Wonji Doh

Wonji Doh

Auto-generated profile, based on publicly available information

Wonji Doh

University of Michigan – Dearborn, USA

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 September 2025 in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
Verified
Journal Article
Closed access
WDWonji Doh
Wonji Doh
YGYoungnoh Goh
Youngnoh Goh
SKSang-Hwan Kim
Sang-Hwan Kim
Unchecked overreliance on AI systems in human-AI teaming (HAT) can erode critical human judgment and compromise accountability, especially in high-stakes decisions. This paper introduces two integrated studies aimed at preserving balanced human agency in effective AI-assisted decision-making. First, it presents the Human-AI-System Concordance (HASC) Matrix, a diagnostic framework designed to systematically identify vulnerabilities to overreliance within human-AI collaboration (HAC) scenarios. Second, it then empirically evaluates Cognitive Forcing Functions (CFFs)—targeted interventions designed to mitigate overreliance and maintain calibrated trust—through a controlled experimental setting. These studies deliver actionable insights for developing balanced HAC environments, highlighting the conditions under when overreliance occurs and demonstrating how and which cognitive interventions can effectively mitigate it. This comprehensive approach can help to advance understanding of designing AI systems to better support effective human-centered decision-making.


7 September 2025 in Proceedings of the Human Factors and Ergonomics Society Annual Meeting
Verified
Journal Article
Closed access
SMShruthi Venkatesha Murthy
Shruthi Venkatesha Murthy
YLYejin Lee
Yejin Lee
WDWonji Doh
Wonji Doh
YGYoungnoh Goh
Youngnoh Goh
SKSang-Hwan Kim
Sang-Hwan Kim
This study investigates how various task allocation strategies and communication modalities influence Human-AI Team performance in dynamic dual-task environments. Utilizing a modified Ballas Task simulator, the research involved 32 participants with military or similar backgrounds to evaluate four task allocation strategies – Operator Monitoring, Action Split, Target Split, and Take Over. These strategies were assessed under different levels of transparency and communication modes while participants engaged in concurrent tracking and tactical assessment tasks. The results indicate that the effectiveness of task allocation is contingent upon both transparency and the communication modality employed. Operator monitoring consistently reduced workload and assessment time, while enhancing tracking performance, target accuracy, and situational awareness – particularly when paired with verbal communication under low transparency. In contrast, non-verbal communication improved situational awareness and reduced tracking errors in the action split and target split strategies. Although take over yielded fewer advantages compared to the other strategies, it still surpassed the performance of an all-human baseline. These findings provide actionable insights for enhancing human-AI collaboration in high-stakes, time-sensitive contexts, with relevance extending beyond military operations to a variety of dynamic dual-task environments.