Reyansh Badhwar, Zehaan Walji, Parshva Dave, Junho Park
Abstract Trust is a determinant of effective human-AI collaboration, shaping whether users appropriately accept or reject system recommendations. In safety-critical domains, miscalibrated trust can lead to unsafe overrides or uncritical acceptance of erroneous outputs. However, existing trust measurement approaches are often static, rely on subjective self-reports, and struggle to capture dynamic fluctuations during interactions. This study introduces a quantum-inspired computational framework for real-time trust estimation in human-AI teamwork. The model represents trust as a probabilistic state that evolves in response to multiple interaction variables (e.g., sentiment, task progress, response clarity, and accuracy proxies), enabling context-sensitive assessment without assuming fixed linear relationships. We instantiated the framework in a text-based interaction setting and compared model-estimated trust with 349 self-reported trust ratings from 15 participants. The quantum-based estimates closely tracked human trust ratings, with a mean bias of − 0.78 (on a 0-100 scale) and a strong correspondence with self-reported trust ratings. This work demonstrates the feasibility of applying quantum-inspired modeling to real-time trust assessment for human-AI interaction. The framework provides a proof of concept for AI systems that monitor and adapt to evolving user trust in domains such as autonomous driving, clinical decision support, and other high-stakes environments.