Academia in Silico

An agent-based model (ABM) of the scientific community, designed to simulate the production, verification (replication), and selection of scientific knowledge. This project explores how institutional incentives (selection for novelty vs. truth) and publication biases affect the reliability of the scientific record.

Project Structure

• code/: Contains the Python source code for the simulation.
  • main.py: The primary entry point for running single simulations.
  • experiments.py: A script for running batch experiments with custom configurations.
  • simulation/: The core logic package (agents, environment, peer review).
• text/: Contains the Quarto source files for the accompanying thesis.
  • text.qmd: The main manuscript.

Installation

The simulation requires Python 3. It depends on scipy for statistical distributions and matplotlib for visualization.

It is recommended to use the provided virtual environment or create a new one:

cd code
python3 -m venv venv
source venv/bin/activate
pip install scipy matplotlib

Running the Simulation

You can run a single simulation instance using main.py. This is useful for testing specific parameters or watching the system evolve in real-time.

cd code
# Run via the virtual environment
./venv/bin/python main.py --visualize

Key Configuration Flags

Flag	Description	Default
--researchers	Number of active agents	500
--effects	Number of discoverable effects	100000
--timesteps	Duration of simulation	300
--selection	Selection Pressure (0=Truth, 1=Novelty)	1
--bias	Publication Bias (0=None, 1=Weak, 2=Strong)	1
--replication_journal	Enable specialized replication journal (0=No, 1=Yes)	1
--max_replications	Hard limit on replications per effect	None
--visualize	Enable real-time dashboard	False

Example: Run a "Truth-Seeking" community with no publication bias for 500 steps:

./venv/bin/python main.py --selection=0 --bias=0 --timesteps=500 --visualize

Running Experiments

To run multiple conditions consecutively (e.g., to reproduce the results in the paper), use experiments.py.

1. Open code/experiments.py and define your experiments in the experiments list:

   experiments = [
       {
           "name": "baseline_novelty",
           "flags": "--selection=1 --bias=1 --timesteps=300"
       },
       {
           "name": "intervention_truth",
           "flags": "--selection=0 --bias=1 --timesteps=300"
       }
   ]

2. Run the script:

   ./venv/bin/python experiments.py --output=my_results

Results (CSV data and plot images) will be saved in subdirectories under my_results/ (e.g., my_results/baseline_novelty/).