Methodos Series 13
Methodological
Investigations
in Agent-Based
Modelling
Eric Silverman
With Applications for the Social Sciences
Methodos Series
Methodological Prospects in the Social Sciences
Volume 13
Editors
Daniel Courgeau, Institut National d’Études Démographiques
Robert Franck, Université Catholique de Louvain
Editorial Advisory Board
Peter Abell, London School of Economics
Patrick Doreian, University of Pittsburgh
Sander Greenland, UCLA School of Public Health
Ray Pawson, Leeds University
Cees van der Eijk, University of Amsterdam
Bernard Walliser, Ecole Nationale des Ponts et Chaussées, Paris
Björn Wittrock, Uppsala University
Guillaume Wunsch, Université Catholique de Louvain
This Book Series is devoted to examining and solving the major methodological
problems social sciences are facing. Take for example the gap between empirical and
theoretical research, the explanatory power of models, the relevance of multilevel
analysis, the weakness of cumulative knowledge, the role of ordinary knowledge in
the research process, or the place which should be reserved to “time, change and
history” when explaining social facts. These problems are well known and yet they
are seldom treated in depth in scientiﬁc literature because of their general nature.
So that these problems may be examined and solutions found, the series prompts
and fosters the settingup of international multidisciplinary research teams, and it
is work by these teams that appears in the Book Series. The series can also host
books produced by a single author which follow the same objectives. Proposals for
manuscripts and plans for collective books will be carefully examined.
The epistemological scope of these methodological problems is obvious and
resorting to Philosophy of Science becomes a necessity. The main objective of the
Series remains however the methodological solutions that can be applied to the
problems in hand. Therefore the books of the Series are closely connected to the
research practices.
More information about this series at http://www.springer.com/series/6279
Eric Silverman
Methodological
Investigations in
Agent-Based Modelling
With Applications for the Social Sciences
With contribution by
Daniel Courgeau • Robert Franck
Jakub Bijak • Jason Hilton
Jason Noble • John Bryden
Eric Silverman
MRC/CSO Social and Public Health
Sciences Unit
University of Glasgow
Glasgow, UK
Methodos Series
ISBN 978-3-319-72406-5 ISBN 978-3-319-72408-9 (eBook)
https://doi.org/10.1007/978-3-319-72408-9
Library of Congress Control Number: 2017962298
© The Author(s) 2018. This book is an open access publication.
Open Access This book is licensed under the terms of the Creative Commons Attribution 4.0 International
License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation,
distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons license and indicate if changes
were made.
The images or other third party material in this book are included in the book’s Creative Commons
license, unless indicated otherwise in a credit line to the material. If material is not included in the book’s
Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the
permitted use, you will need to obtain permission directly from the copyright holder.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional afﬁliations.
Printed on acid-free paper
This Springer imprint is published by Springer Nature
The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
For Takako:
You make it all worthwhile.
Foreword
Following the aims of the Methodos Series perfectly, this 13th volume on agentbased
models provides a general view of the problems raised by this approach and
shows how these problems may be solved.
These methods are derived from computer simulation studies used by mathematicians
and physicists. They are now applied in many social disciplines such as
artiﬁcial life (Alife), political sciences, evolutionary psychology, demography, and
many others. Those who introduced them often took care not to consider each social
science separately but to view them as a whole, incorporating a wide spectrum of
social processes – demographic, economic, sociological, political, and so on.
Rather than modelling speciﬁc data, this approach models theoretical ideas and
is based on computer simulation. Its aim is to understand how the behaviour of
biological, social, or more complex systems arises from the characteristics of the
individuals or agents composing the said system. As Billari and Prskawetz (2003,
p. 42) said,
Different to the approach of experimental economics and other ﬁelds of behavioural science
that aim to understand why speciﬁc rules are applied by humans, agent-based computational
models pre-suppose rules of behaviour and verify whether these micro based rules can
explain macroscopic regularities.
This is, therefore, a bottom-up approach, with population-level behaviour emerging
from rules of behaviour of autonomous individuals. These rules need to be
clearly discussed; unfortunately, this approach is now used without sufﬁcient
discussions in many social sciences. It eliminates the need for empirical data on
personal or social characteristics to explain a phenomenon, as it is based on simple
decision-making rules followed by individuals, which can explain some real-world
phenomena. But how can we ﬁnd these rules? As Burch (2003, p. 251) puts it,
A model explains some real-world phenomenon if a) the model is appropriate to the realworld
system [...] and b) if the model logically implies the phenomenon, in other words, if
the phenomenon follows logically from the model as speciﬁed to ﬁt a particular part of the
real world.
vii
viii Foreword
Also, a theoretical model of this kind cannot be validated in the same way as an
empirical model with the “covering law” approach, which hinders social research
and leads to a pessimistic view of the explanatory power of the social sciences. In
Franck’s words (Franck 2002, p. 289),
But, one has ceased to credit deduction with the power of explaining phenomena.
Explaining phenomena means discovering principles which are implied by the phenomena.
It does not mean discovering phenomena which are implied by the principles.
As the agent-based approach focuses on the mechanisms driving the actions
of individuals or agents, it will simulate the evolution of such a population from
simple rules of behaviour. It may thus use game theory, complex systems theory,
emergence, evolutionary programming and – to introduce randomness – Monte
Carlo methods. It may also use survey data, not to explain the phenomenon studied,
but only to verify if the parameters used in the simulation lead to a behaviour similar
to the one observed in the survey.
As we have already said, such an approach raises many problems which this
volume will try to answer. We will present here these main problems, letting the
reader see how Silverman has treated it.
The ﬁrst problem is that these models “are intended to represent the import and
impact of individual actions on the macro-level patterns observed in a complex
system” (Courgeau et al. 2017, p. 38). This implies that a phenomenon emerging
at the aggregate level can be entirely explained by individual behaviour. Holland
(2012, p. 48), however, states that agent-based models include “little provision for
agent conglomerates that provide building blocks and behaviour at a higher level
of organisation.” For instance, a multilevel study on the effects of an individual
characteristic (being a farmer) and the corresponding aggregate characteristic (the
proportion of farmers living in an area) on the probability of internal migration in
Norway shows that the effects are contradictory (Courgeau 2007): it seems hard
to explain a macro-characteristic acting positively by a micro-characteristic acting
negatively. In fact, micro-level rules are often hard to link to aggregate-level rules,
and I believe that aggregate-level rules cannot be modelled with a purely micro
approach, for they transcend the behaviours of the component agents.
The second problem is that this approach is basically bottom-up. However, it
seems important to take into consideration simultaneously a top-down process from
higher-level properties to lower-level entities. More speciﬁcally, we should speak
of a micro-macro link (Conte et al. 2012, p. 336) that “is the loop process by
which behaviour at the individual level generates higher-level structures (bottomup
process), which feedback to the lower level (top-down), sometimes reinforcing
the producing behaviour either directly or indirectly”. The bottom-up approach of
a standard agent-based model cannot take such a reciprocal micro-macro link into
account, given that it only simulates one level of analysis.
The third problem concerns the validation of an agent-based model. Such an
approach imitates human behaviour using some well-chosen mechanisms. It may
be judged successful when it accurately reproduces the structure of this behaviour.
Foreword ix
Determining success, however, requires a method very different from the standard
tests used to verify the validity of the effects of different characteristics in the
other approaches. Such tests can be performed in the natural sciences but are more
difﬁcult in the social sciences. As Küppers and Lenhard observe (Günter et al. 2005,
paragraph 1.3),
The reliability of the knowledge produced by computer simulation is taken for granted if
the physical model is correct. In the second case of social simulations in general there is
no theoretical model on which one could rely. The knowledge produced in this case seems
to be valid if some characteristic of the social dynamics known from experience with the
social world are reproduced by the simulation.
To determine if such an exploration has been successful, we need to consider
different aspects. First, how do we test that there are no other models offering a
better explanation of the observed phenomenon? Researchers often try out different
kinds of models so they can choose the one most consistent with empirical data. But
this hardly solves the problem, as there is an inﬁnity of models that can predict the
same empirical result as well or even better. Second, how do we test that the chosen
model has a good ﬁt with the observed data? Unfortunately, there is no clearly
deﬁned procedure for testing the ﬁt of a simulation model, such as signiﬁcance
tests for the approaches described earlier. We can conclude that there are no clear
veriﬁcation and validation procedures for agent-based models in the social sciences.
While the agent-based approach appears to resemble event-history analysis, for it
focuses on individual behaviour, it nevertheless aims to explain collective behaviour.
At that point, the key question is: how do we generate macroscopic regularity using
simple individual rules? Conte et al. (2012, p. 340) perfectly describe the difﬁculties
encountered:
First, how to ﬁnd out the simple local rules? How to avoid ad hoc and arbitrary
explanations? As already observed, one criterion has often been used, i.e., choose the
conditions that are sufﬁcient to generate a given effect. However, this leads to a great deal
of alternative options, all of which are to some extent arbitrary.
Without factoring in the inﬂuence of networks on individual behaviour, we can
hardly obtain a macro behaviour merely by aggregating individual behaviours.
To obtain more satisfactory models, we must introduce decision-making theories.
Unfortunately, the choice of theory is inﬂuenced by the researcher’s discipline and
can produce highly divergent results for the same phenomenon studied.
In order to go further, Chap. 9, co-authored by Jakub Bijak, Daniel Courgeau,
Robert Franck and Eric Silverman, proposes for demography the enlargement of
agent-based models to a model-based research. This will not be a new paradigm in
the traditional sense, as with the cross-sectional, the cohort, the event-history and the
multilevel approaches, but a new way to overcome the limitations of demographic
knowledge. It is a research programme which adds a new avenue of empirical
relevance to demographic research. The examples given in the following chapters,
despite the simplicity of the models used, give us a glimpse of the importance of
model-based demography.
x Foreword
I hope I have given to the reader of this volume a clear idea of its importance for
social sciences.
Mougins, France Daniel Courgeau
August 2017
References
Billari, F., & Prskawetz, A. (2003). Agent-based computational demography: Using simulation to
improve our understanding of demographic behaviour. Heidelberg: Physica Verlag.
Burch, T. K. (2003). Data, models, theory and reality: The structure of demographic knowledge. In
F. Billari & A. Prskawetz (Eds.), Agent-based computational demography: Using simulation to
improve our understanding of demographic behaviour (pp. 19–40). Heidelberg: Physica Verlag.
Conte, R., Gilbert, N., Bonelli, G., Ciofﬁ-Revilla, C., Deffuant, G., Kertesz, V., Loreto, V., Moat,
S., Nadal, J.-p., Sanchez, A., Nowak, A., Flache, A., San Miguel, M., & Helbing, D. (2012).
Manifesto of computational social science. European Physical Journal Special Topics, 214,
325–346.
Courgeau, D. (2007). Multilevel synthesis: From the group to the individual. Dordrecht: Springer.
Courgeau, D., Bijak, J., Franck, R., & Silverman, E. (2017). Model-based demography: Towards
a research agenda. In A. Grow & J. Van Bavel (Ed.), Agent-based modelling and population
studies (Springer series on demographic methods and population analysis, pp. 29–51).
Berlin/Heidelberg: Springer.
Franck, R. (Ed.). (2002). The explanatory power of models: Bridging the gap between empirical
and theoretical research in the social sciences (Methodos series 1). Boston/Dordrecht/London:
Kluwer Academic.
Holland, J. H. (2012). Signals and boundaries. Cambridge: MIT Press.
Küppers, G., & Lenhard, J. (2005). Validation of simulation: Patterns in the social and natural
sciences. Journal of Artiﬁcial Societies and Social Simulation, 8(4), 3.
Acknowledgements
This volume covers more than 10 years of research, stretching back to the start
of my PhD in 2003. Over this lengthy period, this work has been supported by a
number of institutions and funders. Chapters 1 through 8 are substantially based
on my PhD research, completed at the School of Computing at the University of
Leeds from 2003 to 2007. Chapters 9 through 11 are based on a work completed
at the University of Southampton between 2010 and 2015 during the Care Life
Cycle Project, funded by Engineering and Physical Sciences Research Council
grant EP/H021698/1. Teesside University supported my work on the manuscript
between 2015 and 2017. The completion of the manuscript was also supported by
the Complexity in Health Improvement programme at the MRC/CSO Social and
Public Health Sciences Unit at the University of Glasgow, with QQR funding code
MC_UU_12017/14.
Of course this work has also beneﬁtted from the collaboration, vital input and
feedback from a host of colleagues and friends. At Leeds, my thanks go out to
Jason Noble for being such a good friend, close colleague and trusted conﬁdant
throughout my years in research; Seth Bullock for his supervision and collaboration
during my PhD; David Hogg for taking on supervision duties and supporting me
during a critical period in my PhD; John Bryden for being a close collaborator and
dear friend and for introducing me to Babylon 5; and Roger Boyle and Peter Jimack
for being so supportive throughout my PhD studies.
At Southampton, thanks once again go to Jason and Seth, for supporting me
throughout my postdoc experience and for fostering a great community within the
research group. Special mention must go to Jakub Bijak, close collaborator on
key chapters of this volume, trusted friend and a spectacularly talented researcher.
Without his support and encouragement, this volume would never have come into
being. Part III of this volume came to life thanks to exciting collaborations with
Daniel Courgeau, Robert Franck and Jakub, and I feel very privileged to have been
part of those conversations. I was also fortunate to be able to discuss the future
of model-based demography with Thomas Burch, whose work I hold in very high
esteem.
xi
xii Acknowledgements
At Teesside, I was lucky to share ofﬁces with Ian Wood, who was a great source
of support and a great colleague, without whom my time there would have been
much poorer. Thanks also to Mark Cavazza for his support during his time there
and his dedication to the union. Thanks as well to colleagues Julie Porteous, Fred
Charles and The Anh Han – I am glad we will continue working together in the
coming years.
For all their encouragement and kindness, I am indebted to my parents, Lewis
and Linda Silverman, who have never failed to be a continuous source of inspiration
throughout my academic career. Takashi Ikegami has been an inspiration to me
since we ﬁrst met in Boston in 2004; his enthusiasm and belief in Alife and his
willingness to break boundaries in research have been a pleasure to experience. A
heartfelt thanks also go to my friend and former housemate David Mitchell, who
deﬁned my years in Leeds and made my time there a true joy.
Last but certainly not least, thank you to my wife, Takako, who has showed
such endless patience with my strange working hours, bizarre sleeping schedule
and rambling monologues on obscure topics. Thank you for sticking with me, even
when my brain was wandering off somewhere else entirely.
Contents
Part I Agent-Based Models
1 Introduction ................................................................. 3
1.1 Overview.............................................................. 3
1.2 Artiﬁcial Life as Digital Biology .................................... 4
1.2.1 Artiﬁcial Life as Empirical Data-Point .................... 4
1.3 Social Simulation and Sociological Relevance ..................... 5
1.3.1 Methodological Concerns in Social Simulation........... 5
1.4 Case Study: Schelling’s Residential Segregation Model ........... 6
1.4.1 Implications of Schelling’s Model ......................... 6
1.5 Social Simulation in Application: The Case of Demography ...... 7
1.5.1 Building Model-Based Demography ...................... 7
1.6 General Summary .................................................... 7
1.6.1 Alife Modelling............................................. 8
1.6.2 Simulation for the Social Sciences......................... 8
1.6.3 Schelling’s Model as a Case Study in Modelling ......... 8
1.6.4 Developing a Model-Based Demography ................. 9
1.6.5 General Conclusions of the Text: Messages for the
Modeller..................................................... 9
1.6.6 Chapter Summaries......................................... 10
1.6.7 Contributions ............................................... 13
References.................................................................... 14
2 Simulation and Artiﬁcial Life ............................................. 17
2.1 Overview.............................................................. 17
2.2 Introduction to Simulation Methodology ........................... 18
2.2.1 The Goals of Scientiﬁc Modelling ......................... 18
2.2.2 Mathematical Models ...................................... 18
2.2.3 Computational Models ..................................... 19
2.2.4 The Science Versus Engineering Distinction.............. 20
xiii
xiv Contents
2.2.5 Connectionism: Scientiﬁc Modelling in Psychology ..... 21
2.2.6 Bottom-Up Modelling and Emergence .................... 22
2.3 Evolutionary Simulation Models and Artiﬁcial Life ............... 22
2.3.1 Genetic Algorithms and Genetic Programming ........... 22
2.3.2 Evolutionary Simulations and Artiﬁcial Life.............. 23
2.3.3 Bedau and the Challenges Facing ALife .................. 24
2.4 Truth in Simulation: The Validation Problem....................... 26
2.4.1 Validation and Veriﬁcation in Simulation ................. 26
2.4.2 The Validation Process in Engineering Simulations ...... 26
2.4.3 Validation in Scientiﬁc Simulations: Concepts of Truth .. 27
2.4.4 Validation in Scientiﬁc Models: Kuppers and
Lenhard Case Study ........................................ 28
2.5 The Connection Between Theory and Simulation .................. 29
2.5.1 Simulation as ‘Miniature Theories’........................ 29
2.5.2 Simulations as Theory and Popperian Falsiﬁcationism ... 29
2.5.3 The Quinean View of Science.............................. 30
2.5.4 Simulation and the Quinean View ......................... 31
2.6 ALife and Scientiﬁc Explanation .................................... 32
2.6.1 Explanation Through Emergence .......................... 32
2.6.2 Strong vs Weak Emergence ................................ 33
2.6.3 Simulation as Thought Experiment ........................ 34
2.6.4 Explanation Compared: Simulations vs
Mathematical Models ...................................... 35
2.7 Summary and Conclusions .......................................... 36
References.................................................................... 36
3 Making the Artiﬁcial Real ................................................. 39
3.1 Overview.............................................................. 39
3.2 Strong vs. Weak Alife and AI ....................................... 40
3.2.1 Strong vs. Weak AI: Creating Intelligence ................ 40
3.2.2 Strong vs. Weak Alife: Creating Life?..................... 40
3.2.3 Deﬁning Life and Mind .................................... 40
3.3 Levels of Artiﬁciality ................................................ 41
3.3.1 The Need for Deﬁnitions of Artiﬁciality .................. 41
3.3.2 Artiﬁcial1
: Examples and Analysis ........................ 42
3.3.3 Artiﬁcial2
: Examples and Analysis ........................ 42
3.3.4 Keeley’s Relationships Between Entities .................. 43
3.4 ‘Real’ AI: Embodiment and Real-World Functionality ............ 43
3.4.1 Rodney Brooks and ‘Intelligence Without Reason’....... 43
3.4.2 Real-World Functionality in Vision and Cognitive
Research..................................................... 44
3.4.3 The Differing Goals of AI and Alife: Real-World
Constraints .................................................. 44
Contents xv
3.5 ‘Real’ Alife: Langton and the Information Ecology................ 45
3.5.1 Early Alife Work and Justiﬁcations for Research ......... 45
3.5.2 Ray and Langton: Creating Digital Life?.................. 45
3.5.3 Langton’s Information Ecology............................ 46
3.6 Toward a Framework for Empirical Alife ........................... 47
3.6.1 A Framework for Empirical Science in AI ................ 47
3.6.2 Newell and Simon Lead the Way .......................... 48
3.6.3 Theory-Dependence in Empirical Science ................ 49
3.6.4 Artiﬁcial Data in Empirical Science ....................... 50
3.6.5 Artiﬁcial Data and the ‘Backstory’ ........................ 53
3.6.6 Silverman and Bullock’s Framework: A PSS
Hypothesis for Life ......................................... 54
3.6.7 The Importance of Backstory for the Modeller ........... 55
3.6.8 Where to Go from Here .................................... 56
3.7 Summary and Conclusions .......................................... 57
References.................................................................... 58
4 Modelling in Population Biology .......................................... 61
4.1 Overview.............................................................. 61
4.2 Levins’ Framework: Precision, Generality, and Realism ........... 62
4.2.1 Description of Levins’ Three Dimensions................. 62
4.3 Levins’ L1, L2 and L3 Models: Examples and Analysis ........... 63
4.3.1 L1 Models: Sacriﬁcing Generality......................... 63
4.3.2 L2 Models: Sacriﬁcing Realism ........................... 64
4.3.3 L3 Models: Sacriﬁcing Precision .......................... 64
4.4 Orzack and Sober’s Rebuttal......................................... 64
4.4.1 The Fallacy of Clearly Delineated Model Dimensions ... 64
4.4.2 Special Cases: The Inseparability of Levins’ Three
Factors....................................................... 65
4.5 Resolving the Debate: Intractability as the Fourth Factor .......... 66
4.5.1 Missing the Point? Levins’ Framework as
Pragmatic Guideline ........................................ 66
4.5.2 Odenbaugh’s Defence of Levins ........................... 66
4.5.3 Intractability as the Fourth Factor: A Reﬁnement......... 67
4.6 A Levinsian Framework for Alife ................................... 69
4.6.1 Population Biology vs. Alife: A Lack of Data ............ 69
4.6.2 Levinsian Alife: A Framework for Artiﬁcial Data? ....... 70
4.6.3 Resembling Reality and Sites of Sociality ................ 70
4.6.4 Theory-Dependence Revisited ............................. 72
4.7 Tractability Revisited ................................................ 72
4.7.1 Tractability and Braitenberg’s Law ........................ 72
4.7.2 David Marr’s Classical Cascade ........................... 74
4.7.3 Recovering Algorithmic Understanding................... 75
4.7.4 Randall Beer and Recovering Algorithmic
Understanding .............................................. 75
4.7.5 The Lure of Artiﬁcial Worlds .............................. 76
xvi Contents
4.8 Saving Simulation: Finding a Place for Artiﬁcial Worlds .......... 77
4.8.1 Shifting the Tractability Ceiling ........................... 77
4.8.2 Simulation as Hypothesis-Testing ......................... 78
4.9 Summary and Conclusion............................................ 79
References.................................................................... 80
Part II Modelling Social Systems
5 Modelling for the Social Sciences ......................................... 85
Eric Silverman and John Bryden
5.1 Overview.............................................................. 85
5.2 Agent-Based Models in Political Science ........................... 86
5.2.1 Simulation in Social Science: The Role of Models ....... 86
5.2.2 Axelrod’s Complexity of Cooperation..................... 87
5.3 Lars-Erik Cederman and Political Actors as Agents ............... 87
5.3.1 Emergent Actors in World Politics a Modelling
Manifesto ................................................... 87
5.3.2 Criticism from the Political Science Community ......... 88
5.3.3 Areas of Contention: The Lack of ‘Real’ Data............ 89
5.4 Cederman’s Model Types: Examples and Analysis ................ 89
5.4.1 Type 1: Behavioural Aspects of Social Systems .......... 89
5.4.2 Type 2: Emerging Conﬁgurations.......................... 90
5.4.3 Type 3: Interaction Networks .............................. 91
5.4.4 Overlap in Cederman’s Categories......................... 91
5.5 Methodological Peculiarities of the Political Sciences ............. 92
5.5.1 A Lack of Data: Relating Results to the Real World...... 92
5.5.2 A Lack of Hierarchy: Interdependence of Levels
of Analysis .................................................. 93
5.5.3 A Lack of Clarity: Problematic Theories .................. 93
5.6 In Search of a Fundamental Theory of Society ..................... 93
5.6.1 The Need for a Fundamental Theory ...................... 93
5.6.2 Modelling the Fundamentals ............................... 94
5.7 Systems Sociology: A New Approach for Social Simulation?..... 95
5.7.1 Niklas Luhmann and Social Systems ...................... 95
5.7.2 Systems Sociology vs. Social Simulation ................. 96
5.8 Promises and Pitfalls of the Systems Sociology Approach ........ 97
5.8.1 Digital Societies? ........................................... 97
5.8.2 Rejecting the PSS Hypothesis for Society................. 98
5.9 Social Explanation and Social Simulation .......................... 98
5.9.1 Sawyer’s Analysis of Social Explanation ................. 99
5.9.2 Non-reductive Individualism............................... 99
5.9.3 Macy and Miller’s View of Explanation................... 101
5.9.4 Alife and Strong Emergence ............................... 102
5.9.5 Synthesis .................................................... 102
5.10 Summary and Conclusion............................................ 103
References.................................................................... 104
Contents xvii
6 Analysis: Frameworks and Theories for Social Simulation............ 107
6.1 Overview.............................................................. 107
6.2 Frameworks and ALife: Strong ALife .............................. 108
6.2.1 Strong ALife and the Lack of ‘Real’ Data ................ 108
6.2.2 Artiﬁcial1
vs Artiﬁcial2
: Avoiding the Distinction ........ 108
6.2.3 Information Ecologies: The Importance of
Back-stories ................................................. 108
6.3 Frameworks and ALife: Weak ALife................................ 109
6.3.1 Artiﬁcial1
vs. Artiﬁcial2
: Embracing the Distinction ..... 109
6.3.2 Integration of Real Data: Case Studies .................... 110
6.3.3 Backstory: Allowing the Artiﬁcial ......................... 110
6.4 The Legacy of Levins ................................................ 111
6.4.1 The 3 Types: A Useful Hierarchy?......................... 111
6.4.2 Constraints of the Fourth Factor ........................... 112
6.5 Frameworks and Social Science ..................................... 112
6.5.1 Artiﬁcial1
vs. Artiﬁcial2
: A Useful Distinction? .......... 112
6.5.2 Levins: Still Useful for Social Scientists? ................. 113
6.5.3 Cederman’s 3 Types: Restating the Problem .............. 115
6.5.4 Building the Framework: Unifying Principles for
Biology and Social Science Models ....................... 116
6.5.5 Integration of Real Data .................................... 117
6.6 Views from Within Social Simulation............................... 118
6.6.1 Finding a Direction for Social Simulation................. 118
6.6.2 Doran’s Perspective on the Methodology
of Artiﬁcial Societies ....................................... 119
6.6.3 Axelrod and Tesfatsion’s Perspective: The
Beginner’s Guide to Social Simulation .................... 120
6.7 Summary and Conclusions .......................................... 121
References.................................................................... 122
7 Schelling’s Model: A Success for Simplicity ............................. 125
7.1 Overview.............................................................. 125
7.2 The Problem of Residential Segregation ............................ 126
7.2.1 Residential Segregation as a Social Phenomenon......... 126
7.2.2 Theories Regarding Residential Segregation .............. 126
7.3 The Chequerboard Model: Individual Motives in Segregation..... 127
7.3.1 The Rules and Justiﬁcations of the Model................. 127
7.3.2 Results of the Model: Looking to the Individual.......... 127
7.3.3 Problems of the Model: A Lack of Social Structure ...... 128
7.4 Emergence by Any Other Name: Micromotives
and Macrobehaviour ................................................. 128
7.4.1 Schelling’s Justiﬁcations: A Valid View of Social
Behaviour? .................................................. 128
7.4.2 Limiting the Domain: The Acceptance
of Schelling’s Result........................................ 129
xviii Contents
7.4.3 Taylor’s Sites of Sociality: One View of the
Acceptance of Models...................................... 130
7.4.4 The Signiﬁcance of Taylor: Communicability
and Impact .................................................. 130
7.5 Fitting Schelling to the Modelling Frameworks .................... 131
7.5.1 Schelling and Silverman-Bullock: Backstory ............. 131
7.5.2 Schelling and Levins-Silverman: Tractability ............. 131
7.5.3 Schelling and Cederman: Avoiding Complexity .......... 132
7.6 Lessons from Schelling .............................................. 132
7.6.1 Frameworks: Varying in Usefulness ....................... 132
7.6.2 Tractability: A Useful Constraint .......................... 133
7.6.3 Backstory: Providing a Basis .............................. 133
7.6.4 Artiﬁciality: When it Matters .............................. 134
7.6.5 The Practical Advantages of Simplicity ................... 135
7.7 Schelling vs Doran and Axelrod..................................... 136
7.7.1 Doran’s View: Supportive of Schelling?................... 136
7.7.2 Schelling vs Axelrod: Heuristic Modelling ............... 137
7.8 Schelling and Social Simulation: General Criticisms .............. 137
7.8.1 Lack of ‘Real’ Data......................................... 137
7.8.2 Difﬁculties of Social Theory ............................... 138
7.8.3 Schelling and the Luhmannian Perspective ............... 138
7.8.4 Ramiﬁcations for Social Simulation and Theory.......... 139
7.9 The Final Hurdle: Social Explanation ............................... 140
7.9.1 Schelling’s Model and the Explanation Problem.......... 140
7.9.2 Implications for Luhmannian Social Explanation......... 141
7.10 Summary and Conclusions .......................................... 142
References.................................................................... 143
8 Conclusions.................................................................. 145
8.1 Overview.............................................................. 145
8.2 Lessons from Alife: Backstory and Empiricism .................... 145
8.2.1 Backstory in Alife .......................................... 145
8.2.2 Schelling’s Avoidance of the Issue ........................ 146
8.3 The Lure of Artiﬁcial Worlds ........................................ 147
8.3.1 Levinsian Modelling........................................ 147
8.3.2 Levins and Artiﬁcial Worlds ............................... 147
8.4 Modelling in Alife: Thoughts and Conclusions..................... 148
8.4.1 Lessons from Schelling..................................... 148
8.4.2 The Power of Simplicity ................................... 149
8.4.3 The Scope of Models ....................................... 149
8.5 The Difﬁculties of Social Simulation ............................... 150
8.5.1 Social Simulation and the Backstory ...................... 150
8.5.2 Social Simulation and Theory-Dependence ............... 150
8.5.3 Social Simulation and Explanation ........................ 151
Contents xix
8.6 Schelling’s Approach ................................................ 151
8.6.1 Schelling’s Methodological and Theoretical Stance ...... 151
8.6.2 Difﬁculties in Social Explanation.......................... 152
8.6.3 Schelling and Theory-Dependence ........................ 153
8.7 Luhmannian Modelling .............................................. 153
8.7.1 Luhmann and Theory-Dependence ........................ 153
8.7.2 Luhmann and New Social Theory ......................... 154
8.7.3 Luhmann and Artiﬁcial Worlds ............................ 154
8.8 Future Directions for Social Simulation............................. 155
8.8.1 Luhmannian Modelling as a Way Forward ................ 155
8.8.2 What Luhmann is Missing: Non-reductive
Individualism ............................................... 155
8.8.3 Integrating Lessons from ALife Modelling ............... 156
8.8.4 Using the Power of Simplicity ............................. 157
8.9 Conclusions and Future Directions .................................. 157
8.9.1 Forming a Framework ...................................... 157
8.9.2 Messages for the Modeller ................................. 159
8.9.3 The Tension Between Theory-Dependence
and Theoretical Backstory ................................. 160
8.9.4 Putting Agent-Based Modelling for the Social
Sciences into Practice ...................................... 161
References.................................................................... 162
Part III Case Study: Simulation in Demography
9 Modelling in Demography: From Statistics to Simulations............ 167
Jakub Bijak, Daniel Courgeau, Robert Franck, and Eric Silverman
9.1 An Introduction to Demography..................................... 167
9.2 The Historical Development of Demography ....................... 168
9.2.1 Early Statistical Tools in Demography .................... 169
9.2.2 Cohort Analysis............................................. 170
9.2.3 Event-History Analysis..................................... 170
9.2.4 Multilevel Approaches ..................................... 171
9.2.5 Cumulativity ................................................ 171
9.3 Uncertainty, Complexity and Interactions in Population
Systems ............................................................... 173
9.3.1 Uncertainty in Demographic Forecasts .................... 173
9.3.2 The Problem of Aggregation ............................... 173
9.3.3 Complexity vs. Simplicity.................................. 174
9.3.4 Addressing the Challenges ................................. 174
9.4 Moving Toward a Model-Based Demography ...................... 175
9.4.1 The Explanatory Capacity of Simulation.................. 176
9.4.2 The Difﬁculties of Demographic Simulation.............. 177
9.5 Demography and the Classical Scientiﬁc Programme.............. 177
9.6 Stages of Model-Based Demographic Research .................... 179
xx Contents
9.7 Overcoming the Limitations of Demographic Knowledge ......... 180
9.8 The Pragmatic Beneﬁts of Model-Based Demography............. 181
9.9 Beneﬁts of the Classical Scientiﬁc Approach....................... 182
9.10 Conclusions........................................................... 183
References.................................................................... 185
10 Model-Based Demography in Practice: I ................................ 189
Eric Silverman, Jakub Bijak, and Jason Hilton
10.1 Introduction........................................................... 189
10.2 Demographic Modelling Case Studies .............................. 189
10.2.1 The Decline of the Anasazi ................................ 190
10.2.2 The Wedding Ring.......................................... 191
10.3 Extending the Wedding Ring ........................................ 193
10.3.1 Model Motivations ......................................... 193
10.3.2 Simulated Individuals ...................................... 194
10.4 Extension Details..................................................... 196
10.4.1 Spatial and Demographic Extensions...................... 196
10.4.2 Health Status Component .................................. 197
10.4.3 Agent Behaviours and Characteristics ..................... 197
10.4.4 Simulation Properties....................................... 198
10.5 Simulation Results ................................................... 199
10.5.1 Population Change ......................................... 199
10.5.2 Simple Sensitivity Analysis ................................ 199
10.5.3 Scenario Generation Example ............................. 201
10.5.4 Spatial Distribution of Informal Care...................... 203
10.5.5 Sensitivity Analysis Using Emulators ..................... 204
10.6 The Wedding Doughnut: Discussion and Future Directions ....... 206
10.7 General Conclusions ................................................. 207
References.................................................................... 209
11 Model-Based Demography in Practice: II ............................... 211
Eric Silverman, Jason Noble, Jason Hilton, and Jakub Bijak
11.1 Introduction........................................................... 211
11.2 Model Motivations ................................................... 211
11.3 The ‘Linked Lives’ Model ........................................... 212
11.3.1 Basic Model Characteristics ............................... 212
11.3.2 Health Status Component .................................. 213
11.3.3 Agent Behaviours and Demographic Elements ........... 215
11.4 Results ................................................................ 216
11.4.1 Parameter Sweeps .......................................... 216
11.4.2 Sensitivity Analysis with Emulators ....................... 219
11.5 The Power of Scenario-Based Approaches ......................... 221
References.................................................................... 222
12 Conclusions.................................................................. 225
12.1 Model-Based Demography: The Story so Far ...................... 225
12.2 The Practice of Model-Based Demography......................... 226
Contents xxi
12.3 Limitations of the Model-Based Approach ......................... 227
12.3.1 Demographic Social Simulation ........................... 227
12.3.2 Demographic Systems Sociology.......................... 228
12.4 The Future of Model-Based Demography .......................... 229
12.5 Model-Based Demography as an Exemplar for Social
Science Modelling.................................................... 230
12.5.1 The Advantages of Developing a Framework ............. 231
12.5.2 Model-Based Social Sciences.............................. 232
12.6 A Manifesto for Social Science Modelling? ........................ 233
References.................................................................... 234
Acronyms
In general I have attempted to keep this volume free of excessive abbreviations, but
the acronyms below will appear at times given their widespread usage in related
ﬁelds of research.
ABCD Agent-Based Computational Demography. This term describes an
approach to the discipline of demography which incorporates agent-based
modelling.
ABM Agent-Based Model. These are computer simulations designed to examine
the behaviour and interactions of autonomous agents.
ABSS Agent-Based Social Simulation. An approach to social simulation which
explicitly focuses on the use of agent-based models.
xxiii
Part I
Agent-Based Models
This ﬁrst part of the text examines the theory and practice of computational
modelling, much of it viewed through the lens of artiﬁcial life. Artiﬁcial life, or
Alife for short, is a discipline that focuses on ‘the simulation and synthesis of living
systems’, most frequently through the use of simulation. Using agent-based models,
evolutionary algorithms, and other recent innovations in simulation and robotics,
Alife researchers hope to unravel the mysteries of how life develops and evolves in
the real world by studying digital manifestations of ‘life as it could be’.
As Alife has itself evolved over the years, researchers have sought closer
connections to the real world and to real biology. This has brought about difﬁcult
questions concerning the integration of real data into simulated worlds, and the
relationship between digital biology and physical biology. As early Alife has slowly
given way to a greater desire for empirical relevance, it has become increasingly
important to understand the potential role Alife and Alife-inspired approaches can
play in understanding real biological systems.
Of course the difﬁculties inherent in modelling complex biological processes
and populations are familiar to population biologists just as much as digital ones
– perhaps even more so, given the short history of Alife. We will examine in
detail some of the theoretical frameworks developed by population biologists in
order to develop their models and position them as a valid form of enquiry,
and investigate how we might use these frameworks as a way to understand and
categorise computational modelling efforts in disciplines such as Alife. In so doing,
we will lay the foundations for Part II, in which we will investigate the additional
modelling complexities we encounter when we begin to model social systems.
