General Intellectual Humility Is Malleable Through AI-Mediated Reflective Dialogue

A new study from researchers Mohammad Ratul Mahjabin and Raiyan Abdul Baten, published on arXiv, presents a landmark finding: General Intellectual Humility (GIH)—the trait of recognizing one's beliefs may be wrong—can be durably improved through structured AI conversation. In a randomized controlled trial with 400 participants, an intervention using an LLM to guide a reflective, Socratic dialogue tripled the rate of reliable individual improvement in GIH compared to a control group. Crucially, these gains showed no detectable decay over a two-week follow-up period, challenging the long-held assumption that such traits are largely stable and resistant to change.

The AI-mediated dialogue was carefully structured, progressing from conceptual understanding to applying, analyzing, and evaluating intellectual humility, culminating in participants generating novel, self-relevant scenarios. This scaffolded approach proved broadly effective, working across political affiliations and regardless of participants' baseline personality profiles. The study also found the intervention reduced the 'rank-order stability' of GIH, meaning it reshuffled individuals' relative standing on this trait, further evidence of its malleability.

This research opens a significant new frontier for applied AI, moving beyond task completion toward fostering foundational cognitive and social virtues. It suggests that carefully designed LLM interactions could be deployed at scale for educational, professional development, or even therapeutic purposes, aiming to improve reasoning, learning, and the quality of discourse in polarized environments. The success of this 'cognitive scaffolding' model points to a future where AI doesn't just provide answers, but actively helps build the intellectual character necessary to evaluate them.