Each glowing sphere is a simulated social-media user placed on a two-axis opinion map. Every simulation step, a recommender algorithm decides which other users' posts each agent sees — represented by the amber lines flashing between nodes. When that algorithm favours content from opinion-similar users, agents drift further toward their neighbours and away from everyone else, and the population fractures into isolated filter bubbles: dense, opinion-homogeneous clusters that rarely see each other's content at all.
The term "filter bubble" was coined by Eli Pariser in 2011. Later empirical work found the effect is real but not universal — it depends heavily on how strongly a platform's ranking system weights engagement/similarity versus deliberately injecting diverse content, which is exactly the trade-off this simulation lets you tune.
A 3D social network where every agent's position and colour encode a two-dimensional opinion, and a tunable recommender algorithm decides which peers' views each agent gets shown — watch the population fracture into isolated filter bubbles as bias increases.
Amber lines flash between agents to show live "recommendations." When the recommender favours opinion-similar peers, agents keep updating toward people who already agree with them, so clusters compact and drift apart — a visible feedback loop between ranking algorithms and opinion polarization.
Raise recommender bias to see homophily-driven clustering accelerate; raise diversity injection to counteract it with random exposure. Adjust social influence and simulation speed, then watch the polarization and bubble-count stats update live.
Researchers studying real recommender systems find that a small amount of injected randomness can meaningfully reduce echo-chamber formation without much hurting engagement — exactly the trade-off the diversity slider models here.