Chapter 1
Introduction
1.1 Overview
As computer simulation has developed as a methodology, so its range of applications
has grown across different ﬁelds. Beyond the use of mathematical models for
physics and engineering, simulation is now used to investigate ﬁelds as varied and
disparate as political science, psychology, evolutionary biology, and many other
disciplines.
With simulation becoming such a common adjunct to conventional empirical
research, debate regarding the methodological merits of computer simulation
continues to develop. Some ﬁelds, artiﬁcial life being the primary example used
in this text, have developed using computer simulation as a central driving force.
In such a case, researchers have developed theoretical frameworks to delineate the
function and purpose of computer simulation within their ﬁeld of study.
However, the expansion of computer simulation into ﬁelds which use empirical
study as a central methodology means that new frameworks for the appropriate
use of simulation must develop. How might simulation enhance one’s use of
conventional empirical data? Can simulations provide additions to empiricallycollected
data-sets, or must simulation data be treated entirely differently? How
does theoretical bias inﬂuence the results of a simulation, and how can such biases
be investigated and accounted for?
The central goal of this text is to investigate these increasingly important
concerns within the context of simulation for the social sciences. Agent-based
models in particular have become a popular method for testing sociological
hypotheses that are otherwise difﬁcult or impossible to analyse empirically, and
as such a methodological examination of social simulations becomes critical as
social scientists begin to use such models to inﬂuence social policy. Without a clear
understanding of the relationship between social simulation and social sciences as a
whole, the use of models to explain social phenomena becomes difﬁcult to justify.
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_1
3
4 1 Introduction
Bearing in mind this central theme, this text will utilise a modelling example
which will be revisited regularly in Parts I and II. This example will serve as a
means for illustrating the important concepts described in the various modelling
frameworks under discussion, and for tying together these frameworks by showing
the effect of each upon the construction and implementation of a simulation. This
central example takes the form of a model of bird migration; this example seemed
most appropriate as this sort of problem can be examined through various modelling
means, from mathematical to agent-based computational models. The context and
purpose of this hypothetical model will vary from example to example, but the
central concern of developing an understanding of the behaviour of migratory birds
will remain throughout.
Toward the latter half of Part II, we will use the classic example of Schelling’s
residential segregation model (Schelling 1971) to discuss some particular methodological
points in detail. Part III will delve deeply into speciﬁc examples of
agent-based modelling work in the ﬁeld of demography in order to illustrate how
the modelling concepts discussed in Parts I and II can inﬂuence the practice of
modelling in social science.
1.2 Artiﬁcial Life as Digital Biology
The ﬁeld of artiﬁcial life provides a useful example of the development of theoretical
frameworks to underwrite the use of simulation models in research. The Artiﬁcial
Life conference bills itself as a gathering to discuss ‘the simulation and synthesis
of living systems’; with such potentially grandiose claims about the importance of
artiﬁcial life simulations, theoretical debate within the ﬁeld has been both frequent
and ﬁerce.
In the early days of Alife, Langton and other progenitors of this novel research
movement viewed simulation as a means to develop actual digital instantiations
of living systems. Beyond being an adjunct to biology, Alife was viewed as
digital biology, most famously described as the investigation of ‘life-as-it-could-be’
(Langton 1992). Ray boasted of his Tierra simulation’s explosion of varied digital
organisms (Ray 1994), and theorists proposed this sort of digital biology as a means
for divining the nature of living systems.
1.2.1 Artiﬁcial Life as Empirical Data-Point
Since these heady days Artiﬁcial life has sought more conventional forms of
methodological justiﬁcation, seeking to link simulation with more conventional
means of data-gathering in biology. This has lead to varying forms of theoretical
justiﬁcation within Alife, ranging from further explorations of Langton’s early ideas
1.3 Social Simulation and Sociological Relevance 5
(Bedau 1998; Silverman and Bullock 2004) to the use of Alife simulation as a form
of ‘opaque thought experiment’ (Di Paolo et al. 2000).
Within this text, these varying theoretical frameworks for Alife will be examined
in turn, both within the context of biology and within Alife itself. Once Alife
seeks direct links with conventional biology, theoretical justiﬁcation becomes
correspondingly more difﬁcult, and thus the debate must branch out into more
in-depth discussions of biological modelling methodology. An investigation of the
use of modelling in population biology, beginning with the somewhat-controversial
ideas of Levins (1966, 1968) provides a means for describing and categorising the
most important methodological elements of biological models. Having developed
an understanding of the complex relationship between biology and Alife, we can
then proceed to a discussion of the future of modelling within the social sciences.
1.3 Social Simulation and Sociological Relevance
Social simulation has appeared in the limelight within social science quite recently,
starting with Schelling’s well-known residential segregation model (Schelling 1978)
and continuing into Axelrod’s explorations of cooperative behaviour (Axelrod
1984). The development of simple algorithms and rules that can describe elements
of social behaviour has led to an increasing drive to produce simulations of social
systems, in the hopes that such systems can provide insight into the complexity of
human society.
The current state-of-the-art within social simulation relies upon the use of agentbased
models similar to those popularised in Alife. Cederman’s inﬂuential book
describing the use of such models in political science has helped to bolster an
increasing community of modellers who hope that such individual-based simulations
can reveal the emergence of higher-order complexity that we see around us in
human society (Cederman 1997). Social science being a ﬁeld where the empirical
collection of data is already a signiﬁcant difﬁculty, the prospect of using simulation
to produce insights regarding the formation and evolution of human society is an
enticing one for many.
1.3.1 Methodological Concerns in Social Simulation
Of course, with such possibilities comes great debate from within the social
science community. Proponents offer varying justiﬁcations of the potential power
of simulation in social science; Epstein echoes the Alife viewpoint by proposing
that social simulation can provide ‘generative social science,’ a means to generate
new empirical data-points (Epstein 1999). Similarly, Axelrod stresses the ability of
social simulation to enhance conventional empirical studies (Axelrod and Tesfatsion
2006).
6 1 Introduction
Others however are more cautious with their endorsement of social simulation.
Kluver and Stoica stress the difﬁculty in creating models consistent with social
theory (Klüver et al. 2003), noting that social systems do not lend themselves to the
same hierarchical deconstruction as some other complex systems. Others theorise
that social simulation faces the danger of incorporating vast theoretical biases into
its models, eliminating one of the potential strengths of social models: a means for
developing more general social theory (Silverman and Bryden 2007).
Further examinations of these questions within this text will seek to link such
ideas with the methodological frameworks developed within Alife modelling and
biology. While both ﬁelds display obvious differences in both methodological and
theoretical objectives, the philosophical difﬁculties facing agent-based modelling
in these contexts are much the same. In both cases the link between empirical datagathering
and simulated data-generation is difﬁcult to develop, and as a consequence
the use of simulation can be difﬁcult to justify without a suitable theoretical
justiﬁcation.
1.4 Case Study: Schelling’s Residential Segregation Model
Having developed a detailed comparison between the use of models in biology and
social science, this text will use Schelling’s residential segregation model as a case
study for examining the implications of the theoretical frameworks discussed and
outlined in that comparison. Schelling’s model is famously simple, its initial version
running on nothing more than a chequerboard, but its conclusions had a far-reaching
impact on social theory at the time (Schelling 1978). Schelling’s ideas regarding the
‘micromotives’ of individuals within a society, and the resulting effects upon that
larger society, sparked extensive discussion of the role of individuals in collective
social behaviour.
1.4.1 Implications of Schelling’s Model
With this in mind, our investigation will explore the reasons for Schelling’s great
success with such a simple model, and its ramiﬁcations for future modelling
endeavours. How did such an abstract formulation of the residential segregation
phenomenon become so powerful? What theoretical importance did Schelling
attribute to his model’s construction, and how did that inﬂuence his interpretation
of the results? Finally, how does his model illuminate both the strengths and
weaknesses of social simulation used for the purpose of developing social theory?
All of these questions bear upon our ﬁnal examination of the most appropriate
theoretical framework for social simulation as a whole.
1.6 General Summary 7
1.5 Social Simulation in Application: The Case of
Demography
Having developed some theoretical approaches to social simulation, we will need to
move on to discuss the establishment of these methods as a trusted and functional
element of the social scientist’s toolbox. We will take on this problem by investigating
the ﬁeld of demography, the study of human population change. Demography
is a fundamentally data-focused discipline, relying on at times vast amounts of
complicated survey data to understand and predict the future development of
populations (Silverman et al. 2011). We will investigate the core assumptions
underlying demographic research, discuss and analyse the methodological shifts
that have occurred in the ﬁeld over the last 350 years (Courgeau et al. 2017), and
develop a framework for a model-based demography that incorporates simulation as
a central conceit.
1.5.1 Building Model-Based Demography
In order to understand the challenges facing a model-based social science, we
will discuss several examples of agent-based approaches to demography. Starting
with some inspirational work from the early 2000s (Billari and Prskawetz 2003;
Axtell et al. 2002; Billari et al. 2007), we will move on to current work integrating
statistical demographic modelling directly into an agent-based approach. We will
examine the beneﬁts and the shortcomings of these models, and in the process
develop an understanding of the power of a scenario-based approach to the study
of future population change. Finally, we will evaluate the progress of model-based
demography thus far, and present some conclusions about the lessons we can take
from this in our future research efforts.
1.6 General Summary
This text is organised as essentially a three-part argument. In Part I, the theoretical
underpinnings of Alife are examined, and their relationship to similar modelling
frameworks within population biology. Part II reviews the current state-of-the-art
in simulation for the social sciences, with a view toward drawing comparisons
with Alife methodology. A subsequent analysis of theoretical frameworks for social
simulation as applied to a speciﬁc case study provides a means to draw these
disparate ideas together, and develop insight into the fundamental philosophical and
methodological concerns of simulation for the social sciences. Finally, in Part III we
take the speciﬁc example of demographic research and attempt to build a cohesive
8 1 Introduction
theoretical framework through which social simulation approaches can be integrated
productively with empirically-focused social science.
1.6.1 Alife Modelling
This portion of the text aims ﬁrst to describe the relatively new ﬁeld of artiﬁcial
life, and discuss its goals and implications. Once the background and import of
Alife is established, then the shortcomings and theoretical pitfalls of such models
are discussed. Given the strong association of Alife with biology and biological
modelling, the theoretical discussion includes in-depth analysis of a framework for
modelling in population biology proposed by Levins (1966, 1968). This analysis
allows the theoretical implications of Alife to be placed in a broader context in
preparation for the incorporation of further ideas from social science simulation.
1.6.2 Simulation for the Social Sciences
Agent-based modelling in the social sciences is a rather new development, similar
to Alife. Social scientists may protest that modelling of various types has been
ongoing in social science for centuries, and this is indeed true; however, this more
recent methodology presents some similarly novel methodological and theoretical
difﬁculties. This section of the text begins by describing the past and present
of agent-based modelling in the social sciences, discussing the contributions and
implications of each major development. Then, a discussion of current theoretical
concerns in agent-based models for social science proceeds, describing modelling
frameworks which attempt to categorise the various types of social simulations
evident thus far in the ﬁeld. Finally, an analysis of the problems of explanation via
simulation which are particularly critical for the social sciences allows us to develop
a broader understanding of these in a philosophical context.
1.6.3 Schelling’s Model as a Case Study in Modelling
Schelling’s model of residential segregation is notable for its impact and inﬂuence
amongst social scientists and modellers (Schelling 1978). Despite the model’s
simplicity, the illustration it provided of a problematic social issue provoked a
great deal of interest, both from social scientists interested in modelling and those
formulating empirical studies. This investigation of Schelling will focus on how
his model surpassed its simplicity to become so inﬂuential, and how this success
can inform our discussion of agent-based modelling as a potentially powerful
methodology in social science.
1.6 General Summary 9
1.6.4 Developing a Model-Based Demography
Demography is an old discipline, originating from a major conceptual shift in
the treatment of demographic events like birth, death and reproduction in the
seventeenth century (Graunt 1662). In the years since, demography has gone
through a series of methodological shifts, going from relatively straightforward
early statistical work to present-day microsimulation and multilevel modelling
approaches (Courgeau 2012). Simulation approaches to demography are now
gaining popularity, particularly in areas such as migration, where simulation offers
an opportunity to better understand the individual decision-making that plays a
vital role in such processes (Anna Klabunde and Frans Willekens 2016). In Part III
of this book, we will examine the methodological foundations of demography in
detail, and investigate how simulation approaches can contribute to this highly
empirical social science. We will present a proposal for a model-based approach to
demography which attempts to resolve the conceptual gaps between the empirical
focus of statistical demography and the explanatory and exploratory tendencies of
social simulation. We will then discuss some applied examples of model-based
demographic research and evaluate how these studies can inﬂuence our future efforts
both in demography and in the social sciences more generally.
1.6.5 General Conclusions of the Text: Messages for the
Modeller
By its nature, this text encompasses a number of different threads related to agentbased
modelling to bring the reader to an understanding of both the positives and
the negatives of this approach for the researcher who wishes to use simulation in
the social sciences. Each of the three portions of the text builds upon the previous,
with the goal of presenting modellers with both theoretical and practical concepts
they can apply in their own work. Part I of the text demonstrates the problems
and limitations of biologically-oriented agent-based models; such an approach is
inherently theory-dependent, and modellers must be aware of this fact and justify
the use of their model as a means to test and enhance their theories.
Part II of the text, focusing on simulation for the social sciences, describes the
current state of this ﬁeld and the various major disputes regarding its usefulness
to the social scientist. This new type of modelling approach provides both new
possibilities and new problems for the social scientist; the use of simulation can be a
difﬁcult balancing act for the researcher who wishes to provide useful conclusions.
Thus, the social scientist interested in modelling must be knowledgable regarding
these methodological difﬁculties, as analysed here, and avoid the impulse to produce
highly complex models which may fall foul of the guidelines discussed.
In order to reinforce these points, we discuss an example of a powerful,
successful, and simple model used within the social sciences: Schelling’s residential
10 1 Introduction
segregation model (Schelling 1971, 1978). In the context of the modelling frameworks
discussed in the previous portions, Schelling’s model provides a platform for
examining those frameworks in a detailed fashion. Schelling’s model demonstrates
that the most useful models are not the most complex; simplicity and analysability
are much more valuable than complexity for those who wish to understand the
phenomena being modelled. In essence, no model can do it all, and a knowledge of
the modelling frameworks under discussion here and their implications allows one
to understand the necessary balancing act of designing and implementing a model
in much greater depth.
Perhaps the most important balancing act related here is the tension between the
need for a modeller to provide a theoretical backstory and the desire to minimise a
model’s theory-dependent nature. This is a common thread running throughout the
text, whether the model in question is related to biology or social science. Modellers
who create a model without a theoretical backstory that provides a context may ﬁnd
themselves creating a model with no relevance except to itself, while those who
create a model with too great a degree of theory-dependence may ﬁnd themselves
warping their model into one restricted by theoretical bias, once again moving the
model further from real-world applicability. The notion of balancing acts in model
creation and implementation is often practiced intuitively by modellers, but yet this
tension between backstory and theory dependence is rarely discussed explicitly by
modellers in the literature.
Part III of the text brings us to the speciﬁc example of demography, a discipline
where agent-based modelling approaches have begun to take hold in certain areas
of enquiry. Building upon the foundations laid in previous chapters, the modelbased
demography framework described here presents a positive case-study for the
integration of simulation with empirically-focused social science. Example models
demonstrate how considered choices during model construction, development and
implementation produces results that add to demographic knowledge without letting
the simulations became unmanageable. The intention is for these models to serve as
positive examples of pragmatic, considered modelling practices; each of them has
limitations, but are still able to provide insight on the research questions they target.
1.6.6 Chapter Summaries
The analysis begins with an overall review of the philosophical issues and debates
facing simulation science in general. Chapter 2 focuses on these general concerns,
providing a summation of current thinking regarding issues of simulation methodology.
A large portion of this chapter focuses upon the problem of validation of
simulation results, which is an issue that is of great importance to the theoretical
frameworks under examination. A further discussion of the difﬁculties inherent in
linking the artiﬁcial with the natural provides a broader philosophical context for
the discussion.
1.6 General Summary 11
Chapter 3 picks up at this point, focusing on the efforts of Alife researchers to
make the artiﬁcial become ‘real.’ After introducing the concepts of ‘strong’ and
‘weak’ artiﬁcial life, the signiﬁcance of these two perspectives is discussed in the
context of the still-developing philosophical debates of Alife practitioners. A central
theme in this chapter is the drive to develop empirical Alife: simulations which can
supplement datasets derived from real-world data. Taking into account the problems
of validation discussed earlier and the two varying streams of Alife theory, a possible
theoretical framework for underwriting empirical Alife is developed.
Chapter 4 moves on to population biology, drawing upon modelling frameworks
developed within that discipline to strengthen our burgeoning theoretical backstory
for Alife. Levins’ three types of models, described in his seminal 1966 paper,
provoked a great deal of debate regarding the strengths and weaknesses of modelling
in biology, a debate which continues to rage today. After an analysis of Levins’ three
types, an expanded version of his framework is developed in the hope of providing
a more pragmatic theoretical position for the model-builder.
Chapter 5 focuses mainly upon a review of the current state-of-the-art in
simulation for the social sciences. Beginning with a look at early models, such as
Schelling’s residential segregation model (Schelling 1978) and Axelrod’s iterated
prisoner’s dilemma (Axelrod 1984), we move on to more current work including
Cederman’s work within political science (Cederman 1997). This leads to a review
of common criticisms of this growing ﬁeld and the methodological peculiarities
facing social-science modellers. These peculiarities are not limited to social simulation,
of course; social science as a whole has unique aspects to its theory and
practice which are an important consideration for the modeller.
Chapter 6 then proceeds with an analysis of social simulation in the context of
the theoretical frameworks and issues laid out thus far. First, an overall analysis
of Alife and related modelling issues in population biology gives us a set of
frameworks useful for that particular ﬁeld. Next, these theoretical concerns are
applied to social simulation in the hope of discovering the commonalities between
these two varieties of simulation science. This leads to a discussion of the possibility
of using social simulation to drive innovations in social theory as a whole; the work
of Luhmann is used as an example of one perspective that may prove valuable
in that respect (Luhmann 1995). Finally, having placed social simulation within a
theoretical framework, the debate regarding the usefulness of social simulation for
social explanation is summarised and discussed.
Chapter 7 extends the analysis begun in Chap. 5 by utilising a case study:
Schelling’s well-known residential segregation model (Schelling 1978). Schelling’s
model is noted for its simplicity: residential segregation is illustrated by a single
rule applied to individual agents on a simple two-dimensional grid. This chapter
investigates the reasons behind the powerful impact of Schelling’s abstract formulation,
placing the model in the theoretical constructs described thus far. The
implications of Schelling’s model on social theory is also discussed, with reference
to the Luhmannian modelling perspective described in the previous chapter.
12 1 Introduction
Chapter 8 offers a conclusion to the arguments laid out in Parts I and II.
Having examined Alife modelling, modelling in biology, and social simulation,
future directions for substantive modelling works are proposed. In the context of
social simulation speciﬁcally, the problems of validation and explanation introduced
earlier are revisited. The overall questions of methodological individualism in social
simulation are investigated as well, with an eye toward developing methods of
simulation which can transcend the perceived limitations on the explanatory power
of social science models. Having used Schelling as a case study for the modelling
frameworks under discussion, this chapter will also discuss how other modelling
methodologies may ﬁt cohesively into these frameworks.
Chapter 9 marks the beginning of Part III, in which we delve into the application
of agent-based modelling to the speciﬁc discipline of demography. This
chapter describes the historical evolution of the ﬁeld, detailing the cumulative
development of four successive methodological paradigms. From there we propose
a methodological framework for a model-based demography, in which simulation
helps demographers to overcome three key epistemological challenges within
the discipline and helps avoid the insatiable ‘beast’ of over-reliance on detailed
demographic data.
Chapter 10 moves beyond theoretical aspects of demography and dives into the
practice of agent-based modelling in the ﬁeld. We begin by discussing two examples
in brief: Axtell et al.’s model of the decline of the Anasazi (Axtell et al. 2002);
and Billari’s Wedding Ring model of partnership formation (Billari et al. 2007).
For our third, more detailed example, we will examine the Wedding Doughnut
– an extended version of the Wedding Ring model which incorporates statistical
demographic methods and adds a simple representation of individual health status
(Silverman et al. 2013a; Bijak et al. 2013). Sensitivity analysis using Gaussian
process emulators is also introduced as a means of understanding the impact of
model parameters on their interactions on the ﬁnal output of interest.
Chapter 11 focuses exclusively on a single model: the Linked Lives model
of social care supply and demand (Noble et al. 2012; Silverman et al. 2013b).
This model is a signiﬁcant leap forward in complexity compared to the Wedding
Doughnut, incorporating a simple economic system, spatial elements, partnership
formation/dissolution, social care need and provision, and migration. We examine
the model in detail, including another sensitivity analysis using Gaussian process
emulators, and discuss how the strengths of this model can serve as a useful
exemplar for future modelling efforts in demography.
Finally, Chap. 12 summarises our ﬁndings in Part III and links them to the
theoretical discussions presented earlier in the volume. We evaluate the current state
of model-based demography, and discuss how the development of this approach
can inform efforts to bring agent-based modelling to other areas of the social
sciences. Ultimately we will take model-based demography as a positive example of
a discipline taking new methods and weaving them gradually and thoughtfully into
the broader tapestry of demographic research. Demography beneﬁts particularly
from having a cumulative approach to methodology over the last three and a half
1.6 General Summary 13
centuries. Other disciplines can beneﬁt from the insights presented by model-based
demography, and in turn develop new approaches to simulation that may strengthen
other areas of social science alongside demography’s focus on empirical relevance.
1.6.7 Contributions
The major contributions of this text lie within its philosophical and methodological
study of modelling within both artiﬁcial life and the social sciences. These analyses
provide a novel perspective on agent-based modelling methodologies and their
relationship to more conventional empirical science. Other elements of the text
present a sort of anthropological study of modelling within these areas of science,
in the hope of providing a more cohesive view of the use and impact of simulation
in a broader context.
Elements of Chap. 3 were based upon a work published in the proceedings
for Artiﬁcial Life IX; this work aimed to develop a theoretical framework for
empirical studies in Alife by providing comparison with other, more established
ﬁelds of science. Chapter 4 was based substantially on a paper written by myself
and Seth Bullock describing the pitfalls of an approach to modelling that relies
upon ‘artiﬁcial worlds’; this work draws upon the papers of Levins, Braitenberg
and others. Elements of Chaps. 4 and 5 were drawn from a paper by myself and
John Bryden which was published in the proceedings for The European Conference
on Artiﬁcial Life in 2007. This paper proposed a new means of social simulation
which could provide a deeper insight into a fundamental social theory. Chapter 9
is based upon a collaborative paper written with Daniel Courgeau, Jakub Bijak and
Robert Franck which was published in an edited volume on agent-based modelling
for demography. Chapters 10 and 11 are based largely upon two collaborative
papers written with members of the Care Life Cycle Project at the University of
Southampton, which ran from 2010 to 2015.
In summary, this text provides a new synthesis of theoretical and practical
approaches to simulation science across different disciplines of the social sciences.
By integrating perspectives from Alife, biology and social science into a single
approach, this text provides a potential means to underwrite the use of simulation
within these ﬁelds as a means to generate new theory and new insight. Particularly
in ﬁelds relatively new to simulation, such as social science, the acceptance of this
methodology as a valid means of enquiry is a slow process; this text hopes to accelerate
the growth of simulation with this ﬁeld by providing a coherent theoretical
background to illustrate the unique strengths of computational modelling, while
simultaneously delineating its unique pitfalls. The detailed treatment of simulation
modelling in demography will further illustrate how relatively disparate frameworks
– in this case the data-centric demographic approach and the explanatory focus of
agent-based modelling – can be combined to produce new avenues of productive
enquiry.
14 1 Introduction
References
Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books.
Axelrod, R., & Tesfatsion, L. (2006). A beginner’s guide to agent-based modelling in the social
sciences. In L. Tesfatsion, & K. L. Judd (Eds.), Handbook of computational economics, vol. 2:
Agent-based computational economics, page Appendix A. Amsterdam: North Holland.
Axtell, R. L., Epstein, J. M., Dean, J. S., Gumerman, G. J., Swedlund, A. C., Harbeger, J.,
Chakravarty, S., Hammond, R., Parker, J., & Parker, M. (2002). Population growth and collapse
in a multi–agent model of the Kayenta Anasazi in Long House Valley. Proceedings of the
National Academy of Sciences, 99(3), 7275–7279.
Bedau, M. A. (1998). Philosophical content and method of artiﬁcial life. In T. W. Bynum, &
J. H. Moor (Eds.), The digital Phoenix: How computers are changing philosophy (pp. 135–
152). Oxford/Malden: Blackwell
Bijak, J., Hilton, J., Silverman, E., & Cao, V. (2013). Reforging the wedding ring: Exploring a
semi-artiﬁcial model of population for the united kingdom with Gaussian process emulators.
Demographic Research, 29(27), 729–766.
Billari, F., & Prskawetz, A. (2003). Agent-based computational demography: Using simulation to
improve our understanding of demographic behaviour. Heidelberg: Physica Verlag.
Billari, F., Aparicio Diaz, B., Fent, T., & Prskawetz, A. (2007). The “Wedding Ring”: An agentbased
marriage model based on social interaction. Demographic Research, 17, 3.
Cederman, L. E. (1997). Emergent actors in world politics: How states and nations develop and
dissolve. Princeton: Princeton University Press.
Courgeau, D. (2012). Probability and social science: Methodological relationships between the
two approaches (Methodos series, Vol. 10). Dordrecht: Springer.
Courgeau, D., Bijak, J., Franck, R., & Silverman, E. (2017). Model-based demography: Towards
a research agenda. In A. Grow & J. Van Bavel (Eds.), Agent-based modelling and population
studies (Springer series on demographic methods and population analysis, pp. 29–51).
Berlin/Heidelberg: Springer.
Di Paolo, E. A., Noble, J., & Bullock, S. (2000). Simulation models as opaque thought experiments.
In M. A. Bedau, et al. (Ed.), Artiﬁcial life VII (pp. 497–506). Cambridge, MA: MIT Press.
Epstein, J. (1999). Agent-based computational models and generative social science. Complexity,
4(5), 41–60.
Graunt, J. (1662). Natural and political observations mentioned in a following index, and made
upon the bills of mortality. London: Thomas Roycroft.
Klabunde, A., & Willekens, F. (2016). Decision-making in agent-based models of migration: State
of the art and challenges. European Journal of Population, 32(1), 73–97.
Klüver, J., Stoica, C., & Schmidt, J. (2003). Formal models, social theory and computer
simulations: Some methodical reﬂections. Journal of Artiﬁcial Societies and Social Simulation,
6(2), 8.
Langton, C. G. (1992). Life at the edge of chaos. In C. G. Langton, C. Taylor, J. D. Farmer, &
S. Rasmussen (Eds.), Artiﬁcial life II (pp. 41–91). New York: MacMillan.
Levins, R. (1966). The strategy of model-building in population biology. American Scientist, 54,
421–431.
Levins, R. (1968). Evolution in changing environments. Princeton: Princeton University Press.
Luhmann, N. (1995). Social systems. Stanford: Stanford University Press.
Noble, J., Silverman, E., Bijak, J., Rossiter, S., Evandrou, M., Bullock, S., Vlachantoni, A., &
Falkingham, J. (2012). Linked lives: The utility of an agent-based approach to modelling partnership
and household formation in the context of social care. In C. Laroque, J. Himmelspach,
R. Pasupathy, O. Rose, & J. M. Uhrmacher (Eds.), Proceedings of the 2012 Winter Simulation
Conference. Piscataway: IEEE.
Ray, T. S. (1994). An evolutionary approach to synthetic biology: Zen and the art of creating life.
Artiﬁcial Life, 1, 179–209.
References 15
Schelling, T. C. (1971). Dynamic models of segregation. Journal of Mathematical Sociology, 1,
143–186.
Schelling, T. C. (1978). Micromotives and macrobehavior. New York: W. W. Norton.
Silverman, E., & Bryden, J. (2007, to appear). From artiﬁcial societies to new social science theory.
In The Proceedings of ECAL 2007.
Silverman, E., & Bullock, S. (2004). Empiricism and artiﬁcial life. In J. Pollack, M. Bedau,
P. Husbands, T. Ikegami, & R. A. Watson (Eds.), Artiﬁcial Life IX: Proceedings of the
Ninth International Conference on the Simulation and Synthesis of Living Systems. MIT
Press/Bradford Books.
Silverman, E., Bijak, J., & Noble, J. (2011). Feeding the beast: Can computational demographic
models free us from the tyranny of data? In Advances in artiﬁcial life: ECAL 2011
(pp. 747–754). Cambridge, MA: MIT Press.
Silverman, E., Bijak, J., Hilton, J., Cao, V., & Noble, J. (2013a). When demography met social
simulation: A tale of two modelling approaches. Journal of Artiﬁcial Societies and Social
Simulation, 16(4), article 9.
Silverman, E., Hilton, J., Noble, J., & Bijak, J. (2013b). Simulating the cost of social care in an
ageing society. In W. Rekdalsbakken, R. T. Bye, & H. Zhang (Eds.), Proceedings of the 27th
European Conference on Modelling and Simulation (pp. 689–695). Dudweiler: Digitaldruck
Pirrot.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
Chapter 2
Simulation and Artiﬁcial Life
2.1 Overview
Before beginning this extensive analysis of agent-based modelling, the philosophical
and methodological background of simulation science must be discussed.
Scientiﬁc modelling is far from new; conceptual and mathematical models have
driven research across many ﬁelds for centuries. However, the advent of computational
modelling techniques has created a new set of challenges for theorists as they
seek to describe the advantages and limitations of this approach.
After a brief discussion of the historical foundations of scientiﬁc modelling,
both mathematical and computational, some distinctions that characterise modelling
endeavours will be described. The distinction between models for science and
engineering problems provides a useful framework in which to discuss both the
goals of modelling and the difﬁcult problem of validating models. These discussions
are framed in the context of artiﬁcial life, a ﬁeld which depends upon computational
modelling as a central methodology.
This chapter lays the groundwork for the next two chapters to come, and by
extension Part I of this text as a whole. The discussion here of emergence, and the
related methodologies of ‘bottom-up’ modelling, will allow us to understand the
major philosophical issues apparent in the ﬁeld of computational modelling. This,
in turn, will prepare us for an in-depth discussion of the ﬁeld of Artiﬁcial Life in
the following chapter, which will provide a backdrop for the discussion of general
issues in biological modelling in Chap. 4. Similarly, the discussion here of the debate
regarding the explanatory capacity of simulation models will be a recurring thread
throughout this text, in all three parts of the argument.
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_2
17
18 2 Simulation and Artiﬁcial Life
2.2 Introduction to Simulation Methodology
2.2.1 The Goals of Scientiﬁc Modelling
The construction of models of natural phenomena has been a continuous feature
of human scientiﬁc endeavour. In order to understand the behaviour of the systems
we observe around us, we choose to construct models to allow us to describe that
behaviour in a simpliﬁed form. To use our central bird migration example, imagine
that a researcher wishes to describe the general pattern of a speciﬁc bird species’
yearly migration. That researcher may choose to sit and observe the migrations of
that species year-on-year for a lengthy period of time, allowing for the development
of a database showing that species movement over time.
However, a model of that species’ migration patterns could save our researcher a
great deal of empirical data collection. A model which takes the collected empirical
data and derives from it a description of the bird species’ general behaviour during
each migration season could provide a means of prediction outside of a constant
real-world observation of that species. Further, one can easily imagine models of the
migration problem which allow for detailed representation of the birds’ environment
and behaviour, allowing for a potentially deeper understanding of the causes of these
migrations.
Thus, in the context of this discussion, the goals of scientiﬁc modelling encompass
both description and explanation. A simpliﬁed description of a phenomenon
is inherently useful, reducing the researcher’s dependence on a continuous ﬂow
of collected empirical data, but explanation is the larger and more complex goal.
Explanation implies understanding: a cohesive view of the forces and factors that
drive the behaviour of a system. To develop that level of understanding, we must
ﬁrst understand the nature and limitations of the tools we choose to employ.
2.2.2 Mathematical Models
Throughout the history of science, mathematical models have been a vital part
of developing theories and explanations of natural phenomena. From Newton’s
laws of motion to Einstein’s general relativity, mathematical models have provided
explicit treatments of the natural laws that govern the world around us. As we shall
come to understand, however, these models are particularly well-suited to certain
classes of phenomena; the concise description of natural laws that follows from a
mathematical model becomes ever more difﬁcult to develop as the phenomena in
question becomes more complex.
Even with the appealing simplicity of a mathematical description of a phenomenon
however, certain methodological difﬁculties come into play. Some models
may require a very complex system of linked equations to describe a system,
creating a vast number of parameters which are not known a priori (see Chap. 4 for
2.2 Introduction to Simulation Methodology 19
discussion of this in relation to population biology). These black-box simulations
create difﬁculties for both theorists and experimentalists; theorists struggle to set
values for those parameters, while experimentalists likewise struggle to validate
models with such a range of potential parameters.
There is also the question of the accuracy of a given mathematical model, which
can be difﬁcult to determine without repeated testing of that model’s predictive
capacity. For example, the motion of objects in space can be described by Newton’s
laws of motion; however, for very large bodies which are affected by the tug of
gravity, we must use Einstein’s general relativity to describe that motion. If we turn
the other way and wish to describe the motions of atoms and particles, then we
must use quantum mechanics. The intersections of those models, particularly that
of Einstein’s relativity and quantum mechanics, are far from easy to develop; the
fabled union of these two theories has occupied physicists for decades and will
likely continue to do so for some time (Gribbin 1992).
2.2.3 Computational Models
The advent of easily-available computing power has revolutionised the process
of scientiﬁc modelling. Previously intractable mathematical models have been
tractable through the sheer brute-force calculating power of today’s supercomputers.
Supercomputers now allow physicists to run immense n-body simulations of
interacting celestial bodies (Cox and Loeb 2007), model the formation of blackhole
event horizons (Brugmann et al. 2004) and develop complex models of global
climate and geophysics (Gregory et al. 2005).
Beyond the sheer number-crunching power of computational methods, the
ﬂexibility of computational modelling has resulted in the development of new
varieties of models. While the speciﬁc characteristics of each simulated celestial
body in a model of colliding galaxies is relatively unimportant, given that each of
those galaxies contains billions of massive bodies with similar gravitational impacts
on surrounding bodies, certain other phenomena depend on complex individual
variation to be modelled accurately. Evolution, for example, depends upon the
development of new species through processes of individual mutation and variation
mediated by natural selection (Darwin 1859); thus, a model of evolving populations
requires a description of that individual variation to be effective.
Take once again our central example. If our hypothetical bird-migration
researcher hypothesizes that migration behaviour is due to evolutionary factors
that he may be able to represent in a model, then he must be able to represent the
effects of biological evolution in some form within that model. If he wishes to
represent those effects, the digital birds within his model would require individual
complexity that can produce variation within the simulated species. In contrast, if he
were modelling only the patterns of the bird movements themselves, then he need
only represent the impact of those movements on the other agents in the simulation,
as in the colliding galaxy model above; individual variation in those agents is not
20 2 Simulation and Artiﬁcial Life
such a driving force behind the patterns of bird movements as it would be in an
evolutionary model of the development of those movements.
2.2.4 The Science Versus Engineering Distinction
As computational modelling has developed, so has a distinction between varying
computational approaches. Some models seek to provide predictive power related
to a speciﬁc physical or natural phenomenon, while others seek to test scientiﬁc
hypotheses related to speciﬁc theories. The ﬁrst type has been characterised as
modelling for engineering, and the second has been described as modelling for
science (Di Paolo et al. 2000; Law and Kelton 2000).
Models for engineering are dependent on empirical data to produce predictions.
For example, a transportation engineer may wish to examine the most efﬁcient
means for setting trafﬁc signals (Maher 2007). To determine this, the modeller
will examine current signal settings, note the average arrival time of vehicles at
each junction, analyse the anticipated demand for each service, and other similar
factors. With this information in hand the engineer can produce a model of the
current operating trafﬁc pathways, and alter parameters of those simulated services
to attempt to produce an optimum scheduling algorithm for the new signals.
