Human-in-the-loop: how managers can govern AI decisions in HR practices

Purpose Research is increasingly governed by artificial intelligence (AI) algorithms that operate as “black boxes” within human resource (HR) practices. The purpose of this paper is to examine how human-in-the loop (HITL) models incorporating explainable AI (XAI) strategies, specifically SHAP and LIME, empower human resource managers to effectively supervise and govern AIdriven decisions in critical HR scenarios. Design/methodology/approach It used a mixed methods experimental design, with 240 HR managers of various industries. They constitute retention/no explanation, retention/LIME explanations, retention/SHAP explanations and synthesis retention. Four machine learning models were trained on both simulated and real alternative anonymized HR data (Logistic Regression, Random Forest, XGBoost and Neural Network). These measurable metrics (accuracy, confidence, trust) were evaluated in light of qualitative findings of semi-structured interviews. Findings Findings demonstrate that XAI substantially enhances quality of decisions made, measurement increases accuracy by 12.1% by use of LIME and 14.8% by use of SHAP, as opposed to passing information to AI recommendation explanations. SHAP was most accurate and most managerially confident. When AI systems became explained, trust in them also rose significantly. Qualitative responses have identified that transparency minimized amount of thought required and facilitated sufficient HR decision-making. Originality/value To the best of the authors’ knowledge, the proposed framework is one of first empirically verified HITL governance frameworks that can be applied to HR; it is a blend of technical techniques for XAI-based finding issues and human parties to draw final conclusions. It provides evidence-based practical guidance on foundation for balancing automation efficacy and ethical human care, as well as revitalizing AI governance theory and ethical human resource technology applications.