Trust Calibration in AI: The Role of Transparency and User Performance

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

A pilot study was conducted to investigate whether trust and reliability in an object detection system could be predicted by factors related to system transparency, user performance, and familiarity with AI technology. We examined the impact of perceived transparency—measured by the total number of clicks to access more information about how the system works and time spent on task—alongside the number of correct responses, AI knowledge, object detection AI familiarity, and perception of AI on users’ trust and reliability judgments. Twenty-seven undergraduate students interacted with an object detection system simulating AI functionality with varying interactive explainability. A multivariate analysis of variance (MANOVA) assessed the predictive value of these factors on trust and reliability outcomes. The omnibus MANOVA tests were insignificant for all predictors, indicating that the predictors did not collectively explain variance in trust and reliability. Correlational analysis revealed a significant negative correlation between accurately calibrated trust and the number of correct responses (r = -0.45, p < .05), suggesting that as participants performed better, their trust was less accurately calibrated to the system’s performance. In other words, participants with higher accuracy tended to have trust levels that did not appropriately match the system's actual performance. Despite the limited sample size, these results highlight the need for further research to understand factors that influence trust in AI systems. Future studies with larger samples should investigate these dynamics to optimize explainability and foster appropriate levels of trust in such systems.