Durably reducing conspiracy beliefs through dialogues with AI

Conspiracy theory beliefs are notoriously persistent. Influential theories propose they fulfill important psychological needs, thus resisting counterevidence. Yet previous failures in correcting conspiracy beliefs may be due to counterevidence being insufficiently compelling and tailored. To evaluate this possibility, we leverage developments in generative artificial intelligence and engaged 2,190 conspiracy believers in personalized evidence-based dialogues with GPT-4 Turbo. The intervention reduced conspiracy belief by ~20%. The effect remained 2 months later, generalized across a wide range of conspiracy theories, and occurred even among participants with deeply entrenched beliefs. Although the dialogues focused on a single conspiracy, they nonetheless diminished belief in unrelated conspiracies and shifted conspiracy-related behavioral intentions. These findings suggest that many conspiracy theory believers can revise their views if presented with sufficiently compelling evidence., , , # Durably reducing conspiracy beliefs through dialogues with AI

[https://doi.org/10.5061/dryad.v6wwpzh4h](https://doi.org/10.5061/dryad.v6wwpzh4h)

This repository contains data, code, and supplementary information. 

### Datasets

These data are required to run the RMarkdown file containing the analytic code associated with this manuscript. All data reported in the manuscript are included other than the raw conversational data, which potentially contains personally identifiable information and is therefore edited to remove identifiable features.

* **AllDataForPublication.PPI.csv**: Dataset containing all participant responses and results of NLP analyses (with personally identifying responses removed)
* **AllDataFiltered.Cluster.PPI.csv**: Dataset containing all participant responses and results of NLP analyses, only provided to exactly replicate cluster analysis in RMarkdown 
* **embeddings_allData.rds**: Text embeddings of conspiracy theories (do not change, merely output of text-e...