Explainable AI for Well-Being Prediction From Lifestyle Data: 2-Study Design

Area of Science:

• Artificial Intelligence
• Public Health
• Behavioral Science

Background:

• Well-being is crucial for public health and social progress, influenced by dynamic determinants.
• Increasing behavioral data and AI prominence enable scalable well-being assessments.
• Explainable AI (XAI) is vital for user trust and reflection in AI-driven well-being tools.

Purpose of the Study:

• To assess the predictive power of modifiable lifestyle and contextual factors on subjective well-being.
• To investigate how different explanation modalities impact user satisfaction with AI-generated well-being feedback.

Main Methods:

• Developed a parsimonious regularized linear model for estimating individual well-being from lifestyle predictors.
• Conducted an experiment comparing explanation modalities (visual, interactive, textual, quantitative, population-comparison) against a control group.
• Evaluated user satisfaction with AI-generated assessments across different explanation formats with 1252 participants.

Main Results:

• Any form of explanation significantly increased user satisfaction compared to no explanation.
• Visual and interactive explanation modalities yielded the highest user satisfaction scores.
• Population-comparison feedback was least preferred and least effective in explaining AI assessments.

Conclusions:

• Well-being tools should integrate interpretable AI models with visual or interactive explanations.
• Focusing on actionable behavioral levers is more effective than emphasizing population norms.
• Findings provide design guidance for deploying XAI in well-being applications to boost user satisfaction.