A Survey of Explainable Knowledge Tracing

Explainability is critical for educational AI adoption: The paper emphasizes that while AI systems for knowledge tracing (predicting student learning) show strong performance, their lack of transparency reduces stakeholder trust and acceptance. This suggests that accuracy alone is insufficient—educators and students need to understand why the system makes its predictions.

Two main approaches exist for making AI systems interpretable: The survey identifies "transparent models" (designed to be understandable from the start) and "black box models" (complex systems with explanations added afterward). This distinction matters because transparent models may be inherently more trustworthy, while black box approaches require additional explanation techniques.

Current evaluation methods for explainable AI in education are inadequate: The paper notes a significant gap—there are insufficient standards for assessing whether explanations in educational AI systems are actually good or useful. The authors conduct experiments to address this, suggesting the field needs better frameworks for measuring explanation quality.

Explanations must be evaluated from multiple stakeholder perspectives: The paper emphasizes that evaluation methods should consider the needs of different users (teachers, students, administrators), not just technical metrics. This highlights that responsible AI requires understanding what different groups actually need to understand and trust the system.

Multiple explanation techniques exist but lack standardization: The survey reviews various methods for generating explanations at different stages (before and after model predictions), but the diversity of approaches without clear best practices suggests the field would benefit from more established guidelines for responsible implementation.