Governing the Algorithmic Black Box in Talent Acquisition: Towards a Multi-Level Framework of Contestable Accountability

Abstract

Artificial intelligence now mediates the earliest and most consequential stage of the employment relationship, yet the systems that rank, score and reject applicants remain substantially opaque to the candidates they judge and, this paper argues, often to the organisations that deploy them. This conceptual paper interrogates the dominant governance repertoire that has grown up around this “algorithmic black box” in talent acquisition (transparency disclosure, explainability, technical debiasing, human-in-the-loop review and algorithmic audit) and argues that each instrument rests on assumptions that the recruitment context systematically violates. Building on Bovens’s relational conception of accountability, organisational justice theory and the institutional literature on decoupling, the paper develops a multi-level framework of contestable accountability comprising four nested levels (epistemic, procedural, institutional and contestatory) intersected by three temporal moments of ex ante design, in itinere operation and ex post decision. The paper further distinguishes model, system and decision opacity, and identifies procurement opacity as a fourth form generated by the vendor market. Two diagnostic constructs are advanced to explain persistent governance failure: vertical displacement, whereby demands for redress are answered with instruments of disclosure, and temporal displacement, whereby governance effort concentrates before deployment while candidate grievance arises after it. The framework is elaborated through nine testable propositions and situated against a volatile regulatory landscape spanning the European Union, the United States and India. The paper contributes a governance-theoretic vocabulary for human resource management scholarship, and reframes the policy problem from making algorithms explicable to making hiring decisions answerable.