AI Explained

PETE RAI is an independent AI researcher and principal engineer with experience across a range of technology companies. Working at the frontier of applied AI in enterprise contexts, he holds more than two dozen patents and is active in responsible AI initiatives. He brings a practitioner’s perspective on how AI is built, deployed, operated, and governed in real-world settings. DAME WENDY HALL is Regius Professor of Computer Science at the University of Southampton, a Fellow of the Royal Society and the Royal Academy of Engineering, and a member of the United Nations high-level advisory body on AI. She frequently advises governments and international organizations on the societal and policy impact of AI.

• About the Author xvIntroduction 1Behind the Curtain 2Who Is This Book For? 2Demystifying AI 3The Road Ahead 5Chapter 1: The Evolution of AI 7The Birth of Electronic Computation 7Alan Turing and Computability 8Claude Shannon and Information Theory 9John von Neumann and Stored Programmes 10A New Science 11The Dartmouth Conference 11A Seminal Conference 12The Infancy of Computing 12AI Visionaries 12The Gathering 13Legacy 13The Timeline of AI 14Early Foundations (1940–1956) 14The Birth of AI (1956–1969) 15The First AI Winter (1970–1980) 16AI Research Resurgence (1980–1987) 16The Second AI Winter (1987–1997) 17Rise of Machine Learning (1998–2015) 18Renaissance and Commercialisation (2015–Present) 19Generative AI and Machine Creativity (2022–Present) 20Summary 20Chapter 2: Symbolic AI 23Overview 23Enabling Machine Understanding 24Core Concepts and Mechanisms 26In-depth 28Computable Information 28Encoding 29Symbolic Representation 30Knowledge Representation 32Information Classification 33Classes, Instances, and Schemas 34Rules 35Imperative vs. Declarative Rules 36Conditional Rules 36First-Order Logic 37Reasoning 38Inference 39Wider Reasoning 40Common Sense Reasoning 40Knowledge Growth 41Triples 42Ontologies 43Common Tools and Techniques 45Decision Trees 45Rules-Based Engines 46Expert Systems 48Knowledge Graphs 49Natural Language Processing 51Use Cases 53Automated Medical Diagnosis 53Expert Systems and Decision Trees 54Personalised Medical Assessments 54Outcome 54Legal Case Analysis 55Rules-Based Engines 55Enhanced Legal Strategy and Decision-Making 55Outcome 56Financial Fraud Detection 56Decision Tree 56Enhancing Security and Trust 57Outcome 57Drug Discovery and Development 57Ontologies and Knowledge Graphs 57Accelerating Drug Development 58Outcome 58Conclusion 58Chapter 3: Statistical AI 61Overview 62Data-Driven Approaches 62Statistics 63Core Concepts and Mechanisms 65In-depth 67Data, Data, Data . . . 67Data, Information, Knowledge, Wisdom 68Structured vs. Unstructured Data 69Space and Time 70Wider Data Classification Schemes 72Sampling 73Types of Sampling 74Considerations and Challenges 74Probability Theory 75Independent, Dependent, and Correlated Events 76Trends, Patterns, and Predictions 77Distributions 79Hypothesis Testing 80Clusters and Classifications 82Clustering vs. Classifying 83Discovering Object Clusters 83Features, Underfitting, and Overfitting 84Regression Analysis 85Data Assumptions for Regression 86Simple vs. Multiple Linear Regression 86Challenges and Considerations 87Bayesian Methods 88Handling Uncertainties 89Bayesian Networks 90Challenges and Considerations 90Use Cases 91Movie Recommendations 91Clustering and Classifying 92Personalised Recommendations 92Outcome 92Health Risk Assessment 92Sampling and Regression Analysis 93Personalised Risk Profiles 93Outcome 93Financial Market Forecasting 93Probability Theory and Bayesian Methods 94Comparative Analysis, Uncertainty, and Volatility 94Outcome 94Credit Scoring 95Regression Analysis and Bayesian Methods 95Comparative Analysis