Similarly, to stretch our bird example to the engineering realm, imagine that
our migration researcher has decided to model a the dissemination of information
via messenger pigeon. If he wishes to ﬁnd an optimum schedule on which to
release and retrieve these pigeons, he could use an engineering-type model to
solve this problem. He could examine current and past messenger-pigeon services,
note the average transit time for the delivery and retrieval of those messages, and
the level of rest and recuperation needed by each bird. With an understanding of
these factors, the researcher could develop a model which would provide optimum
release schedules, given different potential numbers of birds and varying demand
for message delivery.
Models for science, in contrast, focus instead on explanation and hypothesistesting.
A model of the development of animal signalling by its very nature cannot
depend on the availability of empirical data; after all, we cannot simply watch
random evolving populations in the hope that one of them may develop signalling
behaviours while we wait. Instead, the model is based upon a hypothesis regarding
the contributing factors that may produce the development of signalling behaviours;
if the model produces a simulated population which displays those behaviours, then
the modeller may attribute more validity to that hypothesis. This approach is the
focus of this text.
Returning to the bird example, our researcher would be taking a similarly
scientiﬁc modelling perspective if he wished to construct a model which illustrates
the inﬂuence of individual bird movements on migrating ﬂocks. He hypothesizes
that individual birds within the ﬂock assume controlling roles to drive the timeliness
of the ﬂock’s migratory behaviour. He could test such a hypothesis by developing a
2.2 Introduction to Simulation Methodology 21
simulated population of migrating birds in which certain individual agents within
the simulation can affect the movement of multiple other migrating agents; if
the presence of those controlling agents appears to conﬁrm the necessity of such
individuals to keep migrations moving in an appropriate timeframe, then he may
propose that such mechanisms are important to real-world migratory behaviour.
2.2.5 Connectionism: Scientiﬁc Modelling in Psychology
The advent of this sort of scientiﬁc modelling has produced not only new types
of models, but new ﬁelds of enquiry within established ﬁelds. The development of
the connectionist approach to the study of behaviour and cognition provides one
example of the scientiﬁc modelling perspective, and introduces us to the concept of
emergent explanations.
The development of computational modelling techniques together with advances
in neuroscience led some researchers to investigate models of neural function. These
neural network models consist of simpliﬁed neuronal units, with speciﬁed activation
thresholds and means of strengthening or weakening synaptic connections, which
aim to reproduce the neural basis of behaviours (Rumelhart and McClelland 1986).
The idea that models of this nature could demonstrate the emergence of cognitive
behaviour brought with it related ideas concerning the mind that caused some
controversy.
In order for the connectionist to assert that their model can represent cognitive
behaviour, one must assume that mental states correspond to states of activation
and connection strengths in a given neural network. This concept was derided by
some who viewed this as an overly reductionist stance, and that in fact the symbolic
manipulation capability of the mind is crucial to understanding cognitive function
(Fodor and Pylyshyn 1988). This is related to the perspective espoused by many
in the ﬁeld of artiﬁcial intelligence, in which the manipulation of symbols was
considered essential to the development of intelligence (Newell and Simon 1976).
The connectionist thus forms scientiﬁc models which aim to test the hypothesis
that learning mechanisms in neural networks can lead to the development of
cognition. While psychologists are certain that the human brain functions via the
interaction of billions of individual neurons, the degree of correspondence between
these neural-network models and actual brain function is debatable (see Pinker and
Mehler 1988 for a damning critique of the unrealistic results of a connectionist
model of language, as one example). Many models of this type use idealised
neuronal units to investigate possible explanations of behaviour, such as creating
non-functional ‘lesion’ areas in a network designed to perform visual search tasks
as a means of theorising about possible causes of visual deﬁcits (Humphreys et al.
1992). In this respect, these sorts of connectionist models are designed to test
hypotheses and develop theories rather than generate predictions based on empirical
data.
22 2 Simulation and Artiﬁcial Life
2.2.6 Bottom-Up Modelling and Emergence
As noted above, the controversy surrounding connectionist modelling hinged upon
one of its base assumptions: the idea that the low-level interaction of individual
neuronal units could produce high-level complex behaviour. The cognitivists and
computationalists of psychology found this distressing, as reducing cognition to
collections of neural activations eliminates the concept of symbolic manipulation
as a precursor to thought and cognition; the ‘distributed representation’ concept
proposed by connectionists would remove the necessity for such higher-level
concepts of discrete symbol manipulation by the brain (Fodor and Pylyshyn 1988).
Of course, such perspectives need not necessarily be diametrically opposed.
One can certainly imagine connectionism forming a useful element of the study
of cognition, with the cognitivists continuing the study of mental representation
and symbol manipulation in relation to larger concepts of mental behaviour that
are less well-suited to the connectionist modelling perspective. Indeed, given
that connectionist systems can implement symbol-manipulation systems, these
philosophical differences seem minor (Rowlands 1994).
However, the idea of this type of ‘bottom-up’ modelling is crucial, and as
a consequence the debate over this type of modelling bears great relevance to
our discussion. The view proffered by connectionists that models of low-level
interacting units can produce the emergence of higher-level complexity is a central
element of the modelling perspectives being analysed in this text. The particular
relevance of this controversy when considering problems of validation and scientiﬁc
explanation will be examined further both in this chapter and in Chaps. 6 and 7 in
particular.
2.3 Evolutionary Simulation Models and Artiﬁcial Life
2.3.1 Genetic Algorithms and Genetic Programming
Connectionism was far from the only prominent example of a computational
innovation taking cues from biology. The use of genetic algorithms, modelled
on the processes of biological evolution, has a long history in the computational
sciences. Given the extensive study of natural selection in biological systems as
an optimisation process, and the need for increasingly innovative optimisation
techniques within computer science, the use of an analogue of that process for
computational applications seemed a natural ﬁt. Indeed, as early as the 1950s the
computers available were being put to use on just these sorts of problems (Fraser
1957).
Since these early days of experimentation, the genetic algorithm became an
established method of optimisation in certain problem spaces. Such algorithms
seek to encode potential solutions to the problem at hand in forms analogous to
2.3 Evolutionary Simulation Models and Artiﬁcial Life 23
a biological ‘genotype’; each solution is then examined to determine its suitability
as a solution, and evaluated according to a speciﬁed ﬁtness function. The most ﬁt
solutions can then be combined, individual mutations can be generated if desired,
and the next generation of potential solutions is subjected to the same process.
Over many generations, the genetic algorithm may ﬁnd a solution suitable for the
problem, and in some cases the solution comes in an unexpected and novel form
due to the inﬂuence of this selection pressure. Such systems came into the spotlight
quite prominently in the 1970s, when John Holland’s book on the topic provided a
strong introduction (Holland 1975); later works by Goldberg (1989), Fogel (1988),
and Mitchell (1996) cemented the position of genetic algorithms as a useful method
for optimisation and search problems.
Genetic algorithms do suffer from methodological difﬁculties, of course; certain
problems are not well-suited to genetic algorithms as a means of ﬁnding appropriate
solutions. In addition, the design of a useful ﬁtness function can be extremely
difﬁcult, as the programmer must be careful to avoid solutions which cluster around
local optima (Mitchell 1996). Incorporating an appropriate amount of variation in
the generated population can be vital for certain applications as well, as the right
level of random mutation can provide a useful means to escape those local optima.
2.3.2 Evolutionary Simulations and Artiﬁcial Life
While genetic algorithms became popular amongst certain elements of the computer
science community, they also drew great interest from those interested in the
biological function of evolution. As the artiﬁcial intelligence community sought to
model the fundamentals of human intelligence and cognition, others sought to use
computational methods to examine the fundamentals of life itself.
The ﬁeld of artiﬁcial life, or ALife, has complex beginnings, but is most often
attributed to Langton (2006) who ﬁrst christened the ﬁeld with this title. ALife
however has strong links with the artiﬁcial intelligence community (Brooks 1991),
as well as with the earlier modelling traditions of ecology and population biology.
The inﬂuence of the artiﬁcial intelligence community, the interest in bottom-up
modelling as alluded to earlier in our review of connectionism, and the development
of new techniques to produce adaptive behaviour in computational systems all seem
to have had a hand in the development of ALife.
ALife work to date has revolved around a number of related themes, but all of
them share some method of reproducing the mechanics of biological adaptation in
computational form. Genetic algorithms as described above are perhaps the most
prominent example, with a great number of evolutionary simulations using such
algorithms or some version thereof to provide that element of adaptation. While the
members of this growing research community moved forward with these methods
of simulating evolutionary systems, a related set of new challenges faced that
community.
24 2 Simulation and Artiﬁcial Life
2.3.3 Bedau and the Challenges Facing ALife
Mark Bedau’s 2003 (Bedau 2003) summary of the ﬁeld of artiﬁcial life provides one
view of the array of potential challenges facing the ALife researcher. In Bedau’s
view, ALife clearly displays the potential to enhance our understanding of the
processes of life, and numerous fundamental questions that spring from those
processes:
How does life arise from the non-living?
1) Generate a molecular proto-organism in vitro.
2) Achieve the transition to life in an artiﬁcial chemistry in silico.
3) Determine whether fundamentally novel living organisations can arise from inanimate
matter.
4) Simulate a unicellular organism over its entire lifecycle.
5) Explain how rules and symbols are generated from physical dynamics in living systems.
Bedau begins his extensive list of ALife challenges with a look at the potential
for these new methods of simulation to simulate the origins of life. Of course there
is great debate over the best means to simulate such early beginnings. Simulating
the development of cell structures has been an important theme in ALife (e.g.,
Sasahara and Ikegami 2004), as well as the development of simple self-replicating
structures (Langton 1990). This is not entirely surprising, given that Von Neumann’s
self-replicating cellular automaton was clearly an inﬂuence on those seeking to
understand the development of such forms in silico (Von Neumann and Burks 1966).
However, at what point might we agree that such self-replicating digital organisms
have achieved a ‘transition to life’ as proposed by Bedau? At what point
does that simulated organism become an instantiation of the laws governing the
development of natural life? Agreement here is hard to come by; some argue that
the status of ‘alive’ is best conferred on organisms that can self-reproduce (see Luisi
(1998) for an evaluation of this and other deﬁnitions), while others argue that self-
motility1
is a more important determining factor (Hiroki et al. 2007), and still others
appeal to the concepts of self-organisation and autopoiesis2
(Maturana and Varela
1973). This issue hinges upon the theoretical perspective of the modeller to a large
degree: if one believes that the properties of life are just as easily realised in the
digital substrate as they are in the biological substrate, then an ALife simulation can
easily achieve life (given an appropriate deﬁnition of such) regardless of its inherent
artiﬁciality. The issue of artiﬁciality in ALife research and its import for the theorist
and experimentalist are explored in detail in Chap. 3.
1
The ability to move spontaneously or non-reactively. This is considered a vital capability for
biological life – self-motility allows living things to move in pursuit of food sources, for example.
See Froese et al. (2014) for a detailed exploration.
2
Autopoietic systems are systems that can produce and sustain themselves through their own
internal processes, such as the biological cell. The concept was originally described in relation
to biological systems, but has since been adapted to characterise cognitive and social systems as
well.
2.3 Evolutionary Simulation Models and Artiﬁcial Life 25
What are the potentials and limits of living systems?
6) Determine what is inevitable in the open-ended evolution of life.
7) Determine minimal conditions for evolutionary transitions from speciﬁc to generic
response systems.
8) Create a formal framework for synthesizing dynamical hierarchies at all scales.
9) Determine the predictability of evolutionary manipulations of organisms and ecosys-
tems.
10) Develop a theory of information processing, information ﬂow, and information
generation for evolving systems.
Bedau’s next set of challenges are reminiscent of one of Chris Langton’s more
famous descriptions of artiﬁcial life, in which he stated that ALife could seek
to examine ‘life-as-it-could-be’ rather than simply ‘life-as-we-know-it’ (Langton
1992). In other words, given that the ALife researcher can construct an enormous
variety of possible models, and thus living systems if we agree that life can be
realised in silico, then ALife can be used as a platform to understand the vast variety
of potential forms that life can create, rather than only examine the forms of life we
currently perceive in the natural world.
Bedau is alluding to similar ideas, proposing that ALife researchers can use their
work to probe the boundaries of the evolution of life. By simulating evolutionary
systems, he posits that we may be able to investigate the mechanics of evolution
itself in a way impossible in conventional biology. The researcher is able to freely
tweak and direct the evolutionary processes at work in his simulation, and if we
accept that the simulation adequately represents the function of evolution in the real
world, then such research may allow for a greater understanding of the limits of the
evolutionary process.
How is life related to mind, machines and culture?
11) Demonstrate the emergence of intelligence and mind in an artiﬁcial living system.
12) Evaluate the inﬂuence of machines on the next major evolutionary transition of life.
13) Provide a quantitative model of the interplay between cultural and biological evolution.
14) Establish ethical principles for artiﬁcial life. (Bedau 2003, p. 506)
Finally, Bedau closes his list of ALife ‘grand challenges’ with more speculative
notions of relating the development of life with the development of mind and
society. He posits that cognition originates from similar roots as life, in that such
mental activity is a biological adaptation like any other seen in evolving systems,
and that in this context artiﬁcial life may provide insight into the origins of mind as
well as life.
The idea that mind and culture follow similar rules of adaptation to life is not
a new one; the ﬁeld of evolutionary psychology is well-established, if controversial
(Buss 2004), and Dawkin’s discussion of the ‘meme’ in relation to cultural evolution
is one of the more prominent examples of such thinking in sociology (Dawkins
1995). The question of whether simulation can become sufﬁciently sophisticated to
allow for the emergence of these higher-order phenomena is critical to our upcoming
examination of simulation in the social sciences; in fact, the same philosophical
difﬁculties that face modellers of cognition have been linked with similar difﬁculties
26 2 Simulation and Artiﬁcial Life
in using simulation to model the roots of society and culture (see Sawyer 2002,
2003, 2004 and the accompanying discussion in Chap. 5).
2.4 Truth in Simulation: The Validation Problem
2.4.1 Validation and Veriﬁcation in Simulation
While Bedau has provided a useful summary of the possible challenges facing
artiﬁcial life in the research realm, numerous other challenges also loom in the area
of methodology for ALife modellers. The difﬁculty of tying simulation results to
the system being simulated is one not conﬁned to ALife, but instead is common to
all varieties of simulation endeavour.
The validation and veriﬁcation of simulations is often most troublesome for
modellers of all types. Once a model is designed and run, the researcher must be
able to express conﬁdence that the results of that model bear a direct relation to
the system of interest. This is validation, and in relation to computational models
speciﬁcally, Schlesinger’s description of this process as a ‘substantiation that a
computerized model within its domain of applicability possesses a satisfactory range
of accuracy consistent with the intended application of the model’ (Schlesinger et al.
1979) is often cited. In other words, the modeller must demonstrate that the model
displays a measure of accuracy within the domain to which the model has been
applied.
Going hand-in-hand with validation is the concept of veriﬁcation. A veriﬁed
model is one in which the construction of the model is known to be accurate
according to the framework in which that model is designed (Law and Kelton 2000).
In relation to computational models speciﬁcally, this means that the model must be
programmed appropriately to produce results that reﬂect the intent of the model’s
construction; given the inherent complexity of the software design process, this is
not necessarily a simple task to complete.
2.4.2 The Validation Process in Engineering Simulations
Validation in simulations for engineering purposes, as described earlier, would
tend to follow a certain pattern of verifying assumptions and comparing model
predictions to empirical data (Sargent 1982, 1985). To illustrate these concepts we
may return to the bird migration example. If our migration researcher wished to
construct a model which provides an illustration of the migration behaviour of a
certain bird species, he would ﬁrst need to discuss the assumptions inherent in the
conceptual model leading to the simulation’s construction. For example, the speed
and direction of movement of his simulated migrating populations should match
2.4 Truth in Simulation: The Validation Problem 27
those values gleaned from empirical observation; in general, his conceptual model
should be informed by the available data. This veriﬁcation of the conceptual model
provides conﬁdence that the assumptions made to produce the model have a solid
foundation in theory related to the migrations observed in that bird species.
Having veriﬁed the conceptual model, the researcher would then need to verify
the model itself. While conﬁdence in the assumptions in the conceptual model has
been established, the researcher must conﬁrm that these assumptions have been
implemented appropriately in the model itself. Has the code for the simulation
been written correctly? Are the correct parameters in place in the model, given the
parameters required by the conceptual model? If the researcher can demonstrate that
the simulation has been implemented correctly, then validation can proceed.
The validation step is the most difﬁcult, requiring a comparison of model data
with real data. With our model migrating birds in place, the simulation should
provide predictions of the behaviour of those birds when it is run. Do these
simulation runs provide data that correlates appropriately with observational studies
of that bird species’ migration? Does empirical data related to bird migrations
in general imply that the model’s results are believable? Such questions can be
complicated to answer, but nevertheless can be answered with access to appropriate
empirically-collected data (see Law and Kelton 2000 for more discussion).
2.4.3 Validation in Scientiﬁc Simulations: Concepts of Truth
The procedure outlined above is undoubtedly complex, but achievable for the
engineer. For the scientist, however, the procedure becomes more complex still. In
a simulation which is designed to test hypotheses, and in which a clear relation to
empirical data is not always obvious are indeed possible, the prospect of validating
the results of that simulation depends on different sorts of relations between theory,
data and model.
Appealing to the philosophy of science, Alex Schmid describes three theories of
truth for simulations in science: the correspondence theory, the consensus theory,
and the coherence theory (Alex Schmid 2005). The correspondence theory holds
that the simulation must correspond directly with facts in reality; the consensus
theory holds that the simulation must be acceptable under idealised conditions; and
the coherence theory holds that the simulation must form a part of a coherent set of
theories related to the system of interest.
The correspondence theory of truth relates closely to the methods of validation
discussed in relation to engineering: our example bird migration simulation must
display results that correspond directly to empirical data regarding migrating birds,
otherwise the predictions of that simulation are of little use. Such a view coincides
with the prevailing views of validation present in engineering models: validated
models in the engineering perspective must demonstrate a close relationship to
known empirical data about the problem under study.
28 2 Simulation and Artiﬁcial Life
The consensus theory, however, is much less deﬁned, depending as it does on
a more communal evaluation of the truth of the model. Our bird migration model
may not provide entirely accurate predictions for future migration behaviours, but
under this view we may still consider the simulation to be validated if the model
is generally illustrative of bird migration behaviour. A more general version of our
migration simulation, one developed as an abstract model not tied to any particular
bird species, could fall into this category.
The coherence theory moves even further from the concept of truth most
applicable to engineering, requiring only that the model in question ﬁt into a given
set of coherent beliefs. If our bird migration model ﬁts cohesively into the general
body of animal behaviour theory relating to migrating populations, then the model
may provide a useful and valid addition to that theory. However, as Schmid points
out, there is no reason that a coherent system of beliefs cannot also be completely
false (Alex Schmid 2005); a model of our migrating birds travelling across a ﬂat
planet may be accurate given the belief structures of the Flat Earth Society, but
is nevertheless completely separate from the truth of birds migrating over a round
planet.
2.4.4 Validation in Scientiﬁc Models: Kuppers and Lenhard
Case Study
Kuppers and Lenhard’s evaluation of validation of simulation in the natural and
social sciences sought to demonstrate the relation between theory and validation
in models (Günter et al. 2005). As a case study, they focused upon the infamous
scenario in climate modelling of ‘Arakawa’s trick.’
Norman Phillis’ model of atmospheric dynamics (Phillips 1956) was a very
ambitious step toward climate modelling on a grand scale. The results of his
simulation were viewed with respect by the research community of the time,
demonstrating as they did a direct correspondence to empirically-observed patterns
of airﬂow in the atmosphere, but his results were hampered by an unfortunate
consequence of the equations: numerical instability prevented the model from
making any long-term predictions.
Arakawa sought to solve this problem, and eventually did so by virtue of his
notable trick: he altered the equations of state, incorporating assumptions that were
not derived from conventional atmospheric theory, and in the process ensured longterm
stability in the model. Understandably this technique was met with substantial
skepticism, but eventually was accepted as further empirical data showed the
accuracy of Arakawa’s revised model despite the theoretical inadequacies.
Kuppers and Lenhard use this as a demonstration that ‘performance beats
theoretical accuracy’ (Günter et al. 2005). In other words, a simulation can
provide successful data without having a completely accurate representation of the
phenomenon at hand. This certainly seems to spell trouble for Schmid’s descriptions
2.5 The Connection Between Theory and Simulation 29
of relating truth in simulation to the theoretical background of that simulation.
If we agree that simulations may achieve empirical accuracy despite theoretical
inaccuracy, how does this affect our view of truth in simulation?
A different approach to the relation between model and theory seems required.
The status of a successful, validated model relies on more than a simple correspondence
between the model and the research community, or the model and its
surrounding theoretical assumptions, as Arakawa’s computational gambit demon-
strates.
2.5 The Connection Between Theory and Simulation
2.5.1 Simulation as ‘Miniature Theories’
The relations described thus far between theory and simulation clearly lack important
elements. For example, while the coherence theory of truth is appealing in that,
say, an evolutionary model may ﬁnd validation by ﬁtting cohesively into the existing
theory of biological evolution, how that model relates to theory itself remains an
open question. Likewise, if a model must ﬁt into an existing set of beliefs, are we
suddenly restricting the ability of simulation to generate new theory? Is there room
for innovation in such concepts of validation?
One means to escape from this difﬁcult connection between simulation and
theory is to reform our deﬁnition completely: we may consider a simulation as a
theory in itself. Within the simulation community this view is not uncommon:
The validation problem in simulation is an explicit recognition that simulation models are
like miniature scientiﬁc theories. Each of them is a set of propositions about how a particular
manufacturing or service system works. As such, the warrant we give for these models can
be discussed in the same terms that we use in scientiﬁc theorizing in general. (Kleindorfer
et al. 1998, p. 1087)
Similarly, Colburn describes simulation as a means to ‘test a hypothesis in a
computer model of reality’ (Colburn 2000, p. 172). In the context of Arakawa’s
trick, this perspective is attractive: in this view Arakawa’s model can serve as a
miniature theory of its own, and the perceived disconnect between the assumptions
of his model and the accepted atmospheric theory are of no consequence to the
validity of the model.
2.5.2 Simulations as Theory and Popperian Falsiﬁcationism
If we accept that simulations can take the form of such ‘miniature theories,’ then
perhaps the question of validation becomes instead a question of the validity of
scientiﬁc theories. Herskovitz suggests that the process of validating simulation
30 2 Simulation and Artiﬁcial Life
models is at its root a Popperian process of falsiﬁcation (Herskovitz 1991).
In essence, given that a simulation model is considered validated if its results
correspond to the behaviour of real-world systems, then a system must likewise
be falsiﬁed if its results do not correspond to the real-world system.
However, Arakawa’s trick once again throws this view into question. Arakawa’s
model by its very nature incorporates assumptions that are contrary to physical
theory: as one example, he assumes that energy is conserved in his model of the
atmosphere, while the real atmosphere does not display such conservation (Günter
et al. 2005; Arakawa 1966). In this respect, is his model subject to Popperian
falsiﬁcation? If a central assumption of this model, one which informs every
calculation of that model, is demonstrably false, does his model likewise lose all
validity?
Kuppers and Lenhard argue that it does not: that the theory presented by
Arakawa’s model stands apart from the physical theory upon which it was initially
based, and its performance speaks more to its validity than the accuracy of its
conceptual assumptions. Likewise, we might imagine an evolutionary model falling
victim to the same Popperian plight: assumptions made to simplify the process of
evolution within the model may contradict observed facts about evolving species
in nature. However, if those assumptions allow the model to display accuracy in
another respect, either theoretical or empirical in nature, should we still decry the
assumptions of that model?
2.5.3 The Quinean View of Science
In this context the Popperian view of falsiﬁcation seems quite at odds with the
potential nature of scientiﬁc models. In contrast to what Herskovitz seems to believe,
simulations are far more than mere collections of assumptions designed to imitate
and calculate the properties of natural phenomena. Indeed, simulations can often
contain a rich backdrop of internal assumptions and theories, and the measure of
a simulation’s success seems likewise more rich than a simple comparison with
accepted data and theory.
The Quinean view of science seems much more suited to the simulation
endeavour (and indeed, many would argue, more suited to science of all varieties).
The Duhem-Quine problem stands famously at odds with the Popperian view,
asserting that scientiﬁc theories can in fact never be proved conclusively false
on their own (Quine 1951, 1975). Given that theories depend on one or more
(often many) related auxiliary assumptions, theories can be saved from deﬁnitive
falsiﬁcation by adjusting those auxiliary assumptions.
For example, Newton’s laws of gravitation were able to explain a great deal
of natural phenomena, and are still used extensively in modern physics. However,
Newton’s laws could not explain some clearly evident and anomalous behaviours in
astronomical bodies: the perihelion of Mercury’s orbit being a prime example. Yet,
rather than simply disposing of Newton’s theory as inadequate, scientists instead
2.5 The Connection Between Theory and Simulation 31
strove for another explanation in addition to Newton’s theories, which later arrived
in the guise of Einstein’s general relativity. Now Newton’s laws are presented as
essentially a subset of Einstein’s theory, displaying correct results within a certain
set of reference frames. Likewise, points where Einstein’s theories break down (i.e.,
at the Big Bang or at the event horizon of a singularity) are not taken as a falsiﬁcation
of Einstein’s views, but rather an indication of the need for additional theories and
assumptions to explain those anomalies in detail.
2.5.4 Simulation and the Quinean View
Having rightly discarded the Popperian view of simulation and embraced Quine’s
notion of ﬂexible and interconnected scientiﬁc theories, we can revise our initial
view of simulation in the context of this understanding. Noble (1998) provides a
summary of one view of simulation in a Quinean context, speciﬁc to artiﬁcial life:
he argues that the Quinean view implies that new models are generated according
to a requirement to incorporate new information in an existing theory without
completely reorganising that theory.
As an example Noble posits that a new simulation in artiﬁcial life may seek to
explain a behaviour in biology as a consequence of an emergent process (Noble
1998). The modeler may then implement a model incorporating appropriate lowlevel
assumptions with the intention of running the simulation to determine whether
the expected behaviour does indeed emerge. In this respect the model is based
upon pre-existing conceptual frameworks concerning the high-level behaviour, and
contributes to the addition of this new behaviour into the overall biological theory
by providing an explanation of that behaviour in terms of an emergent phenomenon.
More generally, we may add to this characterisation by referring back to the
earlier discussion of simulation-as-theory. When constructing a model to allow
for the integration of new information into an overall conceptual framework, a
simulation model can function as an auxiliary hypothesis in and of itself: that
simulation forms a theory, and thus is subject to the same standards as the larger
conceptual framework. In this case, even if the model does not achieve validation
in comparison to empirical data, all is not lost; in the appropriate Quinean fashion,
the auxiliary hypotheses linked to that simulation may be revised to present a new
version of the simulation-theory (perhaps by revising certain parameter values or
similar).
Simulation then is not simply a means to simplify calculations within a preexisting
theoretical framework, it is a means to modify that theoretical framework.
The validity of a simulation is not easy to determine by any means, but a simulation
based in an existing framework that adds sensible assumptions to that framework
may go a long way toward justifying its existence as a substantive part of the
32 2 Simulation and Artiﬁcial Life
overall theory. Unlike in the Popperian view, an invalidated simulation need not be
discarded, but instead revised; assumptions used in a simulation are pliable, and an
alteration of same could allow that model to produce insights it originally appeared
to lack.
2.6 ALife and Scientiﬁc Explanation
2.6.1 Explanation Through Emergence
Having established that simulation can perform a valuable role in the development
of scientiﬁc theory, this analysis now turns to the role of simulation in scientiﬁc
explanation speciﬁcally. The ability of simulation to provide complete and coherent
scientiﬁc explanation will impact the strength with which this methodology can
develop scientiﬁc theories; bearing this in mind, we require an understanding of the
limits of simulation in developing explanations. This explanatory role for simulation
is often hotly debated, particularly in the case of scientiﬁc models as described here
(see the exchange between O’reilly and Farah 1999 and Burton and Young 1999
for one example, as the authors debate the explanatory coherence of distributed
representations in psychological models).
Within ALife, which depends upon frequently abstract simulations of complex
emergent systems, the explanation problem takes on a new dimension. As Noble
posits (Noble 1998), ALife models provide the unique mechanism of emergence
which can provide new elements of an explanation of a phenomenon; however,
debate continues as to whether explanations of higher-order phenomena through
emergence can capture a complete explanation of those phenomena. This debate
is exempliﬁed once more by the debate within cognitive science regarding connectionism
and distributed representations: the reductive character of connectionist
explanation of mental phenomena is seen as overly restrictive, removing the
possibility of higher-order explanations of mental states.
As noted earlier in our discussion of connectionism, this debate is easily
avoidable in one sense: if we accept that consciousness, for example, is a natural
emergent property of neuronal activity, then the acceptance of this fact does not
preclude the use of higher-order discussions of mental states as a means to explain
the characteristics of that emergent consciousness. This does, however, seem to
preclude the notion of that emergent explanation being a complete explanation;
even if one can show that consciousness does indeed emerge from that lowerlevel
activity, by the nature of emergent phenomena that consciousness is not easily
reducible to those component activities.
2.6 ALife and Scientiﬁc Explanation 33
2.6.2 Strong vs Weak Emergence
The variety of emergence discussed in the previous section is often referred to as
‘strong emergence’: the concept that not only is an emergent phenomenon difﬁcult
to reduce directly to its component parts, but in fact the emergent phenomenon
can display downward causation, or supervenience (inﬂuencing its own component
parts), thus making the cause of that emergent phenomenon very difﬁcult to deﬁne.
In essence, the whole is a consequence of its component parts, but is irreducible to
the actions of those components (see O’Conner 1994; Nagel 1961).
Weak emergence, by contrast, is a means proposed by Mark Bedau to capture
the emergent character of natural phenomena without the troublesome irreducibility
(Bedau 1997). Bedau deﬁnes weak emergence thus:
Macrostate P of S with microdynamic D is weakly emergent iff P can be derived from D
and S ’s external conditions but only by simulation. (Bedau 1997, p. 6)
Thus, similar to strong emergence, the macro-level behaviour of the emergent
system cannot be predicted merely by knowledge of its micro-components. Crucially
however, those macro-level properties can be derived by allowing those
micro-components to perform their function in simulation. Under strong emergence,
an evolutionary simulation constructed in bottom-up ALife fashion would be unable
to capture the complete behaviour of the resultant emergent phenomenon. Under
weak emergence, that simulation could indeed provide a derivation of that higherlevel
behaviour.
Bedau takes great pains to point out however that such weakly emergent
behaviours are still, much like strongly emergent behaviours, essentially
autonomous from their micro-level components. While in theory one could predict
exactly the behaviour of a weakly emergent system with a perfectly accurate
simulation of its micro-level components, in practice such simulations will be
impossible to achieve. Instead, the micro-level explanation via simulation provides
a means to observe the general properties of the macro-level weakly emergent
result.
In this sense there appears to be a certain circularity to weak emergence:
simulation can provide a micro-level explanation of an empirical phenomena, but
in practice ‘we can formulate and investigate the basic principles of weak emergent
phenomena only by empirically observing them at the macro-level’ (Bedau (1997),
p. 25). Some may argue in fact that this constitutes an explanation in only a weak
sense: one could point to a simulation of this type and note that the given microcomponents
lead to the speciﬁed macro-behaviour, but the level of insight into that
macro-behaviour is still fundamentally limited despite intimate knowledge of the
behaviour of its components.
This objection becomes important once more in the context of social simulation
and the concept of non-reductive individualism, which is explored in Chaps. 5, 6
and 7. While Bedau’s concept of weak emergence is less metaphysically tricky
than classical strong emergence, the difﬁculties that remain in explanation by
34 2 Simulation and Artiﬁcial Life
simulation despite this new categorisation of phenomena still allow for criticism
of the simulation approach. Such criticisms will inform our discussion of social
simulation in the second section of this text as well as our discussion of Alife in the
current section.
2.6.3 Simulation as Thought Experiment
Unsurprisingly these difﬁculties in using simulation for scientiﬁc explanation have
generated much discussion within the research community. Di Paolo, Noble and
Bullock approach this thorny issue by proposing that simulations are best viewed
as opaque thought experiments (Di Paolo et al. 2000). This proposal draws upon
Bedau’s earlier description of ALife models, describing them as ‘computational
thought experiments.’
A traditional thought experiment in this view constitutes ‘in itself an explanation
of its own conclusion and its implication’ (Di Paolo et al. 2000, p. 6). In
other words a thought experiment provides a self-contained means with which to
probe the boundaries of the theory which informs that experiment. A successful
thought experiment can provoke a reorganisation of an existing theory as it brings
previously-known elements of that theory into a novel focus.
Simulation experiments, it is argued, can fulﬁll a similar purpose. However,
simulations suffer from an inherent opacity: as noted in our discussion of emergence,
the modeler’s knowledge of the workings of the simulation do not imply
an understanding of the simulation’s results. Unlike in a conventional thought
experiment, the modeler must spend time unraveling the result of his simulation,
probing the consequences to determine the implications for theory.
As a result of this view, the authors propose a different methodology in simulation
research than the conventional view. Firstly, they contend that the successful
replication of a result given some mechanism described in the simulation does not
constitute an explanation (a misconception common to simulation work, and clearly
debunked by the characteristics of emergence mentioned earlier). In consequence
the explanation which may be drawn from simulation work is likely to incorporate
an ‘explanatory organisation,’ in which some elements of the problem are explained
through micro-level behaviour, others may be explained at the macro-level, and still
others in relations between the two.
In essence, they advocate an additional step in the modelling process in which
the modeler performs experiments on the simulation, as one might do in a laboratory
environment. The systematic exploration of the model itself is intended to provide
a greater understanding of its inner workings, and in turn this theory of the model’s
behaviour must then be related to theories about the natural world which provide
the inspiration for the model. So the ALife researcher can accept the view that
emergent behaviours are difﬁcult to explain via simulation, but at the same time
forming theories about the simulation that relate to theories about those behaviours
may produce a new insight into pre-existing theories, as with a successful thought
experiment.
2.6 ALife and Scientiﬁc Explanation 35
2.6.4 Explanation Compared: Simulations vs Mathematical
Models
Taking the thought-experiment perspective into more depth, Bryden and Noble
(2006) contrast the explanatory capacity of simulation models with that of mathematical
models. They seek to explore what is required of an explanation derived
from simulation, noting that the unfortunately commonly accepted view that a
simple qualitative similarity between the simulation result and the behaviour of the
real system is sufﬁcient to provide any sort of explanation.
Bryden and Noble note another element of the inherent opacity of simulation
research: the analytical incompleteness of such models. Mathematical treatments,
when ﬂawed, are easily revealed as such. Simulations in contrast can be run many
times, with different parameter values, and ﬂaws in the coding of the simulation
may not be immediately apparent. Similarly, those simulation runs represent only
isolated data points in the entire possible space of runs allowable in that model; since
no researcher can spare the time to browse the entire space of parameter values for
a simulation, the results we see are only a fraction of what is possible.
The authors go on to advocate a means for decomposing a simulation system
into component mechanistic subsystems which are more amenable to mathematical
explanation. A working model which is decomposed in this way may still not
provide the complete analytical package of an exclusively mathematical treatment,
but it is argued that this brings the researcher closer to a full analytical solution of
the target system. Thus the computational model is seen as a means to generate the
tools necessary to reach a cohesive mathematical explanation of the phenomenon
under study.
In a broader context this approach is quite close to that proposed by Di Paolo,
Noble and Bullock (Di Paolo et al. 2000). In both cases the modeler spends time
exploring the conﬁnes of the model in question, probing its inner workings to deﬁne
the parameters in which that model operates. Having done this, the modeler may
begin to relate those theories about the model to theories about the world; in Bryden
and Noble’s view, for maximum explanatory power those relations should take the
form of mathematical treatments. From both however we draw the conclusion that
models which are able to replicate an emergent behaviour through the simulation
of a system’s micro-level component interactions is still very far from providing
an explanation of that system. The simulation itself must be further deconstructed,
its parameters understood, and its workings probed in order to relate that model
successfully to the real system to which it relates. Thus simulation becomes a
valuable tool in the scientist’s repertoire, but one that must be supplemented by
more traditional means of enquiry as well.
36 2 Simulation and Artiﬁcial Life
2.7 Summary and Conclusions
The problems facing the simulation methodology are clearly far from philosophically
transparent. From mathematical models to evolutionary simulations, these
tools display potential explanatory power and are quick to develop in comparison
to traditional empirical studies. However, with that relative ease of use comes a
correspondingly high difﬁculty of analysis.
The type of simulation discussed here, particularly in the context of ALife,
focuses on the investigation of emergent phenomena in the natural world. These
phenomena by their very nature are difﬁcult to explain; simulation provides a means
to view the origin of these phenomena from the behaviour of low-level component
parts, but still lacks in its ability to explain the overall behaviour of that higher-level
emergent order.
