Understanding XAI Requirements: A Comparative Study of Repetitive and Unique Decision Contexts

Abstract

This paper examines how explanation requirements vary between repetitive and unique AI decision contexts through an empirical study of two XAI prototypes. We analyze user interactions with an e-commerce moderation system and a communication monitoring assistant, finding that standardized visual explanations benefit routine tasks while adaptive approaches suit context-specific decisions. Our results suggest design patterns for balancing transparency with usability across different usage scenarios. While our small-scale study (n=8) and prototype-based methodology limit generalizability, particularly regarding real-world implementation challenges and long-term user behavior, our findings provide valuable initial insights into context-dependent explanation design. Further research is needed to validate these patterns at scale and address open questions about optimal confidence communication and security-transparency tradeoffs.