True knowledge of LLMs fuels trust

Large Language Models (LLMs) are increasingly integrated into human decision-making, information-seeking, and social communication. However, human trust in these systems is dynamic and context-sensitive, often shaped not only by performance accuracy but also by the model’s calibration ability—its confidence and self-assessment regarding correctness. While prior research has examined trust in AI from behavioral and ethical perspectives, the cognitive and neural mechanisms underlying trust calibration remain largely unexplored. The present study combines behavioral and neuroimaging methods to investigate how LLMs’ accuracy and calibration styles influence human trust dynamics. We propose that trust adjustment toward LLMs is governed by distinct neural circuits integrating emotional, cognitive control, and social evaluative processes. By using a within-subjects design and event-related fMRI, this study aims to identify the behavioral patterns and brain network dynamics that underlie human trust recalibration. update:To extend our investigation of these mechanisms, we have supplemented the original experimental paradigm with additional behavioral measures administered following the fMRI session. While maintaining the core within-subjects design and event-related fMRI protocol, participants now complete a comprehensive behavioral task battery that replicates the original trust calibration procedure while incorporating new subjective state assessments. Specifically, we measure participants' perceived control, happiness, and stress levels during trust interactions with LLMs exhibiting varying accuracy and calibration styles. These additional behavioral indices allow us to examine how affective and agency-related states mediate the relationship between LLM calibration behavior and subsequent trust recalibration. By integrating these multidimensional behavioral measurements with neural data, this study aims to identify both the behavioral patterns and brain network dynamics that underlie human trust recalibration, with particular attention to how emotional and control states modulate trust adjustment processes.