海角黑料

Using AI to reduce conspiracy beliefs around the globe

A central puzzle in behavioral science is the persistence of conspiracy theories despite disconfirmatory evidence. While personalized AI dialogues reduce these beliefs, the underlying mechanisms—belief updating via dialectical argumentation—reflect WEIRD cognitive norms. We tested the cross-cultural generalizability of this paradigm with 6,887 participants from six of the world’s most populous countries: the United States, Pakistan, India, Indonesia, Brazil, and Nigeria. Participants engaged in LLM-led debunking conversations or a neutral control. Results reveal a significant negative effect of the treatment on conspiracy belief in all six countries. Furthermore, the effect is relatively small in the United States, suggesting that AI-mediated debunking is broadly effective and may be even more impactful in non-Western contexts.