Trust in Artificial Intelligence (AI): An Integrative and Meta-Analytic Review

Abstract

Human trust is believed to be a critical precondition of successful AI implementations in the age of artificial intelligence (AI). However, there is a lack of consensus on the definition, operationalization, and measurement of AI-based trust across different associated fields of research. In this paper, we aim to add more clarity to current research on trust in AI with an integrative review and a meta-analysis. First, an integrative review based on 113 papers was conducted. The review summarizes conceptualizations and measurements of trust in AI across communities of practice, as well as generates a framework of antecedents of trust in AI. Then, a meta-analysis was conducted to quantitatively summarize the empirical findings relating trust in AI to these antecedents. Our analysis differentiates between AI-, human-, and context-related factors. A total of 48 included studies comprising 243 effect sizes were meta-analyzed. The moderating effects of AI embodiment (robotic, virtual, embedded), measurement (attitude, intention, behavior), and two study descriptors (i.e., study design and study setting) were also explored. Implications, limitations, and future directions of the review efforts are further discussed.

Ähnliche Arbeiten