Who is this Explanation for? Human Intelligence and Knowledge Graphs for eXplainable AI

One-size-fits-all explanations don't work: The paper emphasizes that AI explanations must be tailored to the specific audience receiving them—what works for a doctor may not work for a patient or a loan officer. Effective explainable AI requires understanding "what constitutes a human explanation" rather than simply translating algorithm outputs into words.

Human expertise should shape AI explanations: The authors argue for integrating insights from social sciences and human computation into AI systems, suggesting that understanding how humans naturally explain decisions can improve how machines communicate their reasoning. This requires collaboration between AI researchers and experts in human behavior and communication.

Knowledge graphs can bridge the gap between machines and humans: The paper proposes that Knowledge Graphs—structured representations of information and relationships—can help make AI decisions more interpretable by organizing information in ways that align with how humans understand and reason about the world.

AI explanations must be actionable, not just understandable: Beyond simply being interpretable, explanations need to enable human decision-makers to actually act on them—whether that means accepting, rejecting, or modifying the AI's recommendation. This requires a "Human-in-the-Loop" approach where humans remain actively involved in the decision-making process.

Responsible AI requires interdisciplinary collaboration: The authors call for breaking down silos between AI research, knowledge representation, social sciences, and human-computer interaction to create AI systems that genuinely serve human decision-makers rather than just appearing to explain themselves.