A number of researchers and philosophers have attempted to justify the use of
simulation as a means for scientiﬁc explanation in a variety of ways; our synthesis up
to this point indicates that simulation is certainly a useful element in the explanatory
toolbox. Simulation however cannot stand alone: a simulation which displays an
emergent behaviour still requires a theoretical framework to describe that behaviour.
Within ALife, different streams of thought have developed in response to these
difﬁculties; questions surrounding the appropriate use of simulation in the ﬁeld have
led to extensive debate on the validity of simulation as a means to generate data. A
further investigation into the theoretical underpinnings of ALife in the following
chapter will provide insight into the fundamental aspects of this debate, and lead
us further into this analysis of simulation methodology. This investigation will also
provide important theoretical background for future discussion of Alife models and
their relation to more general methodological concerns in modelling amongst the
broader biology community.
References
Arakawa, A. (1966). Computational design for long-term numerical integration of the equations
of ﬂuid motion: Two-dimensional incompressible ﬂow. Journal of Computational Physics, 1,
119–143.
Bedau, M. A. (1997). Weak emergence. In J. Tomberlin (Ed.), Philosophical perspectives: Mind,
causation, and world (Vol. 11, pp. 375–399). Oxford: Blackwell.
Bedau, M. A. (2003). Artiﬁcial life: Organization, adaptation and complexity from the bottom up.
Trends in Cognitive Sciences, 7, 505–512.
Brooks, R. A. (1991). Intelligence without reason. In J. Myopoulos, & R. Reiter (Eds.), Proceedings
of the Twelfth International Joint Conference on Artiﬁcial Intelligence (pp. 569–595). San
Mateo, CA: Morgan Kaufmann.
Brugmann, B., Tichy, W., & Jansen, N. (2004). Numerical simulation of orbiting black holes.
Physical Review Letters, 92, 211101.1–211101.4.
Bryden, J., & Noble, J. (2006). Computational modelling, explicit mathematical treatments, and
scientiﬁc explanation. In L. M. Rocha, et al. (Ed.), Artiﬁcial Life X: Proceedings of the Tenth
International Conference on the Simulation and Synthesis of Living Systems (pp. 520–526).
Cambridge: MIT Press.
Buss, D. M. (2004). Evolutionary psychology: The new science of the mind. Boston, MA: Pearson.
References 37
Colburn, T. R. (2000). Philosophy and computer science. Armonk, NY: M.E. Sharpe.
Cox, T. J., & Loeb, A. (2007). The collision between the Milky Way and Andromeda. Monthly
Notices of the Royal Astronomical Society, 386(1), 461–474.
Darwin, C. (1859). The origin of species by means of natural selection. London: John Murray.
Dawkins, R. (1995). River out of Eden: A Darwinian view of life. New York: Basic Books.
Di Paolo, E. A., Noble, J., & Bullock, S. (2000). Simulation models as opaque thought experiments.
In M. A. Bedau, et al. (Ed.), Artiﬁcial life VII (pp. 497–506). Cambridge, MA: MIT Press.
Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and cognitive architecture. Cognition, 28,
3–71.
Fogel, D. B. (1988). An evolutionary approach to the travelling salesman problem. Biological
Cybernetics, 60(2), 139–144.
Fraser, A. S. (1957). Simulation of genetic systems by automatic digital computers. Australian
Journal of Biological Science, 10, 492–499.
Froese, T., Virgo, N., & Ikegami, T. (2014). Motility at the origin of life: Its characterization and a
model. Artiﬁcial Life, 20(1), 55–76.
Goldberg, D. E. (1989). Genetic algorithms in search. Reading: Addison Wesley.
Gregory, J. M., et al. (2005). A model intercomparison of changes in the Atlantic thermohaline
circulation in response to increasing atmospheric Co2 concentration. Geophysical Research
Letters, 32, L12703.1–L12703.5.
Gribbin, J. (1992). Unveiling the edge of time. New York: Harmony Books.
Herskovitz, P. J. (1991). A theoretical framework for simulation validation: Popper’s falsiﬁcationism.
International Journal of Modelling and Simulation, 11, 56–58.
Holland, J. H. (1975). Adaptation in natural and artiﬁcial systems. Ann Arbor: University of
Michigan Press.
Humphreys, G. W., Freeman, G. W., & Muller, H. J. (1992). Lesioning a connectionist model of
visual search: Selective effects on distractor grouping. Canadian Journal of Psychology, 46,
417–460.
Kleindorfer, G. B., O’Neill, L., & Ganeshan, R. (1998). Validation in simulation: Various positions
in the philosophy of science. Management Science, 44(8), 1087–1099.
Küppers, G., & Lenhard, J. (2005). Validation of simulation: Patterns in the social and natural
sciences. Journal of Artiﬁcial Societies and Social Simulation, 8(4), 3.
Langton, C. (2006). Artiﬁcial life. In C. Langton (Ed.), Artiﬁcial life (pp. 1–47). Reading: Addison-
Wesley.
Langton, C. G. (1990). Computation at the edge of chaos: Phase transitions and emergent
computation. In CNLS’89: Proceedings of the Ninth Annual International Conference of the
Center for Nonlinear Studies on Self-Organizing, Collective, and Cooperative Phenomena in
Natural and Artiﬁcial Computing Networks on Emergent Computation (pp. 12–37). NorthHolland
Publishing Co.
Langton, C. G. (1992). Life at the edge of chaos. In C. G. Langton, C. Taylor, J. D. Farmer, & S.
Rasmussen (Eds.), Artiﬁcial life II (pp. 41–91). Reading: Addison-Wesley.
Law, A. M., & Kelton, W. D. (2000). Simulation modeling and analysis third edition. New York:
McGraw Hill.
Luisi, P. L. (1998). About various deﬁnitions of life. Origins of life and the evolution of the
biosphere, 28, 613–622.
Maher, J. (2007). Reconﬁgurable hardware evolution platform for a spiking neural-network
controller. In Reconﬁgurable Computing: Architectures, Tools and Applications, Third International
Workshop, ARC 2007, Mangaratiba, 27–29 Mar 2007 (Lecture notes in computer
science, Vol. 4419). Springer.
Matsuno, H., Hanczyc, M. M., & Ikegami, T. (2007). Self-maintained movements of droplets with
convection ﬂow. In M. Randall, H. A. Abbass, & J. Wiles (Eds.), Progress in artiﬁcial life.
ACAL 2007 (Lecture notes in computer science, Vol. 4828). Berlin/Heidelberg: Springer.
Maturana, H., & Varela, F. (1973). Autopoiesis and cognition: The realization of the living. In R.
S. Cohen, & M. W. Wartofsky (Eds.), Boston studies in the philosophy of science (Vol. 42).
Dordrecht: Springer.
38 2 Simulation and Artiﬁcial Life
Mitchell, M. (1996). An introduction to genetic algorithms. Cambridge: MIT Press.
Nagel, E. (1961). The structure of science. San Diego: Harcourt, Brace and Company.
Newell, A., & Simon, H. A. (1976). Computer science as empirical enquiry: Symbols and search.
Communications of the Association for Computing Machinery, 19, 113–126.
Noble, J. (1998). The evolution of animal communication systems: Questions of functions examined
through simulation. PhD Thesis, University of Sussex, Brighton.
O’Conner, T. (1994). Emergent properties. American Philosophical Quarterly, 31, 91–104.
Phillips, N. (1956). The general circulation of the atmosphere: A numerical experiment. Quarterly
Journal of the Royal Meteorological Society, 82, 123–164.
Pinker, S., & Mehler, J. (1988). Connections and symbols. Cambridge, MA: MIT Press.
Quine, W. V. O. (1951). Two dogmas of empiricism. Philosophical Review, 60(1), 20–43.
Quine, W. V. O. (1975). On empirically equivalent systems of the world. Erkenntnis, 9, 313–328.
Rowlands, M. (1994). Connectionism and the language of thought. British Journal for the
Philosophy of Science, 45(2), 485–503.
Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing. Cambridge: MIT
Press.
Sargent, R. G. (1982). Veriﬁcation and validation of simulation models. In F. E. Cellier (Ed.),
Progress in modeling and simulation (pp. 159–169). London: Academic Press.
Sargent, R. G. (1985). An expository on veriﬁcation and validation of simulation models. In
Proceedings of the 17th Conference on Winter Simulation (pp. 15–22). New York, NY: ACM
Press.
Sasahara, K., & Ikegami, T. (2004). Song grammars as complex sexual displays. In J. Pollack, M.
Bedau, P. Husbands, T. Ikegami, & R.A. Watson (Eds.), Artiﬁcial life IX: Proceedings of the
Ninth International Conference on the Simulation and Synthesis of Living Systems (pp. 194–
199). Cambridge, MA: MIT Press.
Sawyer, R. K. (2002). Nonreductive individualism: Part i: Supervenience and wild disjunction.
Philosophy of the Social Sciences, 32, 537–561.
Sawyer, R. K. (2003). Artiﬁcial societies: Multi-agent systems and the micro-macro link in
sociological theory. Sociological Methods and Research, 31(3), 37–75.
Sawyer, R. K. (2004). Social explanation and computational simulation. Philosophical Explorations,
7(3), 219–231.
Schlesinger, S., et al. (1979). Terminology for model credibility. SCS Technical Communications
for Modelling and Credibility in Simulation, 32(3), 103–104.
Schmid, A. (2005). What is the truth of simulation? Journal of Artiﬁcial Societies and Social
Simulation, 8(4), 5.
Von Neumann, J., & Burks, A. W. (1966). Theory of self-reproducing automata. Urbana: University
of Illinois Press.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
Chapter 3
Making the Artiﬁcial Real
3.1 Overview
Having established in Chaps. 1 and 2 a working understanding of the philosophical
underpinnings of computer simulation across disciplines, we turn now to a relatively
new ﬁeld which has created a stir throughout the computer science community to
investigate questions of artiﬁciality and its effect upon this type of inquiry. Can a
simulation create novel datasets which allow us to discover new things about natural
systems, or are simulations based on the natural world destined to be mere facsimiles
of the systems that inspire them?
For a possible answer we turn to Artiﬁcial Life, a ﬁeld which jumped to
prominence in the late 1980s and early 1990s following the ﬁrst international
conferences on the topic. Proponents of ‘Strong’ Alife argue that this new science
provides a means for studying ‘life-as-it-could-be’ through creating digital systems
that are nevertheless alive (Langton et al. 1989; Ray 1994).
Of course, such a bold claim invites healthy skepticism, so in this chapter we
will investigate the difﬁculties with strong Alife. By discussing ﬁrst the nature
of artiﬁciality in science and simulation, then investigating related theoretical and
methodological frameworks in Artiﬁcial Intelligence and more traditional sciences,
we attempt to uncover a way in which the strong Alife community might justify
their ﬁeld as a bona ﬁde method for producing novel living systems.
This discussion bears signiﬁcant importance in the further discussion to come.
Providing a theoretical background for any given simulation can greatly impact
the modeller’s ability to link a simulation to conventional empirical science, and
also serves to illuminate the assumptions inherent in the model and their possible
impacts. These issues will be discussed further in the following chapter, in which
this ﬁrst section of the text incorporates broader concerns regarding modelling from
the biological community in preparation for creating a framework that incorporates
social science in the second section.
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_3
39
40 3 Making the Artiﬁcial Real
3.2 Strong vs. Weak Alife and AI
3.2.1 Strong vs. Weak AI: Creating Intelligence
The AI community has often been characterised as encompassing two separate
strands of research: Strong AI, and Weak AI. Strong AI aims to develop computer
programmes or devices which exhibit true intelligence; these machines would be
aware and sentient in the same way as a human being. Weak AI, in contrast, aims to
develop systems which display a facsimile of intelligence; these researchers do not
attempt to create an intelligent being electronically, but instead a digital presence
which displays the abilities and advantages of an intelligent being, such as natural
language comprehension and ﬂexible problem-solving skills.
3.2.2 Strong vs. Weak Alife: Creating Life?
The title of the ﬁrst international artiﬁcial life conference immediately identiﬁed
two streams of Alife research: the synthesis of living systems as opposed to their
simulation. Perhaps designed to echo the distinction made between strong and weak
artiﬁcial intelligence, this division has been readily adapted by the Alife community
in the following years. While strong Alife attempts to create real organisms in a
digital substrate, the weak strand of Alife instead attempts to emulate certain aspects
of life (such as adaptability, evolution, and group behaviours) in order to improve
our understanding of natural living systems.
3.2.3 Deﬁning Life and Mind
Both intelligence and life suffer from inherent difﬁculties in formalisation; debate
has raged in psychology and biology about what factors might constitute intelligent
or living beings. AI has attempted to deﬁne intelligence in innumerable ways since
its inception, resulting in such notable concepts as the Turing Test, in which a
computer programme or device which can fool a blind observer into believing that
it is human is judged to be genuinely intelligent (Turing 1950).
Meanwhile, prominent philosophers of mind have pointed out the ﬂaws in such
arguments, as exempliﬁed in the oft-cited Chinese Room dilemma posed by Searle
(1980). In this thought experiment, an individual is locked inside a room with
an encyclopaedic volume delineating a set of rules allowing for perfect discourse
in Chinese. This individual engages in a dialogue with a native Chinese speaker
outside the room merely by following his rulebook. Searle contends that since
3.3 Levels of Artiﬁciality 41
the speaker himself does not have any innate knowledge of Chinese, his ability to
speak Chinese ﬂuently and convincingly through his rulebook does not prove his
intelligence. Counter-arguments to Searle’s Chinese Room have attempted to evade
this apparent conundrum by claiming that, while the individual inside the room has
no comprehension of Chinese, the entire system (encompassing the room, rulebook,
and individual) does have an intelligent comprehension of Chinese, thus making
the system an intelligent whole; Searle of course offered his own rejoinder to this
idea (Searle 1980, 1982), though the controversy continues with connectionists and
other philosophers continuing to weigh in with their own analyses (Churchland and
Churchland 1990; Harnad 2005).
Similarly life as a phenomenon is perhaps equally difﬁcult to deﬁne. While most
researchers in relevant ﬁelds agree that living systems must be able to reproduce
independently and display self-directed behaviour, this deﬁnition can fall apart
when one is presented with exceptional organisms such as viruses, which display
that reproductive component but consist of a bare minimum of materials necessary
to for that action. Are such organisms still alive, or are they no more than selfreproducing
protein strands? Alternatively, are even those protein strands in some
way ‘alive’? Researchers and philosophers alike continue to dispute the particulars.
Alife researchers’ claims that they can produce ‘real,’ digital life become more
problematic in this regard, as under such a nebulous framework for what constitutes
life, those researchers have quite a lot of latitude under which to make such
statements.
3.3 Levels of Artiﬁciality
3.3.1 The Need for Deﬁnitions of Artiﬁciality
The root of the problem of strong Alife is hinted at by the suggestion of unreality
or falsity that can be connoted by the terms thus far used to characterise Alife:
synthetic or simulated. A clariﬁcation of the type of artiﬁciality under consideration
could provide a more coherent picture of the type of systems under examination
in this type of inquiry, rather than leaving Alife researchers mired in a morass of
ill-deﬁned terminology.
Silverman and Bullock (2004) outline a simple two-part deﬁnition of the term
‘artiﬁcial,’ each intended to illuminate the disparity between the natural system and
the artiﬁcial system under consideration. First, the word artiﬁcial can be used to
denote a man-made example of something natural (hereafter denoted Artiﬁcial1
).
Second, the word can be used to describe something that has been designed to
closely resemble something else (hereafter denoted Artiﬁcial2
).
42 3 Making the Artiﬁcial Real
3.3.2 Artiﬁcial1
: Examples and Analysis
Artiﬁcal1
systems are frequently apparent in everyday reality. For example, artiﬁcial
light sources produce real light which consists of photons in exactly the same
manner as natural light sources, but that light is indeed manufactured rather than
being produced by the sun or bioluminescence. A major advantage for the ‘strong
artiﬁcial light’ researcher is that our current scientiﬁc understanding provides that
researcher with a physical theory which allows us to combine phenomena such as
sunlight, ﬁrelight, light-bulb light, and other forms of light into the singular category
of real light.
Brian Keeley’s example of artiﬁcial ﬂavourings (Keeley 1997) shows the limitations
of this deﬁnition. While an artiﬁcial strawberry ﬂavouring might produce
sensations in human tastebuds which are indistinguishable from real strawberry
ﬂavouring, this artiﬁcially-produced compound (which we shall assume has a
different molecular structure from the natural compound) not only originates from
a different source than the natural compound, but is also a different compound
altogether. In this case, while initially appearing to be a real instance of strawberry
ﬂavouring, one can make a convincing argument for the inherent artiﬁciality of the
manufactured strawberry ﬂavour.
3.3.3 Artiﬁcial2
: Examples and Analysis
Artiﬁcial2
systems, those designed to closely resemble something else, are similarly
plentiful, but soon show the inherent difﬁculties of relating natural and artiﬁcial
systems in this context. Returning to the artiﬁcial light example, we could imagine
an Artiﬁcial2
system which attempts to investigate the properties of light without
producing light itself; perhaps by constructing a computational model of optical
apparatus, for example, or developing means for replicating the effects of light
upon a room using certain architectural and design mechanisms. In this case, our
Artiﬁcial2
system would allow us learn about how light works and why it appears
to our senses in the ways that it does, but it would not produce real light as in an
Artiﬁcial1
system.
Returning to the case of Keeley’s strawberry ﬂavouring, we can place the manufactured
strawberry ﬂavouring more comfortably into the category of Artiﬁcial2
.
While the compound is inherently useful in that it may provide a great deal of insight
into the chemical and biological factors that produce a sensation of strawberry
ﬂavour, the compound itself is demonstrably different from the natural ﬂavouring,
and therefore cannot be used as a replacement for studying the natural ﬂavouring.
3.4 ‘Real’ AI: Embodiment and Real-World Functionality 43
3.3.4 Keeley’s Relationships Between Entities
In an effort to clarify these complex relationships between natural and artiﬁcial
systems, Brian Keeley (1997) describes three fundamental ways in which natural
and artiﬁcial entities can be related:
...(1) entities can be genetically related, that is, they can share a common origin, (2) entities
can be functionally related in that they share properties when described at some level of
abstraction, and (3) entities can be compositionally related; that is, they can be made of
similar parts constructed in similar ways. (Keeley 1997, p. 3, original emphasis)
This description seems to make the Alife researcher’s position even more
intractable. The ﬁrst category seems highly improbable as a potential relationship
between natural systems and Alife, given that natural life and digital life cannot
share genetic origins. The third category is more useful in the ﬁeld of robotics perhaps,
in which entities could conceivably be constructed which are compositionally
similar, or perhaps even identical, to biological systems. The second category, as
Keeley notes, seems most important to Alife simulation; establishing a functional
relationship between natural life and Alife seems crucial to the acceptance of Alife
as empirical enquiry.
3.4 ‘Real’ AI: Embodiment and Real-World Functionality
3.4.1 Rodney Brooks and ‘Intelligence Without Reason’
Rodney Brooks began a movement in robotics research toward a new methodology
for robot construction with his landmark paper ‘Intelligence Without Reason’
(Brooks 1991). He advocated a shift towards a focus on embodied systems, or
systems that function directly in a complex environment, as opposed to systems
designed to emulate intelligent behaviours on a ‘higher’ level. Further, he posited
that such embodiment could produce seemingly intelligent behaviour without highlevel
control structures at all; mobile robots, for example, could use simple rules to
walk which when combined with the complexities of real-world environments may
produce remarkably adaptable walking behaviours.
For Brooks and his contemporaries, the environment is not something to be
abstracted away in an AI task, but something which must be dealt with directly
and efﬁciently. In a manner somewhat analogous to the Turing Test, AI systems
must in a sense ‘prove’ their intelligence not through success in the digital realm
but through accomplishments rooted in real-world endeavour.
44 3 Making the Artiﬁcial Real
3.4.2 Real-World Functionality in Vision and Cognitive
Research
While embodiment may appear to be less of a concern in AI research related to
vision and cognition, as these behaviours can be separated from the embodied
organism more readily, such research is often still rooted in real-world environments.
Winograd’s well-known SHRDLU program (Winograd 1972) could answer
questions addressed in natural language that related to a ‘block world’ that it was
able to manipulate; the program could demonstrate knowledge of the properties of
this world and the relationships of the various blocks to one another. While the
‘block world’ itself was an artiﬁcial construct, the success of SHRDLU was based
upon its ability to interact with the human experimenter about its knowledge of that
virtual world, rather than just its ability to manipulate the virtual blocks and function
within that world.
Computer vision researchers follow a similar pattern to the natural-language
community, focusing on systems which can display a marked degree of competence
while encountering realistic visual stimuli. Objection recognition is a frequent
example of such a problem, testing a system’s ability to perceive and recognize
images culled from real-world stimuli, such as recognising moving objects in teamsports
footage (Bennett et al. 2004). Similarly, the current popularity of CCTV
systems has lead to great interest in cognitive systems which can analyse the wealth
of footage provided from many cameras simultaneously (Dee and Hogg 2006).
In the extreme, modern humanoid roboticists must integrate ideas from multiple
disciplines related to human behaviour and physiology as well as AI in order to
make these constructions viable in a real-world environment (Brooks et al. 1999).
Within the AI research community, the enduring legacy of the Turing Test
has created a research environment in which real-world performance must be the
end goal; merely understanding basic problems or dealing exclusively in idealised
versions of real-world situations are not sufﬁcient to prove a system’s capacity
for intelligent behaviour. The question of artiﬁciality is less important than that of
practicality; as in engineering disciplines, a functional system is the end goal.
3.4.3 The Differing Goals of AI and Alife: Real-World
Constraints
Clearly, real-world constraints are vital to the success of most research endeavours
in AI. Intelligent systems, Strong or Weak, must be able to distinguish and respond
to physical, visual, or linguistic stimuli, among others, in a manner sufﬁcient to
provide a useful real-world response. Without this, such systems would be markedly
inferior next to the notable processing abilities of even the most rudimentary
mammalian brain.
3.5 ‘Real’ Alife: Langton and the Information Ecology 45
For Alife, however, the landscape is far more muddled. Most simulations take
place in an idealised virtual context, with a minimum of complex and interacting
factors, in the hopes of isolating or displaying certain key properties in a clear
fashion. Given the disparity between the biological and digital substrates, and
the difﬁculties in deﬁning life itself, the relationship between the idealised virtual
context of the simulated organisms and any comparable biological organism seems
quite wide.
In this sense, artiﬁcial intelligence has a great advantage, as ‘human’ intelligence
and reasoning is a property of a certain subset of living beings, but can be viewed
as a property apart from the biological nature of those living beings to a degree.
Artiﬁcial life, by its very nature, attempts to investigate properties which depend
upon that living substrate in a much more direct fashion.
3.5 ‘Real’ Alife: Langton and the Information Ecology
3.5.1 Early Alife Work and Justiﬁcations for Research
The beginnings of Alife research stemmed from a number of examinations into
the properties of life using a series of relatively recent computational methods.
Genetic algorithms, which allow programs to follow a process of evolution in
line with a deﬁned ‘ﬁtness function’ to produce more capable programs, were
applied to digital organisms which attempted to compete and reproduce rather than
simply solve a task (Ray 1994). Similarly, cellular automata displayed remarkable
complexity despite being derived from very simple rules; such a concept of
‘emergent behaviour’ came to underwrite much of Alife in the years to come.
Creating simulations which display life-like behaviours using only simple rule-sets
seemed a powerful metaphor for the complexity of life deriving from the interactions
of genes and proteins.
3.5.2 Ray and Langton: Creating Digital Life?
Ray (1994) and Langton (1992) were early proponents of the Strong Alife view.
Ray contended that his Tierra simulation, in which small programs competed for
memory space in a virtual CPU, displayed an incredible array of varied ‘species’ of
digital organisms. He further posited that such simulations might hail the beginnings
of a new ﬁeld of digital biology, in which the study of digital organisms may teach
researchers about new properties of life that may be difﬁcult to study in a natural
context; he argued that his digital biosphere was performing fundamentally the same
functions as the natural biosphere:
46 3 Making the Artiﬁcial Real
Organic life is viewed as utilising energy, mostly derived from the Sun, to organize
matter. By analogy, digital life can be viewed as using CPU (central processing unit)
time, to organize memory. Organic life evolves through natural selection as individuals
compete for resources (light, food, space, etc.) such that genotypes which leave the
most descendants increase in frequency. Digital life evolves through the same process, as
replicating algorithms compete for CPU time and memory space, and organisms evolve
strategies to exploit one another. (Ray 1996, p. 373-4)
For Ray, then, an environment providing limited resources as a mechanism
for driving natural selection and an open-ended evolutionary process is sufﬁcient
to produce ‘increasing diversity and complexity in a parallel to the Cambrian
explosion.’ He goes on to describe the potential utility of such artiﬁcial worlds
for a new variety of synthetic biology, comparing these new digital forms of ‘real’
artiﬁcial life to established biological life.
Langton (1992), in his investigation of cellular automata, takes this idea one
step further, describing how ‘hardware’ computer systems may be designed to
achieve the same dynamical behaviours as biological ‘wetware.’ He posits that
properly organised synthetic systems can provide these same seemingly unattainable
properties (such as life and intelligence), given that each of these types of systems
can exhibit similar dynamics:
...if it is properly understood that hardness, wetness, or gaseousness are properties of the
organization of matter, rather than properties of the matter itself, then it is only a matter of
organization to turn ‘hardware’ into ‘wetware’ and, ultimately, for ‘hardware’ to achieve
everything that has been achieved by wetware, and more. (Langton 1992, p. 84)
For Langton, life is a dynamical system which strives to avoid stagnation,
proceeding in a series of phase transitions from one higher-level evolved state to the
next. This property can be investigated and replicated in computational form, which
in his view seems to provide sufﬁcient potential for hardware to develop identical
properties to wetware in the appropriate conditions (such as an appropriatelydesigned
cellular automata space).
3.5.3 Langton’s Information Ecology
Langton (1992) and Langton et al. (1989) attempted to justify his views regarding
artiﬁcial life by proposing a new deﬁnition for biological life. Given that natural life
depends upon the exchange and modiﬁcation of genetic information through natural
selection, Langton suggests that these dynamics of information exchange are in fact
the essential components of life:
...in living systems, a dynamics of information has gained control over the dynamics of
energy, which determines the behavior of most non- living systems. (Langton 1992, p. 41)
Thus, the comparatively simple thermodynamically regulated behaviours of nonliving
systems give way to living systems regulated by the dynamics of gene
exchange. From this premise, Langton proposes that if this ‘information ecology’
3.6 Toward a Framework for Empirical Alife 47
were accepted as the deﬁning conditions for life, then a computer simulation could
conceivably create an information ecology displaying the same properties.
3.6 Toward a Framework for Empirical Alife
3.6.1 A Framework for Empirical Science in AI
If we seek the construction of a theoretical framework to underwrite empirical
exploration in Alife, we can gather inspiration from Newell and Simon’s seminal
lecture (Newell and Simon 1976). The authors sought to establish AI as a potential
means for the empirical examination of intelligence and its origins. They argue that
computer science is fundamentally an empirical pursuit:
Computer science is an empirical discipline.... Each new machine that is built is an
experiment. Actually constructing the machine poses a question to nature; and we listen
for the answer by observing the machine in operation and analyzing it by all analytical and
measurement means available. (Newell and Simon 1976, p. 114)
However, the argument that computer science is fundamentally empirical due to
its experimental interactions with replicable, physical systems is not sufﬁcient to
claim that an AI can be used as a means to study intelligence empirically. Newell
and Simon address this by proposing a deﬁnition of a physical symbol system, or
‘a machine that produces through time an evolving collection of symbol structures’
(Newell and Simon 1976, p. 116). The details of the deﬁnition and its full import are
beyond the scope of this text, but in essence, the authors argue that physical symbol
systems are capable of exhibiting ‘general intelligent action’ (Newell and Simon
1976, p. 116), and further, that studying any generally intelligent system will prove
that it is, in fact, a physical symbol system.
Newell and Simon then suggest that physical symbol systems can, by deﬁnition,
be replicated by a universal computer. This leads us to their famous Physical Symbol
System Hypothesis – or PSS Hypothesis – which we can summarise as follows:
1. ‘A Physical Symbol System has the necessary and sufﬁcient means for general
intelligent action.’ (Newell and Simon 1976, p. 116)
2. A computer is capable of replicating a Physical Symbol System.
Thus, by establishing general intelligence as a process of manipulating symbols
and symbol expressions, and that computers are capable of replicating and performing
identical functions – and indeed are quite good at doing so – Newell and Simon
present AI as an empirical study of real, physical systems capable of intelligence.
AI researchers are not merely manipulating software for the sake of curiosity, but
are developing real examples of intelligent systems following the same principles
as biological intelligence.
48 3 Making the Artiﬁcial Real
3.6.2 Newell and Simon Lead the Way
Newell and Simon’s PSS Hypothesis (Newell and Simon 1976) largely succeeded
in providing a framework for AI researchers at the time, and one that continues to be
referenced and used today. Such a framework is notably absent in Alife, however,
given the difﬁculties in both deﬁning Alife and in specifying a uniﬁed ﬁeld of Alife,
which necessarily spans quite a few methodologies and theoretical backgrounds.
With Langton’s extension of the ﬂedgling ﬁeld of Alife into the study of ‘life-asit-could-be,’
a more uniﬁed theoretical approach seems vital to understanding the
relationship between natural and artiﬁcial life:
Artiﬁcial life is the study of artiﬁcial systems that exhibit behavior characteristic of natural
living systems. It is the quest to explain life in any of its possible manifestations, without
restriction to the particular examples that have evolved on Earth. This includes biological
and chemical experiments, computer simulations, and purely theoretical endeavors. Processes
occurring on molecular, social and evolutionary scales are subject to investigation.
The ultimate goal is to extract the logical form of living systems. (Langton, announcement
of Artiﬁcial Life: First International Conference on the Simulation and Synthesis of Living
Systems)
By likening Alife to the study of the very nature of living systems, Langton
appeals to the apparent ﬂexibility and power of computer simulations. The simulation
designer has the opportunity to create artiﬁcial worlds that run orders of
magnitude faster than our own and watch thousands of generations of evolution pass
in a short space of time, and thus seems to provide an unprecedented opportunity
to observe the grand machinery of life in a manner that is impossible in traditional
biology.
This, in turn, leads to an enticing prospect: can such broad-stroke simulations be
used to answer pressing empirical questions about natural living systems? Bedau
(1998), for example, sees a role for Alife in a thought experiment originally
proposed by Gould (1989). Gould asks what might happen if we were able to
rewind evolutionary history to a point preceding the ﬁrst developments of terrestrial
life. If we changed some of those initial conditions, perhaps merely by interfering
slightly with the ‘primordial soup’ of self-replicating molecules, what would we see
upon returning to our own time? Gould suggests that while we might very well see
organisms much the same as ourselves, there is no reason to assume that this would
be the case; we may resume life in our usual time frame to discover that evolutionary
history has completely rearranged itself as a result of these manipulations.
For Bedau, this thought experiment presents an opening for Alife to settle a
question fundamentally closed to traditional biology. By constructing a suitable
simulation which replicates the most important elements of biological life and
the evolutionary process, and running through numerous simulations based on
variously-perturbed primordial soups, we could observe the resultant artiﬁcial
organisms and see for ourselves the level of diversity which results.
Other authors have proposed Alife simulations that fall along similar lines
(Bonabeau and Theraulaz 1994; Ray 1994; Miller 1995), noting that evolutionary
biologists are burdened with a paucity of evidence with which to reconstruct the
3.6 Toward a Framework for Empirical Alife 49
evolutionary course of life on Earth. The fossil record is notoriously incomplete,
and our vanishingly small time on Earth has allowed precious few opportunities
to observe even the species which exist around us today. In addition, as Gould’s
thought experiment highlights, we only have the opportunity to observe those
organisms which have evolved on Earth, leaving us entirely uncertain of which
properties we observe in that life are particular to life on Earth, and which are
particular to life in any form.
Quite obviously such an experiment could be potentially revolutionary, and yet
the methodological problems are vast despite the inherent ﬂexibility of computer
simulation. How could we possibly conﬁrm that these artiﬁcial organisms are
Artiﬁcial1
rather than Artiﬁcial2
? Given our lack of a deﬁnition for biological life,
determining whether a digital organism is alive seems a matter of guesswork at best.
Further, how detailed must the simulation be to be considered a reasonable facsimile
of real-world evolutionary dynamics? Should they select on the level of individuals,
or genes, or perhaps even artiﬁcial molecules of some sort? If by some measure we
determine the simulation to be Artiﬁcial2
rather than Artiﬁcial1
, can we truly claim
that this simulation in any way represents the development of natural living systems,
or is it merely a fanciful exploration of poorly-understood evolutionary dynamics?
3.6.3 Theory-Dependence in Empirical Science
Beyond these methodological considerations, some troubling philosophical issues
appear when considering such large-scale applications of the Alife approach. Some
have argued that Alife simulations should not, and in fact cannot, be considered
useful sources of empirical data given that they are loaded with inherent biases from
their creators (Di Paolo et al. 2000). Simulations must necessarily adopt varying
levels of abstraction in order to produce simulations which are both computable and
analysable; after all, simulations which replicate the complexities of real biology in
exacting detail would gain the experimenter very little time-saving and eliminate
one of the major beneﬁts of computational modelling. These abstractions are
tacitly inﬂuenced by the experimenter’s own biases, relying upon the simulation
creator’s perception of which aspects of the biological system can be simpliﬁed or
even removed from the simulation entirely. Similarly, the parameter settings for
that simulation will come with their own biases, resulting in simulations which
could produce vastly different results depending on the theoretical leanings of the
programmer.
While one might claim that conventional empirical science suffers from very
similar shortfalls, Chalmers points out that biases within such ﬁelds must necessarily
end when the results of an experiment contradict those biases:
...however informed by theory an experiment is, there is a strong sense in which the results
of an experiment are determined by the world and not by theories... we cannot make [the]
outcomes conform to our theories. (Chalmers 1999, p. 39–40)
50 3 Making the Artiﬁcial Real
In the case of computer simulations this conclusion seems more difﬁcult. After
all, we are in essence adding an additional layer – in addition to the two layers of
the physical world and the theory that describes it, we now have a model layer as
well, which approximates the world as described by our theory. Clearly this adds
additional complications: when not only the experiment but the very world in which
that experiment takes place are designed by that biased experimenter, how can one
be sure that pre-existing theoretical biases have not completely contaminated the
simulation in question?
To return to our central bird migration example, one can imagine various ways in
which our migration researcher could introduce theoretical bias into the simulation.
That researcher’s ideas regarding the progression of bird migration, the importance
of individual behaviours or environmental factors in migration, and other preexisting
theoretical frameworks in use by the researcher will inform the assumptions
made during construction of the model. In such a scenario, removing such biases is
difﬁcult for the modeller, as on some level the model requires certain assumptions
to function; the researcher needs to develop a theoretical framework in which this
artiﬁcial data remains usable despite these issues.
3.6.4 Artiﬁcial Data in Empirical Science
An examination of more orthodox methods in empirical science may shed some
light on how some methodologies use artiﬁcially-generated data to answer empirical
questions. While these methods are more based in the natural world than an Alife
simulation, they still rely upon creating a sort of artiﬁcial world in which to examine
speciﬁc elements of a process or phenomenon.
Despite the artiﬁciality of this generated data, such disciplines have achieved a
high degree of acceptance amongst the research community. A brief look at some
examples of the use of such data in empirical research will provide some insight
into possible means for using artiﬁcial data derived from simulation.
3.6.4.1 Trans-Cranial Magnetic Stimulation
Research into brain function often employs patients who have suffered brain damage
through strokes or head injuries. Brains are examined to determine which areas are
damaged, and associations between these damaged areas and the functional deﬁcits
exhibited by the patients are postulated. The technique of trans-cranial magnetic
stimulation, known as TCMS or TMS, has allowed psychology researchers to extend
the scope of this approach by generating temporary artiﬁcial strokes in otherwise
healthy individuals.