and Market Adaptation 95Outcome 95Conclusion 96Chapter 4: Connected AI 99Overview 99Connections 100The Human Brain 100Simulating Biological Neurons 101The Perceptron 101Problems and Limitations 102Resurgence and Later Developments 102Core Concepts and Mechanisms 103In-depth 105Graph Theory 105Nodes, Links, Graphs, and Networks 107Relationships, Rules, and Signals 109Neural Nets 109Contrasts with Biological Brains 112Modern Neural Networks in Practice 113Convolutional Neural Networks 114Recurrent Neural Networks 116Advanced Connected Techniques 118Use Cases 119Inventory Management Systems 119Basic Neural Nets and Knowledge Graphs 120Optimised Stock Levels and Movements 120Outcome 120Financial Fraud Detection 121Advanced Neural Networks 121Safer Financial Ecosystems 121Outcome 122Healthcare Image Diagnostics 122Convolutional Image Analysis 122Early Diagnosis of Critical Conditions 122Outcome 123Autonomous Driving 123Recurrent Sequential Data Processing 123Safer Roads and Safer Journeys 123Outcome 124Conclusion 124Chapter 5: Learning AI 127Overview 128Machine Learning 128Contrasts with Human Learning 129The Learning Process 130Core Concepts and Mechanisms 131In-depth 134Learning from Data 134Learning Strategies 134Applying Digital Learning 135Supervised Learning 136Gathering Labelled Data 138Neural Nets and Feature Selection 140Weights and Biases in Neural Networks 140Backpropagation—Learning from Mistakes 142Quality and Accuracy 143Unsupervised Learning 144Uncovering Patterns in Data 145Working in Multidimensions 147Clustering 149Dimensionality Reduction 151Relationships and Associations 152Anomaly Detection 153Quality and Accuracy 155Challenges and Considerations 156Reinforcement Learning 156Contrasting Machine Learning Approaches 159Advanced Learning Techniques 160Deep Learning 160Transfer Learning 161Ensemble Learning 162Meta-Learning 162Federated Learning 162Use Cases 163Email Spam Detection 163Supervised Learning 164Improved Email Experience and Security 164Outcome 164Customer Segmentation in Marketing 165Unsupervised Learning 165Personalised Marketing Strategies 165Outcome 166Game-Playing AI 166Reinforcement Learning 166Achieving Strategic Mastery 166Outcome 167Voice Recognition 167Deep Learning 167Enhanced User Interaction 168Outcome 168Conclusion 168Chapter 6: Generative AI 171Overview 171Machine Creativity 172Emergence of Non-Human Methodologies 173Creativity Algorithms 174Core Concepts and Mechanisms 175In-depth 178Mathematical Representations 178Autoencoders 179Variational Autoencoders 182Generative Adversarial Networks 183Transformers and Large Language Models 185Chunking 188Embedding 190Text Generation 192Challenges and Limitations 194Diffusion Models and Image Generation 196The Forward Process: Adding Noise 197The Reverse Process: Learning to Denoise with Conditions 197Generating Images from User Prompts 198Orchestration and Agentic AI 199AI Orchestration 199Agentic AI 202Multimodal Generative AI 203Use Cases 204Medical Image Synthesis 204Variational Autoencoders 204Enhanced Patient Outcomes 205Outcome 205AI Assistants 205Transformers 205Assistance via Natural Language 206Outcome 206Images from Text 206Diffusion Models 207Enabling Visual Communication 207Outcome 207Digital Actor Resurrection and De-aging 208Generative Adversarial Networks (GANs) 208Epic Sagas with Consistent Actors 208Outcome 208Conclusion 209Afterword 211The Winding Road of AI 212Charting the Road Ahead 214Scaling and Applying AI 215AGI and Superintelligence 216Emerging Directions 218International Perspectives 220The AI Experience 221Opportunities and Challenges 222Indispensable Intelligence 223Transformational Impact 224Education 224Healthcare 225Defence 226The Price of Progress 227Choosing the Future We Want 229Index 231