The Effect of Type of Explanation on Algorithm Appreciation: The Role of Risk Perceptions in Healthcare Decision-Making

Abstract

This study examines the impact of various types of Artificial Intelligence (AI) explanations—local, counterfactual, and global—on individuals' appreciation of algorithms in healthcare decision-making contexts. Using a scenario-based experiment involving 611 US-based participants, we take a risk perspective to examine how eXplainable (XAI) system credibility (risk probability) and perceived condition severity (risk severity) mediate the relationship between the type of explanation and algorithm appreciation. We also explore how decision-makers’ risk-taking propensity (risk perception) moderates these relationships. Participants assessed diabetes risk predictions for a hypothetical relative based on explanations generated by an XAI system. Findings reveal that the type of explanation significantly influences algorithm appreciation through the perceived severity of the condition, but not through the credibility of the XAI system. Importantly, the effects of the type of explanation vary with participants' risk-taking propensity. Hence, this research highlights the need for personalized, XAI strategies to maximize algorithm appreciation in high-risk healthcare decision-making contexts involving non-expert decision-makers.

Ähnliche Arbeiten