AI Providing Explanations to Users Versus Requesting Explanations from Users & Uncertainty Avoidance

Abstract

Despite the growing availability of algorithm-augmented work, algorithm aversion is prevalent among employees, hindering successful implementations of powerful Artificial Intelligence (AI) aids. Here, we examined the effect of two distinct aspects of powerful AI-based Decision-Support Systems (DSSs): Providing explanations to the user, and Requesting explanations from the user, on employee supportive attitudes towards the system, and the moderating role of employee uncertainty avoidance (UA). Two experiments provided consistent causal evidence for the positive effects of (1) deploying an AI-based DSS that provides explanations, on all employees, but in particular, on high-UA employees; (2) a distinct preference, especially by high UA employees, for complete versus partial explanations; and (3) an enhanced preference by high UA employees, for AI-based DSSs that request users to provide explanations used only for internal/training system purposes (versus for explanations that will be documented). These effects were demonstrated by manipulating the explanation provision features and the requesting feature. Our studies suggest several implications for understanding, mitigating, and managing employee aversion towards the integration of powerful algorithmic aids as well as the design of such systems.

Ähnliche Arbeiten