Knowledge Graphs and Explainable AI as Complementary Resources for Urban Mining

Human accountability must remain central to AI-assisted decisions: The paper emphasizes that in pre-demolition assessment, qualified auditors must stay accountable for decisions, meaning AI should support rather than replace human judgment—a principle applicable across regulated industries where decisions have significant consequences.

Prediction accuracy alone is insufficient for responsible AI: The research identifies that defensibility of decisions matters more than raw prediction accuracy, requiring that supported decisions be legible (understandable), plausible (reasonable), well-sourced, and contestable (open to challenge)—a framework relevant for any high-stakes AI application.

Knowledge graphs and explainable AI work best together: The paper demonstrates that neither knowledge graphs nor explainable AI techniques alone fully address the need for defensible decisions; their integration creates capabilities neither possesses independently, suggesting that responsible AI often requires combining multiple technical approaches rather than relying on single solutions.

Specific integration modes unlock different types of accountability: The four integration modes identified (Lifting, Constraining, Typing, and Revising) each contribute distinct properties to decision defensibility, implying that responsible AI implementation requires thoughtful architectural choices about how different technical components interact.

Regulatory requirements should drive AI system design: The paper grounds its analysis in actual regulatory demands of pre-demolition assessment, suggesting that responsible AI systems should be designed backward from specific accountability and transparency requirements rather than forward from technical capabilities.