Bridging the Disciplinary Gap in Explainable AI: From Abstract Desiderata to Concrete Tasks

XAI research is fragmented across disciplines because different fields prioritize different goals (fairness, accountability, trust) without clearly defining what they mean or how they relate to each other, making it nearly impossible for any single explanation system to satisfy everyone simultaneously.

Higher-level goals like "trust" and "accountability" actually depend on more foundational technical properties like "faithfulness" and "robustness"—the paper maps these dependency structures to show that researchers need to build explanations from the ground up rather than trying to achieve all desiderata at once.

The paper proposes breaking down vague, abstract goals into concrete, measurable tasks using a three-part framework (examining the target, functional role, and mode of justification), which allows researchers to create benchmarks and actually test whether their explanations work rather than making unverifiable claims.

Context matters critically—what counts as a good explanation depends heavily on who needs it (regulators, users, developers) and why they need it, so the same XAI system cannot be universally "fair" or "accountable" without first clarifying these specific use cases.

This structured approach could help bridge the gap between what XAI researchers promise and what stakeholders actually need, by forcing clearer communication about which specific problems each explanation method solves and under what conditions.