DebunkBot
A free, public tool that allows anyone to have a conversation with an AI trained to engage thoughtfully with conspiracy theories. Over 150,000 users and featured in the New York Times, Guardian, and Wall Street Journal.
Try DebunkBotWe investigate the psychology of belief—what makes people adopt, maintain, and occasionally revise their views on contested issues. Our work combines rigorous behavioral science with emerging AI technologies to develop scalable interventions for misinformation, polarization, and democratic health.
Research Areas
AI-Assisted Persuasion
Can AI systems engage productively with misinformation? We develop dialogue-based interventions using large language models to address conspiracy theories, vaccine hesitancy, and political misperceptions.
Conspiracy Beliefs
Why do people believe conspiracy theories? We study the cognitive, motivational, and social factors that predict susceptibility—and test interventions that actually work.
Political Psychology
The psychology of ideology, authoritarianism, and polarization. We're especially interested in whether psychological differences between left and right are real or artifacts of how we measure them.
Methods & Measurement
Better measurement for better science. We develop new scales, test existing ones, and advocate for adversarial collaboration as a norm in contested research areas.
Current Projects
Real-Time Conspiracy Intervention
Testing whether AI dialogues can reduce belief in conspiracy theories as they emerge and spread—before they become entrenched. Funded by DARPA.
Vaccine Dialogue
Personalized AI conversations that address parents' specific concerns about vaccines (HPV, COVID, childhood immunizations). Funded by the Physicians Foundation.
AI Deep Canvassing
Can AI replicate the remarkable effects of human deep canvassing conversations? We're testing this for political attitudes and policy preferences.
Climate Dialogue
Addressing climate skepticism and inaction through personalized AI conversations that meet people where they are.
AI Safety & Persuasion
Evaluating the persuasive capabilities of frontier LLMs and developing safeguards against misuse. Funded by Schmidt Sciences.
Lab Members
The lab is new and growing. Check back for updates!
TC
Thomas Costello
Director
?
Your Name Here
Prospective Member
Interested in joining?
I'm recruiting PhD students, postdocs, and research assistants who are excited about the intersection of AI and behavioral science. I value intellectual curiosity, methodological rigor, and a collaborative spirit.
Key Collaborators
David Rand (Cornell) · Gordon Pennycook (Cornell) · Hause Lin (MIT)
Funding & Support
DARPA
$329,485
Schmidt Sciences
$298,925
CSET
$800,000
Physicians Foundation
$150,000
Anti-Defamation League
$40,000