TMS machinery produces this effect through a set of electrodes which are
placed on the outside of the subject’s skull. The researcher ﬁrst maps the subject’s
brain using an MRI scan, and then begins using the TMS apparatus to ﬁre brief
3.6 Toward a Framework for Empirical Alife 51
magnetic pulses at the surface of the brain. These pulses in effect overstimulate the
affected neurons, causing small areas of the brain to ‘short-circuit,’ which replicates
the effects of certain types of brain injury. TMS researchers have managed to
replicate the effects of certain types of seizure (Fujiki and Steward 1997), and have
examined the effects of stimulation of the occipital cortex in patients with earlyonset
blindness (Kujala et al. 2000). TMS has even produced anomalous emotional
responses; after inhibition of the prefrontal cortex via TMS, visual stimuli that might
normally trigger a negative response were much more likely to cause a positive
response (Padberg 2001).
Such methods provide a way for neuroscientists and psychologists to circumvent
the lack of sufﬁcient data from lesion studies. Much of cognitive neuropsychology
suffers from this paucity of raw data, forcing researchers to spend inordinate
amounts of time searching through lengthy hospital records and medical journals
for patients suffering from potentially appropriate brain injuries. Even when such
subjects are found, normal hospital testing may not have revealed the true nature of
the subject’s injury, meaning that potentially valuable subjects go unnoticed as some
ﬁnely differentiated neurological deﬁcits are unlikely to be discovered by hospital
staff. Assuming that all of these mitigating factors are bypassed, the researcher is
still faced with a vanishingly small subject pool which may adversely affect the
generalisability of their conclusions.
TMS thus seems a remarkable innovation, one which suddenly widens the
potential pool of subjects for cognitive neuropsychology to such a degree that any
human being may become a viable subject regardless of the presence or lack of
brain injury. However, there are a number of shortcomings to TMS which could
cause one to question the validity of associated results. The inhibition of neural
activity produced by TMS causes neurons to explode into such activity that normal
ﬁring patterns become impossible; while this certainly effectively disrupts activity
in the brain area underneath the pulse, TMS is not necessarily capable of replicating
the many and varied ways in which brain injuries may damage the neural tissue.
Similarly, the electromagnetic pulse used to cause this inhibition of activity is only
intended to affect brain areas which are near to the skull surface; the pulse does
penetrate beyond those areas, but the effects of this excess inhibition are not fully
understood. Despite these shortcomings, and the numerous areas in which standard
examinations of brain-injury patients are superior, TMS has continued to become a
rapidly-growing area of research in cognitive neuropsychology.
The data derived from TMS is strongly theory-dependent in nature, in a similar
fashion to computer simulation. In order to use this data as a reasonable method
of answering empirical questions requires that TMS researchers adopt a theoretical
‘backstory.’ This backstory is a framework that categorises TMS data and lesion data
together as examples of ‘real’ brain-damage data. Despite the fact that TMS brain
damage is generated artiﬁcially, it is deemed admissible as ‘real’ data as long as
neuroscientists consider this data Artiﬁcial1
rather than Artiﬁcial2
in classiﬁcation.
52 3 Making the Artiﬁcial Real
3.6.4.2 Neuroscience Studies of Rats
Studies of rats have long been common within the ﬁeld of neuroscience, given the
complex methodological and ethical difﬁculties involved in studying the human
brain. Though such studies allow much greater ﬂexibility due to their more
uncontroversial participants, certain fundamental limitations still prohibit certain
techniques; while ideally, researchers would prefer to take non-invasive, in situ
recordings of the neural activity of normal, free-living rats, such recordings are well
beyond the current state of the art.
Instead, neuroscientists must settle for studies of artiﬁcially prepared rat brains
or portions thereof; such a technique would be useful if the researcher wishes to
determine the activity of speciﬁc neural pathways given a predetermined stimulus,
for example. One study of the GABA-containing neurons within the medial
geniculate body of the rat brain required that the rats in question were anaesthetised
and dissected, with slices of the brain then prepared and stimulated directly (Peruzzi
et al. 1997). Through this preparatory process, the researchers hoped to determine
the arrangement of neural connections within the rat’s main auditory pathway, and
by extension gain some insight into the possible arrangement of the analogous
pathway in humans.
Given how commonplace such techniques have become within modern neuroscience,
most researchers in that community would not question the empirical
validity of this type of procedure; further, the inherent difﬁculties involved in
attempting to identify such neural pathways within a living brain makes such
procedures the only known viable way to take such measurements. However,
one might argue that the behaviour of cortical cells in such a preserved culture,
entirely separate from the original brain, would certainly differ signiﬁcantly from
the behaviour of those same cells when functioning in their usual context. The entire
brain would provide various types of stimulus to the area of cortex under study, with
these stimuli in turn varying in response to both the external environment and the
rat’s own cognitive behaviour.
With modern robotics and studies of adaptive behaviour focusing so strongly
upon such notions of embodiment and situatedness, the prevailing wisdom within
those ﬁelds holds that an organism’s coupling to its external environment is
fundamental to that organism’s cognition and behaviour (Brooks 1991). In that
case, neuroscience studies of this sort might very well be accused of generating
and recording ‘artiﬁcial’ neural data, producing datasets that differ fundamentally
from the real behaviour of such neurons. How then do neuroscientists apply such
data to the behaviour of real rats, and in turn to humans and other higher mammals?
In fact, research of this type proceeds on the assumption that such isolation of
these cortical slices from normal cognitive activity actually increases the experimental
validity of the study. By removing the neurons from their natural environment,
the effects of external stimuli can be eliminated, meaning that the experimenters
have total control over the stimulus provided to those neurons. In addition, the
chemical and electrical responses of each neuron to those precise stimuli can be
measured precisely, and the inﬂuence of experimenter error can be minimised.
3.6 Toward a Framework for Empirical Alife 53
With respect to the artiﬁciality introduced into the study by using such means,
neuroscience as a whole appears to have reached a consensus. Though removing
these cortical slices from the rat is likely to fundamentally change the behaviour
of those neurons as compared to their normal activation patterns, the individual
behaviour of each neuron can be measured with much greater precision in this way;
thus, the cortical slicing procedure can be viewed as Artiﬁcial1
, as the individual
neurons are behaving precisely as they should, despite the overall change in the
behaviour of that cortical slice when isolated from its natural context. Given the
(perhaps tacit) agreement that these methods are Artiﬁcial1
in nature, the data from
such studies can be agreed to consist of ‘real’ neuroscience data.
3.6.5 Artiﬁcial Data and the ‘Backstory’
The two examples above drawn from empirical science demonstrate that even commonplace
tools from relatively orthodox ﬁelds are nevertheless theory-dependent in
their application, and further that such dependence is not necessarily immediately
obvious or straightforward. The tacit assumptions underlying the use of TMS and rat
studies in neuroscience are not formalised, but they do provide a working hypothesis
which allows for the use of data derived from these methods as ‘real’ empirical data.
The lack of a conclusive framework showing the validity of these techniques does
not exclude them from use in the research community.
This is of course welcome news for the strong Alife community, given that a
tangible and formal deﬁnition of what constitutes a living system is quite far from
being completed. However, strong Alife also lacks this tacit ‘backstory’ that is
evident in the described empirical methods; without this backstory those methods
may fall afoul of the Artiﬁcial1
/Artiﬁcial2
distinction. With this in mind, how might
the Alife community describe a similar backstory which underwrites this research
methodology as a valid method for producing new empirical datasets?
The examples of TMS and rat neuroscience offer one possibility. In each of
those cases, the investigative procedure begins with an uncontroversial example of
the class of system under investigation (i.e., in the case of TMS, a human brain).
This system is then prepared in some way in order to become more amenable
to a particular brand of investigation; rat neuroscientists, by slicing and treating
the cortical tissue, allow themselves much greater access to the functioning of
individual neurons. In order to justify these modiﬁcations to the original system,
these preparatory procedures must be seen as neutral in that they will not distort the
resulting data to such a degree that the data becomes worthless. Indeed, in both TMS
and rat neuroscience, the research community might argue that these preparatory
procedures actually reduce some fairly signiﬁcant limitations placed upon them by
other empirical methodologies.
At ﬁrst blush such a theoretical framework seems reasonable for Alife. After
all, other forms of ‘artiﬁcial life’ such as clones or recombinant bacteria begin
with such an uncontroversial living system and ‘prepares’ it while still producing
54 3 Making the Artiﬁcial Real
a result universally accepted to be another living system. However, this does fall
substantially short of one of the central goals of strong artiﬁcial life, as these systems
certainly produce augmented datasets regarding the living systems involved, but
they do not generate entirely novel datasets. While recombinant bacteria and
mutated Drosophila may illuminate elements of those particular species that we
may have been unable to uncover otherwise, the investigation of ‘life-as-it-couldbe’
remains untouched by these efforts.
In addition to these shortcomings, the preparatory procedures involved in a
standard Alife simulation are quite far removed from those we see in standard
biology (or the neuroscience examples discussed earlier). These simulated systems
exist entirely in the digital realm, completely removed from the biological substrate
of living systems; though many of these simulated systems may be based upon the
form or behaviour of natural living systems, those systems remain separate from
their simulated counterparts.
Further, this computer simulation is ‘prepared’ in this digital substrate through
a process of programming to produce this artiﬁcial system. Programming is
inherently a highly variable process, the practice of which differs enormously from
one practitioner to the next, in contrast to the highly standardised procedures of
neuroscience and biology. The result of these preparations is to make the system
amenable to creating life, rather than simply taking a previously-existing system
and making it more amenable to a particular type of empirical observation. While
characterising this extensive preparatory process as benign, as in neuroscience and
biology, would be immensely appealing to the Alife community, the argument that
preparing a computer and somehow creating life upon its digital substrate is a benign
preparatory procedure is a difﬁcult one to make.
3.6.6 Silverman and Bullock’s Framework: A PSS Hypothesis
for Life
Thus, while we have seen that even orthodox empirical science can be considered
strongly theory-dependent, the gap between natural living systems and Alife
systems remains intimidatingly wide. From this perspective, characterising Alife
as a useful window onto ‘life-as-it-could-be’ is far from easy; in fact, Alife appears
dangerously close to being a quintessentially Artiﬁcial2
enterprise.
Newell and Simon’s seminal paper regarding the Physical Symbol System
Hypothesis offers a possible answer to this dilemma (Newell and Simon 1976).
Faced with a similar separation between the real system of interest (the intelligence
displayed by the human brain) and their own attempt to replicate it digitally,
the PSS Hypothesis offered a means of justifying their ﬁeld. By establishing a
framework under which their computers offer the ‘necessary and sufﬁcient means’
for intelligent action, Newell and Simon also establish that any computer is only a
short (albeit immensely complicated) step away from becoming an example of real,
Artiﬁcial1
intelligence.
3.6 Toward a Framework for Empirical Alife 55
The PSS Hypothesis also escapes from a potential theoretical conundrum by
avoiding discussion of any perceived similarities between the behaviour of AI
systems and natural intelligence. Instead Newell and Simon attempt to provide
a base-level equivalence between computation and intelligence, arguing that the
fundamental symbol-processing abilities displayed by natural intelligence are replicable
in any symbol-processing system of sufﬁcient capability. This neatly avoids
the shaky ground of equating AI systems to human brains by instead equating
intelligence to a form of computation; in this context, the idea that computers can
produce a form of intelligence follows naturally.
With this in mind, we return to Langton’s concept of the information ecology
(1992). If we accept the premise that living systems have this ecology of information
as their basis, then can we also argue that information-processing machines may also
lead to the development of living systems? Following Newell and Simon’s lead,
Silverman and Bullock (2004) offer a PSS Hypothesis for life:
1. An information ecology provides the necessary and sufﬁcient conditions for life.
2. A suitably-programmed computer is an example of an information ecology. (Silverman
and Bullock 2004, p. 5)
Thus, assuming that the computer in question was appropriately programmed
to take advantage of this property, then an Alife simulation could be regarded as
a true living system, rather than an Artiﬁcial2
imitation of life. As in Newell and
Simon, the computer becomes a system that is a sort of blank slate, needing only
the appropriate programming to become a repository for digital life.
This PSS Hypothesis handily removes the gap between the living systems
which provide the inspiration for Alife and the systems produced by an Alife
simulation. Given that computers inherently provide an information ecology, an
Alife simulation can harness that property to create a living system within that
computer. Strong Alife then becomes a means for producing entirely new datasets
derived from genuine digital lifeforms rather than simply a method for creating
behaviours that are reminiscent of natural life.
3.6.7 The Importance of Backstory for the Modeller
As discussed in earlier sections about examples of the theoretical backstory in
empirical disciplines, the presence of such a backstory allows the experimenter to
justify the usefulness of serious alterations to the system under study. In the case
of rat neuroscience and TMS research, their back-stories allow them to describe the
changes and preparations they make to their subjects as necessary means for data
collection.
In the case of Alife, such a claim is difﬁcult to make, as noted previously, given
that the simulation is creating data rather than collecting it from a pre-existing source
as is the case in our empirical examples. The PSS Hypothesis for Life avoids this
difﬁculty by stating that any given computer hardware substrate forms the necessary
56 3 Making the Artiﬁcial Real
raw material for life; in a sense, our simulation is merely activating this potential to
create an information ecology, and then collecting data from the result. Thus, we
may say that our programming of the simulation has prepared this digital substrate
for empirical data-collection in a manner similar to that of empirical studies.
Of course, such a backstory is signiﬁcant in scope, and could invite further
criticism. However, such a statement is difﬁcult to refute, given the unsophisticated
nature of our current deﬁnitions for biological life, and a lack of agreement on
whether life may exist in other substrates. Either way, the importance for the
modeller is that such a simulation takes on a different character. The focus of
such a theoretical justiﬁcation is on implementing a model which produces some
empirical data collection that may bear on our overall understanding of life, not on
simply tweaking an interesting computational system to probe the results. The PSS
Hypothesis for Life gives us some basis on which to state that this view of Alife
models has some theoretical validity.
3.6.8 Where to Go from Here
Now that we have produced a theoretical backstory which underwrites Alife
research as a means for generating new empirical data points, what is missing? The
PSS Hypothesis for Life allows us to pursue simulations of this type as a means
for understanding life as a new form of biology, investigating the properties and
behaviours of entirely novel organisms. One can imagine how such endeavours
could provide interesting insight for those interested in the broader questions of
what makes something alive.
However, none of this allows us to provide any direct relevance in our results in
Alife to the biological sciences. The biologist who views our bird migration model,
justiﬁed as an information ecology under the PSS Hypothesis for Life, as interesting,
but irrelevant to the concerns of the empirical ornithologist. Indeed, if our simulation
is producing merely a bird-like manifestation of digital life, then we have no basis
on which to state that what we see in our results can tell us anything about real birds,
completely removed from the virtual information ecology we have implemented.
This issue becomes the focus of the next chapter, the ﬁnal chapter of Part I.
How might we apply Alife modelling techniques and agent-based models to the
broader canvas of biological science? An in-depth discussion of methodological and
theoretical issues related to modelling in population biology will provide us with the
context necessary to begin to answer this question. The concerns of the strong Alife
modeller as depicted in this chapter differ in several important respects to those of a
modeller with a weak Alife orientation, and those concerns focus largely on creating
a relationship to external empirical data rather than creating their own empirical data
as the PSS Hypothesis for Life describes.
3.7 Summary and Conclusions 57
3.7 Summary and Conclusions
ALife by its very nature is a ﬁeld which depends upon the use and acceptance
of artiﬁcially-generated data to investigate aspects of living systems. While some
members of this research community have posited that artiﬁcial systems can be
alive in a manner identical to biological life, there are numerous philosophical and
methodological concerns inherent in such a viewpoint.
The comparison with artiﬁcial intelligence laid out in this chapter illustrates
some of the particular methodological difﬁculties apparent in ALife. ALife seeks to
replicate properties of life which are heavily dependent on the biological substrate,
in contrast with AI, which seeks to emulate higher-level properties of living
organisms which seem more easily replicable outside of the biological realm.
AI does not entirely escape the problem of artiﬁciality in its data and methods
however, nor for that matter does conventional empirical science. Some disciplines
are able to make use of a great deal of artiﬁcial data by tying it to a theoretical
‘backstory’ of a sort. ALife up to now has lacked this backstory, while AI has Newell
and Simon’s PSS Hypothesis (Newell and Simon 1976) as one example of such a
theoretical endeavour.
Silverman and Bullock (2004) used Newell and Simon’s account as an inspiration
for a PSS Hypothesis for life, a framework which could be used to underwrite
strong ALife as a form of empirical endeavour. By accepting that life is a form of
information ecology which does not depend exclusively on a biological substrate, a
researcher signing up to this backstory may argue for the use of strong ALife data
as a form of empirical data. Initially such an idea is appealing; after all, the strong
ALife proponent seeks to create forms of ‘life-as-it-could-be’ in a digital space, and
this variant of the PSS Hypothesis describes such instances of digital life as a natural
consequence of life itself being an information ecology.
However, this framework does not account for the more pragmatic methodological
concerns which affect modelling endeavours of this type, and in fact modelling
in general. Constructing a computational model requires a great deal more than a
useful theoretical justiﬁcation to function appropriately and provide useful data. As
one example, accepting ALife simulations as a form of empirical enquiry does not
simplify the task of relating that data to similar data derived from entirely natural
systems. As exciting as the prospect of digital life may be, creating self-contained
digital ecosystems which are unrelatable to natural ecosystems seems of limited
utility for the biologist. Such issues, examined in the context of mathematical
models in population biology in the following chapter, will provide greater insight
into these crucial elements of our growing theoretical framework for simulation
research. This framework can then be expanded to bear on our upcoming discussion
of simulation for the social sciences in the second overall section of this text.
58 3 Making the Artiﬁcial Real
References
Bedau, M. A. (1998). Philosophical content and method of artiﬁcial life. In T. W. Bynum, & J.
H. Moor (Eds.), The digital Phoenix: How computers are changing philosophy (pp. 135–152).
Oxford/Malden, MA: Blackwell Publishers.
Bennett, B., Magee, D. R., Cohn, A. G., & Hogg, D. C. (2004). Using spatio-temporal continuity
constraints to enhance visual tracking of moving objects. European Conference on Artiﬁcial
Intelligence.
Bonabeau, E. W., & Theraulaz, G. (1994). Why do we need artiﬁcial life? Artiﬁcial Life, 1, 303–
325.
Brooks, R. A. (1991). Intelligence without reason. In J. Myopoulos, & R. Reiter (Eds.), Proceedings
of the Twelfth International Joint Conference on Artiﬁcial Intelligence (pp. 569–595). San
Mateo, CA: Morgan Kaufmann.
Brooks, R. A., Breazeal, C., Marjanovic, M., Scassellati, B., & Williamson, M. M. (1999). The
Cog project: Building a humanoid robot. In Chrystopher L. Nehaniv (Ed.), Computation for
metaphors, analogy, and agents. Berlin: Springer.
Chalmers, A. F. (1999). What is this thing called science? Buckingham: Open University Press.
Churchland, P. M., & Churchland, P. S. (1990). Could a machine think? Scientiﬁc American, 262,
32–37.
Dee, H., & Hogg, D. (2006). Navigational strategies and surveillance. In Sixth IEEE International
Workshop on Visual Surveillance, Graz.
Di Paolo, E. A., Noble, J., & Bullock, S. (2000). Simulation models as opaque thought experiments.
In M. A. Bedau, et al. (Eds.), Artiﬁcial life VII (pp. 497–506). Cambridge, MA: MIT Press.
Fujiki, M., & Steward, O. (1997). High frequency transcranial magnetic stimulation mimics the
effects of ECS in upregulating astroglial gene expression in the murine CNS. Molecular Brain
Research, 44, 301–308.
Gould, S. J. (1989). Wonderful life: The burgess shale and the nature of history. New York: W.W.
Norton.
Harnad, S. (2005). What is consciousness? New York Review, 52(11).
Keeley, B. L. (1997). Evaluating artiﬁcial life and artiﬁcial organisms. In C. G. Langton, &
K. Shimohara (Eds.), Artiﬁcial life V (pp. 264–271). Cambridge: MIT Press.
Kujala, T., Alho, K., & Naatanen, R. (2000). Cross-modal reorganization of human cortical
functions. Trends in Neuroscience, 23, 115–120.
Langton, C. G. (1992). Life at the edge of chaos. In C. G. Langton, C. Taylor, J. D. Farmer, & S.
Rasmussen (Eds.), Artiﬁcial life II (pp. 41–91). Reading: Addison-Wesley.
Langton, C. G., et al. (1989). Artiﬁcial life. Redwood City: Addison-Wesley.
Miller, G. F. (1995). Artiﬁcial life as theoretical biology: How to do real science with computer
simulation. Cognitive Science Research Paper 378, University of Sussex, Brighton.
Newell, A., & Simon, H. A. (1976). Computer science as empirical enquiry: Symbols and search.
Communications of the Association for Computing Machinery, 19, 113–126.
Padberg, F. (2001). Prefrontal cortex modulation of mood and emotionally induced facial expressions:
A transcranial magnetic stimulation study. Journal of Neuropsychiatry and Clinical
Neurosciences, 13, 206–212.
Peruzzi, D., Bartlett, E. L., Smith, P. H., & Oliver, D. L. (1997). A monosynaptic GABAergic input
from the inferior colliculus to the medial geniculate body in rat. Journal of Neuroscience, 17,
3766–3777.
Ray, T. S. (1994). An evolutionary approach to synthetic biology: Zen and the art of creating life.
Artiﬁcial Life, 1, 179–209.
Ray, T. S. (1996). An approach to the synthesis of life. In M. A. Boden (Ed.), The philosophy of
artiﬁcial life (pp. 111–145). Oxford: Oxford University Press.
Searle, J. R. (1980). Minds, brains and programs. Behavioral and Brain Sciences, 3(3), 417–457.
Searle, J. R. (1982). The chinese room revisited. Behavioral and Brain Sciences, 5, 345–348.
References 59
Silverman, E., & Bullock, S. (2004). Empiricism and artiﬁcial life. In J. Pollack, M. Bedau,
P. Husbands, T. Ikegami, & R. A. Watson (Eds.), Artiﬁcial Life IX: Proceedings of the
Ninth International Conference on the Simulation and Synthesis of Living Systems. MIT
Press/Bradford Books.
Turing, A. M. (1950). Computing machinery and intelligence. Mind, 49, 433–460.
Winograd, T. (1972). Understanding natural language. Cognitive Psychology, 3(1), 1–191.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
Part II
Modelling Social Systems
The advent of Alife lead to a period of great excitement, as the power and ﬂexibility
of simulation suggested whole new ways to study the processes of biology. So
too with social simulation, a relatively recent development in the social sciences
in which simulation methods – most commonly agent-based models – are used
to model social systems and their behaviours. The prospect of using simulation to
model the emergence of population-level effects from individual social interactions
at the agent level.
In Part II we will investigate social simulation as a methodology, uncovering both
the strengths and weaknesses of this approach for revealing the processes underlying
human society and its evolution. We will examine the complex relationships
between social simulations, real-world population data, and social theory. We will
also expand the methodological analysis begun in Part I and discover how the
modelling frameworks we studied may be applied to social simulation, and how
they compare to similar frameworks developed by social modellers and theorists
themselves.
In order to examine these frameworks and their potential utility for social
simulation, we will take inspiration from the early days of the ﬁeld. The beginning
of social simulation is frequently credited to Thomas Schelling and his residential
segregation model, an elegantly simple investigation of how even minor preferences
amongst individuals for living near to others similar to themselves can lead to
residential segregation emerging at the population level. We will use Schelling’s
model to compare the methodological frameworks uncovered through Parts I and II,
and use the insights gained to propose a way forward for social simulation as a
whole.
Chapter 5
Modelling for the Social Sciences
Eric Silverman and John Bryden
5.1 Overview
The examination of the use of ‘artiﬁcial worlds’ in the previous chapter seemed
to produce some rather damning concerns for those involved in agent-based
simulation. While such models can provide a nicely compartmentalised and distilled
view of a vastly more complicated real-world system, such a model can create
tremendous difﬁculty when the researcher must begin to analyse the resultant data.
In the context of our overall discussion, however, the examples and analysis
presented so far have focused largely on models based upon or inspired by biology.
Agent-based modelling is far from conﬁned to this singular area of study, so this
chapter will introduce a major theme of this text: the uses and limitations of agentbased
models in the social sciences.
As agent-based modelling spreads through various disciplines, there are certain
applications which seem particularly promising due to an attendant lack of empirical
data underwriting those disciplines. Social science seems an especially relevant
example of this situation; by its very nature, the study of social systems makes
data-gathering a difﬁcult proposition.
In such an instance, can agent-based models provide a means for developing
social theories despite the lack of empirical data to validate the models themselves?
This chapter will examine this question in detail, ﬁrst by describing the current
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_5
85
86 5 Modelling for the Social Sciences
state-of-the-art in simulation in the ﬁeld, and second by examining the methodological
and philosophical implications of applying simulation techniques to the social
sciences.
This chapter forms the ﬁrst section of Part II of our analysis. The discussion in
this chapter places the current state of social science simulation into the context of
our analysis of modelling for Alife and the biological sciences. This allows us to
develop a contrast between simulation approaches between these two ﬁelds, and in
turn discover those points at which the unique problems of social science research
impact upon the relevance of agent-based models to the social science researcher.
5.2 Agent-Based Models in Political Science
5.2.1 Simulation in Social Science: The Role of Models
In the past, models within social sciences such as economics, archaeology and
political science have focused on mathematical approaches, though as modelling
techniques within the ﬁeld progressed, some began to criticise this focus. Read
(1990) observes that mathematical models within archaeology may have produced
sets of modelling assumptions that do not produce useful insights into human
behaviour. He posits that the transition of mathematical models to social science
from traditional uses in physics and other natural sciences had actually restricted
the progression of archaeology by focusing on these inappropriate assumptions.
Similarly, the traditionally statistically-focused discipline of demography has
begun to embrace new methodologies as the limitations of these methods for certain
types of research questions has become evident (Billari and Prskawetz 2003). The
advent of microsimulation in the demographic community has led to some in-depth
examination of the foundations of demographic knowledge (e.g., Courgeau 2007),
with some suggesting that agent-based modelling could form the foundations of a
new, systems-based demography (Courgeau et al. 2017). The development of these
modelling methods has been met with signiﬁcant enthusiasm, though the marriage
between data-focused demographic statistical modelling and abstract, individualbased
modelling is an uneasy one.
Looking at social sciences more broadly, McKelvey (2001) notes that an
increasing number of social scientists and economists propose that the traditional
perspective of dynamics in these areas focused upon changing equilibria are outdated,
and that the agent-based perspective of interacting autonomous social actors
will provide greater insight into social behaviour. Indeed, Billari and Prskawetz
(2003) argue that demography may only be able to surpass some of the challenges
facing the study of population dynamics by embracing agent-based models. Henrickson
and McKelvey (2002) even propose that an increased focus on agent-based
methodologies could give social science greater legitimacy within the scientiﬁc
community, allowing for a greater degree of experimentation and analysis across
social science as a whole.
5.3 Lars-Erik Cederman and Political Actors as Agents 87
5.2.2 Axelrod’s Complexity of Cooperation
Axelrod’s 1997 book “The Complexity of Cooperation” led the charge for this
increasing number of social scientists looking towards agent-based models as
a method for examining sociological structures. In the same year, Cederman’s
“Emergent Actors in World Politics” provided an intriguing look at potential
applications of such models to the political sciences. In the years that followed,
the early mathematical models of political and social problems began to transfer
to agent-based modelling methodologies: Epstein et al. (2001) used agent-based
models to examine civil violence; Lustick (2002) used similar techniques to study
theories of political identity in populations; Kollman et al. (1997) model the
movements and shifting alliances of voters; and Schreiber (Donald et al. 1999)
modelled the emergence of political parties within a population, to provide just a
few examples. Thus, an increasing number of political scientists seem interested
in modelling the emergence of social structures and institutions from the level of
individuals or small sub-populations; however, the community remains divided as
to the usefulness of such methodologies.
5.3 Lars-Erik Cederman and Political Actors as Agents
5.3.1 Emergent Actors in World Politics a Modelling Manifesto
Cederman’s initial work in his 1997 book-length treatise focused on the simulation
of inter-state interactions. Each model represented nation-states as agents, each
with differing drives that altered their pattern of interaction with other states in
the region. He presents these simulations as a means for building an aggregatelevel
understanding of the function of political structures by understanding the
interactions of micro-level features which produce those effects (although, notably,
he did not begin modelling these interactions using smaller units until a few years
later, after the publication of Emergent Actors).
The excitement apparent in the pages of Emergent Actors seemed infectious,
bringing other social and political scientists into the fold with an increasing number
of agent-based models ﬁnding publication in political science journals. Lustick
(2000) proposed that agent-based models could alleviate some of the shortcomings
of previous modelling approaches:
Difﬁculties of amassing and manipulating collective identity data into theoretically potent
comparisons are among the reasons that agent-based modelling can play an important role
in the elaboration, reﬁnement, and testing of the kind of speciﬁc and logically-connected
theoretical claims that constructivists have been faulted for not producing. Because the
models run on computers there is no room for ambiguity in the speciﬁcation of the model’s
underlying rules.
88 5 Modelling for the Social Sciences
Kenneth Benoit went so far as to argue that ‘because simulations give the
researchers ultimate control, simulations may be far better than experiments in
addition to being cheaper, faster, and easier to replicate’ (Benoit 2001). In a
discipline where solid ﬁeld-collected data is both complex to analyse and expensive
to collect (or even impossible depending on the related political conditions), the
prospect of being able to generate useful data from low-level simulated interactions
seemed quite promising.
5.3.2 Criticism from the Political Science Community
Criticism of agent-based models in political science has come from a number of
different sources, but a large portion of those criticisms focus on the difﬁculty of
making sensible abstractions for social and political structures within such a model.
As described earlier in relation to Alife models, one can potentially view all of
scientiﬁc inquiry as reﬂective of the inherent biases of the experimenter, and this
problem of theory-dependence is even more acute in models requiring the level of
abstraction that political models necessitate.
Of course, even the most abstract of Alife models may reference both the
real-life behaviour of natural biological systems and the wealth of knowledge
obtained from many decades of observation and experimentation in evolutionary
biology. Political science, however, does not have that luxury. Highly complex social
structures and situations, such as Cederman’s models of nationalist insurgency
(Cederman 2002, 2008) involve further layers of abstraction, including factors
which do not immediately lend themselves to quantiﬁcation such as cultural and
national identities.
Evolutionary Alife simulations also beneﬁt from an extensive backdrop of both
theoretical and empirical work on innumerable species, allowing for the basic
functions of evolutionary dynamics within biological systems to be modeled fairly
competently. In contrast, political systems involve multiple layers of interacting
components, each of which is understood primarily as an abstracted entity; frequently
only the end results of political change or transition are easily observable,
and even then the observer will have great difﬁculty pinpointing speciﬁc low-level
effects or drives which may have inﬂuenced those results.
As Klüver et al. (2003) describe, sociological theory may not beneﬁt from the
micro/macro distinction of levels of analysis that beneﬁts researchers of evolution
and other large-scale processes. A micro/macro distinction allows the simulation
researcher to create a hierarchical relation between elements, making for simpler
analysis. The interacting social levels present in a political system however cannot
be so clearly differentiated into a hierarchy of processes, making simulation a
difﬁcult, and highly theory-dependent, exercise. Due to these elements, theorydependence
in social simulation becomes a more acute problem than in ALife;
Sects. 5.6 and 5.7 examine this difﬁculty in detail, and propose some possible
solutions for the social simulation community based upon the systems sociology
approach of Luhmann (1995).
5.4 Cederman’s Model Types: Examples and Analysis 89
5.3.3 Areas of Contention: The Lack of ‘Real’ Data
Donald Sylvan’s review of Cederman’s Emergent Actors in World Politics highlights
another common complaint leveled at agent-based models by conventional
political science:
Moreover, ‘data,’ in the way this term is understood by most statistically-based modelling
procedures, is largely absent from Emergent Actors. This feature is very much in line with
the rational- choice modelling tradition, of which the author is so critical. Many readers
will ﬁnd nothing problematic about this feature in a theoretical work such as this. However,
it is important that readers understand that the lack of ‘data’ is a standard feature of CAS
simulation as they evaluate the ‘results’ reported. (Sylvan 1998, p. 378)
As Sylvan points out, Cederman’s ‘data’ only relates to the interactions of virtual
states in an idealised grid-world; applying such data to real-life political events or
transitions seems suspect. The level of complexity at work in large-scale political
events may be very difﬁcult to capture in an agent-based model, and knowing when
to draw a speciﬁc conclusion from a model of such an inherently difﬁcult-to-analyse
situation is quite difﬁcult.
5.4 Cederman’s Model Types: Examples and Analysis
Despite these objections, Cederman, as evidenced by his extensive book-length
work on agent-based modelling and subsequent methodological papers, sees agentbased
modelling as a promising means of investigation for political scientists
(Cederman 1997, 2001). His attempt to present a framework to describe the various
potential goals of models in this discipline provides an opportunity to contrast this
proposed social simulation approach with the other modelling frameworks analysed
thus far.
5.4.1 Type 1: Behavioural Aspects of Social Systems
Cederman, in describing his three-part categorisation of social simulation (see
Table 5.1), credits Axelrod’s early work on the iterated prisoner’s dilemma as the
ﬁrst major foray into modelling behavioural aspects of social systems (Cederman
2001; Axelrod and Hamilton 1981; Axelrod 1984). Axelrod’s work aimed to show
the emergence of cooperation, and with the iterated prisoner’s dilemma showed
that cooperation is possible in social settings as long as the interactions of involved
agents are iterated.
This variety of work has continued in the years since, beginning with modiﬁcations
to Axelrod’s original model, such as spatially-embedded versions which
show the emergence of cooperative clusters in a social system (Lomborg 1996).
By incorporating more complex elements into the original models, researchers
have attempted to draw further conclusions about the evolution of cooperative
90 5 Modelling for the Social Sciences
Table 5.1 Summary of Cederman’s three modelling types
Cederman’s model classiﬁcation
C1 Focus on behavioural elements of social systems
C2 Focus on emergence of agent conﬁgurations
C3 Focus on emergence of interaction networks between agents
behaviours; such a focus on these aspects of a social system is characteristic of a
Type 1 model (hereafter referred to as C1) under Cederman’s classiﬁcation.
A major beneﬁt of this type of model is computational simplicity. The prisoner’s
dilemma example noted here is a well-known problem in game theory, and has
been implemented and studied countless times over the past few decades. For the
modeller, reproducing such a game computationally is a relatively simple task
compared to more complex models, due to the lack of excessive numbers of
parameters and the inherent compartmentalised nature of the interactions between
players of the game.
To use our bird migration example, imagine that a certain bird species has
demonstrated a social behaviour which can produce greater nest productivity
between cooperating females, but at the expense of more frequent reproduction.
In this case a C1 model might be useful, as a game-theoretic model could be
designed to probe the ramiﬁcations of this behaviour in different situations. While
our researcher would be pleased in one sense, given the greater simplicity of modelconstruction
in this case, the model would also be quite narrow in its focus, and the
abstractions made in such a model would be signiﬁcant.
5.4.2 Type 2: Emerging Conﬁgurations
Cederman identiﬁes Type 2 models (C2) as those which attempt to explain the
emergence of particular conﬁgurations in a model due to properties of the agents
(or ‘actors,’ to use Cederman’s terminology) involved (Cederman 2001). Models of
cultural evolution, ﬁt this description, as they rely upon the interaction and exchange
of agent properties (often identiﬁed as ‘arguments or ‘attitudes’) and examine the
resultant conﬁgurations of agents (March 1991; Axelrod 2001).
Ian Lustick’s Agent-Based Identity Repertoire Model (ABIR) is a suitable example
of a modern C2 model, as it provides agents with potential ‘identities’ relating
to different groups of agents which can be modiﬁed through interactions between
those agents (Lustick 2006). C2 models such as ABIR focus on demonstrating the
emergence of larger conﬁgurations within the social systems they simulate; in this
case, the properties of each agent in the ABIR model have been used to study
the development of clusters of ethnic and religious groups under various social
situations.
These C2 models offer greater complexity for the modeller than C1 models, but
they also offer the possibility of examining another category of questions about
5.4 Cederman’s Model Types: Examples and Analysis 91
social systems. To use our bird migration example, imagine that our researcher
wished to examine the interactions of members of a ﬂock upon arrival at a
destination and reaching a suitable colony site. A C2 model may be a useful
approach in this context, as our researcher could construct a model which assigns
each agent certain properties (such as gender, behaviour mode, and so on) which
could then allow the agents to interact using these properties and delegate roles and
responsibilities in establishing a colony.
Of course, this model is far more complex than the C1 example above, but the
problem in question is also very different. Not all problems are easily broken down
into variations or extensions of highly-studied game-theoretic situations, so in this
case our bird researcher may prefer to construct a novel model in the C2 style which
suits this problem, despite the greater difﬁculties inherent in doing so.
