A chatbot can reduce prejudice against trans people — at least temporarily. Because prejudice toward marginalized groups can be reduced through storytelling and empathetic, one-on-one conversations.
This is according to a study, “Can AI help reduce prejudice? Evaluating the effectiveness of AI-powered personalized persuasion on support for transgender rights”, that was published in PNAS Nexus.
The researchers noted that prior work has shown that these approaches are especially effective when paired with moral matching, or tailoring messages to reflect an individual’s core values. However, such interventions are time-intensive and difficult to scale.
John Holbein and colleagues, therefore, tested whether artificial intelligence can overcome these barriers by delivering personalized, value-aligned conversations at scale.
In a preregistered experiment with more than 2,800 American adults, participants first completed a Moral Foundations Questionnaire to identify their core values. Roughly half of the participants were assigned to interact with a GPT-4o-powered chatbot designed to deliver persuasive messages tailored to those values, while the rest served as a control group.
The chatbot was instructed to follow structured prompts that emphasized empathy, perspective-taking, and moral alignment, for example, highlighting shared values such as fairness or compassion.
Participants who engaged with the chatbot showed significant short-term increases in support for transgender rights across multiple measures compared to controls. On a 0–100 feeling thermometer, support increased by approximately 3.94–5.19 points relative to the control group.
The intervention also improved broader attitudes and willingness to take supportive actions. However, these effects were not durable. Follow-up results one week later showed substantial attenuation, with most effects no longer statistically significant.
The results demonstrate that brief, AI-mediated conversations grounded in moral values can reduce prejudice, at least in the short term, relative to no conversation at all. As such, AI may serve as a scalable complement to community-based and structural approaches aimed at fostering inclusion and reducing bias, according to the authors.
