The Flow of Trust: A Visualization Framework to Externalize, Explore, and Explain Trust in ML Applications.

Area of Science:

• Human-Computer Interaction
• Machine Learning
• Trustworthy AI

Background:

• Trust in machine learning (ML) applications is currently an implicit cognitive process.
• Lack of explicit trust mechanisms hinders feedback and communication in ML systems.
• Existing ML workflows lack methods to formalize or externalize user trust.

Purpose of the Study:

• To propose a conceptual framework for developing visual interactive techniques.
• To enable formalization and externalization of trust in ML workflows.
• To facilitate effective building and communication of trust across ML stages.

Main Methods:

• Formulating research questions on trust in ML.
• Developing a typology of trust objects, issues, and (mis)trust reasons.
• Designing formalisms for machine-readable trust representation.
• Investigating user interaction methods for trust expression (text, drawing, marking).
• Creating visual interactive techniques for trust representation and exploration.

Main Results:

• A conceptual framework for visual interactive trust formalization in ML.
• Identification of key research directions for trust in ML.
• Proposed methods for users to express trust states interactively.
• System-facilitated communication of trust states.
• Techniques for visualizing and exploring trust across the ML pipeline.

Conclusions:

• The proposed framework provides a foundation for developing novel interactive visualizations.
• Externalizing trust can enhance user understanding and interaction with ML systems.
• Further research is needed to implement and evaluate these techniques in real-world ML applications.