Leveraging artificial intelligence to identify the psychological factors associated with conspiracy theory beliefs online

Jonas R. Kunst, Aleksander B. Gundersen, Izabela Krysińska, Jan Piasecki, Tomi Wójtowicz, Rafał Ryguła, Sander van der Linden, Mikołaj MorzyNature Communications 2024 · DOI

TL;DR

Researchers combined psychological surveys from 2,506 Twitter users with an AI-driven analysis of their 7.7 million real-world engagements. They discovered that while political extremity (both far-left and far-right) and belief in false information are significant risk factors, older age is the most consistent predictor of sharing COVID-19 conspiracy theories online.

What Problem Does This Paper Address?

Psychological research into conspiracy theories has historically suffered from an "intention-behavior gap." Most studies rely on self-reported surveys where participants rate their belief in a theory in a controlled lab setting. However, believing a theory is not the same as actively amplifying it in the "wild" of social media.

Furthermore, large-scale behavioral studies often lack the psychological "why"—they can see what is being shared but don't know the personality traits or political leanings of the people sharing it. This paper bridges these two worlds, linking deep psychological profiles with massive behavioral datasets to identify who is actually at risk of spreading misinformation.

Methods

The authors employed a sophisticated multi-stage pipeline to connect psychology to "Big Data":

  1. Psychological Profiling: 2,506 U.S. users completed surveys measuring demographics, political alignment, and traits like narcissism, "need for chaos," and denialism.
  2. Data Harvesting: With consent, the team collected 7.7 million engagements (likes, posts, replies, reposts) from these users’ timelines spanning the COVID-19 pandemic.
  3. AI Filtering & Classification:
    • Step A: They used a Sentence Transformer model (NLP) to calculate semantic similarity, filtering the 7.7M engagements down to those topically related to six specific COVID-19 conspiracy theories.
  4. Step B: They used GPT-3.5 to perform binary classification, determining if a tweet actually supported the theory or merely mentioned it.
  5. Statistical Modeling: The team used Bernoulli generalized linear mixed models (GLMMs) to account for the fact that data was nested within individuals (some users are much more active than others).

Datasets

  • Primary Behavioral Dataset: 7,713,506 Twitter (now X) engagements from Dec 2019 to Dec 2021.
  • Psychological Dataset: Self-reports from 2,506 U.S.-based participants recruited via CloudResearch.
  • Ground Truth Validation: A subset of 1,019 tweets was manually annotated by human experts to validate the AI’s classification accuracy (achieving high F1 scores across most theories).

Key Insights and Ideas

  • The Age Factor: Contrary to the stereotype of the "digital native" being more susceptible to online trends, older age was the most consistent predictor of conspiracy theory support across almost all models.
  • The "U-Shaped" Political Curve: Support for conspiracy theories wasn't exclusive to the right. For several theories, researchers found a curvilinear relationship: individuals at the extreme ends of both the left and right were more likely to engage with conspiratorial content.
  • The Intention-Behavior Disconnect: Interestingly, traits like narcissism and a "need for chaos"—which are frequently linked to conspiracy beliefs in lab settings—did not emerge as significant predictors of actual online sharing behavior in this study. This suggests that what drives a belief in private may differ from what drives amplification in public.
  • Overconfidence as a Risk: A "perceived ability to recognize misinformation" was positively associated with sharing one of the theories, suggesting that those who think they are "immune" to fake news might actually be more vulnerable to it.

Key Results

  • Frequency: Support was highest for the theory that "governments and politicians intentionally spread misinformation" (20,705 engagements).
  • Age Impact: In one model, older participants were 136% more likely to support the theory that governments were spreading false information compared to younger cohorts.
  • Political Extremity: Those self-identifying as "very left" or "very right" showed significantly elevated support for three out of the six conspiracy theories.
  • Predictive Power: The models explained between 7% and 22% of the variance in sharing behavior, indicating that while psychology matters, other factors (like platform algorithms or social network structure) likely play a massive role.

Limitations and Future Work

The authors acknowledge several key constraints:

  • Platform Specificity: The study focused exclusively on Twitter/X; behavior might differ on more visual platforms like Instagram or TikTok.
  • Causality: As a correlational study, it cannot prove that being "far-right" causes one to share conspiracy theories; it only shows they happen together.
  • The "Shadow" of Moderation: The data was collected while Twitter was actively banning COVID-19 misinformation. This likely "thinned" the dataset, meaning the results represent a conservative estimate of conspiracy support.
  • Future Directions: The researchers suggest that future interventions should specifically target older populations with digital literacy programs and explore how platform architecture (like "likes" and "shares") interacts with these psychological risk factors.

Reference: Kunst, J. R., et al. (2024). Leveraging artificial intelligence to identify the psychological factors associated with conspiracy theory beliefs online. Nature Communications, 15:7497. https://doi.org/10.1038/s41467-024-51740-9

Summary updated

Abstract

Given the profound societal impact of conspiracy theories, probing the psychological factors associated with their spread is paramount. Most research lacks large-scale behavioral outcomes, leaving factors related to actual online support for conspiracy theories uncertain. We bridge this gap by combining the psychological self-reports of 2506 Twitter (currently X) users with machine-learning classification of whether the textual data from their 7.7 million social media engagements throughout the pandemic supported six common COVID-19 conspiracy theories. We assess demographic factors, political alignment, factors derived from theory of reasoned action, and individual psychological differences. Here, we show that being older, self-identifying as very left or right on the political spectrum, and believing in false information constitute the most consistent risk factors; denialist tendencies, confidence in one's ability to spot misinformation, and political conservativism are positively associated with support for one conspiracy theory. Combining artificial intelligence analyses of big behavioral data with self-report surveys can effectively identify and validate risk factors for phenomena evident in large-scale online behaviors.
BibTeX
@inproceedings{kunst2024leveraging,
  author = {Jonas R. Kunst and Aleksander B. Gundersen and Izabela Krysińska and Jan Piasecki and Tomi Wójtowicz and Rafał Ryguła and Sander van der Linden and Mikołaj Morzy},
  title = {{Leveraging artificial intelligence to identify the psychological factors associated with conspiracy theory beliefs online}},
  booktitle = {Nature Communications},
  year = {2024},
  doi = {10.1038/s41467-024-51740-9},
}
Figure