Integrating Evidence into the Design of XAI and AI-based Decision Support Systems: A Means-End Framework for End-users in Construction

Explanations alone are insufficient for trustworthy AI: The paper identifies a critical gap in current explainable AI systems—they often provide explanations without the supporting evidence needed to verify reliability and accountability. In construction decision-making, this absence of evidence "undermines the validity of explanations and the trustworthiness of system recommendations."

Evidence quality matters as much as transparency: The framework evaluates three dimensions of supporting evidence—strength, relevance, and utility—recognizing that not all evidence is equally valuable. This means responsible AI design requires assessing what evidence backs up AI recommendations, not just why the AI made a decision.

Different users need different types of evidence: The paper emphasizes that explanations should be "tailored to users' knowledge needs and decision contexts." Construction workers, project managers, developers, and regulators have different epistemic goals, meaning a one-size-fits-all explanation approach fails to serve responsible AI implementation across stakeholder groups.

Construction industry reveals broader AI governance gaps: By focusing on construction—where AI adoption is increasing but oversight is limited—the paper highlights how industries beyond tech may lack frameworks for ensuring AI accountability, suggesting this is a systemic problem requiring sector-specific solutions.

Responsible AI requires foundational design principles: Rather than adding explanations as an afterthought, the framework proposes building evidence integration into AI system design from the start, establishing an "epistemic foundation" that makes accountability and trustworthiness central to how systems are developed.