Explainable Artificial Intelligence (XAI): An Introduction

Bibliographic reference / Référence bibliographique (RDA/ISBD)

BRITO, João B. G. Explainable Artificial Intelligence (XAI): An Introduction — From Black Boxes
to Transparent Decision Models. 1st ed. – Ghent : Zenodo, 2024. 82 p. ; 29,7 cm. — ( Data Science
Series ; vol. 3 ). DOI 10.5281/zenodo.17545855. ISBN 978-65-01-79603-1.

Download

This book, Explainable Artificial Intelligence (XAI): From Black Boxes to Transparent Decision Models, was authored entirely in Markdown, allowing the entire authoring process to be managed with the Git version control system.

The complete book is available for download at: https://gitlab.com/joao17/Book_XAI/-/raw/main/Published/Book_XAI_FullBook.pdf

Data Science Series

The Data Science Series offers a complete journey through the world of data science,  from data engineering and preparation to interpretation and visual communication. Each volume is designed to build a solid and progressive foundation of knowledge, combining theory, practice, and real-world applications. The journey begins with Data Engineering: An Introduction, which explores data collection, transformation, and integration — the cornerstones of reliable analytics. It continues with Causal Inference, Explainable Artificial Intelligence (XAI), and Reinforcement Learning, expanding the reader’s understanding of how to uncover cause–effect relationships, interpret complex models, and learn optimal policies through interaction and reward. The series culminates with Data Visualization Art, a volume that merges science and creativity to transform data into expressive and impactful visual narratives. Written in clear and accessible language, the collection is intended for students, researchers, and professionals seeking to grasp both the logic and the beauty that unite data, algorithms, and decision-making. Each book balances conceptual rigor and practical application, showing how data science can serve as both a powerful technical tool and a form of intellectual and aesthetic expression.

Explainable Artificial Intelligence (XAI)

This book presents a concise and accessible introduction to Explainable Artificial Intelligence (XAI), a field that has become essential for understanding, auditing, and trusting modern machine learning systems. As predictive models grow in complexity and influence, the need to interpret their decisions becomes fundamental for ethical, legal, and technical reasons. Aimed at students, researchers, and professionals entering data science or artificial intelligence, this text provides the conceptual foundations and methodological structure required to navigate the landscape of model explainability.

Grounded in applied knowledge and supported by established academic literature, the book introduces the motivation behind XAI, clarifies its conceptual framework, and presents the main families of methods used to explain machine learning models. Rather than offering an exhaustive catalog of algorithms, it focuses on the essential principles that enable readers to understand model behavior, assess risks, detect biases, and make informed decisions based on predictive systems.

The content is organized into seven chapters, each dedicated to a key component of the XAI discipline.

Chapter 1 — Introduction contextualizes the rise of explainability in AI, describes the limitations of black-box models, and identifies the gap that motivates the need for transparent decision systems. It defines XAI, explains the logic behind grouping methods into families, and introduces the main categories that structure the remainder of the book.

Chapter 2 — Explanatory Model Analysis (EMA) presents post-hoc techniques designed to interpret complex models after training. Readers are introduced to global and local approaches such as feature importance, partial dependence, ICE curves, surrogate models, LIME, and SHAP, learning how these methods uncover patterns and reasoning embedded within predictive algorithms.

Chapter 3 — Interpretable Machine Learning (IML) explores models that are transparent by design. It covers linear and additive models, decision rules, explainable boosting machines, and other structures that balance interpretability and performance. The chapter distinguishes global interpretability, linked to model structure, from local interpretability, tied to individual predictions.

Chapter 4 — Causal XAI integrates explainability with causal inference. It introduces causal diagrams, structural causal models, counterfactual reasoning, and methods for actionable recourse. The chapter emphasizes how causal perspectives strengthen explanations by distinguishing true causal effects from statistical associations.

Chapter 5 — Transparent and Glass-Box Deep Learning focuses on methods that reveal the internal mechanisms of neural networks. It presents saliency maps, Grad-CAM, Integrated Gradients, Layerwise Relevance Propagation, and concept-based explanations, showing how these techniques clarify the behavior of deep models at both global and local scales.

Chapter 6 — Fairness, Robustness, and Reliability examines how models and explanations can be evaluated for stability, bias, and trustworthiness. Topics include fairness metrics, robustness testing, interpretability evaluation frameworks, and documentation standards such as model cards. The chapter highlights why explanation quality matters as much as accuracy.

Chapter 7 — Human-Centered XAI addresses the interaction between explanations and their users. It discusses principles for designing human-aligned explanations, communication strategies, dashboard design, and cognitive considerations that shape how individuals interpret model outputs. The chapter connects XAI to broader issues of usability and decision support.

Chapter 8 — Synthesis and Future Directions summarizes the key concepts developed across the book and highlights how the different families of XAI complement each other. It discusses current challenges such as evaluating explanations, balancing transparency and performance, and integrating causality and human-centered design. The chapter concludes by outlining emerging directions for building more robust, interpretable, and socially aligned AI systems.

With a practical and pedagogical approach, supported by illustrative examples and references to leading literature, Explainable Artificial Intelligence (XAI): From Black Boxes to Transparent Decision Models provides a structured and coherent path for understanding how AI systems can be made accountable, interpretable, and trustworthy. Rather than aiming for encyclopedic coverage, the book focuses on core concepts and methodological clarity, offering readers the foundations needed to think critically about model explanations and to use them responsibly in modern data-driven environments.