Explanation Debt in the Responsible Governance of Artificial Intelligence Systems

Transparency has become the most frequently invoked and least operationalised principle in responsible artificial intelligence. Organisations deploy opaque models under the assumption that explanatory obligations can be discharged later, yet the cost of discharging them grows as evidence, personnel and institutional memory decay. This paper introduces explanation debt: the accumulated shortfall between the explanatory capacity an artificial intelligence system delivers to a stakeholder at a decision point and the sufficiency that decision demands. Extending the technical-debt metaphor from software engineering into governance, the paper contributes four elements. First, a taxonomy of four debt classes — design, documentation, drift and reception — distinguished by origin and visibility. Second, an accrual model in which debt compounds at an evidentiary decay rate and is bounded by a stake-sensitive proportionality frontier. Third, an explanatory sufficiency criterion comprising four conjunctive conditions: actionability, contestability, fidelity and timeliness. Fourth, the Explanation Debt Audit Protocol, a six-phase procedure producing a persistent debt register. The framework is mapped onto the European Union Artificial Intelligence Act, the NIST AI Risk Management Framework and the National Artificial Intelligence Document of Iran, and is illustrated through three hypothetical deployments. The paper is conceptual; a three-study empirical validation agenda is specified.

کلیدواژه ها:

explainable artificial intelligence ، algorithmic transparency ، responsible AI ، AI governance ، technical debt ، contestability ، AI regulation ، auditability