulation/: Helper functions for initializing and running network simulations.
  • results/: Output directory for plots generated by reproduce_experiments.py.

Citation

If you use this code, please cite the original paper:

Albarracin, M., Demekas, D., Ramstead, M. J. D., & Heins, C. (2022). Epistemic Communities under Active Inference. Entropy, 24(4), 476. https://doi.org/10.3390/e24040476