SalienTrack: providing salient information for semi-automated self-tracking feedback with model explanations

AI explanations improve user trust in automated feedback: The paper demonstrates that by explaining why certain tracked events are salient (informative), users can better understand and trust automated recommendations for behavior change, rather than receiving unexplained suggestions.

Not all data is equally useful for reflection: The research found that large volumes of self-tracked data often overwhelm users with irrelevant information. SalienTrack addresses this by filtering to show only the most meaningful tracking events, reducing cognitive burden during reflection on health behaviors.

Human judgment remains essential in AI systems: The framework combines both manual feedback (based on human expertise) and automatically generated feedback (from AI models), suggesting that responsible AI implementation in health contexts should preserve human oversight rather than fully automating decisions about what matters to users.

Explainability is critical for health applications: By making the AI model's reasoning transparent—showing users not just what information is salient but why—the system supports informed decision-making about behavior change, which is particularly important when AI influences personal health choices.

Real-world testing reveals practical design needs: The formative user study highlighted that implementing AI-driven feedback requires careful interface design and integration into existing user workflows, emphasizing that responsible AI deployment demands iterative testing with actual users, not just technical validation.