A Multidisciplinary Survey and Framework for Design and Evaluation of Explainable AI Systems

[subinium]

• https://arxiv.org/pdf/1811.11839.pdf
• https://github.com/SinaMohseni/Awesome-XAI-Evaluation

[subinium]

XAI Design Goals

• AI Novices
  • G1: Algorithmic Transparency
  • G2: User Trust and Reliance
  • G3: Bias Mitigation
  • G4: Privacy Awareness
• Data Experts
  • G5: Model Visualization and Inspection
  • G6: Model Tuning and Selection
• AI Experts
  • G7: Model Interpretability
  • G8: Model Debugging

Evaluation Measures

• M1: Mental Model
• M2: Explanation Usefulness and Satisfaction
• M3: User Trust and Reliance
• M4: Human-AI Task Performance
• M5: Computational Measures

XAI Design and Evaluation Framework

• Guideline1: Determine XAI System Goals
• Guideline2: Decide What to Explain
• Guideline3: Evaluate System Outcomes
• Guideline4: Decide How to Explain
• Guideline5: Evaluate Explanation Usefulness
• Guideline6: Design Interpretability Technique
• Guideline7: Evaluate Trustworthiness