Your Model Is Unfair, Are You Even Aware? Inverse Relationship Between Comprehension and Trust in Explainability Visualizations of Biased ML Models
2025-08-04
Summary
The study examines how the design of explainability visualizations affects user comprehension, perception of bias, and trust in machine learning (ML) models. An inverse relationship was found: as comprehension increases, trust decreases, primarily because more comprehensible visualizations reveal biases in ML models that reduce trust. This relationship was confirmed through experiments manipulating visualization design and model fairness.
Why This Matters
As ML systems become integrated into critical areas like healthcare and finance, understanding their behavior and biases is crucial for stakeholders. Explainability visualizations play a vital role in this process, but their design can significantly impact user perception and trust. Recognizing the trade-offs between comprehension and trust can guide the development of more effective visualization tools, ensuring that users are both informed and appropriately critical of ML decisions.
How You Can Use This Info
Professionals using ML systems should focus on explainability tools that balance clarity and simplicity to improve understanding while being aware of potential biases. This knowledge can inform training sessions and communication strategies when introducing ML tools to non-expert users. Additionally, when evaluating ML models, consider how visualization designs might influence perceptions of fairness and trust, and adjust accordingly to facilitate more responsible decision-making.