Thomas H. Costello
Carnegie Mellon University
5 works in GUIDE
In GUIDE
2026
Dialogues with Large Language Models Reduce Conspiracy Beliefs Even When the AI Is Perceived as HumanBoissin et al. · Forthcoming in PNAS Nexus
2026
Just the Facts: How Dialogues with AI Reduce Conspiracy BeliefsCostello et al. · Working paper
2026
Reducing belief in conspiracy theories as they unfold using large language modelsCostello et al. · National Academy of Sciences
2026
Addressing climate change skepticism and inaction using human-AI dialoguesCzarnek et al. · PsyArXiv Preprints
2026
Personalized AI Dialogues and Parents' HPV Vaccine Intent: A Randomized TrialXu et al. · PsyArXiv Preprints
Thomas H. Costello is a researcher at Carnegie Mellon University's Department of Social and Decision Sciences. His research focuses on belief formation, conspiracy theories, and the psychology of reasoning, particularly how generative AI can help individuals update false beliefs through personalized dialogue and evidence-based counterargument.
Sources in this wiki
- Xu et al. (2026) — Personalized AI Dialogues and Parents' HPV Vaccine Intent: A Randomized Trial — Demonstrates that a single dialogue with GPT-4 Turbo significantly increases parental HPV vaccination intentions; the AI's responses were highly accurate (99.6% of 15,559 claims fact-checked correct)
- Czarnek et al. (2026) — Addressing climate change skepticism and inaction using human-AI dialogues — Demonstrates that personalized LLM-driven dialogues addressing individuals' specific climate concerns substantially reduce skepticism and increase willingness to take climate action, outperforming consensus messaging
- Costello et al. (2026) — Reducing belief in conspiracy theories as they unfold using large language models — Demonstrates that LLM-based dialogue effectively reduces conspiracy belief about emerging events in real-time; identifies seven persuasion strategies employed by the model and provides evidence for downstream effects on related conspiracy beliefs
- Boissin et al. (2026) — Dialogues with Large Language Models Reduce Conspiracy Beliefs Even When the AI Is Perceived as Human — Tests whether the debunking effect depends on perceived messenger identity (AI vs. human expert); finds no significant difference, suggesting the power of AI dialogue comes from evidence quality, not source characteristics
- Costello, Pennycook & Rand (2026) — Just the Facts: How Dialogues with AI Reduce Conspiracy Beliefs — Large-scale experiments (N≈2,100) showing that AI-mediated dialogue significantly reduces conspiracy beliefs through factual counterargument; demonstrates that facts are necessary and sufficient (rapport and narrative are negligible); shows conspiracy beliefs are durably updatable contrary to decades of prior assumptions
Areas of research
- Conspiracy theories and conspiracy thinking
- Belief updating and resistance to evidence
- Generative AI and personalized interventions
- Psychology of reasoning and misinformation