Explainable AI in High-Stakes Decision Making: Beyond Accuracy

Artificial Intelligence (AI) is increasingly shaping high-stakes decision-making across healthcare, finance, criminal justice, defense, and autonomous systems. Traditionally, model evaluation has been dominated by accuracy-centric metrics; however, these are insufficient in contexts where decisions can directly affect human life, liberty, or well-being. Black-box models, despite high predictive performance, often fail to provide transparent reasoning, undermining accountability, fairness, and stakeholder trust. Explainable AI (XAI) has emerged as a paradigm shift that emphasizes interpretability and human-centered accountability over raw statistical accuracy. This paper critically examines the limitations of accuracy as a sole benchmark and investigates how explainability functions as a safeguard against bias, ethical lapses, and systemic risks. Drawing upon a mixed-methods design, we integrate quantitative survey data from healthcare, finance, and justice professionals with qualitative case analyses of real-world AI deployment failures. Statistical evidence demonstrates that stakeholders consistently prioritize interpretability, fairness, and trustworthiness over marginal accuracy improvements.