Measuring and Understanding Trust Calibrations for Automated Systems: A Survey of the State-Of-The-Art and Future Directions

Document Title

Measuring and Understanding Trust Calibrations for Automated Systems: A Survey of the State-Of-The-Art and Future Directions

Document Information

• Subject Area: Human-Computer Interaction and Trust Calibration in Automated Systems
• Keywords: Trust Calibration, Automation, Empirical Research, Appropriate Trust, Trust Adjustment, User Studies

Research Background and Issues

• Identified Issues:

  1. Users may exhibit distrust (below system capability) or overtrust (exceeding system capability) toward automated systems.
  2. There is no unified understanding of how to align user trust with the actual reliability of systems (i.e., trust calibration) across different contexts.
  3. Research on trust calibration lacks standardized methodologies and has not been systematically reviewed.

• Importance of the Issues:

  • Trust calibration is critical to ensuring proper use of automated systems, avoiding inefficiency, unsafe practices, or erroneous reliance on systems.
  • Mismatched trust can lead to misuse or disuse of systems, negatively impacting system performance.

• Research Motivation and Related Work:

  • Automation has been widely applied in various fields, such as medical diagnosis, autonomous driving, and workforce evaluation.
  • Although there is extensive literature on the topic (over 1,000 papers), a structured and comprehensive review summarizing methods, challenges, and future research directions for trust calibration is lacking.

Proposed Solution

• Proposed Approach:

  • Conduct a review of 96 empirical studies from various fields, focusing on interventions, experimental designs, operationalization of system capabilities, measurement methods, and calibration outcomes related to trust calibration.
  • Derive four key dimensions of trust calibration: external/internal calibration, appropriate/inappropriate trust calibration, static/adaptive calibration, and capability-oriented/process-oriented calibration.

• Innovative Aspects of the Solution:

  1. Provides an overview of the current state of trust calibration research and proposes a clear classification system.
  2. Differentiates between static and dynamic trust calibration methods and explores their effectiveness.
  3. Highlights potential pitfalls of trust interventions leading to miscalibration and suggests improvements for future research.

• Implementation Steps and Key Techniques:

  • Perform extensive literature searches across multiple databases, including Google Scholar, Scopus, and ACM.
  • Select relevant studies, with a particular focus on whether they explicitly address the alignment between trust and actual system capabilities.
  • Conduct quantitative and qualitative analyses of the selected studies, examining experimental design, sample composition, task types, trust measurement methods, etc.

Research Outcomes

• Specific Findings:

  1. Trust Calibration Interventions: Identifies common interventions, such as prior information, feedback, transparency, alerts, and explanatory mechanisms.
  2. Diversity of Calibration Methods: Summarizes three main approaches to trust measurement and calibration outcome evaluation: relative measurement (comparing trust levels), correlation analysis (relationship between trust and system capability), and behavioral measurement.
  3. Proposed Four Calibration Dimensions: External-internal calibration, appropriate-inappropriate calibration, static-dynamic calibration, and capability-oriented-process-oriented calibration.

• Advantages Compared to Existing Research:

  • Provides a comprehensive synthesis of scattered trust calibration studies.
  • Introduces trust calibration dimensions, offering a clearer description of different intervention types and their effects.
  • Emphasizes and distinguishes between appropriate and inappropriate trust, enhancing the depth of academic discussions.

• Experimental or Evaluation Results:

  • Users are generally able to perceive system capability levels but often respond slowly or irrationally to changes in capability, necessitating specific interventions.
  • Increasing system transparency (e.g., displaying system confidence, providing explanations) can enhance trust but may also increase cognitive load or lead to misleading trust.

• Limitations and Future Directions:

  • Limitations:
    1. Research methods are limited to matching system capability with user trust perception, without fully exploring the dynamics of trust behavior.
    2. Definitions and measurements of trust calibration are not yet standardized.
    3. The review is restricted to published English-language literature.
  • Future Directions:
    • Conduct more applied research, such as in high-risk medical contexts.
    • Explore adaptive trust calibration mechanisms to better align user expectations with actual system use.
    • Incorporate trust calibration into more complex, multi-task environments, optimizing algorithms based on user behavior feedback.

In summary, this document provides a solid foundation for trust calibration research through a systematic literature review. It also highlights numerous directions for future research to meet the demands of rapidly evolving automation applications.