Human-centered Explainable AI: Towards a Reflective Sociotechnical Approach

Explanations alone aren't enough for responsible AI: The paper emphasizes that post-hoc explanations (explanations added after a system is built) are just one tool for making AI accessible, and their effectiveness depends heavily on understanding the specific humans who will use them—their values, relationships, and social context.

"Who is the human?" matters more than we typically assume: Rather than designing explanations for a generic user, HCXAI requires understanding the diverse characteristics of actual people who interact with AI systems, including their interpersonal dynamics and how they're situated within broader social systems, not just their technical literacy.

Technical and human understanding must develop together: The paper's case study demonstrates that advances in explanation technology and insights about human needs co-evolve—meaning responsible AI design requires ongoing dialogue between technologists and the people affected by these systems, rather than building first and explaining later.

Current human-computer interaction models are too narrow: The paper identifies a gap in focusing only on one-to-one interactions between individuals and AI systems, suggesting that responsible AI design must also account for how explanations function in group settings and broader organizational contexts.

Critical reflection on design choices is essential: By incorporating perspectives like Critical Technical Practice, value-sensitive design, and participatory design, organizations can identify their own blind spots in AI development and create space for more equitable and accountable AI systems.