Changes in Perceptions of Trust in Human-AI Interactions

This study examined how educational briefings about Artificial Intelligence (AI) hallucinations influence user trust, caution, and willingness to continue using AI systems. We recruited 51 participants through the online platform Prolific for an online questionnaire evaluating AI use and task-focused AI scenarios. After questions about trust and AI, participants read about common inaccuracies in Large Language Models. Participants showed a significant decrease in self-reported trust, indicating increased awareness of AI limitations. Contrary to expectations, perceptions of AI reliability and competence did not predict willingness to reuse AI in academic scenarios. Affective trust played a stronger role than cognitive trust, with perceived emotional support significantly predicting forgiveness of AI errors. Increased caution also predicted greater recalibration of trust, although willingness to continue using AI remained high. These findings suggest that educational interventions can promote calibrated trust in AI without discouraging continued use, highlighting the importance of transparency and context-sensitive design.