Advancing Transparent and Human-Centered Artificial Intelligence: A Comprehensive Review of Explainable AI Theories, Methods, and Applications

Abstract

Explainable Artificial Intelligence (XAI) has emerged as a central paradigm in contemporary artificial intelligence research, driven by the growing deployment of machine learning systems in high-stakes domains such as healthcare, finance, governance, and autonomous systems. While predictive accuracy has traditionally dominated the evaluation of machine learning models, increasing concerns regarding opacity, accountability, fairness, trust, and ethical compliance have exposed fundamental limitations of black-box approaches. This article presents a comprehensive, publication-ready research study that synthesizes and critically elaborates on the theoretical foundations, taxonomies, methodologies, and application-driven implications of XAI, based strictly on established scholarly literature. Drawing from foundational surveys, conceptual frameworks, and domain-specific studies, the article examines explainability from multiple perspectives, including technical model interpretability, human-centered explanation effectiveness, causality and counterfactual reasoning, knowledge-based representations, and stakeholder-oriented requirements. Particular attention is given to the tension between model complexity and interpretability, the distinction between intrinsic and post-hoc explanations, and the evolving role of XAI in regulated and safety-critical environments. Methodologically, the study adopts a structured qualitative synthesis approach, integrating comparative analysis and conceptual reasoning to uncover patterns, gaps, and unresolved challenges within the existing body of work. The results highlight that explainability is not a singular technical property but a socio-technical construct shaped by context, audience, and purpose. The discussion extends these findings by addressing limitations of current XAI methods, including evaluation ambiguity, potential for misleading explanations, and insufficient alignment with human reasoning. The article concludes by proposing future research directions toward responsible, human-aligned, and causally grounded XAI systems. Overall, this work contributes an in-depth, theoretically rich, and integrative perspective intended to guide researchers, practitioners, and policymakers toward more transparent and trustworthy artificial intelligence.