5.4.3 Type 3: Interaction Networks
Cederman classiﬁes Type 3 models (C3) as perhaps being the most ambitious:
this type of system attempts to model both the individual agents themselves and
their interaction networks as emergent features of the simulation (Cederman 2001).
Interestingly, Cederman cites the ﬁeld of artiﬁcial life as one likely to inform this
area of computational work in political science, given that Alife focuses on such
emergent features. He also acknowledges that some overlap can occur between C1
and C3 models by allowing agents more latitude in choosing interaction partners for
example (Cederman 2001; Axelrod 2000).
Cederman argues that C3 models may provide very powerful tools for the
political scientist, allowing for profound conclusions to be drawn regarding the
development of political institutions. This approach does seem the most methodologically
difﬁcult of the three types in this classiﬁcation, however, as the already
signiﬁcant abstractions necessary to create C1 and C2 models must be relaxed even
further to allow for such ambitious examinations of emergent features at multiple
levels.
To once again use our bird migration example, we could imagine any number of
possible ALife-type models which would fall under the C3 categorisation. Our bird
researcher would encounter the same difﬁculties with these models that we have
described in previous chapters. The C3 classiﬁcation as provided by Cederman is
quite broad indeed – presumably due to the relatively recent appearance of this type
of model within political science.
5.4.4 Overlap in Cederman’s Categories
As mentioned above, Cederman acknowledges that there is some overlap between
his C1 and C3 categories (Cederman 2001). However, given the complex nature
92 5 Modelling for the Social Sciences
of social interaction, none of his categories provide a hard distinction that makes
categorisation of simulations obvious in every case.
Cederman points to the possibility of a C1 model straying into C3 territory by
simply allowing its agents more greater choice in choosing interaction partners. In a
sense, C3 seems to be a superset of C1 and C2; a C3 model could provide insight into
similar issues to those examined by a C1 or C2 model. Given that the C3 approach
is naturally more broad, then the border separating either of the other types from C3
becomes more fuzzy.
Thus, the utility of Cederman’s categories is slightly different from the pragmatic
nature of Levins’ modelling dimensions (Levins 1966). Deﬁning the position of
a model along Levins’ dimensions is difﬁcult due to the problems inherent in
specifying the exact meaning of generality, realism and precision, but the framework
as a whole remains useful as a pragmatic guideline for modellers (see Chap. 4
for further discussion). Cederman’s framework is not intended to serve this same
purpose, but does provide a means to classify and discuss models in terms of social
science research questions. For this reason, Cederman’s framework will be useful
to us as we investigate modelling in the social sciences in greater detail in the
remainder of the text.
5.5 Methodological Peculiarities of the Political Sciences
5.5.1 A Lack of Data: Relating Results to the Real World
As Sylvan’s review of Cederman emphasizes, Cederman’s ‘data’ only relates to the
interactions of virtual states in an idealised grid-world; applying such data to reallife
political events or transitions seems suspect at best. The levels of complexity
at work in large-scale political events may be very difﬁcult to capture in an agentbased
model, and knowing when to draw a speciﬁc conclusion from a model of such
an inherently difﬁcult-to-analyse situation is quite difﬁcult.
In essence the lack of real ‘data’ produced by such simulations is an issue critical
to the acceptance of such models in mainstream political science. While some
accept the potential for social simulations to illuminate the emergence of certain
properties of political structures (Epstein et al. 2001; Axelrod 2001), the difﬁculty in
connecting these abstracted simulations to real-world political systems is signiﬁcant.
Weidmann and Gerardin, with their GROWLab simulation toolkit, have attempted
to sidestep these concerns by making their framework compatible with GIS
(geographic information system) data in order to allow ‘calibration with empirical
facts to reach an appropriate level of realism’ (Weidmann and Girardin 2006).
They also emphasize the relational and spatially-embedded aspects of GROWLab
simulations, presumably a nod to the importance of spatial considerations and social
interactions in a real-world political context.
5.6 In Search of a Fundamental Theory of Society 93
5.5.2 A Lack of Hierarchy: Interdependence of Levels
of Analysis
While even abstract Alife models may reference the real-life behaviour of natural
biological systems, and the wealth of related empirical data, political models do
not necessarily have that luxury. Highly complex social structures and situations,
such as Cederman’s models of nationalist insurgency and civil war (Cederman
and Girardin 2005; Cederman 2008) involve further layers of abstraction, often
involving factors which do not immediately lend themselves to quantiﬁcation, such
as cultural and national identities.
In addition, sociological theory is notoriously difﬁcult to formalise, incorporating
as it does a number of both higher- and lower-level cognitive and behavioural
interactions. In fact, sociological theory may not beneﬁt from the micro/macro
distinction of levels of analysis that beneﬁts researchers of evolution and other largescale
processes (Klüver et al. 2003). These interacting social levels cannot be clearly
differentiated into a hierarchy of processes, making simulation a very difﬁcult, and
highly theory-dependent, exercise.
5.5.3 A Lack of Clarity: Problematic Theories
Doran (2000) identiﬁes a number of problems facing social scientists who wish
to ‘validate’ their simulation work. He maintains that social scientists need not
provide ‘speciﬁc validation,’ a direct connection to a target system, but instead face
a more nebulous difﬁculty in demonstrating relevance of the assumptions within that
simulation to social systems at large. He notes the immediate difﬁculties of ﬁnding
an appropriate parameter space and method for searching that space, of analysing
the simulation results in a way that does not produce an ‘intractable level of detail,’
and the problem of instability in simulations and the necessity of detailed sensitivity
analyses. In his conclusion he argues convincingly for sensible constraints in social
simulations which do not add confounding cultural biases to the behaviour of agents
within the simulation. While Doran’s examples provide a useful illustration of this
concept, simulation architects may ﬁnd great difﬁculty in ensuring that such biases
are absent from their work, particularly in more complex multi-agent simulations.
5.6 In Search of a Fundamental Theory of Society
5.6.1 The Need for a Fundamental Theory
As we have seen, social science presents a few particularly thorny methodological
problems for the social simulator. Despite this, can social simulation be used to
94 5 Modelling for the Social Sciences
illuminate the underlying factors which lead to the development and evolution of
human society? Social simulation certainly presents a new approach for examining
societal structures, and perhaps could serve as a novel method for testing hypotheses
about the very origins of society itself.
However, using social simulation for this purpose seems fraught with difﬁculties.
Human society is an incredibly complex system, consisting as it does of billions
of individuals, each making hundreds of individual decisions and participating in
numerous interactions every day. There are vast numbers of factors at play and many
of them are inherently unanalysable given that we cannot examine the contents of
a human’s brain during a decision or interaction which appears to make the already
monumental task of the social simulator nearly impossible in such a case.
The problem of how life has evolved on Earth would seem also to be one of
insurmountable complexity as well if it weren’t for the theories of Charles Darwin
(1859). Perhaps there is hope for a theory in social science of similar explanatory
power to evolution, not necessarily one that fully explains society, but instead
provides us with a holistic framework to push forward our understanding of society
– in a similar way that evolution does for biology.
5.6.2 Modelling the Fundamentals
While Cederman describes a broad framework in which social simulation can
operate, a fundamental social theory seems difﬁcult to develop under his description
of C1, C2, and C3 models (Cederman 2001). C3 models are designed to allow for
the development of broad-stroke models which can illuminate more fundamental
theories about political systems; however, to extend that to societal structures and
interactions as a whole requires a new level of abstraction.
In essence an extension of the C3 categorisation becomes necessary when
seeking a fundamental theory. In the context of political science, agents would
be operating under a framework developed from that particular ﬁeld of social
science; while political decisions amongst agents will by their nature require
the incorporation of elements of psychology and sociology, a more fundamental
approach requires an even more abstract method for allowing all varieties of social
behaviour to emerge from the simulated system.
As part of this new approach, we also need a new perspective on the development
of human society. How do individual actors grow to communicate? How does that
communication then become structured? How do these structured communications
then grow into a societal-level framework guiding interactions between members of
a population? To see the fundamentals of human society develop, the model would
need to set a stage upon which society may grow without a pre-set communicative
framework already in place.
5.7 Systems Sociology: A New Approach for Social Simulation? 95
5.7 Systems Sociology: A New Approach for Social
Simulation?
As we have seen, the advent of social simulation has proved inﬂuential in the
social sciences, provoking new questions regarding the origin and nature of society.
While the examples discussed thus far demonstrate the potential impact of social
simulation, they also illustrate the inherent difﬁculties involved in generalising the
conclusions drawn from a social simulation. More generalised models of society
may provide a means for investigating aspects of society which elude the empirical
data-collector and in turn inform our search for a fundamental social theory, but in
order for this to occur we need to establish a method of examining society on a
broad theoretical scale through simulation.
5.7.1 Niklas Luhmann and Social Systems
The well-known social systems theory of Niklas Luhmann provides one example
of an attempt to develop an understanding of the foundations for social behavior.
Luhmann classiﬁes social systems as systems of communication which attempt to
reduce complexity by presenting only a fraction of the total available information
(Luhmann 1995).
One of the fundamental issues facing the systems sociology theorist is solving
the problem of double contingency, an issue Luhmann describes as central to the
development of social order. Put simply, if two entities meet, how do they decide
how to behave without a pre-existing social order to govern their actions? How
might these entities decide to develop a common means of interaction, and through
those interactions develop a shared social history?
As Dittrich, Kron and Banzhaf describe, Luhmann described a method for
resolving this contingency problem which was far more elemental than previous
approaches, relying as it does on ‘self-organization processes in the dimension of
time’ rather than through more standard social processes. The entities in question
would perform initial contingency-reducing actions during an encounter to allow for
each to develop an understanding of the expectations of each party in the interaction
(Dittrich et al. 2003).
In Luhman’s view, the social order develops as a consequence of these
contingency-reducing actions on a large scale. As elements of the developing
society develop their expectations about the social expectations of others (described
as ‘expectation-expectation’ by Luhmann), a system of social interaction develops
around this mutual social history. This system then produces as a consequence the
social institutions which can further inﬂuence the development of the social order.
These social institutions perform a similar function by reducing the amount of
information disseminated amongst the members of a society, essentially providing
contingency-reducing services on a much larger scale.
96 5 Modelling for the Social Sciences
Agent-based models in the context of artiﬁcial life have certainly proved useful in
the examination of other autopoietic systems; however, recent attempts to formalize
Luhman’s theories into a usable model, while producing interesting results, have
highlighted the inherent difﬁculties of encapsulating the many disparate elements of
Luhman’s theories of social systems into a single model (Fleischmann 2005).
5.7.2 Systems Sociology vs. Social Simulation
As we can see from Luhman’s analysis, while there may indeed be a lack of ‘data’
inherent to the study of artiﬁcial societies, there still exists a theoretical framework
for understanding the fundamental mechanisms which drive the creation of a larger
social order. While some social simulation researchers may seek to strengthen their
models through establishing direct connections with empirically-collected data from
social science, the systems sociology perspective could provide a different path to
more useful examinations of human society.
The social simulation stream is oriented towards speciﬁc elements of social
behaviour; simulations of cooperation (Axelrod 1997), nationalist insurgency (Cederman
1997), or the spatial patterning of individuals or opinions within a society
(Lustick 2006). Social simulation’s stronger links with empirical data may make
validation of such models much easer, but further restricts the domain of those models
to focus on social problems for which usable data exists. Given the difﬁculties
inherent in collecting social science data, these problems tend to be a subset of those
social problems for which models could prove potentially illuminating.
This very restriction into particular domains prevents the social simulation
approach from reaching a more general perspective; this approach is constrained
by approaching social phenomena from the top-down. These top-down approaches
are necessarily rooted in the societies they model. In essence, looking for a feature
in society and then attempting to reproduce it in a model is not sufﬁcient to develop
a fundamental theory.
In contrast, the systems sociology stream abstracts outside of the standard view
of society. Luhmann’s perspective aims to describe interactions which can lead
to the development of social order, in a sense examining the development of
human society through an ‘outside perspective.’ Luhmann essentially moves beyond
standard sociology, attempting to describe what occurs prior to the existence of
social order, rather than operating within those bounds as with social simulation.
Returning for a moment to our bird migration example, imagine that our
migration researcher wishes to construct a model to investigate the beginnings of
migration behaviour. One obvious approach may be to model individual agents,
each of which is given the choice of moving to follow changing resources or
environmental conditions. However, in a Luhmannian context, we could remove
that element of pre-existing ideas concerning the origins of migration. A model
which features individual agents which can move of their own accord, and have
5.8 Promises and Pitfalls of the Systems Sociology Approach 97
basic requirements for survival in the simulated environment, may provide differing
explanations of migration behaviour if that behaviour is seen to emerge from this
very basic scenario. In this way, we allow for the possibility of other means for
migration to emerge: perhaps through a developing social structure which drives
movement of groups of birds, for example. In essence we would seek to move the
migration model to a stage in which we assume as little as possible about the origins
of migration behaviour, rather than assuming that certain factors will produce that
effect.
Similarly, by viewing society from its earliest beginnings prior to the existence
of any societally-deﬁned modes of interaction and communication, the systems
sociology approach hopes to develop a theoretical understanding of the fundamental
behavioural characteristics which lead to the formation of social order. In many
ways this approach is reminiscent of the Alife approach to modelling ‘life-asit-could-be’
(Langton et al. 1989); the systems sociology perspective leads us to
examine society-as-it-could-be.
5.8 Promises and Pitfalls of the Systems Sociology Approach
5.8.1 Digital Societies?
Having established this relatively promising outlook on the future prospects of
social-science simulation using Luhmann’s approach, a certain resemblance to the
early philosophy of artiﬁcial life becomes apparent. As in Alife, we may have simply
replaced one troubling set of methodological and philosophical concerns with
another. Strong Alife’s contention that computer simulations can be repositories
for real, digital life provides an escape route for theorists to develop a suitable
theoretical backstory for Alife. As discussed in Chap. 3, such a backstory can
underwrite these computer simulations as a new method for gathering empirical
data, a means for examining processes like evolution in a method that is otherwise
completely impractical. As long as we maintain that a computer simulation can
potentially produce life, then our experiments on that digital biosphere can proceed
apace. However, such a backstory for this ‘artiﬁcial society’ approach to social
science seems a great deal more tenuous. Potentially, we could harken back to
Silverman and Bullock’s Physical Symbol System Hypothesis for Alife (Silverman
and Bullock 2004):
1. An information ecology provides the necessary and sufﬁcient conditions for life.
2. A suitably-programmed computer is an example of an information ecology.
Then, if we further argue that society is a property of living beings, we may
contend that such an information ecology would also provide the necessary and
sufﬁcient conditions for the development of a society.
98 5 Modelling for the Social Sciences
5.8.2 Rejecting the PSS Hypothesis for Society
Ignoring for a moment the philosophically and ethically troubling nature of the
potential theoretical backstory outlined above, those who might ﬁnd such an
account appealing will be forced once again to face the artiﬁcial-world problem.
Additionally, the vastly increased complexity of a population of organisms capable
of developing a society would also increase the troubling aspects of this artiﬁcialworld
approach, creating ever more complex artiﬁcial societies that are increasingly
removed from real-world societies.
As discussed in Chap. 4, the greatest difﬁculty with developing an artiﬁcial world
in which to study such complex systems is the problem of connecting that artiﬁcial
world to the natural one on which it is based. The Strong Alife community may
argue that probing the boundaries of what constitutes life in a virtual world is
inherently a valuable pursuit, allowing for the creation of a new ﬁeld of digital
biology.
For the social scientist, however, the possibility of creating a further ﬁeld of
‘digital sociology’ is less than appealing. In a ﬁeld where empirical data in relation
to the natural world is far more lacking than in biology, and in which simulation
seems to be viewed as a means for enhancing the ability of social scientists to
produce and test sensible theories, then producing and testing those theories in
relation to a virtual society without direct connection to real society is quite a
wasteful pursuit. Indeed, Burch (2002) contends that computer simulations in social
science will revolutionise the ﬁeld by embracing this theoretical complexity and
tying it directly to empirically-relevant questions.
With the appeal of Luhmann’s approach deriving from the potential for examining
the earliest roots of societal development, and from that developing a fundamental
theory of society analogous to evolution in biology, a theoretical backstory along
the lines of strong Alife seems inappropriate. Instead, the Luhmann-inﬂuenced
social simulator would strive for a theoretical framework which emphasizes the
potential role for simulation as a means for social explanation and theory-building,
rather than allowing for the creation of digital forms of society.
5.9 Social Explanation and Social Simulation
The problem of explanation in social science, as in most scientiﬁc endeavours, is
a difﬁcult one. In a ﬁeld with such various methods of data-gathering, prediction,
and theory-construction, developing a method for providing social explanation is no
mean feat.
Proponents of social simulation regard the agent-based simulation methodology
as one potential method for providing an explanation of social phenomena. Before
we establish the veracity of this opinion, however, we must establish the ground
rules for our desired form of social explanation. With social systems involving
5.9 Social Explanation and Social Simulation 99
potentially many millions of participants, we must determine how we will focus
our social explanations to derive the greatest possible theoretical understanding of
the processes underlying human society.
5.9.1 Sawyer’s Analysis of Social Explanation
As noted by R. Keith Sawyer, ‘causal mechanistic accounts of scientiﬁc explanation
can be epistemically demanding’ (Sawyer 2004). A causal mechanistic explanation
of a social system would require a detailed analysis of the large-scale elements of the
system, and their related elements, but also of the individual actions and interactions
of every member of that society.
Of course, this explanation may still be insufﬁcient. On a macroscopic scale,
the behaviour of a human society may be identical despite signiﬁcant variations
in the microscopic actions of individual members of that society. In that case, the
causal mechanist account fails to encompass the larger-scale elements of a social
explanation which could describe these effects.
Oddly enough, this description echoes that of many current agent-based models.
Most of the models discussed thus far in this chapter have displayed a clear
mechanistic bent; agents interact in ways reminiscent of societal actors and produce
complex dynamical behaviour, but there is little to no incorporation of larger-scale
structures such as social institutions. Therefore such models are not only causal
mechanistic accounts, but they are also methodologically individualist (Sawyer
2004; Conte et al. 2001).
With this in mind, Sawyer implies that the current state-of-play in social
simulation is incapable of providing true social explanation. He states that ‘an
accurate simulation of a social system that contains multiply-realised macro-social
properties would have to represent not only individuals in interaction, but also these
higher-level system properties and entities’ (Sawyer 2003, 2004).
5.9.2 Non-reductive Individualism
Revisiting Kluver and Stoica for a moment, we recall their concerns regarding
agent-based models and the difﬁculty of capturing the multi-leveled complexity of
social phenomena within such a structure (Klüver et al. 2003). They argue that social
phenomena do not adhere to a strictly hierarchical structure in which micro-scale
properties result in macro-scale behaviour; in fact, these levels are intertwined, and
separating social systems into that sort of structure as is common in agent-based
models may be extremely difﬁcult.
Sawyer’s take on social explanation and social simulation expands on this topic,
describing the difﬁculty of applying the concept of emergence (familiar to us from
artiﬁcial life) to the social sciences. While the idea that lower-level simplicity leads
100 5 Modelling for the Social Sciences
to higher-level complexity seems intuitively appealing within Alife and other related
ﬁelds, Sawyer presents the idea that in fact social systems do not necessarily obey
this relation (Sawyer 2003, 2004).
In this view, there is a fundamental conﬂict between emergence and the social
sciences. If social systems can display properties are ‘irreducibly complex,’ then
those properties cannot be the result of individual actions or properties, and thus
could not have emerged from those actions or properties. This is clearly quite a
dangerous possibility for the social simulator, as then the causal mechanistic method
of simulating low-level agent interaction to produce high-level complexity would be
a fundamentally ﬂawed approach.
In order to escape this potentially troubling theoretical conﬂict, Sawyer proposes
his own version of emergence for the social sciences which he dubs non-reductive
individualism (see Sawyer 2002, 2003, 2004). In this view, Sawyer concedes to
individualists that their fundamental assumptions about the roles of individuals in
society are correct (i.e., that all social groups are composed of individuals, and
those groups cannot exist without the participation of individuals). However, he also
contends that some social properties are not inherently reducible to the properties
of individuals; in this case, there is reason to present new ideas and theories which
treat the properties of social groups or collectives as a separate entity.
Returning to our bird example, imagine a model which addresses the social
behaviour of a bird species, perhaps the development of birdsong for use in
signalling between individuals or something similar. We could choose to model
these social developments using an agent-based model, which Sawyer would not
ﬁnd objectionable; after all, the actions of individuals do drive society in Sawyer’s
perspective as much as for any other social scientist. We then choose to model the
development of these birdsong behaviours by allowing these agents to signal one
another before performing an action, then observing if those signals begin to ﬁnd
use amongst the simulated population in different circumstances.
However, Sawyer might argue that our model would be insufﬁcient to ever
display the richness inherent in bird social behaviour; the development and spread
of new songs, the separation of songs into differing contexts among different social
groupings and other such factors may be difﬁcult to capture in a simple agentbased
model. As with human society, he might argue that the complexity and
variation of birdsong development requires another layer of simulation beyond
the simple agent-based interactions underlying the drive to communicate between
agents. Perhaps vindicating this perspective, some modelling work has shown that
birdsong grammars and their development can be represented as evolving ﬁnite-state
automata (Sasahara and Ikegami 2004).
Sawyer names this perspective quite appropriately, drawing as it does upon
the philosophy of mind perspective of non-reductive materialism. Non-reductive
materialists accept the primacy of matter in driving the brain, and thus mental
phenomena, but reject the idea that only low-level discourse regarding this brain
matter is valid in the context of studying the mind. In other words, non-reductive
materialists are not dualists, but argue that mental phenomena are worthy of study
5.9 Social Explanation and Social Simulation 101
despite the primacy of matter; or, in Sawyer’s words, ‘the science of mind is
autonomous from the science of neurons’ (Sawyer 2002).
Thus, in the case of social science, Sawyer essentially argues that the science of
society is autonomous from the science of individuals. This leads to his contention
that individualist agent-based models are insufﬁcient to provide social explanation.
Without incorporating both individual effects and irreducible societal effects, the
model would not provide the complete picture of societal complexity, as described
in our example above.
Interestingly, Sawyer does leave one potential door open for individualist
modellers. As he admits, one cannot be certain whether a given social property
can be given an individualist mechanistic explanation, or whether that property will
be proven irreducible to such explanations (Sawyer 2004); presumably, individualbased
models could be used to ﬁll that gap. A suitably rich model of interacting
individuals could possibly provide a testing ground to determine whether a certain
system does display properties independent of individual properties. However,
under this view those simulations would only be able to provide explanation in such
limited cases, and given that not all social systems will display such reducibility to
individual properties, simulating every possible social construct to ﬁnd such systems
is presumably a rather inefﬁcient way to utilise agent-based models in social science.
5.9.3 Macy and Miller’s View of Explanation
While Sawyer points out some potentially troubling methodological difﬁculties for
social simulators, and also proposes entirely new simulation methods to circumvent
those difﬁculties, he does still maintain that social simulation provides a means
for social explanation when implemented appropriately (Sawyer 2004). Macy and
Miller also argue that simulation provides a remarkably useful tool for social
science, and lament the lack of enthusiasm for agent-based models within the social
science community (Macy and Willer 2002).
Macy and Miller propose that agent-based models can provide a perspective
particularly well-suited to sociology, arguing that this methodology ‘bridges
Schumpeter’s (1909) methodological individualism and Durkheim’s rules of a nonreductionist
method’ (Macy and Willer 2002, p. 7). Thus, agent-based models can
produce groups of agents which produce novel and complex higher-level behaviour,
and in this respect reﬂect a potentially appealing method of investigation for the
social scientist seeking to understand the origin of certain societal properties.
However, Macy and Miller join Sawyer in his caution regarding the application of
such bottom-up methods to all social phenomena. Stepping away from the excitement
of the artiﬁcial life theorists, they admit that these methods are not always
inherently useful. In fact, they constrain the application of individualist models
to ‘studying processes that lack central coordination, including the emergence of
institutions that, once established, impose order from the top down’ (Macy and
Willer 2002, p. 8).
102 5 Modelling for the Social Sciences
5.9.4 Alife and Strong Emergence
Sawyer and Macy and Miller’s points are more than reminiscent of the debate
over strong emergence discussed in Chap. 2. In the Alife context, strong emergence
contends that emergent phenomenon can show downward causation, inﬂuencing the
behaviour of its own components. Just as Sawyer discusses, this would result in a
situation in which that strongly emergent behaviour cannot be reduced to the actions
of those component parts (O’Conner 1994; Nagel 1961).
Bedau attempted to get around this restriction by proposing weak emergence, in
which the macro-components of a system, allowed to run in simulation, can demonstrate
the general properties of a weakly-emergent phenomenon (Bedau 1997). As
noted in the previous discussion, however, Bedau’s categorisation requires a certain
circularity of reasoning; Bedau himself states that only empirical observations at the
macro-level of a given system can allow us to develop a means to investigate them
through simulation.
In essence, Bedau alters the problem slightly by contending that a simulation can
provide a general understanding of the properties of a system’s macro-behaviour, but
Sawyer would argue that such an explanation is still incomplete. Bedau’s method,
after all, does not actually propose a means for the simulation researcher to avoid
the difﬁculties caused by downward causation posited by the strong emergence
theorists. Macy and Miller, despite being more positive about simulation than
Sawyer, argue that this very difﬁculty fundamentally limits the utility of simulation
of this type. Without a means for capturing this downward causative inﬂuence
of macro-level social institutions and similar structures, they contend that more
traditional empirical methods would remain more useful in some cases.
Thus, for the simulation researcher who wishes to illuminate some potential
inﬂuencing low-level factors in a given social system, the issues of non-reductive
individualism or strong emergence do not make an enormous difference. Even if
such objections are true, the researcher can still produce results which are indicative
of the importance of those low-level factors in the emergence of the high-level
behaviours under investigation, particularly with the assistance of an appropriate
theoretical backstory. However, for the researcher wishing to use simulation as an
explanation, and thus as a means for generating more powerful social theory, such
objections create more difﬁculty.
5.9.5 Synthesis
Macy and Miller go on to identify two main streams in social simulation: studying
the self-organisation of social structure and studying the emergence of social order
(Macy and Willer 2002). The second stream is quite relevant to our earlier discussion
of the implications of Luhmann’s theories regarding social order to the agent-based
modelling methodology.
5.10 Summary and Conclusion 103
As described in our discussion of Luhmann, a central difﬁculty in social simulation
is the issue of heavily-constrained agent interaction. While agent interactions
can provide an explanation of social behaviours at the individual level, those
interactions may be far too limited to provide a useful explanation of larger-scale
social structures. Sawyer and Macy and Miller provide an afﬁrmation of this idea,
arguing that most agent-based models are inherently individualist and thus limited
in their ability to explain many social structures.
As we describe, the theories of Luhmann provide an intriguing means for
studying the emergence of social order at the most fundamental level. However,
while these models might provide insight into the earliest beginnings of certain
societal interactions and structures, if we believe Sawyer and Macy and Miller then
we would still be lacking critical elements of a complete social explanation.
5.10 Summary and Conclusion
Having taken a tour through the issues of social explanation that bear upon our
proposed uses of agent-based models in the social sciences, we have a more
complete picture of the possible empirical niche that such models may ﬁll within this
ﬁeld. In particular our look at the explanatory deﬁciencies inherent to agent-based
models draws us toward some speciﬁc conclusions regarding their most promising
uses.
First, as indicated by our analysis of Luhmann’s theories, we see that agentbased
models suffer from some inherent constraints due to their status as artefacts
of a society themselves. Given that models constructed based upon our own
understanding of societal structure to date will naturally have certain fundamental
assumptions about the operating parameters of a society, using such models to draw
conclusions about a possible fundamental theory of society is fraught with potential
difﬁculties.
In addition, most agent-based models seen thus far in this analysis have been individualist
constructions which seek a mechanistic explanation for societal properties.
As Sawyer and others have shown, individualist models may lack another essential
portion of information needed to produce a full social explanation. If we accept the
non-reductive individualist contention that some social groups or collectives may
have non-trivial behaviour that cannot be reduced to the actions of individuals, then
we would be unable to model such social constructions using conventional agentbased
modelling techniques.
Thus, our analysis points toward a synthesis between Luhmann-style modelling
of fundamentals combined with top-down elements. However, these elements seem
rather disparate. Is it possible to combine models of the earliest beginnings of social
interaction together with the inﬂuence of established top-down social structures?
Does not the Luhmann view by its very nature preclude the inclusion of such preset
structures, ﬁlled as they are by tacit assumptions regarding the functioning of society
and its structures?
104 5 Modelling for the Social Sciences
Clearly these two views would not mesh particularly well within a single model,
but a combination of these two approaches when looking at different aspects of society
may contribute to the development of the fundamental theory of human society
that we seek. The next chapter will examine the current and future state of social
simulation in relation to the theoretical frameworks elucidated in our analysis thus
far. These comparisons will give us a more nuanced view of the issues facing agentbased
modelling, with the addition of the social science perspective providing some
new considerations. With these elements in mind, and with a view toward the issues
of social explanation discussed in this chapter, we shall begin a more complete
synthesis of theoretical frameworks that may drive future work in social simulation.
References
Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books.
Axelrod, R. (1997). Advancing the art of simulation in the social sciences. In R. Conte, R.
Hegselmann, & P. Terna (Eds.), Simulating social phenomena (Lecture notes in economics
and mathematical systems, Vol. 456, pp. 21–40). Berlin: Springer.
Axelrod, R. (2000). Harnessing complexity: Organizational implications of a scientiﬁc frontier.
New York: Basic Books.
Axelrod, R. (2001). Evolution of cooperation without reciprocity. Nature, 414, 441–443.
Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation. Science, 211, 1390–1396.
Bedau, M. A. (1997). Weak emergence. In J. Tomberlin (Ed.), Philosophical perspectives: mind,
causation, and world (Vol. 11, pp. 375–399). Oxford: Blackwell.
Benoit, K. (2001). Simulation methodologies for political scientists. The Political Methodologist,
10(1), 12–16.
Billari, F., & Prskawetz, A. (2003). Agent-based computational demography: Using simulation to
improve our understanding of demographic behaviour. Heidelberg: Physica Verlag.
Burch, T. K. (2002). Computer modelling of theory: Explanation for the 21st century. In R. Franck
(Ed.) The explanatory power of models (pp. 245–265). Amsterdam: Kluwer Academic.
Cederman, L. E. (1997). Emergent actors in world politics: How states and nations develop and
dissolve. Princeton: Princeton University Press.
Cederman, L. E. (2001). Agent-based modeling in political science. The Political Methodologist,
10, 16–22.
Cederman, L. E. (2002). Endogenizing geopolitical boundaries with agent-based modeling.
Proceedings of the National Academy of Sciences, 99(suppl 3), 7296–7303.
Cederman, L.-E. (2008). Articulating the geo-cultural logic of nationalist insurgency. In Kalyvas,
S. N., Shapiro, I., & Masoud, T. (Eds.), Order, conﬂict, and violence (pp. 242–270). New York:
Cambridge University Press.
Cederman, L. E., & Girardin, L. (2005). Beyond fractionalization: Mapping ethnicity onto nationalist
insurgencies. In Paper Read at Disaggregating the Study of Civil War and Transnational
Violence Conference, San Diego.
Conte, R., Edmonds, B., Moss, S., & Sawyer, K. (2001). The relevance of the social sciences
for multi-agent-based simulation. Computational and Mathematical Organization Theory, 7,
183–205.
Courgeau, D. (2007). Multilevel synthesis: From the group to the individual. Dordrecht: Springer.
Courgeau, D., Bijak, J., Franck, R., & Silverman, E. (2017). Model-based demography: Towards
a research agenda. In A. Grow & J. Van Bavel (Eds.), Agent-based modelling and population
studies (Springer series on demographic methods and population analysis, pp. 29–51).
Berlin/Heidelberg: Springer.
References 105
Darwin, C. (1859). The origin of species by means of natural selection. London: John Murray.
Dittrich, P., Kron, T., & Banzhaf, W. (2003). On the scalability of social order: Modeling the
problem of double and multi contingency following Luhmann. Journal of Artiﬁcial Societies
and Social Simulation, 6(1), 3.
Donald, D., Abell, J., Andreou, N., & Schreiber, R. (1999). The new design: The changing role
of industrial engineers in the design process through the use of simulation. In D. Sturrock,
P. Farrington, H. Nembhard, G. Evans (Eds.), Proceedings of the 1999 Winter Simulation
Conference (pp. 61–67). Piscataway: IEEE.
Doran, J. E. (2000). Trajectories to complexity in artiﬁcial societies. In A. Kohler & G. Gumerman
(Eds.), Dynamics in Human and Primate Societies. New York: Oxford University Press.
Epstein, J. M., Steinbruner, J. D., & Parker, M. T. (2001). Modeling civil violence: An agent-based
computational approach. CSED Working Paper No. 20.
Fleischmann, A. (2005). A model for a simple Luhmann economy. Journal of Artiﬁcial Societies
and Social Simulation, 8(2), 4.
Henrickson, L., & McKelvey, B. (2002). Foundations of a ‘new’ social science: Institutional
legitimacy from philosophy, complexity science, postmodernism, and agent-based modeling.
Proceedings of the National Academy of Sciences, 99, 7288–7295.
Klüver, J., Stoica, C., & Schmidt, J. (2003). Formal models, social theory and computer
simulations: Some methodical reﬂections. Journal of Artiﬁcial Societies and Social Simulation,
6(2): 8.
Kollman, K., Miller, J. H., & Page, S. E. (1997). Political institutions and sorting in a Tiebout
model. The American Economic Review. Available at SSRN: https://ssrn.com/abstract=75548.
Langton, C. G., et al. (1989). Artiﬁcial life. Redwood City: Addison-Wesley.
Levins, R. (1966). The strategy of model-building in population biology. American Scientist, 54,
421–431.
Lomborg, B. (1996). Nucleus and shield: The evolution of social structure in the iterated prisoner’s
dilemma. American Sociological Review, 61, 278–307.
Luhmann, N. (1995). Social systems. Stanford: Stanford University Press.
Lustick, I. (2000). Agent-based modelling of collective identity: Testing constructivist theory.
Journal of Artiﬁcial Societies and Social Simulation, 3(1), 1.
Lustick, I. (2002). PS-I: A user-friendly agent-based modeling platform for testing theories of
political identity and political stability. Journal of Artiﬁcial Societies and Social Simulation,
5(3), 7.
Lustick, I. (2006). Deﬁning violence: A plausibility probe using agent-based modeling. In Paper
Prepared for LiCEP. Princeton University, 12–14 May 2006.
Macy, M. W., & Willer, R. (2002). From factors to actors: Computational sociology and agentbased
modeling. Annual Review of Sociology, 28, 143–166.
March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science,
2(1), 71–87.
McKelvey, B. (2001). Energizing order-creating networks of distributed intelligence. International
Journal of Innovation Management, 5, 181–212.
Nagel, E. (1961). The structure of science. Harcourt: Brace and Company.
O’Conner, T. (1994). Emergent properties. American Philosophical Quarterly, 31, 91–104.
Read, D. W. (1990). In A. Voorrips (Ed.) Mathematics and information science in archaeology
(pp. 29–60). Holos: Bonn.
Sasahara, K., & Ikegami, T. (2004). Song grammars as complex sexual displays. In Artiﬁcial life
IX. Cambridge: MIT Press.
Sawyer, R. K. (2002). Nonreductive individualism: Part I: Supervenience and wild disjunction.
Philosophy of the Social Sciences, 32, 537–561.
Sawyer, R. K. (2003). Artiﬁcial societies: Multi-agent systems and the micro-macro link in
sociological theory. Sociological Methods and Research, 31(3), 37–75.
Sawyer, R. K. (2004). Social explanation and computational simulation. Philosophical Explorations,
7(3), 219–231.
106 5 Modelling for the Social Sciences
Schumpeter, J. (1909). On the concept of social value. The Quarterly Journal of Economics, 23(2),
213–232. https://doi.org/10.2307/1882798.
Silverman, E., & Bullock, S. (2004). Empiricism and artiﬁcial life. In J. Pollack, M. Bedau, P.
Husbands, T. Ikegami, & R. A. Watson (Eds.), Artiﬁcial Life IX: Proceedings of the Ninth
International Conference on the Simulation and Synthesis of Living Systems. Cambridge: MIT
Press/Bradford Books.
Sylvan, D. (1998). Modeling the rise and fall of states, review of Emergent Actors in World Politics
by Lars-Erik Cederman. Mershon International Studies Review, 42, 377–379.
