Table 3.
Assessments of the essential properties of XAI and the measures of explanation effectiveness.
| [29] | [30] | [31] | [32] | [33] | [34] | |||
|---|---|---|---|---|---|---|---|---|
| Essential properties of XAI |
Stakeholders of XAI | AI regulators | ||||||
| AI developers | ||||||||
| AI managers | ||||||||
| AI users | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ||
| Individuals affected by AI-based decisions | ✓ | |||||||
| Objectives of XAI | Explainability to evaluate AI | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | |
| Explainability to justify AI | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ||
| Explainability to improve AI | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ||
| Explainability to learn from AI | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ||
| Explainability to manage AI | ||||||||
| Quality of personalized explanations | Fidelity | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | |
| Generalizability | N/A | N/A | N/A | N/A | N/A | N/A | ||
| Explanatory power | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ||
| Interpretability | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ||
| Comprehensibility | ✓ | |||||||
| Plausibility | ✓ | ✓ | ✓ | ✓ | ✓ | |||
| Effort | ✓ | |||||||
| Privacy | ||||||||
| Fairness | ||||||||
| Measures of explanation effectiveness | Mental model | |||||||
| User satisfaction | ✓ | ✓ | ✓ | ✓ | ✓ | |||
| Trust assessment | ✓ | ✓ | ||||||
| Task performance | ✓ | ✓ | ✓ | |||||
| Correctability | ✓ | ✓ | ✓ | |||||