Generating Process-Centric Explanations to Enable Contestability in Algorithmic Decision-Making: Challenges and Opportunities

Explanations alone are insufficient for fair AI: The paper challenges the common assumption that technical explanations of algorithmic decisions are enough to address fairness concerns. Instead, explanations must be connected to people's ability to actually contest and challenge decisions—a right that requires understanding not just what a system decided, but why it was built and deployed that way in the first place.

Process-centric explanations go beyond "black box" solutions: Traditional AI explainability (XAI) methods typically focus on explaining individual decisions or outcomes. This paper argues for "process-centric explanations" that trace back through the entire lifecycle of an algorithmic system—including design choices, training data selection, and deployment decisions—to help people understand the full context behind decisions affecting them.

Contestability is a fundamental right that requires transparency: The paper emphasizes that for AI systems to be truly accountable, people subject to algorithmic decisions need meaningful opportunities to contest them. This requires explanations that reveal the rationales behind system development and deployment, not just technical details about how a particular decision was made.

Current research treats explanations as isolated technical fixes: Most studies evaluate explanations in isolation without considering whether they actually enable people to challenge decisions or understand systemic choices. The paper calls for research that evaluates explanations in relation to people's real ability to contest algorithmic outcomes.

Significant research gaps remain in generating and evaluating process-centric explanations: The paper identifies major challenges and opportunities for developing explanations that capture the full decision-making process, suggesting this is an emerging area requiring interdisciplinary work beyond traditional computer science approaches.