Knowledge-based XAI through CBR: There is more to explanations than models can tell

Domain expertise is essential for meaningful explanations: The paper demonstrates that AI models can achieve high accuracy using limited data perspectives, but explaining decisions to humans requires incorporating diverse viewpoints. This suggests that responsible AI systems need input from subject matter experts and domain knowledge, not just model performance metrics.

One-size-fits-all explanations are insufficient: The research emphasizes that "the audience of explanations can entail multiple stakeholders and span diverse perspectives," indicating that different users (regulators, patients, engineers, etc.) may need different types of explanations for the same AI decision. Organizations deploying AI should tailor explanations to their specific audiences.

Better explanations require additional data beyond model training data: The paper shows that adding domain knowledge features improves the ability to select appropriate explanation categories. This suggests that achieving responsible AI requires investment in curating explanation-specific data and knowledge bases, not just relying on the data used to train the original AI system.

Explainability should be treated as a separate technical problem: By framing explanation generation as its own supervised classification task (rather than extracting explanations from model internals), the research suggests that organizations should develop dedicated systems for producing explanations, similar to how they develop their primary AI models.