Towards Relatable Explainable AI with the Perceptual Process

Contrastive explanations matter more than raw comparisons: The paper demonstrates that people understand AI decisions better when systems explain why a prediction occurred instead of an alternative outcome, rather than simply comparing raw features or examples—a finding that challenges how many current AI systems provide explanations.

Counterfactual explanations with semantic context are most effective: User studies revealed that "counterfactual" explanations (showing what would need to change for a different outcome) were significantly more useful when paired with semantic cues—meaningful concepts people can relate to—whereas saliency maps alone (highlighting important image regions) did not help users understand decisions.

AI explanations need to connect to human reasoning: Drawing from cognitive psychology, the research shows that effective AI explanations should mirror how humans naturally process information by relating decisions to familiar concepts, hypotheticals, and associations rather than presenting isolated technical features.

Relatable explanations improve trust and usability: By making explanations more semantically meaningful and connected to human understanding, the framework addresses a critical gap in responsible AI: users can better evaluate whether an AI system's reasoning is sound, which is essential for trustworthy deployment in real-world applications like emotion recognition from speech.