Building Trust in AI: Explainable AI in Critical Decision-Making

Artificial intelligence (AI) is increasingly used to support decisions in healthcare, finance, recruitment, transportation, and public services. Although complex machine-learning models can produce accurate predictions, their reasoning is not always easy to understand. This creates practical questions: why did the system recommend a particular outcome, did it use appropriate information, and when should a person challenge its decision? Explainable artificial intelligence (XAI) brings together techniques that help people understand a model's behaviour or the factors behind a prediction. This review discusses interpretable models, feature-attribution methods such as LIME and SHAP, counterfactual explanations, and model-agnostic approaches. It compares their uses and limitations in high-impact settings. An understandable explanation, however, is not proof that a model is accurate, fair, dependable, or safe. The paper uses a narrative review of foundational work and selected research and survey publications on interpretability, responsible AI, and human-centred explanations. No new experiments, participant studies, or quantitative tests were performed. The literature points to unresolved issues involving explanation fidelity, stability, accessibility, privacy, bias, evaluation standards, and over-reliance on automated advice. Future research should use domain-specific evaluations, retain meaningful human oversight, communicate uncertainty clearly, assess fairness across groups, and report limitations transparently. XAI can support accountability and informed supervision, but it cannot guarantee trustworthiness on its own.