Weidmann, N. B., Girardin, L. (2006). GROWLab: A toolbox for social simulation. Paper
presented at the First World Congress on Social Simulation, Kyoto, 21–25 Aug 2006.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
Chapter 6
Analysis: Frameworks and Theories for Social
Simulation
6.1 Overview
The previous chapter focused upon the current methodologies and theoretical
implications of agent-based modelling in the social sciences. While many within
this growing ﬁeld accept that agent-based models provide a potentially powerful
new method for examining social behaviours and structures, a great debate still
continues over the best methods for utilising the strengths of this methodology.
As seen in Part I, such debates are not unique to social simulation. Indeed,
artiﬁcial life and more conventional forms of biological modelling have faced
similar challenges over the past few decades. With this in mind, this chapter begins
by placing artiﬁcial life within the various theoretical frameworks discussed thus
far in Chaps. 3, 4 and 5. In this way the limitations of each framework can be
illuminated.
Social science simulation using agent-based models shares a number of constraints
and methodological difﬁculties with biological modelling using the same
methodology. Thus, having placed artiﬁcial life and biological models within a theoretical
framework, social simulation will be subjected to a similar analysis. Finding
the most appropriate framework for social simulation will lay the groundwork for
Chap. 7, in which one of the more prominent exemplars of social simulation will be
subjected to theoretical and methodological analysis. Chapter 7 lays the foundation
for the conclusions of Part II, in which our analysis of Schelling’s residential
segregation model will provide a means to demonstrate the most important elements
of a useful modelling framework for the social sciences.
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_6
107
108 6 Analysis: Frameworks and Theories for Social Simulation
6.2 Frameworks and ALife: Strong ALife
6.2.1 Strong ALife and the Lack of ‘Real’ Data
As noted by critics of artiﬁcial life, social simulation and related methodologies,
computational simulations suffer from a perceived lack of ‘real’ data, or data derived
from experimental observation. Part of this inherent difﬁculty stems from the need
for abstraction in many such models; for example, connectionist models of cognitive
processes embrace the idea of ‘distributed representation’ and their potential role
in cognition, while generally avoiding integrating those models into larger, more
complex neural structures as seen in the brain (Rumelhart and McClelland 1986).
Strong ALife simulations suffer even more strongly from this shortcoming.
Ray’s Tierra provides an enticing look at a ‘digital ecology’ composed of evolving
computer programmes competing for memory space (Ray 1996), but those creatures
are purely artiﬁcial constructions. While the parasites and hyper-parasites which
eventually evolve in Tierra’s world may provide an analogue to real-life parasitic
behaviour, the specialised nature of their virtual world is such that analyses of
Tierra would be very difﬁcult to apply to the natural world. Ray might argue that
his open-ended evolutionary system, which lacks the standard ﬁtness function of
many genetic algorithms and instead provides selection only through life and death,
evokes real-world interactions in evolving systems. Does the difﬁculty of verifying
such claims conﬁne analyses of Tierra to the realm of mathematical curiosity?
6.2.2 Artiﬁcial1
vs Artiﬁcial2
: Avoiding the Distinction
As noted in Silverman and Bullock’s description of differing varieties of artiﬁciality
in science (Silverman and Bullock 2004), providing a distinction between manmade
instances of natural systems and man-made facsimiles of natural systems is
important to understanding the goals of a simulation. In the case of strong ALife,
researchers aim to produce models that embody Artiﬁcial1
, or a man-made instance
of something natural; these models claim to create digital life, rather than simply
resemble biological life. In this case, the strong ALife researcher falls into the role
of a sort of digital ecologist, studying the behaviour and function of real, albeit
digital, organisms. Of course, such claims seem remarkable, but in the absence of a
complete and veriﬁable deﬁnition of life such claims are difﬁcult to refute.
6.2.3 Information Ecologies: The Importance of Back-stories
The inclination of the strong ALife researcher to study ‘real’ digital organisms
points to the importance of formulating a theoretical backstory for any given
6.3 Frameworks and ALife: Weak ALife 109
simulation model. As per Silverman and Bullock’s PSS Hypothesis for Life,
presuming that:
1) An information ecology provides the necessary and sufﬁcient conditions for life.
2) A suitably-programmed computer is an example of an information ecology.
. . . then the strong ALife researcher may claim, under such a framework, that
their simulation represents an information ecology, and thus is a digital instantiation
of a biological system. Whether or not the low-level functions of that system match
those of real-life, carbon-based biological systems is immaterial; the only criterion
is the presence of an ecology of information in which genetic material competes for
representation, and under this criterion the position stated here is justiﬁable.
To use our central example, if we construct a bird migration model in Alife
fashion using individual interacting agents, we may wish to demonstrate that we
are in fact performing some form of empirical data collection, rather than simply
investigating an interesting mathematical system. So, we claim in this case that are
simulated birds do in fact demonstrate an information ecology; perhaps our agents
evolve and reproduce, this producing a dynamic of information amongst the agent
population. If we follow Langton, and are willing to class ourselves in the strong
Alife camp, then signing up to the PSS Hypothesis for Life may be a good course
of action for us to take. In that case, our bird model becomes a true information
ecology, and thus presents an opportunity for empirical data-collection in this virtual
population of birds.
6.3 Frameworks and ALife: Weak ALife
6.3.1 Artiﬁcial1
vs. Artiﬁcial2
: Embracing the Distinction
In contrast to strong ALife, weak ALife faces some initially more daunting
theoretical prospects. This seems somewhat paradoxical; after all, the strong ALife
researcher seeks to equate digital organisms with natural organisms, whereas the
weak ALife researcher seeks only the replication of certain properties of natural
life. However, while the strong ALife researcher may justify their investigations
into a digital ecology by signing up to an appropriate theoretical backstory, however
far-fetched, proving the relation between a natural system under investigation and a
computational model based on that system is a more difﬁcult problem and requires
more in-depth justiﬁcations.
Returning to our central example, recall our researcher who wishes to model the
ﬂocking behaviour of migrating birds. While a great deal of experimental data exists
upon which one can base such a computational study, the researcher must choose
which elements of that data provide a useful background for the simulation and
which do not. Data about bird migration varies greatly across different species and
climates, and the researcher must identify the most salient forms of collected data
to use as a basis for the model. These choices will in turn inform the construction
110 6 Analysis: Frameworks and Theories for Social Simulation
of the model, and the related points of theory regarding migration that must be
incorporated into that model.
As Chalmers suggests (Chalmers 1999), these abstractions reveal the inherent
theory-dependence of artiﬁcial life as an enterprise; the researcher’s choice of
abstractions may conform to their speciﬁc theoretical biases. In order to make these
choices effectively, and to draw a strong correlation between the digital system
and the natural system, one must ﬁnd a framework which embraces the inherently
artiﬁcial nature of such simulations and uses the advantages of the digital medium
effectively to draw useful experimental conclusions.
6.3.2 Integration of Real Data: Case Studies
6.3.3 Backstory: Allowing the Artiﬁcial
The integration of a suitable theoretical backstory into weak ALife research seems
a more difﬁcult task than for strong ALife. As Bryan Keeley describes, weak ALife
can only hope to be functionally related to natural life, producing behaviours that
are analogous to those displayed in biology (Keeley 1997). However, establishing
a clear relationship to a natural system is not always straightforward, particularly
when the artiﬁcial system under consideration bears less resemblance to the
fundamental make-up of the natural world.
For some researchers and theorists, these artiﬁcial worlds present a tantalising
opportunity to examine characteristic evolutionary behaviours in a simpliﬁed environment,
one amenable to study; given that natural evolution is far more difﬁcult to
observe than an abstracted evolutionary algorithm, this naturally seems an attractive
prospect for those who wish to observe evolution in action. Ray (1994, 1996) goes
so far as to assert that digital forms of evolution can produce the same diversity
of forms that we observe as a result of natural evolution (though perhaps given his
statements regarding Tierra as a whole, this position is, for him, milder than most).
Unfortunately for Ray, while Tierra does display an impressive variety of selfreproducing
digital entities that display unexpected behaviour, the simulation tends
to get caught in an eventual evolutionary cycle in which certain forms repeat. This
contrasts strongly with real-world evolutionary behaviour, in which the general
level of complexity in evolving populations tends to continue to increase over
time. Similarly, Avida and other simulation environments developed by the ALife
community suffer the same problem, preventing the community from replicating
the sort of staggering diversity seen in nature (Adami and Brown 1994). Bearing
this in mind, can the researcher be certain that these artiﬁcial evolutionary systems
are truly functionally related to natural evolutionary systems? Given that the overall
processes of evolution are still under constant debate and revision, and the innate
6.4 The Legacy of Levins 111
difﬁculty of providing appropriate selection pressures in an artiﬁcial environment,
how much do these systems coincide on a ‘given level of abstraction’ (Keeley 1997).
In cases such as this, one must be careful in deﬁning a theoretical backstory
linking the natural system to the artiﬁcial. A clear statement of the mechanisms
at work in the simulation and how they relate to similar theorised mechanisms in
natural evolution seems most helpful; given that there is no fundamental metric
to determine just how abstracted a given model is when compared to reality, a
clear statement of assumptions made and potential confounds in the simulation (i.e.,
difﬁculties in ﬁtness functions and similar issues) could be helpful in attempting to
link the simulation results to empirical results.
In the case of our bird example, such a backstory would need to include
information about the assumptions made when constructing our simulation. We
would have to describe the real-world instances of bird behaviour that we are trying
to replicate, and how these real-world instances have inﬂuenced our implementation
of the model. Where simpliﬁcations have been made, i.e. by simplifying the
structure of the agents to facilitate computability and simplicity of analysis, we
would need to note this fact and mention how these simpliﬁcations may change
the character of the behaviour observed in the simulation. In the ideal situation,
someone reading a paper describing our bird model should be made aware of the
shortcomings of the simulation, where it attempts to reproduce real-world bird
behaviour and physiology, and where it makes abstractions for the purposes of
making the simulation tractable.
6.4 The Legacy of Levins
6.4.1 The 3 Types: A Useful Hierarchy?
As discussed at length in Chap. 4, the Levinsian framework for modelling in
population biology appears generally useful for ALife modelling endeavours. After
all, Levins seemed intent upon creating a pragmatic framework for constructing
biological models, and since ALife often falls within that remit his ideas remain
relevant. However, the extended framework developed from Levins in Chap. 4 seems
perhaps more useful within the context of artiﬁcial life.
With Braitenberg’s Law in mind, the concept of a tractability ceiling placed on
ALife seems appropriate, if vexing. With ALife systems spanning an enormous
variety of biological phenomena, often incorporating versions of vast and complex
biological mechanisms such as evolution, the question of analysis becomes of
paramount importance. While we may comfortably classify ALife models within
Levins’ three types with relatively little difﬁculty, we remain uncertain how
analysable those models will prove to be with only that classiﬁcation in mind.
112 6 Analysis: Frameworks and Theories for Social Simulation
6.4.2 Constraints of the Fourth Factor
Indeed, the fourth Levinsian factor appears to place some serious limitations upon
ALife systems. As noted in Chap. 3, such systems frequently fall into the Type
3 category, oriented as they are toward producing broad-stroke investigations of
generalised populations. Applying those results toward real populations becomes
increasingly problematic as tractability concerns become important; without reasonable
analysis of the data produced by these simulations, the researcher will
have great difﬁculty applying that data to any understanding of a natural biological
system. In essence, even with carefully-designed simulation models, this tractability
ceiling prevents highly complex simulations from being productive of great insight.
Thus, our Alife-type bird migration model may run into difﬁculties if we
incorporate too many real-world complexities. If we use agents with neural network
controllers, for example, then such networks are very difﬁcult to analyse (recall Beer
2003a,b). As modellers we must judge whether the use of such components in the
simulation is justiﬁed given the increase of complexity and analytical difﬁculty. The
more we incorporate added elements in an attempt to capture real-world complexity,
the more we approach the tractability ceiling.
Even assuming that our model manages to balance Levins’ three types appropriately
while maintaining a reasonable level of tractability, signiﬁcant problems
still remain. With Braitenberg’s Law in mind, agent-based models in general
suffer a greater disconnect between invention and analysis than other modelling
methodologies, leading to a greater chance of producing impenetrably opaque
simulations.
If we then apply agent-based methodologies to the ﬁeld of social science, in
which there are already signiﬁcant difﬁculties in constructing models based upon
strongly empirical social data, then these problems become increasingly acute.
Bearing in mind these discussions of theoretical frameworks in relation to ALife
and agent-based models in general, we shall examine how these same frameworks
impact upon the agent-based social simulations detailed in the previous chapter.
6.5 Frameworks and Social Science
6.5.1 Artiﬁcial1
vs. Artiﬁcial2
: A Useful Distinction?
ALife research focuses on living systems and their governing processes, and as such
relies upon deﬁnitions and theories of life as a basis for enquiry. Life being difﬁcult
to deﬁne empirically, strong ALife can, as illustrated earlier, attempt to demonstrate
a digital instantiation of life itself when life is deﬁned appropriately within the
context of that research. However, when expanding beyond processes relating to
simple organisms and populations and attempting to model social structures and
6.5 Frameworks and Social Science 113
processes, additional layers of complexity come into focus. Epstein (1999) provides
a comparison to the famous ‘Boids’ model of ﬂocking behaviour:
Generating collective behaviour that to the naked eye “looks like ﬂocking” can be extremely
valuable, but it is a radically different enterprise from generating, say, a speciﬁc distribution
of wealth with parameters close to those observed in society. Crude qualitative caricature is
a perfectly reasonable goal. But if that is one’s goal, the fact must be stated explicitly....
This will avert needless resistance from other ﬁelds where “normal science” proceeds under
established standards patently not met by cartoon “boid” ﬂocks, however stimulating and
pedagogically valuable these may be. (p. 52–53)
In essence, Epstein presents a position reminiscent of our earlier discussion of
Braitenberg: imitation of a system with a model, as in mimicking the behaviour
of ﬂocking birds, is far simpler than creating a model system which can generate
that behaviour. Further, such models can confuse the research landscape, perhaps
claiming to produce more insight than they are fundamentally capable of providing.
In such cases a model which seeks this sort of qualitative similarity is not
constructed in such a way as to allow any insight into the root causes of that complex
behaviour. Epstein implies later in his discussion that in the case of social science,
which involves interacting layers of actors in a society, this problem becomes more
acute for the computational modeller.
In such a context, any argument for ‘strong social-science simulation’ seems
difﬁcult to justify; in order to accept that a given social model is a digital
instantiation of a social system, one would have to accept that the agents have
sufﬁcient complexity to interact with one another, generate a social structure, and
react and respond to that structure in an appropriately non-trivial manner. There is a
signiﬁcant danger, as noted by Epstein, of a model of a complex social system falling
within the realm of a ‘crude qualitative caricature.’ Social science then may be said
to lie within the domain of Artiﬁcial2
: something made to resemble something else.
Fortunately, our examination of Luhmann has revealed the problematic nature
of building an Artiﬁcial1
social simulation. With our search for a fundamental
social theory relying upon developing a new means for hypothesis-testing and social
explanation, creating instantiations of ‘digital society’ is rather less than useful.
Thus, while remaining within the domain of Artiﬁcial2
may seem initially limiting,
in fact the social simulator is likely to ﬁnd it of far greater utility than the alternative.
6.5.2 Levins: Still Useful for Social Scientists?
As demonstrated earlier in this chapter, Levins’ framework for modelling in
population biology is remarkably applicable to today’s more modern computational
methodologies. Our updated Levinsian framework developed in Chap. 4 and its
consideration of tractability presents an account of a concern common to most
varieties of computational models: efﬁcient utilisation of computing resources relies
on a relatively tractable and analysable problem. However, when considering the
application of such a framework to social simulation, some differing concerns come
to light.
114 6 Analysis: Frameworks and Theories for Social Simulation
As Gilbert and Tierna note (Gilbert and Terna 2000), emergent phenomena in
social science reﬂects a certain additional complexity in comparison to the natural
sciences:
In the physical world, macro-level phenomena, built up from the behaviour of microlevel
components, generally themselves affect the components.... The same is true in the
social world, where an institution such as government self-evidently affects the lives of
individuals. The complication in the social world is that individuals can recognise, reason
about and react to the institutions that their actions have created. Understanding this feature
of human society, variously known as second-order emergence Gilbert (1995), reﬂexivity
Woolgar (1988) and the double hermeneutic, is an area where computational modelling
shows promise. (p. 5)
Thus, Gilbert and Tierna contend that agent-based models can capture this emergent
phenomena more vividly than other methodologies which, while providing
potentially strong and useful predictions of a macro-level system’s behaviour, do
not provide an explanation of that behaviour in terms of its component parts. Epstein
(1999) takes this statement further, arguing that in certain contexts, successful mathematical
models may be ‘devoid of explanatory power despite [their] descriptive
accuracy’ (p. 51). Epstein goes on, however, to acknowledge the difﬁculties inherent
in creating an artiﬁcial society with sufﬁcient ‘generative power, proposing the use
of evolutionary methods to ﬁnd the rule-sets most amenable to complex emergent
behaviour in a given simulation.
Levins’ framework, which depends upon tractability as a key concern for the
modeller, may at ﬁrst blush seem insufﬁcient to deal with the methodological
complexities inherent in the simulation of social structures. After all, the researcher
is dealing with multiple interacting layers of complexity, with some of these
emergent behaviours relying upon not just reactions to a changing environment,
as in an artiﬁcial life simulation, but a cognitive reaction to that environment, which
can then inﬂuence both that agent and others within that artiﬁcial society. With such
high-level abstractions taking place in these simulations, how might one quantify
the realism, generality and precision of a social model?
To clarify, imagine that we have designed and implemented an agent-based
model of our migrating bird population. Each agent is capable of moving through
a simulated spatial environment, reproducing, and thus evolving through simulated
generations. Now, in an effort to capture the social effects at play within a bird
colony upon arrival at its destination, we allow each agent to communicate and
exchange information with its compatriots. In order to capture this we suddenly
need to add all sorts of new elements to our implementation: a means of encoding
commuication between individuals, means for choosing the method and frequency
of those communications, and deciding how these interactions will affect the agent,
its communication partners, and the surrounding environment. How can we capture
such effects? How might we model the birds’ communications, and how they affect
the simulation as a whole? Already we have introduced a number of new factors
into the model for which there is little hard empirical information to guide our
implementation of these factors. If we cannot identify the level of realism of these
new components of the simulation, how is it possible to clarify the position of our
new model amongst Levins’ four dimensions of model-building?
6.5 Frameworks and Social Science 115
In essence, without a solid set of criteria identifying the realism of simulated
social behaviours and cognition, one becomes reliant on qualitative judgements
to decide the validity of a social simulation; in other words, the researcher must
determine that the behaviour of that artiﬁcial society sufﬁciently resembles the
behaviour of a real society to call it a successful result. Perhaps then social
simulations begin skewed away from realism, and further toward generality and
precision than other simulation varieties. With realism so difﬁcult to deﬁne in a
social context, and with the necessary abstractions for social simulation so wideranging,
the researcher seems best served by striving to provide examples of possible
mechanisms for an emergent social behaviour based upon a very general picture
of agent-based interactions and communications. Deﬁnitive statements regarding
which of these mechanisms are responsible for a given social phenomenon may be
impossible, but the model could provide a means for illuminating the possible role
of some mechanisms in the formation and behaviour of social structures (this being
one of Epstein’s suggested roles for simulation in ‘generative social science’).
6.5.3 Cederman’s 3 Types: Restating the Problem
If Levins’ framework is insufﬁcient to capture some of the methodological complexities
particular to modelling for the social sciences, perhaps this framework could
be usefully informed by Cederman’s own framework of C1, C2 and C3 political
models (see Table 5.1). With some examination of this framework, we may be able
to draw useful parallels between these three types and those described by Levins.
Cederman’s C1 models focus on modelling the behavioural aspects of social
systems, or the emergence of general characteristics of certain social systems in a
computational context. He cites Axelrod’s Complexity of Cooperation as a major
foray into this type of modelling, as well as Schelling’s residential segregation
model (Cederman 2001). This type of model seems related to Levins’ L3 models,
which sacriﬁce precision for realism and generality; Cederman’s C1 models do not
attempt to link with empirical data, but instead seek to demonstrate the emergence
and development of behavioural phenomena in a generalised or idealised context.
Cederman’s C2 models aim to explain the emergence of conﬁgurations in a
model due to properties of the agents within the simulation (Cederman 2001).
Examples of this methodology include Lustick’s Agent-Based Argument Repertoire
Model (Lustick 2002, 2006), which provides agents with a complex set of opinions
which they may communicate with other agents. The opinions of agents within
the simulation can alter through these communications, leading to the apparent
generation of social structures within the agent population.
In this case, the comparison with Levins is more difﬁcult; while the agents
themselves are designed to be more complex within a C2 model, can this necessarily
be pinned down as an increase in either precision or realism? The closest analogue
appears to be Levins’ L2 models, which eschew realism in favour of generality and
precision. While a C2 Cederman model or an L2 Levins model is not concerned
116 6 Analysis: Frameworks and Theories for Social Simulation
with comparison to empirical data, these models do seek to provide a useful
framework for describing complex social behaviours in an idealised context, similar
to those population biology models alluded to by Levins. In either discipline, the
modellers hope to illuminate some of the contributing factors that produce the
modelled behaviour, and perhaps stray closer to a useful description of the reallife
instantiation of that behaviour than may be expected initially from such an
abstracted model.
Cederman’s C3 models are the most ambitious in that they attempt to model
the emergence of both agent behaviours and their interaction networks (Cederman
2001). He speciﬁcally cites ALife as a ﬁeld which may provide illuminating
insights in that regard, given that ALife is concerned with such complex emergent
behaviours and structures. As discussed earlier, however, Cederman’s categories
have no hard borders between them, particularly in the case of C3 models; Cederman
himself admits that C1 and C3 can easily overlap. As a consequence, identifying
where the C3 models lie amongst a given crop of social science simulations is not
always such a simple task, though the C3 categorisation does provide useful context
in which to examine how the goals of different varieties of models can affect their
construction and implementation.
Once again, though, the comparison with Levins is difﬁcult; allowing for such
emergent behaviours as described by Cederman in the C3 categorisation does not
correlate directly with Levins’ three categorisations. Perhaps the closest analogue
here is once again Levins’ L3 models, given the focus on emergence of both agentlevel
and societal-level behaviours; in neither case is the modeller overly concerned
with a direct relation to empirical data. Instead the modeller hopes to provide a
cogent explanation for the emergence of social behaviours by allowing agents to
interact and change due to pressures within the model, rather than due to complex
constraints placed upon that model.
6.5.4 Building the Framework: Unifying Principles for Biology
and Social Science Models
We have clearly run into a difﬁculty in comparing Levins and Cederman’s respective
modelling frameworks. Levins reserves the L1 category for those models which
seek a direct relationship to empirical data, leaving generality behind in favour
of realism and precision. However, Cederman’s 3 types leave out this distinction,
instead providing what appear to be two variations on Levins’ more general L3
models. Both Cederman’s C1 and C3 models appear to leave precision behind,
seeking instead to describe social phenomena in an idealised context; C1 models
focus purely on the emergence of social structures, while C3 models focus on the
emergence of both societal and individual structures and behaviours.
Cederman’s view can be used to further qualify Levins’ original framework
however. Levins himself cites L3 models as the most promising, and his preferred
6.5 Frameworks and Social Science 117
Table 6.1 One possible Levins/Cederman framework
Modiﬁed Levinsian framework
L1 Precision and realism at the expense of generality
L2 Generality and precision at the expense of realism
L3A Generality and realism at the expense of precision at one level of simulation
L3B Generality and realism at the expense of precision at multiple levels of simulation
methodology within population biology; similarly, Cederman cites his C3 models
as the most promising within political and social science (though interestingly,
Cederman’s recent work has strayed towards a version of Levins’ L1; see Cederman
2006 and the section below). The question then becomes: how can we characterise
Cederman’s C1 vs. C3 distinction in terms of the Levinsian factors of generality,
precision and realism?
One can envision a further subdivision of Levins’ L3 into a L3A and L3B:
L3A being characterised by a sacriﬁce of precision in one level of the simulation
(i.e., Cederman’s C1 which seeks emergent social behaviours), and L3B being
characterised by a sacriﬁce of precision at multiple levels (as in Cederman’s C3,
which seeks both emergent social behaviours and interaction networks); Table 6.1
provides a summary of this potential framework.
Alternatively, with the incorporation of tractability into the Levinsian framework
as a fourth factor as in Chap. 4, this subdivision may be much simpler. Both Levins
and Cederman acknowledge these L3 models to be both the most useful and the
most challenging; perhaps then Cederman’s C1 and C3 may be characterised by
differing levels of tractability. In this way we can maintain this modiﬁed Levinsian
framework as a more ﬂuid continuum, based upon the determining factor of overall
tractability, rather than introducing additional sub-categories for special cases of
speciﬁc methodologies.
6.5.5 Integration of Real Data
Some researchers involved in computational modelling have attempted to sidestep
the difﬁculties of the inherent lack of ‘real data’ by attempting to integrate experimental
data into their simulations. Cederman’s 2006 paper on geo-referencing in
datasets for studies of civil wars provides a useful example. In this case, Cederman
uses data and maps from the Russian Atlas Narodov Mira, an atlas produced by a
1960s Soviet project aiming to chart the distribution of ethnic groups worldwide.
The project then seeks to formulate agent-based models using this data to examine
the potential causes of ethnic conﬂict. As Cederman notes, ‘there is no substitute
for real-world evidence’ when attempting to understand the causes of such conﬂict;
however, the ethnographic data in this case is both limited and quite old.
118 6 Analysis: Frameworks and Theories for Social Simulation
Of course the worldwide ethnographic distribution has likely changed significantly
since the publication of this atlas in 1964, and updating the information
contained therein is no simple task. The integration of such data into an agentbased
computational model seems like a potentially fruitful method for tying the
results of that model more closely to political and social reality. However, with the
limitations of this dataset and the difﬁculties inherent in collecting future data of
a similar type, is this data integration still useful as a framing mechanism to place
this model on more solid empirical ground, or is it an interesting but ultimately
misguided enterprise?
A further difﬁculty with the Atlas Narodov Mira dataset is that the atlas provides
a static ethnographic picture. While it is a remarkably detailed look at a particular
time in world history and the related distributions of peoples throughout the globe,
the lack of similar data in the following decades leaves us with an inability to directly
associate the intervening political and social changes with ethnic conﬂicts that have
erupted in the years since the atlas’ publication. Some argue that computational
modelling in such a circumstance provides a remarkable capacity for hypothesis
testing; by basing a relatively realistic model on such a solid footing of experimental
and observational data, the researcher can experiment with varying parameters and
initial conditions in an attempt to replicate the ethnic conﬂicts seen since the atlas
was produced. However in such a case the same problems return to haunt the
researcher: deciding which abstractions to make can be critical, and deciding how
to formalise the inﬂuence of complex social and political interactions is far from
trivial.
6.6 Views from Within Social Simulation
6.6.1 Finding a Direction for Social Simulation
While the last chapter provided an in-depth examination of the current state of social
simulation, and an analysis of its ability to explain and interpret real-world social
phenomena, we still have relatively little understanding of the perception of the
utility of social simulation within the social sciences. A number of prominent social
simulators display great enthusiasm for the pursuit, as would be expected, but how
do those viewing this growing trend from other parts of the ﬁeld react?
With this in mind we will examine some views of the general purpose of social
simulation, and descriptions of the perceived problems of the methodology, from
within the social sciences. We can then incorporate these analyses with our own
discussion of the importance of simulation in a fundamental social theory and
further expand our growing methodological and theoretical framework for social
simulation.
6.6 Views from Within Social Simulation 119
6.6.2 Doran’s Perspective on the Methodology
of Artiﬁcial Societies
Doran in his article for Tools and Techniques for Social Science Simulation
provides a coherent examination of the major difﬁculties facing the artiﬁcial society
methodology in social simulation (Doran 2000). Doran describes this method as
follows:
The central method of artiﬁcial societies is to select an abstract social research issue,
and then to create an artiﬁcial society within a computer system in which that issue
may systematically be explored. Building artiﬁcial societies is a matter of designing and
implementing agents and inter-agent communication, in a shared environment, and existing
agent technology provides a wealth of alternatives in support of this process. (p. 18)
Thus, Doran views social simulation as a means of examining the workings of
large-scale, abstract social issues. For Doran simulation provides a way to generate
new ‘world histories,’ allowing the researcher to watch a society grow from a
provided set of initial conditions and assumptions.
Of course he notes the signiﬁcant methodological difﬁculties inherent in this
artiﬁcial society approach. He identiﬁes three main problems: searching the space
of initial conditions and parameters; describing the results of the simulation in a
tractable way; and overcoming problems of instability due to changes in initial
conditions. The ﬁrst and third problems here are highly reminiscent of more
general problems common to agent-based modelling as a whole, while the problem
of analysis and tractability harkens back to Levins and our initial theoretical
explorations of artiﬁcial life.
Interestingly, however, Doran posits that the greatest problem facing the social
simulator is the largely undeﬁned nature of the computational agent itself. He notes
that different uses of agent-based models often incorporate different base properties
for the agents within the model, and that despite the simplistic views of what
constitutes an agent within social science, there is little overall consensus regarding
a deﬁnition of agent architectures.
He argues that agents should be deﬁned in computational terms, and thus should
be able to emerge from a model in the same way as other more complex phenomena.
Of course, this is not the normal course in most agent-based models; in practice
such models are designed to include predeﬁned agent structures based upon certain
theoretical assumptions. As a consequence, this would not be an easy task for
the modeller; constructing a simulation in which agent structures are expected to
emerge seems extremely difﬁcult. There would almost certainly need to be some
sort of precursors to deﬁned agents to allow such structures to develop, given that
the earliest beginnings of our own life and society are so murky to begin with and
can provide little clear inspiration. The purpose, however, is sound, as producing
simulations that at least approach such possibilities would allow the modeller to step
away further from the difﬁculties produced by theory dependence and pre-deﬁned
agent structures.
120 6 Analysis: Frameworks and Theories for Social Simulation
Doran’s view is also interesting in that it meshes with our earlier discussion of
Niklas Luhmann and the search for a fundamental social theory. Since in Doran’s
view, artiﬁcial societies aim to examine the general development of human society in
an abstract fashion, these societies must be based upon valid assumptions. However,
those assumptions are necessarily based upon our own cultural experiences, and thus
will be imprinted upon that model in some fashion. A model which eliminates this
difﬁculty, or at least minimises it, would allow for agents to develop in simulation
which bear far fewer markings of our own societal preconceptions. For the social
scientist looking to develop new social theory, such an approach would be far more
fruitful than the heavily theory-dependent alternative.
As discussed in the previous chapter, Doran agrees that artiﬁcial societies should
strive to develop from the earliest beginnings of societal interaction (‘the lowest
possible level, even below the level of agent’ in Doran’s phrasing [p. 24]), and
that the mechanisms of society should emerge from that basis. This would avoid
producing a simulation constructed around pre-existing assumptions from our own
society.
6.6.3 Axelrod and Tesfatsion’s Perspective: The Beginner’s
Guide to Social Simulation
Axelrod and Tesfatsion (2005) lay out a brief guide to social scientists hoping to
incorporate social simulation into their current work. Given the authors’ prominent
position in the current social simulation ﬁeld, this guide provides an illuminating
look at those aspects of agent-based modelling perceived to be the most valuable by
those within this area of research.
After beginning with a brief introduction to the basic characteristics of agentbased
models (including a brief discussion on creating ‘histories,’ as described by
Doran), Axelrod and Tesfatsion lays out a four-part description of the potential goals
of simulation models in the social sciences. They describe each of these potential
goals in turn:
1. Empirical understanding: ‘ABM researchers seek causal explanations grounded in the
repeated interactions of agents operating in speciﬁed environments.’
2. Normative understanding: ‘ABM researchers pursuing this objective are interested in
evaluating whether designs proposed for social policies, institutions, or processes will
result in socially desirable system performance over time.’
3. Heuristic: ‘How can greater insight be attained about the fundamental causal mechanisms
in social systems?’
4. Methodological advancement: ‘How best to provide ABM researchers with the methods
and tools they need to undertake the rigorous study of social systems through controlled
computational experiments?’ [p. 4–5]
Here we discover yet another framework underwriting agent-based models in
social science. However, unlike the work of Levins or Cederman, Axelrod and
6.7 Summary and Conclusions 121
Tesfatsion prefer not to discuss speciﬁc criteria which may place a given agent-based
model within these categories, preferring instead to provide only examples of each
research goal.
Interestingly, despite their initial mention of Doran’s described ‘artiﬁcial societies’
methodology, and the general goal of generating artiﬁcial ‘world histories,’
the remainder of Axelrod and Tesfatsion’s introduction presents a sharp contrast
to Doran’s ideas. They stress the potential for agent-based models to produce
substantive empirical insights related to real-world societies, rather than produce
more general insights about the initial formation of societies and social interaction
as a whole.
However, given the caveats presented thus far with agent-based social simulation,
Axelrod and Tesfatsion’s research goals seem at odds with the conclusions we have
reached. If we agree, as discussed in the last chapter, that social simulation may
lack some critical elements required to provide social explanation, then hoping to
use such simulations to design real-world social institutions may produce disastrous
results. Similarly, developing empirical understanding and ‘causal explanations’
would be equally problematic, particularly as any non-reductive aspects of the
society under investigation would be difﬁcult to identify.
Even if we disagree with the alleged difﬁculties in social explanation for social
simulation, Doran’s views have illuminated a further difﬁculty for Axelrod and
Tesfatsion’s suggested approaches. Without a clear deﬁnition of what constitutes
an ‘agent’ in an empirical or normative simulation, the level of correspondence
between these agents and human social actors is too ill-deﬁned. What cognitive
capacities would agents require to provide that degree of understanding? Surely
Schelling-esque simplicity is not going to be the best course for all possible
investigations using simulated societies?
6.7 Summary and Conclusions
Thus far we have seen a great variety of proposed theoretical frameworks for
the use of agent-based models. Artiﬁcial life provided a useful starting point,
given its relationship to the long-standing use of mathematical models in various
biological disciplines. Comparing the methodological difﬁculties of artiﬁcial life
with mathematical models in biology allows us to develop a greater understanding
of the problems most particular to agent-based models.
However, applying the resulting methodological discussions to agent-based
models in social science is not so straightforward. As discussed in Chap. 5, social
scientists must cope with some unique difﬁculties: social structures are not clearly
hierarchical in nature; empirical studies of social phenomena are frequently problematic
due to the interacting complexities of individual and collective behaviour;
and social explanation may suffer from difﬁculties similar to those faced by
researchers in mental phenomena, as some social phenomena may be similarly
irreducible to straightforward low-level behaviour.
122 6 Analysis: Frameworks and Theories for Social Simulation
Obviously all of these points impact the social simulator and make the job of
model-building signiﬁcantly more difﬁcult. However, as seen during this analysis,
these difﬁculties shift in emphasis depending on the purpose of the model (as we
might expect from previous discussion regarding the Levins and Cederman frameworks).
Taking an external, Luhmannian perspective, and incorporating Doran’s
related views, social simulation may be able to provide a unique window into certain
aspects of social theory that are otherwise inaccessible through standard empirical
means.
This methodology brings us back to the artiﬁcial world problem of Chap. 4. If
we accept that ‘growing’ artiﬁcial societies from a level even below that of agents is
a promising means for investigating potential fundamental social theories, we must
likewise accept that these artiﬁcial societies may be quite difﬁcult to relate to realworld
societies. Given the gaps in our understanding in social science, ranging from
unanswered questions in individual behaviour through to questions of high-level
social organisation, can such artiﬁcial societies bring us any closer to the desired
fundamental social theory?
The next chapter will bring us closer to answering this critical question. By
examining one of the most prominent exemplars of social simulation, Schelling’s
residential segregation model, we will be able to illuminate this concern in greater
detail. Schelling’s model is the very deﬁnition of abstraction – and yet it is credited
with producing some important insights into the social problem it is based upon.
By determining how Schelling’s model surpassed its abstract limitations, we can
develop a framework which may underwrite future models in social simulation of a
similar character.
This analysis of Schelling marks a conclusion of sorts to the arguments and
frameworks discussed thus far in Parts I and II. Having discussed modelling
frameworks in Alife and biology, and having brought those together with new
elements from simulation in the social sciences, Schelling’s model will provide
a means to demonstrate the importance of these theoretical, methodological, and
pragmatic frameworks for the modeller who wishes to push social science forward
through simulation. As we shall see, Schelling’s simple model was not nearly so
simple in its widespread inﬂuence and overall importance. The elements which
contributed to this success can be seen via its relationships to the frameworks
discussed thus far.
