Key points are not available for this paper at this time.
This study investigates how varying levels of display transparency in an AI system affect human performance, cognitive workload, and trust in a collaborative human–AI task. (1) Background: As AI-assisted systems become increasingly prevalent, understanding how transparency shapes human cognition and trust is critical for effective ergonomic design. The literature presents conflicting findings regarding whether greater AI transparency reduces or redistributes cognitive demand and whether it consistently enhances trust. (2) Methods: Twenty participants completed a Pictionary-style drawing task under two counterbalanced display transparency conditions—low transparency (Top-1 AI prediction) and high transparency (Top-5 predictions with similarity scores). Objective performance, eye-movement data, NASA-TLX workload ratings, and multi-dimensional trust questionnaires were collected. Condition-level measures were analyzed using a mixed ANOVA framework with non-parametric confirmatory tests where normality assumptions were violated. (3) Results: Increased transparency significantly enhanced visual attention to AI information. However, neither subjective workload nor trust showed statistically significant differences between conditions. Task completion time and error rates were likewise unaffected. Directional trends favored higher transparency for competence-related trust dimensions, though these did not reach significance. (4) Conclusions: Rather than simply reducing cognitive burden, display transparency may redistribute cognitive effort—replacing interpretive uncertainty with integrative processing demands. These findings suggest that display transparency alone is insufficient to produce measurable improvements in workload or trust, and that richer forms of explanatory transparency are needed to meaningfully support human–AI collaboration. Design implications for collaborative AI interfaces are discussed.
Park et al. (Sun,) studied this question.