Complex Adaptive Systems

John H. Miller is professor of economics and social sciences at Carnegie Mellon University. Scott E. Page is professor of complex systems, political science, and economics at the University of Michigan. He is the author of "The Difference" (Princeton).

"The use of computational, especially agent-based, models has already shown its value in illuminating the study of economic and other social processes. Miller and Page have written an orientation to this field that is a model of motivation and insight, making clear the underlying thinking and illustrating it by varied and thoughtful examples. It conveys with remarkable clarity the essentials of the complex systems approach to the embarking researcher."—Kenneth J. Arrow, winner of the Nobel Prize in economics

• List of Figures xiii List of Tables xv Preface xvii Part I: Introduction 1 Chapter 1: Introduction 3 Chapter 2: Complexity in Social Worlds 9 2.1 The Standing Ovation Problem 10 2.2 What's the Buzz? 14 2.2.1 Stay Cool 14 2.2.2 Attack of the Killer Bees 15 2.2.3 Averaging Out Average Behavior 16 2.3 A Tale of Two Cities 17 2.3.1 Adding Complexity 20 2.4 New Directions 26 2.5 Complex Social Worlds Redux 27 2.5.1 Questioning Complexity 27 Part II: Preliminaries 33 Chapter 3: Modeling 35 3.1 Models as Maps 36 3.2 A More Formal Approach to Modeling 38 3.3 Modeling Complex Systems 40 3.4 Modeling Modeling 42 Chapter 4: On Emergence 44 4.1 A Theory of Emergence 46 4.2 Beyond Disorganized Complexity 48 4.2.1 Feedback and Organized Complexity 50 Part III: Computational Modeling 55 Chapter 5: Computation as Theory 57 5.1 Theory versus Tools 59 5.1.1 Physics Envy: A Pseudo-Freudian Analysis 62 5.2 Computation and Theory 64 5.2.1 Computation in Theory 64 5.2.2 Computation as Theory 67 5.3 Objections to Computation as Theory 68 5.3.1 Computations Build in Their Results 69 5.3.2 Computations Lack Discipline 70 5.3.3 Computational Models Are Only Approximations to Specific Circumstances 71 5.3.4 Computational Models Are Brittle 72 5.3.5 Computational Models Are Hard to Test 73 5.3.6 Computational Models Are Hard to Understand 76 5.4 New Directions 76 Chapter 6: Why Agent-Based Objects? 78 6.1 Flexibility versus Precision 78 6.2 Process Oriented 80 6.3 Adaptive Agents 81 6.4 Inherently Dynamic 83 6.5 Heterogeneous Agents and Asymmetry 84 6.6 Scalability 85 6.7 Repeatable and Recoverable 86 6.8 Constructive 86 6.9 Low Cost 87 6.10 Economic E. coli (E. coni?) 88 Part IV: Models of Complex Adaptive Social Systems 91 Chapter 7: A Basic Framework 93 7.1 The Eightfold Way 93 7.1.1 Right View 94 7.1.2 Right Intention 95 7.1.3 Right Speech 96 7.1.4 Right Action 96 7.1.5 Right Livelihood 97 7.1.6 Right Effort 98 7.1.7 Right Mindfulness 100 7.1.8 Right Concentration 101 7.2 Smoke and Mirrors: The Forest Fire Model 102 7.2.1 A Simple Model of Forest Fires 102 7.2.2 Fixed, Homogeneous Rules 102 7.2.3 Homogeneous Adaptation 104 7.2.4 Heterogeneous Adaptation 105 7.2.5 Adding More Intelligence: Internal Models 107 7.2.6 Omniscient Closure 108 7.2.7 Banks 109 7.3 Eight Folding into One 110 7.4 Conclusion 113 Chapter 8: Complex Adaptive Social Systems in One Dimension 114 8.1 Cellular Automata 115 8.2 Social Cellular Automata 119 8.2.1 Socially Acceptable Rules 120 8.3 Majority Rules 124 8.3.1 The Zen of Mistakes in Majority Rule 128 8.4 The Edge of Chaos 129 