Uncovering the Hidden Biases in Personal Informatics

Personal Informatics (PI) systems, such as apps and wearables that help users track physical activity, sleep, heart rate, or stress, have become critical tools for self-monitoring and health research. As these systems increasingly drive personal and clinical decisionmaking, it's vital to understand how equitable and representative they really are. Real-world harms have already surfaced in adjacent domains: health sensors like pulse oximeters underperform on darker skin tones [1], and female speakers and non-US nationalities experience significant performance degradation in automated speaker recognition [2]. These failures aren't just technical - they're structural, human-centric, and societal. Yet, despite their growing influence, PI systems remain critically under-researched from a fairness and equity perspective [3]. Our research, detailed in [4], investigates this question by examining when, how, and for whom bias arises in the lifecycle of PI systems.