Beyond the Black Box: Designing Equitable Algorithmic Governance for High-Stakes Institutional Screening

Abstract

Purpose: The rapid integration of artificial intelligence into high-stakes institutional screening threatens procedural justice by obscuring historical demographic biases within opaque algorithmic models. While existing literature extensively diagnoses this "black box" problem, there remains a critical dearth of actionable, legally defensible governance frameworks capable of preventing proxy discrimination. This paper aims to bridge the gap between computer science fairness metrics and administrative jurisprudence.  Design/Methodology/Approach: Grounded in theories of organizational justice and administrative equity, this conceptual paper critically synthesizes recent legal, ethical, and sociotechnical scholarship to construct a comprehensive algorithmic governance architecture. Findings: This study proposes a tripartite structural governance framework. First, it mandates rigorous pre-deployment algorithmic impact assessments (AIAs) to detect and mathematically neutralize proxy variables in training data. Second, it requires the integration of explainable artificial intelligence (AI) (XAI) metrics to guarantee decision contestability and operationalize the right to explanation. Third, it designs stringent " human-in-the-loop " (HITL) protocols, introducing mechanisms, such as blinded baselines and independent algorithmic appeals committees, to actively counteract automation bias and enforce human accountability. Originality/Value: Moving beyond theoretical critique, this research delivers a concrete managerial and policy blueprint. It demonstrates how institutions must fundamentally restructure their technological procurement and internal oversight to align with emerging civil rights legislation, ensuring that AI serves as an instrument of administrative equity rather than an automated architect of systemic marginalization.