An Appraisal-Based Approach to Human-Centred Explanations

Explanations need to match how humans actually think, not just list technical features. The paper argues that traditional methods like "feature importance" fail because they don't reflect the cognitive processes people use when making decisions. By grounding explanations in appraisal theory from cognitive science, the framework aims to make AI reasoning feel more natural and understandable to users.

Context matters more than technical accuracy alone. The proposed framework structures explanations around dimensions like relevance (Is this information important to me?), implications (What are the consequences?), and coping potential (Can I do something about it?). This means the same AI decision might be explained differently depending on what matters most to the person receiving the explanation.

This approach is especially important for high-stakes decisions. The paper emphasizes healthcare, finance, and decision support systems where people need to genuinely understand and trust AI recommendations. A poorly explained medical diagnosis or loan denial can have serious real-world consequences, making human-centered explanations a matter of fairness and accountability.

Bridging cognitive science and AI could improve both trust and usability. Rather than treating explainability as purely a technical problem, this work suggests that AI systems designed with insights from how humans actually process information could be easier to understand and more likely to be appropriately trusted by users.