Faculteit Wetenschappen Center Leo Apostel for Interdisciplinary Studies Design and Control of Self-organizing Systems PhD Dissertation presented by Carlos Gershenson Promoters: Prof. Dr. Bart D’Hooghe and Prof. Dr. Francis Heylighen Advisors: Prof. Dr. Bruce Edmonds Prof. Dr. Bernard Manderick Prof. Dr. Peter McBurney Prof. Dr. Ann Now´e 2007 ii CONTENTS Contents iii List of Figures vi List of Tables viii Abstract ix Acknowledgements xi 1 Introduction 1 1.1 Motivation ............................. 2 1.2 Aims ................................ 3 1.3 Outline ............................... 3 1.3.1 How to Read the Thesis ................. 4 1.3.2 How the Thesis was Written ............... 6 2 Complexity 9 2.1 Introduction ............................ 10 2.2 Classical Thinking ......................... 10 2.3 Complexity ............................. 11 2.4 Indeterminacy ........................... 14 2.5 Nonlinearity and Chaos ..................... 17 2.6 Adapting to Complexity ..................... 19 2.7 Conclusions ............................ 21 3 Self-organization 23 3.1 Introduction ............................ 24 3.2 The Representation-Dependent Dynamics of Entropy .... 24 3.3 The Role of the Observer ..................... 29 3.4 Ontological Issues ......................... 30 3.5 Self-organization: A Practical Notion .............. 32 iii iv CONTENTS 3.5.1 Artificial self-organizing systems ............ 33 3.5.2 Levels of abstraction ................... 34 3.5.3 Coping with the unknown ................ 34 3.6 Conclusions ............................ 35 4 A General Methodology 37 4.1 Introduction ............................ 38 4.2 The Conceptual Framework ................... 38 4.3 The Methodology ......................... 43 4.3.1 Representation ...................... 44 4.3.2 Modeling ......................... 46 4.3.3 Simulation ......................... 53 4.3.4 Application ........................ 54 4.3.5 Evaluation ......................... 55 4.3.6 Notes on the Methodology ............... 55 4.4 Discussion ............................. 56 4.5 Conclusions ............................ 59 5 Self-organizing Traffic Lights 61 5.1 Introduction ............................ 62 5.2 Applying the Methodology I .................. 63 5.3 Experiments: First Results .................... 69 5.4 Applying the Methodology II .................. 76 5.5 Experiments: Second Results .................. 77 5.6 Applying the Methodology III .................. 81 5.7 Experiments: Third Results ................... 83 5.8 Applying the Methodology IV .................. 85 5.9 Discussion ............................. 87 5.9.1 Adaptation or optimization? .............. 88 5.9.2 Practicalities ........................ 89 5.9.3 Environmental benefits ................. 90 5.9.4 Unattended issues .................... 91 5.10 Conclusions ............................ 92 6 Self-organizing Bureaucracies 95 6.1 Introduction ............................ 96 6.2 Designing Self-organizing Bureaucracies ............ 98 6.3 The Role of Communication ...................100 6.3.1 Decision Delays ......................104 6.4 The Role of Sensors ........................104 6.5 The Role of Hierarchies ......................106 CONTENTS v 6.6 The Role of Context ........................109 6.7 A Toy Model: Random Agent Networks ............110 6.7.1 Using self-organization to improve performance . 112 6.7.2 Simulation Results ....................113 6.7.3 RAN Discussion .....................115 6.8 Conclusions ............................123 7 Self-organizing Artifacts 125 7.1 A Scenario .............................126 7.2 Requirements for self-organizing artifacts ...........127 7.3 Achieving self-organization ...................129 7.4 Learning to communicate ....................130 7.5 Learning to cooperate .......................131 7.6 Learning to coordinate ......................133 7.7 Conclusions ............................135 8 Conclusions 137 8.1 Achievements ...........................138 8.1.1 Limitations ........................139 8.2 Future Work ............................140 8.3 Philosophical Implications ....................142 8.3.1 Objectivity or Subjectivity? Contextuality! ......142 8.3.2 The Benefits of Self-organization ............143 Bibliography 145 Glossary 167 Index 171 LIST OF FIGURES 1.1 Thesismap ............................. 5 2.1 Is it a duck, a rabbit, or both? .................. 15 2.2 The same sphere seen from different angles .......... 16 3.1 Entropy increases and decreases for the same system .... 27 4.1 Diagram relating different stages of Methodology. ...... 44 4.2 Detailed diagram of Methodology. ............... 57 5.1 Screenshot of a part of the traffic grid .............. 65 5.2 Results for standard methods .................. 