New study reveals 5 hidden inequity sources in AI wellbeing sensing

A new paper accepted to AIES 2026 digs into the fairness risks of passive sensing for wellbeing—systems that continuously collect smartphone and wearable data to infer psychological states like depression or cognitive load. The authors, led by Han Zhang from the University of Washington, conducted semi-structured interviews with 14 researchers and practitioners across five countries. They found that fairness research in this domain has been limited to post-hoc, identity-based comparisons, missing deeper inequities that arise from the unique characteristics of passive sensing: heterogeneous hardware, indirect behavioral inference, and longitudinal deployment. The study empirically identifies five situated sources of inequity—such as comfort with monitoring and behavioral regularity—that systematically shape fairness risks beyond simple demographic attributes. They also catalog 15 specific fairness risks and corresponding mitigation strategies across the full system lifecycle, from study design to deployment.

The paper goes further by identifying structural barriers that constrain fair practice in reality, such as lack of funding incentives and publication norms. The researchers argue that enabling fair passive sensing requires not only individual researcher effort but also ecosystem-level governance support from funders, publication venues, and deploying institutions like hospitals and universities. This is a critical call to action as these systems are increasingly adopted in high-stakes settings where biased inferences could lead to unequal treatment or resource allocation. The findings provide a concrete framework for practitioners to audit and redesign their sensing pipelines for equity, moving beyond checklist-style fairness audits to a more holistic, lifecycle-aware approach.