Do Mildly Ill Patients Really Need an Offline Doctor? Deciphering Health Choices from Artificial Intelligence Diagnostic Design and Trust Dynamics
基于双过程理论，通过两项实验（676名轻度患者和205名重症患者）发现，AI诊断的智能性、可解释性和独特性通过认知信任和情感信任影响轻度患者对线下医疗的需求，且医疗恐惧调节这一关系；但对重症患者，信任不减少线下医疗依赖。
Artificial intelligence (AI) technology offers huge potential for addressing the shortage of offline healthcare through providing accurate diagnostics and treatment recommendations for mildly ill patients. Based on the dual process theory, this paper theorizes the relative impacts of the three design attributes of AI diagnostics (i.e. intelligence level, explainability, and uniqueness consideration) on two types of trust (i.e. cognition-based trust and affect-based trust), the relative impacts of these two types of trust on people’s demand for offline healthcare, and the roles of medical fear in moderating the influence of two types of trust on demand for offline healthcare. In Study 1, we conduct a field experiment with 676 subjects who have mild illnesses. Using a variety of data analysis techniques, we draw several key conclusions: (1) the three attributes of AI diagnostics differentially influence cognition-based and affect-based trust; (2) cognition-based trust has a stronger negative effect on demand for offline healthcare than affect-based trust; and (3) medical fear weakens the relationship between cognition-based trust and demand for offline healthcare, while strengthening the relationship between affect-based trust and demand for offline healthcare. Robustness tests confirm these findings and rule out the influence of disease severity. In Study 2, using experimental data from 205 critically ill subjects, we find that critically ill patients can develop trust in AI diagnostic systems when these systems are well-designed in terms of intelligence, explainability, and uniqueness. However, this trust does not significantly reduce their reliance on offline medical resources, which means the substitutional relationship between AI diagnostics and offline healthcare cannot apply to critically ill patients. This study contributes to the design of AI diagnostics, trust-building mechanisms, and patients’ decision-making processes.