70 5.3 Results for self-organizing methods .............. 71 5.4 Full synchronization ....................... 75 5.5 Second results for standard methods .............. 78 5.6 Second results for self-organizing methods .......... 79 5.7 Comparison of initial and average number of cars ...... 80 5.8 Simulation of the Wetstraat and intersecting streets ..... 83 5.9 Wetstraat results .......................... 85 5.10 Potential implementation of sotl-platoon. ............ 86 6.1 Asynchronous communication .................102 6.2 Response delay ..........................103 6.3 Hierarchy represented as a network ..............108 6.4 Dynamics of a random agent network of N = 25, K =5 . 111 6.5 RAN self-organization mechanism ...............112 6.6 Results for N = 15, K =1. ....................114 6.7 Results for N = 15, K =2. ....................116 6.8 Results for N = 15, K =5. ....................117 6.9 Results for N = 15, K = 15. ...................118 6.10 Results for N = 100, K =1. ...................119 6.11 Results for N = 100, K =2. ...................120 6.12 Results for N = 100, K =5. ...................121 vi LIST OF FIGURES vii 6.13 Results for N = 100, K = 100. ..................122 LIST OF TABLES 5.1 Parameters of NetLogo simulations. .............. 72 5.2 Vehicle count per hour, Wetstraat ................ 83 5.3 Emissions by idling engines on Wetstraat ........... 91 viii ABSTRACT Complex systems are usually difficult to design and control. There are several particular methods for coping with complexity, but there is no general approach to build complex systems. In this thesis I propose a methodology to aid engineers in the design and control of complex systems. This is based on the description of systems as self-organizing. Starting from the agent metaphor, the methodol- ogy proposes a conceptual framework and a series of steps to follow to find proper mechanisms that will promote elements to find solutions by actively interacting among themselves. The main premise of the methodology claims that reducing the “friction” of interactions be- tween elements of a system will result in a higher “satisfaction” of the system, i.e. better performance. A general introduction to complex thinking is given, since designing self-organizing systems requires a non-classical thought, while practical notions of complexity and self-organization are put forward. To illustrate the methodology, I present three case studies. Self-organizing traffic light controllers are proposed and studied with multi-agent simulations, outperforming traditional methods. Methods for improving communication within self-organizing bureaucracies are advanced, introducing a simple computational model to illustrate the benefits of self-organization. In the last case study, requirements for self-organizing artifacts in an ambient intelligence scenario are discussed. Philosophical implications of the conceptual framework are also put forward. ix x Abstract ACKNOWLEDGEMENTS It takes several years to build up a PhD. And it is impossible to do so alone. Along these years, many mentors, colleagues, and friends have influenced my research, formation, and life in different aspects. I am in debt with my promoters, Francis Heylighen, Diederik Aerts, and Bart D’Hooghe, who have shared their knowledge and experience, and whose support has gone well beyond the academic. Collaborations with Johan Bollen, Jan Broekaert, Paul Cilliers, Seung Bae Cools, Atin Das, Bruce Edmonds, Ang´elica Garc´ıa Vega, Stuart Kauff- man, Tom Lenaerts, Carlos de la Mora, Marko Rodriguez, Ilya Shmule- vich, Dan Steinbock, Jen Watkins, and Andy Wuensche have enriched my research. I have learned a lot from interacting with them. Hugues Bersini, Jean-Louis Denebourg, Marco Dorigo, Bernard Man- derick, Gregoire Nicolis, Ann Nowe, Luc Steels, Ren´eThomas, Jean Paul Van Bendegem, Frank Van Overwalle, and other professors at the VUB and ULB have provided me with great advice and inspiration. Being in Brussels, I had the opportunity to interact with several re- search groups akin to my interests in both VUB and ULB. I am grateful for the discussions and good times I’ve had with the members of the Evolu- tion, Complexity and Cognition Group and the Centrum Leo Apostel the AI Lab, IRIDIA, and CeNoLi. The ideas presented here were influenced especially by Ricardo Barbosa, Vassilios Basios, Bart De Vylder, Anselmo Garc´ıa Cant´uRos, Erden G¨oktepe, Carlos de la Mora, Marko Rodriguez, Francisco Santos, and Cl´ement Vidal. I appreciate the time, support, and advices given by Juan Juli´an Merelo Guerv´os and the GeNeura team at the Universidad de Granada. The five months I spent with them were crucial for the development of this thesis. The possibility of a PhD does not come out of nowhere.