Towards an Explanation Space to Align Humans and Explainable-AI Teamwork

Explainability is not one-size-fits-all: The paper emphasizes that effective AI explanations are "situated interactions" rather than static design principles, meaning explanations must be customized based on the specific user, their context, and their needs—not simply applied uniformly across all users.

Understanding user mental models is critical: Before designing explanations, developers must first understand what end-users already know and how they think about a problem. The paper identifies "end-users mental models" as the first component of their architecture, suggesting that explanations should build on and align with existing user knowledge rather than contradict it.

Explanations serve different purposes at different times: The framework distinguishes between what needs to be explained (content based on mental models), why it's necessary (context based on cognitive processes), how to explain it (format and interface design), and when to provide explanations—recognizing that timing and delivery method matter as much as the explanation itself.

Systematic design process improves human-AI teamwork: The paper proposes the "Abstracted Explanation Space" as a modeling tool to help designers systematically align explanations with users' work practices, needs, and goals, suggesting that structured, evidence-based approaches to explanation design lead to better collaboration between humans and AI systems.

Real-world context matters: The authors demonstrate their framework through a case study in aircraft maintenance, illustrating that explanation requirements vary significantly across different professional domains and work environments.