Is radicalization reinforced by social media censorship? The question reframes a common assumption: that removing extremist content from a platform reduces the extremism. To test it, Justin Lane, 📝Kevin McCaffree, and 📝F. LeRon Shults built an 📝Agent-Based Modeling (ABM) simulation (arXiv:2103.12842, 2021) in which artificial agents hold and exchange beliefs across a social network, and moderation is applied through different mechanisms — softening dissent, suppressing posts, or banning individuals outright. Running the model under varied conditions lets the dynamics of belief change play out at population scale rather than being argued in the abstract.
The finding is counterintuitive and sometimes called the 📝Censorship Effect(s). Censoring content tends to harden radicalized views by decreasing the amount of dissent to which an agent is exposed, and centralized banning of individuals has the strongest radicalizing effect of all. When a platform de-platforms a voice, it does not delete the belief; it removes the friction that might have moderated it, leaving agents in tighter, more homogeneous clusters where extremity compounds. The intervention meant to reduce radicalization can reinforce it.
I built this model because I distrust intuitions about complex social systems, including my own. It is easy to assume that pulling extreme content makes a network healthier, but a simulation forces the assumption to survive contact with mechanism. What the work shows me, as a cognitive scientist, is that dissent is not noise to be filtered — it is the corrective signal a belief system needs. Strip it away and you do not get moderation. You get a quieter, more certain extreme.