home » the workshop » Normalization Sim
Simulation
NEW!Normalization Sim
An agent-based toy model of what happens when AI pulls everyone’s opinions toward the mean. Tracks stability, Gini inequality, opinion diversity and trust over time.
- access:
- public
- year:
- 2025
- built with:
- Python, NumPy, matplotlib
What happens to a society when an AI quietly nudges everyone’s opinions toward the average? What if it does the opposite and amplifies the extremes for engagement? This simulation is a hackable toy for poking at those questions, built to be visual enough for a Primer-style explainer.
Agents are split into a majority and a minority with unequal starting resources, and “outlier” movements push on altruism, trust, polarization and policy. One dial, lambda, sets how hard opinions are pulled toward the mean, and engagement mode flips it to push them apart instead. Every simulated day it records stability, inequality (Gini), opinion and affiliation diversity, average trust, and how many minority agents starve.
It’s a teaching toy, not a prediction engine: something for building intuition about tradeoffs.
~ highlights ~
- Command-line dials for population, days, normalization strength and minority share
- CSV metrics, matplotlib plots and an optional GIF of the agents over time