Adding Why to What? Analyses of an Everyday Explanation

Explaining AI to non-experts requires different approaches than explaining to professionals. The research emphasizes that laypeople cannot be assumed to have technical expertise, so explanations must be fundamentally different from those designed for specialists in the field.

Effective explanations balance two complementary perspectives: the "how it works" (Architecture) and the "why it matters" (Relevance). The study found that good explanations address both the technical logic of algorithms and the real-world implications and severity of decisions, rather than focusing on only one aspect.

Explainers typically start with technical basics before addressing consequences. The research found that people naturally begin by explaining the physical/structural components of a system first, then transition to discussing what those components mean for users—suggesting this may be an intuitive and effective sequencing strategy.

Explanation strategies should adapt based on the listener's needs and emerging confusion. The study found that skilled explainers shifted between technical and practical perspectives depending on their audience's knowledge level, the goals of the explanation, and when misunderstandings arose during the conversation.

The "dual nature theory" provides a practical framework for analyzing and improving AI explanations. By categorizing explanations into Architecture and Relevance components, researchers can more systematically study and compare how different people explain AI systems to laypeople.