Explanation as a process: user-centric construction of multi-level and multi-modal explanations

Explanations need to be tailored to different users: The paper emphasizes that one-size-fits-all explanations don't work—developers, domain experts, and end users all have different needs and levels of technical understanding, requiring customized approaches rather than a single explanation method.

Users should be able to ask follow-up questions: Rather than receiving a single explanation, the research proposes treating explanation as an interactive process where users can drill down into different levels of detail through conversational interaction, getting more information as their understanding grows.

Combining multiple explanation types is more effective: The paper advocates for "multi-modal explanations" (combining text, visualizations, and other formats) alongside "multi-level explanations" (varying complexity), suggesting that different stakeholders benefit from different combinations of these approaches.

Interpretable machine learning enables better explanations: The research uses Inductive Logic Programming—an interpretable approach—rather than relying solely on post-hoc explanations of black-box models, demonstrating that building explainability into the model itself can support more meaningful user interactions.

Explanation should be user-centered, not just technically focused: The paper critiques XAI research for focusing primarily on technical innovation rather than actual user needs, calling for a shift toward understanding what different stakeholders actually require to trust and understand AI systems.