The Centers and Margins of Modeling Humans in Well-being Technologies

Research Background and Issues

• What problems or challenges did the authors identify?
  • This paper explores the challenges of human modeling in health and well-being applications using machine learning technologies, particularly how these technologies embed hidden assumptions that create tensions between user experience and technical design. The core issues include:
    • Current machine learning models in health applications often embed stereotypical assumptions about the human body and experience (e.g., bodily regularity, the binary nature of health and illness).
    • The field of human-centered machine learning tends to overlook the social values of machine learning models and their ability to address the diverse needs of users.
    • Data-driven design focuses solely on algorithmic performance, failing to effectively capture the complex entanglement between human and technical systems.
• Why is this issue important?
  • As health technologies increasingly permeate daily life, their models can influence users' perceptions of their bodies and health states. If these models fail to account for the diversity of users' bodies and experiences, they may exacerbate inequities within technical systems.
• Research motivation and related work:
  • Drawing from critical technical practice and posthumanist philosophy, the study explores how to deconstruct and redesign current machine learning models to better capture the complexity of users' bodies and health experiences inclusively.
  • It proposes a design perspective based on decentering, questioning the issues of universalist assumptions and normative concepts of the body embedded in existing models.

Solutions

• What methods or solutions did the authors propose?
  • Through three case studies of health-related machine learning applications (Clue menstrual tracking, social fitness apps, and the Oura health tracking ring), the authors identified central assumptions and marginalization phenomena in current modeling practices.
  • By integrating critical technical practice methods and posthumanist theory, and using "Agential Realism" as a philosophical foundation, the authors redefined the encoding and design framework of machine learning models for the "human body."
• What is innovative about this solution?
  • Adopting a decentering approach: Reinterpreting and redesigning traditional human-centered models to challenge universalist and stereotypical notions of the body, such as regular cycles and the binary nature of health states.
  • Incorporating key philosophical insights: Introducing the theory of "Agential Realism" to position the entanglement of technology and human experience as central to modeling, rather than treating technology as a neutral tool.
• What are the implementation steps and key technologies used?
  1. Case selection: Choosing three typical health applications to reflect on user experiences and technical design, revealing hidden technical assumptions.
  2. Step 1: Experience collection: Using first-person experiences (e.g., autoethnography) and technical documentation analysis to identify points of disconnection in user-technology interactions.
  3. Step 2: Design reflection: Based on Agential Realism theory, questioning how these assumptions define which "bodies" are recognized and prioritized in technical models.
  4. Step 3: Envisioning alternatives: Proposing new modeling pathways, such as datasets supporting users with irregular cycles and creatively designing state transition models.

Research Outcomes

• What specific outcomes were achieved?
  • The study clarified how current machine learning modeling practices in health technologies embed centralized assumptions about "humans" and redefined the role of the human body within technical systems.
  • It proposed machine learning design directions from a posthumanist perspective, including redesigning predictive models, considering complex bodily rhythms, and designing interactive systems that allow users to actively participate in data interpretation.
• What advantages does it have compared to existing solutions?
  • This study goes beyond traditional data collection strategies, which merely increase database diversity to improve model performance. Instead, it adopts a deconstructive approach to redefine assumptions that classify health states as "normal" or "abnormal."
  • It offers a new perspective, transforming machine learning from a simple representation to a complex interaction with users' real-time and dynamic relationships.
• What were the experimental or evaluation results?
  • Through analysis of the three case studies, the authors demonstrated how the central assumptions of machine learning models lead to disconnections in user experience and showed that redesigning these assumptions can help create more inclusive health technologies.
• Limitations and future directions:
  • Research limitations:
    • As the research methods rely on the author's first-person experiences, the conclusions may not fully generalize to all users.
    • Only three applications were selected as case studies, leaving room for further exploration of a broader range of health technologies.
  • Future directions:
    • Developing collaborative machine learning models capable of handling concept drift to adapt to the co-evolution of user-technology relationships.
    • Exploring how design tools can continuously reveal and address potential exclusivity issues within the posthumanist perspective.