Transparent Adaptive Learning via Data-Centric Multimodal Explainable AI

Adaptive learning systems need better transparency: Current AI-driven educational systems make decisions about what students learn without clearly explaining how or why, creating a "black box" problem that limits trust and understanding from educators, students, and parents.

One-size-fits-all explanations don't work: The paper identifies that traditional explainability techniques focus on technical outputs rather than what different users actually need to understand—teachers need different information than students or parents—suggesting explanations must be customized by user role and goals.

Explainability should be a conversation, not a label: Rather than treating explanations as static technical outputs, the framework redefines explainability as "a dynamic communication process," meaning AI systems should adapt their explanations based on user feedback and comprehension, similar to how a good teacher adjusts their teaching.

Multiple formats improve understanding: By combining traditional XAI techniques with generative AI and personalization, the framework generates "multimodal" explanations (text, visuals, examples tailored to individual learners), which can help different people grasp how educational AI decisions are being made.

Key challenges remain unresolved: The paper highlights that critical questions about accuracy, fairness, and personalization in explainable educational AI still need research, indicating this is an emerging area where responsible AI practices are still being developed.