Predicting Trust Dynamics with Dynamic SEM in Human-AI Cooperation

TL;DR

This work tackles the challenge of predicting human trust dynamics in human-AI collaboration to prevent over-trust and under-trust. It introduces dynamic structural equation modeling (DSEM) with an exploratory-static path design followed by time-series optimization to produce interpretable dynamic path diagrams, enabling direct prediction of Over/Under-Trust. Across drone- and autonomous driving–simulation tasks, the approach yields high accuracy (e.g., ACC around $0.90$–$0.98$) and outperforms conventional baselines such as $O(2^n)$-space Brute-force-inspired models like AR/ARMA/SARIMA. The method supports proactive trust calibration via cues and offers explainable insights into the factors driving trust dynamics, suggesting practical impact for safer and more efficient human-AI cooperation; the time-series optimization leverages a constrained search with hyperparameter $\eta$ to manage the exponential complexity $O(2^n)$.

Abstract

Humans' trust in AI constitutes a pivotal element in fostering a synergistic relationship between humans and AI. This is particularly significant in the context of systems that leverage AI technology, such as autonomous driving systems and human-robot interaction. Trust facilitates appropriate utilization of these systems, thereby optimizing their potential benefits. If humans over-trust or under-trust an AI, serious problems such as misuse and accidents occur. To prevent over/under-trust, it is necessary to predict trust dynamics. However, trust is an internal state of humans and hard to directly observe. Therefore, we propose a prediction model for trust dynamics using dynamic structure equation modeling, which extends SEM that can handle time-series data. A path diagram, which shows causalities between variables, is developed in an exploratory way and the resultant path diagram is optimized for effective path structures. Over/under-trust was predicted with 90\% accuracy in a drone simulator task,, and it was predicted with 99\% accuracy in an autonomous driving task. These results show that our proposed method outperformed the conventional method including an auto regression family.

Figures (7)

• Figure 1: Dynamic path diagram with path coefficients in Exp-1 and Exp-2.
• Figure 2: Dron-simulator for human-AI cooperative object recognition okamura2020adaptive.
• Figure 3: Prediction results of over-trust by DSEM. The blue solid line represents the actual proportion of over-trust, and the orange dashed line represents the predicted proportion.
• Figure 4: Results of prediction accuracy in a drone simulator task.
• Figure 5: Screen shot of drive simulation while a user was intervening.