[Paper Review] What Do End-Users Really Want? Investigation of Human-Centered XAI for Mobile Health Apps

Katharina Weitz, Alexander Zellner+1 more|arXiv (Cornell University)|Oct 7, 2022

Persona Design and Applications4 citations

TL;DR

This paper proposes a user-centered persona framework to guide the design of human-centered explainable AI (XAI) for mobile health apps, based on an online survey of end-users' preferences for explanation styles and content in a stress monitoring application. The key contribution is the identification of three prototypical user personas—power, casual, and privacy-oriented—demonstrating that explanation preferences are significantly influenced by demographics, personality, and desired personalization, with interactivity and data transparency being critical for user acceptance.

ABSTRACT

In healthcare, AI systems support clinicians and patients in diagnosis, treatment, and monitoring, but many systems' poor explainability remains challenging for practical application. Overcoming this barrier is the goal of explainable AI (XAI). However, an explanation can be perceived differently and, thus, not solve the black-box problem for everyone. The domain of Human-Centered AI deals with this problem by adapting AI to users. We present a user-centered persona concept to evaluate XAI and use it to investigate end-users preferences for various explanation styles and contents in a mobile health stress monitoring application. The results of our online survey show that users' demographics and personality, as well as the type of explanation, impact explanation preferences, indicating that these are essential features for XAI design. We subsumed the results in three prototypical user personas: power-, casual-, and privacy-oriented users. Our insights bring an interactive, human-centered XAI closer to practical application.

Motivation & Objective

• To address the gap between researcher-driven XAI and end-user needs in mobile health applications.
• To investigate how end-users' demographics, personality traits, and explanation preferences vary across different styles and content types.
• To develop a user-centered persona concept grounded in empirical data to guide practical XAI design.
• To evaluate the impact of personalization, interactivity, and explanation content on user satisfaction and engagement.
• To support the transition of XAI from theoretical research to practical, user-friendly deployment in healthcare apps.

Proposed method

• Conducted an online survey with end-users to collect preferences on explanation styles (e.g., data-based, photo-based, live explanation) and content in a mobile stress monitoring app.
• Applied clustering and qualitative analysis to derive prototypical user personas based on demographic, personality, and preference data.
• Used a taxonomy of XAI techniques—visual, feature-based, knowledge-extraction, and example-based (e.g., counterfactuals)—to structure explanation types in the survey.
• Evaluated user responses on explanation satisfaction, willingness to interact, and preferences for personalization and data transparency.
• Designed and tested explanation types including 'ask-the-app' and 'live explanation' to assess interactivity effects.
• Mapped user responses to three distinct personas: power user (detailed, interactive), casual user (intuitive, concise), and privacy-oriented user (data protection-focused).

Experimental results

Research questions

• RQ1 How do end-users' demographics and personality traits influence their preferences for XAI explanation styles and content in mobile health apps?
• RQ2 Which explanation types (e.g., data-based, photo-based, live explanation) are most preferred, and what factors drive user satisfaction and engagement?
• RQ3 To what extent does personalization and interactivity in XAI systems affect users' willingness to use and trust the system?
• RQ4 How do users' self-reported desires for detailed explanations compare to their actual behavior when interacting with explanations?
• RQ5 Can a persona-based approach effectively represent and guide the design of human-centered XAI for diverse end-user groups in mobile health?

Key findings

• Users preferred data-based explanations over photo-based ones, indicating a stronger preference for factual, transparent information in health contexts.
• While users initially expressed a desire for detailed explanations, in practice, overly lengthy explanations were perceived as disruptive and annoying, especially when information was repeated.
• The 'ask-the-app' and 'live explanation' types received the highest approval, indicating strong user preference for interactive, dynamic explanation systems that allow parameter adjustment and question-asking.
• Personalization of explanations significantly increased user engagement, with participants willing to invest more time when explanations were tailored to their needs.
• Three distinct user personas emerged: power users (seeking detailed, interactive explanations), casual users (preferring intuitive, concise designs), and privacy-oriented users (prioritizing data protection and minimal data sharing).
• Users frequently asked 'Why?', 'Why not?', and 'How?' when receiving a stress prediction, highlighting the importance of interactive, question-driven explanation systems for user comprehension and trust.