8.4.1 Is There an Edge? 130 8.4.2 Computation at the Edge of Chaos 137 8.4.3 The Edge of Robustness 139 Chapter 9: Social Dynamics 141 9.1 A Roving Agent 141 9.2 Segregation 143 9.3 The Beach Problem 146 9.4 City Formation 151 9.5 Networks 154 9.5.1 Majority Rule and Network Structures 158 9.5.2 Schelling's Segregation Model and Network Structures 163 9.6 Self-Organized Criticality and Power Laws 165 9.6.1 The Sand Pile Model 167 9.6.2 A Minimalist Sand Pile 169 9.6.3 Fat-Tailed Avalanches 171 9.6.4 Purposive Agents 175 9.6.5 The Forest Fire Model Redux 176 9.6.6 Criticality in Social Systems 177 Chapter 10: Evolving Automata 178 10.1 Agent Behavior 178 10.2 Adaptation 180 10.3 A Taxonomy of 2 x 2 Games 185 10.3.1 Methodology 187 10.3.2 Results 189 10.4 Games Theory: One Agent, Many Games 191 10.5 Evolving Communication 192 10.5.1 Results 194 10.5.2 Furthering Communication 197 10.6 The Full Monty 198 Chapter 11: Some Fundamentals of Organizational Decision Making 200 11.1 Organizations and Boolean Functions 201 11.2 Some Results 203 11.3 Do Organizations Just Find Solvable Problems? 206 11.3.1 Imperfection 207 11.4 Future Directions 210 Part V: Conclusions 211 Chapter 12: Social Science in Between 213 12.1 Some Contributions 214 12.2 The Interest in Between 218 12.2.1 In between Simple and Strategic Behavior 219 12.2.2 In between Pairs and Infinities of Agents 221 12.2.3 In between Equilibrium and Chaos 222 12.2.4 In between Richness and Rigor 223 12.2.5 In between Anarchy and Control 225 12.3 Here Be Dragons 225 Epilogue 227 The Interest in Between 227 Social Complexity 228 The Faraway Nearby 230 Appendixes A An Open Agenda For Complex Adaptive Social Systems 231 A.1 Whither Complexity 231 A.2 What Does it Take for a System to Exhibit Complex Behavior? 233 A.3 Is There an Objective Basis for Recognizing Emergence and Complexity? 233 A.4 Is There a Mathematics of Complex Adaptive Social Systems? 234 A.5 What Mechanisms Exist for Tuning the Performance of Complex Systems? 235 A.6 Do Productive Complex Systems Have Unusual Properties? 235 A.7 Do Social Systems Become More Complex over Time 236 A.8 What Makes a System Robust? 236 A.9 Causality in Complex Systems? 237 A.10 When Does Coevolution Work? 237 A.11 When Does Updating Matter? 238 A.12 When Does Heterogeneity Matter? 238 A.13 How Sophisticated Must Agents Be Before They Are Interesting? 239 A.14 What Are the Equivalence Classes of Adaptive Behavior? 240 A.15 When Does Adaptation Lead to Optimization and Equilibrium? 241 A.16 How Important Is Communication to Complex Adaptive Social Systems? 242 A.17 How Do Decentralized Markets Equilibrate? 243 A.18 When Do Organizations Arise? 243 A.19 What Are the Origins of Social Life? 244 B Practices for Computational Modeling 245 B.1 Keep the Model Simple 246 B.2 Focus on the Science, Not the Computer 246 B.3 The Old Computer Test 247 B.4 Avoid Black Boxes 247 B.5 Nest Your Models 248 B.6 Have Tunable Dials 248 B.7 Construct Flexible Frameworks 249 B.8 Create Multiple Implementations 249 B.9 Check the Parameters 250 B.10 Document Code 250 B.11 Know the Source of Random Numbers 251 B.12 Beware of Debugging Bias 251 B.13 Write Good Code 251 B.14 Avoid False Precision 252 B.15 Distribute Your Code 253 B.16 Keep a Lab Notebook 253 B.17 Prove Your Results 253 B.18 Reward the Right Things 254 Bibliography 255 Index 261