Selective Trust: Understanding Human-AI Partnerships in Personal Health Decision-Making Process

Research Background and Issues

• What problems or challenges did the authors identify?
  The authors point out that as artificial intelligence (AI) becomes increasingly embedded in personal health technologies, its potential to improve health decision-making through personalized recommendations has become significant. However, people's understanding of using AI to assist decision-making in personal health contexts remains limited, particularly regarding trust issues in health decisions related to physical activity.

• Why is this issue important?
  AI has the potential to play a crucial role in personal health management by enhancing decision-making processes through personalized recommendations. However, trust in AI determines its actual adoption and effectiveness, especially in sensitive areas involving personal health. Furthermore, miscalibrated trust could lead to misjudgments or reliance on incorrect AI suggestions.

• Research Motivation and Related Work
  This study aims to fill the current gap in understanding how people perceive AI-assisted personal health decision-making and analyze how these processes influence trust in AI. The authors draw on collaborative decision-making models between doctors and patients in clinical settings and trace how personal health technologies have evolved from data-tracking tools to more collaborative and autonomous systems.

Solutions

• What methods or solutions did the authors propose?
  The authors designed a physical activity decision-support tool called MoveAI, based on GPT-4.0, and investigated trust in AI-assisted health decision-making through online surveys and semi-structured interviews. MoveAI simulates the collaborative decision-making process between doctors and patients, employing a model of choice, option recommendation, and decision prompts.

• What are the innovative aspects of this solution?
  The innovations of this study include:

  • Utilizing a shared decision-making model (SDM) as the framework for AI-user interaction, facilitating personalized health decisions through selection, collaboration, and negotiation.
  • Exploring the interaction between traditional decision-making theories (e.g., intuitive and rational decision-making modes) and user trust in AI.
  • Proposing design directions for AI-assisted health technologies that adapt to diverse user decision-making styles.

• What are the implementation steps and key technologies used?

  1. Designing the AI interaction framework: Based on the SDM model, MoveAI guides users through information gathering, option discussion, and reaching final decisions.
  2. Research design and experiments: Conducting an online survey with 184 participants, followed by in-depth interviews with 24 participants.
  3. Data analysis: Combining quantitative methods (e.g., correlation analysis) and qualitative methods (e.g., thematic analysis) to explore user interaction data with MoveAI.

Research Findings

• What specific findings were obtained?

  1. MoveAI achieved a trust level comparable to that of medical professionals, but its recommendations were perceived as lacking personalization and depth.
  2. Trust in MoveAI depended on its ability to adapt to users' physical activity needs and the accuracy and transparency of its recommendations.
  3. Users' decision-making styles (e.g., rational, intuitive, or dependent) played a significant role in how they perceived and trusted AI.

• What advantages does it have compared to existing solutions?

  • MoveAI not only provides general recommendations but also attempts to deliver personalized advice based on dynamic dialogue with user input.
  • By simulating the shared decision-making process between doctors and patients, MoveAI explores how AI can be better integrated into collaborative health decision-making.

• What were the experimental or evaluation results?

  • Participants rated multiple dimensions of trust (e.g., technical competence, reliability, comprehensibility) moderately high.
  • Decision-making styles significantly influenced trust in MoveAI. Rational users valued comprehensibility more, while dependent and intuitive users rated reliability and technical competence higher.

• Limitations and Future Directions

  • Limitations:
    • The sample was overly concentrated on Western societies and highly educated populations, potentially limiting the generalizability to diverse cultural and educational backgrounds.
    • MoveAI failed to fully capture users' complex needs and external factors.
    • The experimental scenarios were simulated and did not measure real-life user behavior.
  • Future Directions:
    • Enhance the personalization and context-awareness of AI recommendations, moving beyond interaction to dynamically adjust content.
    • Develop AI tools capable of explaining the rationale behind recommendations and ensuring users "feel explained to."
    • Better address the specific impacts of decision-making styles on trust and incorporate these insights into the design of AI-assisted technologies to accommodate diverse user needs.

Through this study, the authors provide new perspectives on the field of AI-assisted personal health decision-making, emphasizing the importance of trust and personalization while offering clear directions for future research and practice.