Authors: Dr. M. Senthilkumar, Dr. Kavipriya S, and Dr. A. Balamurugan

ISBN: 978-81-69857-75-8

DOI: https://doi.org/10.59646/829

Date of Publication: September 24, 2026

Preface

Artificial Intelligence (AI) has evolved from systems based on explicitly programmed rules to highly sophisticated machine learning and deep learning models capable of processing enormous volumes of data, recognizing complex patterns, generating predictions, and supporting decisions across diverse domains. However, the increasing complexity of modern AI models has created a significant challenge: although these systems can achieve remarkable predictive performance, their internal reasoning is often difficult for humans to understand. Explainable Artificial Intelligence (XAI) addresses this challenge by developing methods, models, and interaction mechanisms that make AI behaviour more understandable, transparent, interpretable, and accountable. Explainable AI: Trustworthy Intelligence for the Future presents a comprehensive exploration of the principles, methodologies, technologies, and applications required to develop AI systems that can be understood and appropriately trusted by their users.

The book begins with the foundations of Explainable AI, introducing its definitions, scope, human-centred perspectives, cognitive dimensions, and role throughout the AI lifecycle. It examines explainability from data and model development to prediction and decision-making, while considering the requirements of developers, end users, regulators, and society. The discussion then progresses to interpretable machine learning models, covering linear and rule-based models, generalized additive models, explainable boosting machines, Bayesian approaches, symbolic reasoning, sparse models, monotonic constraints, and model simplification. Particular attention is given to the fundamental trade-offs among predictive accuracy, model complexity, transparency, and interpretability.

The book further examines post-hoc explainability techniques that enable explanations to be generated for complex models after training. Important approaches including SHAP, LIME, Integrated Gradients, saliency-based methods, surrogate models, model distillation, Partial Dependence Plots, and Individual Conditional Expectation plots are discussed alongside concepts such as local and global explanations, explanation stability, consistency, and robustness. These techniques provide essential foundations for understanding the behaviour of contemporary black-box models.

As AI increasingly relies on deep neural architectures, the book addresses explainability in deep learning and complex models, including convolutional neural networks, transformers, natural language processing systems, graph neural networks, layer-wise relevance propagation, and concept-based approaches such as TCAV and Concept Activation Vectors. The discussion also considers language-based reasoning and the challenges involved in interpreting highly interconnected and multimodal AI architectures. Beyond technical interpretability, trustworthy AI requires attention to fairness, bias, and responsible AI. The book therefore explores sources of bias in data, models, and deployment environments; fairness metrics such as demographic parity, equalized odds, and calibration; bias detection and mitigation; explainability for fairness auditing; and transparency requirements in high-stakes systems such as recruitment, lending, and healthcare. These discussions emphasize that an explanation should not merely describe a model’s output but should also support meaningful scrutiny of potential discrimination and unintended consequences.

The later units focus on the relationship between humans and intelligent systems. Human–AI interaction and trust calibration introduces psychological foundations of trust, user-specific explanation design, interactive and conversational XAI, visualization, cognitive load, explanation fidelity, usability, and effectiveness. The final unit extends these concepts to safety-critical and regulated domains, including healthcare, finance, legal systems, autonomous vehicles, robotics, cybersecurity, fraud detection, and human-in-the-loop decision-making.

Overall, this book is designed to provide students, researchers, academicians, AI developers, data scientists, and technology professionals with a structured foundation for understanding and applying Explainable AI. By connecting interpretable models, post-hoc explanation techniques, deep learning explainability, fairness, human–AI interaction, and domain-specific requirements, the book presents XAI as an essential component of modern trustworthy intelligence. It aims to support the development of AI systems whose predictions and decisions are not only computationally effective but also understandable, scrutinizable, and appropriately aligned with human needs and societal requirements.