References
Adami, C., & Brown, C. T. (1994). Evolutionary learning in the 2D artiﬁcial life system Avida.
Cambridge: MIT Press.
Axelrod, R., & Tesfatsion, L. (2005). A beginner’s guide to agent-based modelling in the social
sciences. In L. Tesfatsion & K. L. Judd (Eds.), Handbook of computational economics, Vol. 2:
Agent-based computational economics, p. Appendix A.
Beer, R. (2003a). Arches and stones in cognitive architecture. Adaptive Behavior, 11(4), 299–305.
References 123
Beer, R. (2003b). The dynamics of active categorical perception in an evolved model agent.
Adaptive Behavior, 11(4), 209–243.
Cederman, L. E. (2001). Agent-based modeling in political science. The Political Methodologist,
10, 16–22.
Chalmers, A. F. (1999). What is this thing called science? Buckingham: Open University Press.
Doran, J. E. (2000). Trajectories to complexity in artiﬁcial societies. In A. Kohler & G. Gumerman
(Eds.), Dynamics in human and primate societies. New York: Oxford University Press.
Epstein, J. (1999). Agent-based computational models and generative social science. Complexity,
4(5), 41–60.
Gilbert, G. N. (1995) Emergence in social simulation. In G. N. Gilbert & R. Conte (Eds.), Artiﬁcial
societies: The computer simulation of social life, pp. 144–156. London: UCL Press.
Gilbert, N., & Terna, P. (2000). How to build and use agent-based models in social science. Mind
and Society, 1(1), 57–72.
Keeley, B. L. (1997). Evaluating artiﬁcial life and artiﬁcial organisms. In C. G. Langton &
K. Shimohara (Eds.), Artiﬁcial life V (pp. 264–271). Cambridge: MIT Press.
Lustick, I. (2002). PS-I: A user-friendly agent-based modeling platform for testing theories of
political identity and political stability. Journal of Artiﬁcial Societies and Social Simulation,
5(3), 7.
Lustick, I. (2006). Deﬁning violence: A plausibility probe using agent-based modeling. Paper
Prepared for LiCEP, Princeton University, 12–14 May 2006.
Ray, T. S. (1994). An evolutionary approach to synthetic biology: Zen and the art of creating life.
Artiﬁcial Life, 1, 179–209.
Ray, T. S. (1996). An approach to the synthesis of life. In M. A. Boden (Ed.), The philosophy of
artiﬁcial life (pp. 111–145). Oxford: Oxford University Press.
Rumelhart, D. E., & McClelland, J. L. (1986). Parallel distributed processing. Cambridge: MIT
Press.
Silverman, E., & Bullock, S. (2004). Empiricism and artiﬁcial life. In J. Pollack, M. Bedau,
P. Husbands, T. Ikegami, & R. A. Watson (Eds.), Artiﬁcial Life IX: Proceedings of the
Ninth International Conference on the Simulation and Synthesis of Living Systems. MIT
Press/Bradford Books.
Woolgar, S. (1988). Science: The very idea. Chichester: Elias Horwood.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
Chapter 7
Schelling’s Model: A Success for Simplicity
7.1 Overview
In the previous chapter the numerous methodological and theoretical frameworks
elucidated thus far were analysed as a whole in an attempt to develop a useful
synthesis for the social science modeller. Bringing together elements from population
biology, artiﬁcial life, and social science itself, we examined the multiple
dimensions that must be addressed in a successful social simulation model.
In this chapter we will examine one particular example of such a successful
model: Schelling’s residential segregation model. After describing the context in
which this model was developed, we will analyse this relatively simple model under
the constraints of each of the theoretical frameworks described thus far. Despite the
abstractions inherent in Schelling’s construction, the model achieved a remarkable
measure of recognition both inside and outside the social science community; our
analysis will discuss how this ‘chequerboard model’ found such acceptance.
Using Schelling as a case study, we can further develop our ﬂedgling modelling
framework for the social sciences. In the quest to underwrite social simulation
as a more rigorous form of social-scientiﬁc exploration, Schelling will provide a
valuable examination of the issues facing modellers as this methodology grows in
relevance throughout social science. After discussing the issues brought forward by
this analysis, which brings together the frameworks and methodological proposals
discussed thus far, we will move on to the last chapter of Part II and present a
synthesis.
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_7
125
126 7 Schelling’s Model: A Success for Simplicity
7.2 The Problem of Residential Segregation
7.2.1 Residential Segregation as a Social Phenomenon
7.2.1.1 The Importance of the Problem
One of the most puzzling aspects of the sociological study of residential segregation
is the sizeable gap between the individual residential preferences of the majority
of the population and the resultant neighbourhood structure. The General Social
Survey of major US metropolitan areas asked black respondents whether they
preferred to live in all-black, all-white, or half-black/half-white areas; 55.3% of
those surveyed stated a preference for half-black/half-white neighbourhoods (Davis
and Smith 1999). However, census data reveals that very small percentages of black
individuals in major US metropolitan areas actually live in half-black/half-white
neighbourhoods; the 1990 census indicates that for most major cities, less than 5%
of black residents live in such mixed areas.
7.2.2 Theories Regarding Residential Segregation
The phenomenon of residential segregation is a complex one, involving interacting
contributing factors at various levels of societal structure. A deﬁnitive statement of
why residential segregation occurs still troubles researchers; the problem varies so
widely across cultures, societies and ethnic groups that the speciﬁc factors at play
are still elusive.
Freeman’s summary of residential segregation of African-Americans in major
American metropolitan areas (Freeman 2000) provides one example of the difﬁculties
facing social scientists in this area. As Freeman notes, African-Americans
seem particularly prone to residential segregation; other minority populations tend
to ‘spatially assimilate’ after a period of residential segregation, integrating with the
majority population as educational and ﬁnancial disparities decrease (Massey et al.
1985). African-American populations, however, do not display signiﬁcant spatial
assimilation, despite the decrease in socio-economic disparities evident from the
available data; some posit that this may be due to unseen bias in local housing
authorities, cultural cohesion keeping African-Americans in established minority
communities, or decreased access to public services in low-income areas, but none
of these possibilities can fully account for these unusual tendencies in the data.
7.3 The Chequerboard Model: Individual Motives in Segregation 127
7.3 The Chequerboard Model: Individual Motives
in Segregation
7.3.1 The Rules and Justiﬁcations of the Model
Schelling’s ‘chequerboard model’ sought to make a singular point, as illustrated by
the simplistic construction of the model. He argued that if the racial make-up of a
given area was critical to an individual’s choice of housing, then even populations
of individuals tolerant to mixed-race environments would end up segregating into
single-race neighbourhoods (Schelling 1978). Schelling hoped to illustrate that
large-scale factors such as socio-economic or educational disparities between ethnic
populations could not explain the generally puzzling phenomenon of residential
segregation; indeed, without a greater insight into the preferences and thought
processes of individuals within a given population (their ‘micromotives’), some
critical aspects of this social problem may elude the researcher.
Schelling illustrated this idea using a model constructed in a simple fashion,
using a type of cellular automaton model (initially constructed using a physical
chequerboard, hence the model’s nickname). He describes a world composed of
square cells, ﬁlled with agents of one of two types. Each agent interacts only with its
eight direct neighbouring cells, and there are no higher-level structures incorporated
into the model. The agents are given a tolerance threshold: if the number of adjacent
agents of its own type is less than that threshold, the agent will move to a nearby
location on the grid where its tolerance requirements are fulﬁlled. Thus, the model
incorporates a very simple rule set which allows each agent to display a singular
micromotive which is taken to represent that agent’s level of racial tolerance.
Schelling hoped to demonstrate the powerful inﬂuence of this simple micromotive
on the resulting racial structure of the neighbourhood inhabited by the agents.
7.3.2 Results of the Model: Looking to the Individual
The results of Schelling’s model were surprising to social scientists at the time.
The model showed that, for a wide range of tolerance thresholds, these initially
integrated neighbourhoods of tolerant agents would ‘tip’ toward one group or
another, leading the outnumbered group to leave and resulting in segregation
(Schelling 1971). Given that residential segregation was widespread in American
cities prior to the civil rights movement, even the rising tolerance following the
additional rights granted to African-Americans was not sufﬁcient to provoke a
signiﬁcant decrease in residential segregation, even by 1990 (Sander 1998).
128 7 Schelling’s Model: A Success for Simplicity
Schelling’s model demonstrates that a mere improvement in individual preferences
and a lack of housing discrimination (as outlawed by the Civil Rights Act of
1968) are not sufﬁcient to eliminate residential segregation. In fact, largely tolerant
‘microbehaviours’ can still result in such problems, as long as social factors such as
race remain a consideration in housing choice.
7.3.3 Problems of the Model: A Lack of Social Structure
Since Schelling’s original model formulation and the resultant interest of the
social science community, many researchers have since attempted to update the
model with more sophisticated computational techniques and a greater number of
inﬂuential social factors within the model. While Schelling’s model was accepted
as a remarkable illustration of a simple principle regarding the large-scale effects of
individual housing preferences, some modellers sought to create a more ‘realistic’
Schelling-type model which could incorporate socio-economic factors as well
(Sander et al. 2000; Clark 1991; Epstein and Axtell 1996). Given the accepted
complexity of residential segregation as a social problem, and the new insight into
the effects of individual preference illuminated by Schelling, models incorporating
Schelling-style ‘micromotives’ and large-scale social factors were seen as a
potential method for examining the interactions between these two levels of social
structure, something that was very much lacking in Schelling’s original formulation.
7.4 Emergence by Any Other Name: Micromotives
and Macrobehaviour
7.4.1 Schelling’s Justiﬁcations: A Valid View of Social
Behaviour?
Interestingly, despite the general simplicity of Schelling’s model and the lack of a
larger social context for the agents in the model, his discussion of micromotives
quickly gathered momentum among social scientists. His contention that certain
non-obvious conclusions regarding social behaviour may follow from studies that
do not depend upon empirical observation was inﬂuential, leading other modellers
to seek patterns of interaction in generalised social systems [other tipping models
here].
In addition, Schelling’s description of critical thresholds that lead to these ‘tipping’
phenomenon led to an inﬂux of sociological models exploring this possibility
in relation to numerous large-scale social phenomena. Within political science,
7.4 Emergence by Any Other Name: Micromotives and Macrobehaviour 129
Laitin used tipping models to examine the critical points at which populations
choose one language over another, as in the case of Ghana (Laitin 1994), and in the
growing acceptance of the Kazakh native tongue over Russian in Kazakhstan (Laitin
1998). In general, then, Schelling’s justiﬁcations for his residential segregation
model have been widely adapted throughout the social sciences; this simple method
for examining critical points in societal interaction seems to have generated a great
deal of related research in the years following his book’s publication.
Perhaps more importantly for the larger social science community, Schelling’s
model also sparked additional interest in empirical studies over the years as social
scientists wished to conﬁrm his claims. The most prominent example of this
model feeding back new ideas into the empirical domain was W.A.V. Clark’s
study (Clark 1991). Clark used the most recent demographic surveys available at
the time to examine elements of local racial preference in residential segregation
in real communities; while the character of the results did differ from Schelling,
the inﬂuence of local racial preferences was strong, conﬁrming Schelling’s most
important claim. This sort of empirical validation only lends further credence to
Schelling’s methodology and its success.
7.4.2 Limiting the Domain: The Acceptance
of Schelling’s Result
While Schelling’s model did not incorporate any semblance of large-scale social
structure in its simple grid-world, this lack of detail may have contributed to the
general acceptance of his model amongst the social-science community. While
his work did provide a signiﬁcant inﬂuence on later modelling work in the ﬁeld
and some empirical work, as noted above, his initial forays into the residential
segregation problem were very limited in scope (Schelling 1971).
Schelling aimed to illuminate the role of individual behaviours in producing
counter-intuitive results in large-scale social systems, and his simple residential
tipping model produced just such a counter-intuitive result by demonstrating the
inability of individual tolerance for racial mixing to eliminate segregation. In this
way the initial chequerboard model provided a theoretical backdrop for the larger
thesis of his later book-length examination Micro-Motives and Macro-behaviour
(Schelling 1978). Rather than producing one large, complex model which illustrated
the importance of these individual preferences in the behaviours of social systems,
he produced a number of small-scale, simple models which illustrated the same
point in a more easily digestable and analysable way. Perhaps then the lack of
complexity on display was what made his models so inﬂuential; providing such
a basic backdrop made replication and expansion of his result straightforward for
the research community.
130 7 Schelling’s Model: A Success for Simplicity
7.4.3 Taylor’s Sites of Sociality: One View of the Acceptance
of Models
Computational modelling, as described earlier, is inherently a theory-dependent
exercise. Modellers seek to simplify reality in order to examine speciﬁc behaviours
within the system of interest, and those simpliﬁcations and abstractions often betray
a theoretical bias. In addition to this difﬁculty, Taylor describes the concept of
‘sites of sociality’ within modelling communities (Taylor 2000). In Taylor’s view,
these sites correspond to points at which social considerations within a scientiﬁc
discipline begin to deﬁne parameters of a model, rather than considerations brought
about by the subject matter of the model itself.
Thus, if a certain evolutionary theory has become dominant within a speciﬁc
domain of population biology, for example, a model which ignores the conceits of
that theory may not ﬁnd acceptance among the community. Schelling’s modelling
work served to add to current social theory in evidence at that time, but did not
seek to overturn the dominant ideas of the ﬁeld; perhaps, in Taylor’s view, this
was a powerful method for gaining widespread acceptance in the social science
community.
7.4.4 The Signiﬁcance of Taylor: Communicability and Impact
Taylor’s description of sites of sociality within modelling communities brings
an important point to bear. Even if a model is constructed in such a way that
the modeller can justify its relevance to the broader empirical community, that
community may be operating under a different understanding of what is important
to display in a simulation, or the importance of simulations as a whole.
Once again, we may use Vickerstaff and Di Paolo’s model as an example
(Vickerstaff and Di Paolo 2005). Their model was accepted and communicated in
a very popular experimental biology journal, despite being entirely computational
in nature. The editors of the journal still accepted the paper despite its lack of hard
empirical data; the model’s relative simplicity kept it from being too alien a concept
for the biological readership, and the data it presented was novel and relevant despite
its origins. The ability to comprehend the model in a more substantive way is vital
to its acceptance; if the model had been too complex and difﬁcult to replicate, the
results would have been less impactful and interesting for the target audience.
Also like Schelling, the nature of the model makes it very palatable to the
experimental biology readership in a different manner: the model was adding to
the discourse on a particular topic in biology, rather than attempting to make major
alterations to the ﬁeld. If an Alife researcher submitted a paper to an experimental
journal that proclaimed an end to conventional insect biology as we know it, for
example, then the editors are unlikely to be receptive. As with Schelling, this paper
served to illuminate a new idea regarding an issue relevant to the community it
7.5 Fitting Schelling to the Modelling Frameworks 131
addressed; the paper did not ignore the pre-existing conceits of the community, and
the data it produced was of value to the theories established in that community by
decades of difﬁcult data collection and analysis.
Thus, Schelling and Vickerstaff and Di Paolo both show the importance of
communicability in a model which seeks to engage the wider research community.
Schelling had great impact due to the ease of communicating his model results,
and the ease with which members of the social science community could replicate
and engage with those results; similarly, Vickerstaff and Di Paolo achieved
success by crafting a model which demonstrated relevant new ideas that could be
communicated well to the biological community despite the differing methods of
data collection. Moving forward, we will see how the simplicity of a model can not
only assist communicability and impact for a given model, but also its tractability
and ability to produce useful, analysable results.
7.5 Fitting Schelling to the Modelling Frameworks
7.5.1 Schelling and Silverman-Bullock: Backstory
Under Silverman and Bullock’s framework (Silverman and Bullock 2004),
Schelling’s model succeeds due to the integration of a useful theoretical ‘backstory’
for the model. Schelling represents the chequerboard model as an example of
one particular micromotive leading to one particular macrobehaviour, in this case
residential segregation. In the context of this backstory, Schelling is able to present
his model as a suitable example of the impact of these micromotives on one
particularly thorny social problem.
By restricting his inquiry to this singular dimension, the model becomes more
palatable to social scientists, as the signiﬁcant abstractions made within the model
facilitate the portrayal of this aspect of the phenomenon in question. This serves to
emphasize Schelling’s original theoretical formulation by stripping away additional
social factors from the model, rather than allowing multiple interacting social
behaviours to dilute the evidence of a much greater role for individual preference in
residential segregation.
7.5.2 Schelling and Levins-Silverman: Tractability
Under our clariﬁed Levinsian framework from Chap. 4 (see Table 4.1), in which
tractability forms the fourth critical dimension of a model alongside generality,
realism and precision, Schelling’s model appears to fall into the L2 categorisation.
The model sacriﬁces realism for generality and precision, producing an idealised
system which attempts to illustrate the phenomenon of residential segregation in a
broader context.
132 7 Schelling’s Model: A Success for Simplicity
Meanwhile, tractability remains high due to the simplistic rules of the simulation
as well as the general ease of analysis of the model’s results; for many social
scientists, a simple visual examination of the resultant segregated neighbourhoods
from a run of the simulation proves Schelling’s point fairly clearly. More indepth
examinations are also possible, as seen in Zhang’s assessment of the overall
space of Schelling-type models (Zhang 2004); such assessments are rarely possible
with more complex models, which involve much more numerous and complex
parameters.
7.5.3 Schelling and Cederman: Avoiding Complexity
In Cederman’s framework (Cederman 2001, see Table 5.1), Schelling’s model is a
C1 simulation, as it attempts to explain the emergence of a particular conﬁguration
by examining the traits of agents within the simulation. In this way the model
avoids the thorny methodological difﬁculties inherent to C3 models, as well as
the more complex (and hence potentially more intractable) agent properties of C2
models. While C3 models are perhaps more useful to the social science modeller,
due to the possibility of producing agents that determine their own interactions
within the simulation environment, constraints such as those imposed by Schelling
help to maintain tractability. This tractability does however come at the expense of
increased self-organisation within the model.
7.6 Lessons from Schelling
7.6.1 Frameworks: Varying in Usefulness
Having tied Schelling’s work into each of our previously-discussed modelling
frameworks, a certain trend becomes apparent in the placement of Schelling’s model
within each theoretical construction. The simplicity of Schelling’s model places it
toward the extreme ends of each framework: it has an easily-deﬁned theoretical
backstory; lies well within the range of tractability under Silverman and Bullock;
and falls ﬁrmly within the C1 category described by Cederman.
While Cederman’s categorisation may help us to understand the aims and goals
of a Schelling-type model, such ideas are already apparent due to the theoretical
backstory underlying the model (which in turn places the model in good stead
according our modiﬁed Levinsian framework). The pragmatic considerations of
the model itself, as in whether it is amenable to analysis, are more important in
driving our declaration of Schelling as a useful model. After all, a very ambitious
and completely incomprehensible model could quite easily fall into Cederman’s
C3 category; however, its impenetrable nature would be exposed to much greater
7.6 Lessons from Schelling 133
criticism under the more pragmatic views of Levins and our revision of Levins
presented in Chap. 4.
7.6.2 Tractability: A Useful Constraint
As described earlier, Schelling’s model beneﬁts in several ways from its notably
simple construction. Referring back to our revised Levinsian framework, the general
concern of tractability is molliﬁed by the model’s inherent simplicity. Given that the
model can produce a visual demonstration of a segregation phenomenon, the job
of the analyst is made much easier; the qualitative resemblance of that result to an
overview of a segregated, real-world neigbourhood already lends credence to the
results implied by Schelling’s model.
Perhaps more interestingly, the abstract nature of the model also makes further
analysis less enticing for those seeking harder statistics. While the model does
represent agents moving in space in reaction to stimuli, they do so as abstract units
during time and space intervals that bear no set relation to real-world time and space.
Fundamentally, the model seeks only to demonstrate the effects of these agents’
micromotives, and the effects of those micromotives on the question of interest; in
that sense an in-depth analysis of the speed of segregation through the model’s space
and similar measures, while interesting, are not necessary for Schelling to illustrate
the importance of individual motives in residential segregation. As will become
evident in the following section, Schelling’s ideas regarding modelling methodology
drove him to construct his model in this way to maintain both tractability and
transparency.
7.6.3 Backstory: Providing a Basis
Silverman and Bullock’s concept of the importance of a theoretical ‘backstory’ for
any modelling enterprise (Silverman and Bullock 2004) seems supported by the
success of Schelling’s work. His approach to modelling social micromotives derived
from a theoretical backstory which takes in several important points:
1) Within a given research discipline, there are non-obvious analytical truths which may
be discovered by means which do not include standard empirical observation (speciﬁcally
mathematical or computational modelling in this case).
In the case of Schelling’s residential segregation model, the non-obvious result
of the interaction of his tolerant agents is that even high tolerance levels still lead to
segregation; one should note in this case that Schelling’s result was not only nonobvious
in the context of the model, but was also non-obvious to those who studied
the segregation phenomenon empirically.
134 7 Schelling’s Model: A Success for Simplicity
2) The search for general models of phenomena can lead to the important discovery of
general patterns of behaviour which then become evident in examples across disciplines.
Schelling uses ‘critical mass’ models as an example (tipping models being a subclass
of these), arguing that they have proven to be useful in ‘epidemiology, fashion,
survival and extinction of species, language systems, racial integration, jay-walking,
panic behaviour, and political movements’ (Schelling 1978, p. 89). The explosion of
interest in tipping models following Schelling’s success with residential segregation
indicates that such inter-disciplinary usefulness may indeed be a crucial element to
the success of his model.
3) Modellers should seek to demonstrate phenomena in a simple, transparent way. As he
states, a model ‘can be a precise and economical statement of a set of relationships that are
sufﬁcient to produce the phenomena in question, or, a model can be an actual biological,
mechanical, or social system that embodies the relationships in an especially transparent
way, producing the phenomena as an obvious consequence of these relationships’ (Schelling
1978, p. 87).
Schelling’s own models conform to this ideal, utilising very simple agents
governed by very simple rules to illustrate the importance of individual behaviours
in social systems. Zhang’s analysis of Schelling-type models using recent advances
in statistical mechanics shows one of the beneﬁts of this transparency in a modelling
project (Zhang 2004).
7.6.4 Artiﬁciality: When it Matters
The importance of artiﬁciality within a simulation methodology as espoused by
Silverman and Bullock (2004) is especially crucial to an evaluation of Schellingtype
models. Schelling himself posits that models can illuminate important patterns
in a system of interest without requiring recourse to empirical observation (Schelling
1978); in this fashion his work suggests Silverman and Bullock’s Artiﬁcial1
and
Artiﬁcial2
distinction as a sensible path for models to take. However, he further
clariﬁes this idea by proposing that such models must remain transparent and easily
analysable, displaying a clear interaction which leads to the appropriate results in
the system of interest; an overly-complex Artiﬁcial1
model, in his view, cannot
provide that clear link between the forces at play within the model and the resultant
interesting behaviour.
Further, Schelling’s two-part deﬁnition of models goes on to describe the
potential for using an actual biological or social system in a limited context to
illustrate similar points (Schelling 1978); this statement implies that Artiﬁcial2
models, which would be a man-made incidence of these natural behaviours, may be
able to provide that simplicity and transparency that empirical observation of such
complex systems cannot provide. In one sense, then, Schelling appears to dismiss
the question of artiﬁciality in preference to the modellers motivations: in order to
display the effect of micromotives or emergent behaviour, the model must display
7.6 Lessons from Schelling 135
a clear relationship between the resultant behaviour and the contributing factors
alleged to create that behaviour, and whether that model then embodies Artiﬁcial1
or Artiﬁcial2
is not necessarily of any consequence.
7.6.5 The Practical Advantages of Simplicity
Schelling also demonstrates the practical usefulness of creating a simple model of
a phenomenon. So far we have seen how this simplicity allows us to avoid some of
the methodological pitfalls that can trouble those who choose to utilise agent-based
models, and likewise it is easy to demonstrate how this same property can help the
modeller in more pragmatic ways.
Firstly, such simplicity not only allows for higher tractability, but also much
simpler implementation. In the case of Schelling’s model, numerous implementations
of the model exist in a large variety of programming languages. Writing the
code for such a simulation is almost trivial compared to more complex simulations;
indeed, some pre-packaged simulation codebases such as RePast (designed for the
social scientist) can be utilised to produce a Schelling-type model in only a few
lines. Beyond simply the time savings of these ease of implementation, the simple
construction of Schelling’s model vastly reduces the amount of time spent tweaking
a simulation’s parameters. In more complex simulations, such as an evolutionary
model, parameters such as mutation rates can have unexpected impacts on the
simulation results, and ﬁnding appropriate values for those parameters can be time-
consuming.
Secondly, starting from such a simple base allows for much greater extendability.
With the Schelling model being so easily implemented in code, extending some
elements of that model becomes very easy for the researcher. For example,
alterations in the number of agent types, the complexity of the agents themselves, or
the set of rules governing the agents are easy to create. In addition, the simple nature
of the initial model means it is also easy to change one aspect of the model and see
clearly the results of that change; in a more complex formulation, changing one
element of the simulation may produce unexpected changes, and complex elements
in other areas of the simulation could be affected in ways that produce unanticipated
results.
Finally, the modeller beneﬁts from potentially a much larger impact of the
simulation when it is simple to implement. For example, we saw previously how
Zhang was able to probe the properties of the entire space of Schelling-type models
(Zhang 2004). If Schelling’s model were too complex, this would be an impossible
task. Instead, due to its simplicity, dozens of interested researchers could implement
Schelling’s model for themselves with little effort, see the results for themselves,
and then modify the model or introduce new ideas almost immediately. Such ease
of replication and modiﬁcation almost certainly helped Schelling to reach such a
high level of impact from his initial model; the simplicity of the model essentially
lowers the barrier of entry for interested parties to join the discussion.
136 7 Schelling’s Model: A Success for Simplicity
7.7 Schelling vs Doran and Axelrod
7.7.1 Doran’s View: Supportive of Schelling?
Doran’s views of agent-based models in social science (Doran 2000) proposes
that such models can provide a means to generate new ‘world histories,’ or
artiﬁcial explorations of versions of human society that may have come into being
given different circumstances. While Schelling’s model could be construed as an
‘artiﬁcial society’ of the simplest order, the model is oriented less toward reaching
grand conclusions about the structure of society as a whole and more toward a
demonstration of the low-level factors in a society which may produce one particular
phenomenon.
With this in mind, Doran’s further concerns about the undeﬁned role of agents
in social simulation also seem of particular import. Doran argues that agents in
social simulation should be deﬁned on a computational basis to allow those agents
to develop emergent behaviour in the same way as other aspects of the simulation.
Schelling’s model incorporates agents in a most abstract way; each individual in the
model makes only a single decision at any given time-step, based on only a single
rule. Given this simplicity, could we argue that these agents are sufﬁciently advanced
to bring us a greater understanding of the residential segregation problem? If not,
how might Doran’s view inform a more rigorous version of Schelling’s original
vision?
While Schelling’s model is indeed oriented toward a speciﬁc social problem in
an intensely simpliﬁed form, in a sense he is providing a minimal ‘artiﬁcial society’
as Doran describes this methodology. Schelling is able to create alternate ‘world
histories’ for the limited two-dimensional space inhabited by his simple agents; he
can quite easily run and re-run his simulation with different tolerance values for
the agents, and examine the resulting patterns of settlement following each change.
For example, he could determine the result of populating a world with completely
tolerant agents, completely intolerant agents, and all variations in between.
With regard to Doran’s concerns regarding the roles of agents in social simulation,
Schelling’s model suffers more under scrutiny. Despite the simplicity of the
model itself, the agents are built to a pre-deﬁned role: each agent is assumed to be
searching for a place to settle, with its preference for a destination hinging upon
the appearance of its neighbours. This presumes that individuals in a society would
prefer to remain settled, rather than move continuously, and that all individuals will
display a primary preference based upon the characteristics of its neighbours; both
of these assumptions have been placed into the model by Schelling’s own conceptual
model, rather than having those agent properties emerge from the simulation itself.
One could imagine constructing a Luhmannian scenario in which agents are
given only the most base properties: perhaps a means of communication, an ability
to form preferences based on interactions, and the ability to react to its neighbours.
Might these agents, based upon interactions with neighbours of different characteristics,
form preferences independent of the experimenter’s expectations? If so,
then these preferences would emerge from the simulation along with the resultant
7.8 Schelling and Social Simulation: General Criticisms 137
agent conﬁgurations in the simulated world, making the agents more independent
of the experimenter’s biases, though of course never entirely independent of bias.
Such a set-up would certainly alleviate Doran’s concerns about pre-deﬁned agents
and theoretical biases, but whether the results would be more or less useful to the
community than Schelling’s original is still debatable.
7.7.2 Schelling vs Axelrod: Heuristic Modelling
Robert Axelrod and Leigh Tesfatsion’s introduction to agent-based modelling
for social scientists, discussed in the previous chapter, describes four main goals
of social simulation models: empirical understanding, normative understanding,
heuristics, and methodological advancement (Axelrod and Tesfatsion 2005).
Schelling’s model seems to fall most readily into the heuristic category, seeking
as it does a fundamental insight into the mechanisms underlying the phenomenon
of residential segregation.
Axelrod’s view, unlike Doran’s, stops short of examining speciﬁc methodological
difﬁculties in social simulation modelling. Instead, he develops these four general
classiﬁcations of model types, placing agent-based modelling into a framework
oriented more toward empirical study. Given that three of Axelrod’s four categories
are directly concerned with empirical uses of agent-based models, this framework
offers little guidance as to the appropriate use of models like Schelling’s.
Of course, as indicated by the discussion of Axelrod’s categorisations in the
previous chapter, our analysis thus far has indicated a number of theoretical
difﬁculties with this empirical approach to social simulation. With this in mind the
fact that Schelling lies outside the prominent focus of Axelrod’s approach is not
particularly surprising. Along with Doran, Luhmann, and Silverman and Bryden
(Silverman and Bryden 2007), Schelling’s model is more appropriate in the context
of a more general and abstracted modelling perspective, one which seeks general
understanding of social phenomena rather than data relating to speciﬁc aspects of
society.
7.8 Schelling and Social Simulation: General Criticisms
7.8.1 Lack of ‘Real’ Data
While Schelling thus far has held up well under the scrutiny of several major
theoretical frameworks regarding simulation models, there are still concerns levelled
generally at social simulation which must be addressed. The ﬁrst, and potentially
the most troublesome, is the lack of ‘real’ data attributed to models within social
science, and the resultant disconnection between these simulations and empirical
social science.
138 7 Schelling’s Model: A Success for Simplicity
Schelling quite clearly falls afoul of this methodological sticking point. There
is no aspect of the chequerboard model which is based upon empirical data:
the chequerboard itself is designed arbitrarily, with no real-world context or
comparison; the agents are given a tolerance threshold by the experimenter, not one
based upon any empirical study; and there is no representation of the nuances of
human residential areas, such as buildings, other residents, or other interacting social
factors that may effect the residential segregation phenomenon. In other words,
Schelling’s model is very much an abstraction with no real basis in empirical social
science.
However, given the context of Schelling’s work, the abstraction is entirely
justiﬁable. Had Schelling proposed to understand residential segregation in one
particular circumstance, then produced such an abstract model, then he would
have been reaching for conclusions far beyond the scope of the model itself. His
question instead was much more broad: can we illustrate the effect of individual
‘micromotives’ on an otherwise difﬁcult-to-explain social phenomena? His model
answers this question without the need for speciﬁc ties to empirical data-sets, and
indeed ‘real’ data would most likely dilute the strength of his result in this context.
7.8.2 Difﬁculties of Social Theory
Revisiting Kluver and Stoica once more, we recall their assertion that social theory
does not beneﬁt from being easily divisible into interacting hierarchical structures as
in other ﬁelds, such as biology (Klüver et al. 2003). Given that a social system will
encompass potentially millions of individuals, each interacting on multiple levels,
while higher-level social structures impose differing varieties of social order, the
end behaviour of a society through all of these factors can be exceedingly complex.
This inherent difﬁculty in social science makes the prospect of empirically-based
simulation seem ever more distant.
However, Schelling once more illuminates the beneﬁts of a more abstract
approach to examining social systems. Schelling’s model suffers little from the
problem of interacting social structures, as the model itself involves only a set of
agents interacting in a single space: there are no social structures; no imposed social
order in the system; and only a single type of interaction between agents. The model
thus escapes this difﬁculty by quite simply eliminating any social structures; without
these multiple interacting structures to confound analysis, the model’s result remains
clear.
7.8.3 Schelling and the Luhmannian Perspective
With the abstraction and simplicity of Schelling’s model allowing it to escape
from the methodological traps common to most social simulation endeavours,
7.8 Schelling and Social Simulation: General Criticisms 139
we are left with an interesting perspective on our earlier proposed method for
developing fundamental social theory. Similar to Schelling’s tipping model, the
Luhmannian modelling perspective would construct an agent-based model bearing
as few assumptions as possible, allowing the resultant conﬁgurations of agents and
model properties to emerge of their own accord.
In essence, Schelling’s model takes the Luhmannian approach and boils it down
to an exploration of a single aspect of human society. While Luhmann asks what
drives humanity to develop social order (Luhmann 1995), Schelling restricts his
domain to only residential segregation, asking whether individual motives can drive
agents to separate themselves from one another. While Luhmann condenses the
whole of human interaction down to social developments linked to an iterated
process based on our ‘expectation-expectations,’ Schelling similarly condenses the
puzzling behaviour of human segregation down to a series of simple individual
decision-making steps.
Schelling’s success, then, gives the Luhmannian approach a further emphasis.
While the investigation of the overall origins and development of the human social
order is certainly a much more complex endeavour than that of investigating a
single social phenomenon a la Schelling, this residential segregation model provides
an insight into the beneﬁts of the process. Schelling wrote of the importance of
demonstrating the relationships between model properties transparently (Schelling
1978), and with a Luhmannian model the same approach is necessary.
7.8.4 Ramiﬁcations for Social Simulation and Theory
Having established the importance of Schelling’s perspective, and the links between
his modelling paradigm and our proposed means for developing fundamental social
theory, a further re-evaluation of Schelling’s impact on social simulation is required.
We have seen the import of this model’s simplicity and transparency, and even how
its inherent abstraction enables the model to draw intriguing conclusions within the
larger context of general social theory. Does this imply that the empirically-based
approach of Cederman, Axelrod and others is a scientiﬁc dead-end?
Perhaps not: certainly as the ﬁeld of geographic information systems (GIS)
continues to advance, and real-time data collection within social science becomes
more wide-spread and rigorous, then the introduction of real and current data
into social simulation becomes an interesting possibility. This after all is a central
criticism of social simulation of this type: real data is hard to come by, and that
data which is available is often limited in scope or out of date. When this obstacle
disappears, then simulations more closely linked to data produced by real human
society becomes more viable.
However, the other fundamental concerns related to social simulation still
remain. Doran’s concerns regarding the lack of deﬁnition of the roles of agents
in social simulation (Doran 2000) remain important, and as integration with real
data becomes vital to a simulation that concern becomes ever more central. The
140 7 Schelling’s Model: A Success for Simplicity
question of how to develop cognitive agents with an appropriate set of constraints
and assumptions to produce parsimonious results is not one that will be answered
quickly.
Similarly, the inherent abstraction of social processes and structures necessary
in a computer simulation could be problematic in a simulation designed to produce
empirically-relevant results. Axelrod and Tesfatsion (2005) proposes social simulations
that may inform public policy (his ‘normative understanding’ category of
social simulations), and while this is an enticing prospect, we have already seen the
troubling pitfalls the researcher can encounter in this approach (see Chaps. 4 and 5).
This reinforces the importance of Schelling’s model as deconstructed in the
preceding analysis. While initially appearing simplistic, the theoretical implications
of Schelling’s work are rather more complex. The importance of this model in social
science despite its simplicity, complete abstraction, and lack of empirical data shows
the potential of social simulation to stimulate social theory-building. Schelling’s
model stimulated the ﬁeld to view the importance of individual ‘micromotives’ in a
new fashion; a more sweeping model portraying the emergence of a social order in
a similar way could have an equally stimulating effect on social theory.
7.9 The Final Hurdle: Social Explanation
7.9.1 Schelling’s Model and the Explanation Problem
As described in Chap. 5, there is some debate over the explanatory power of computer
simulation within the social sciences. While the predominant idea within such
endeavours is that of emergence, or the development of higher-level organisation in
a system given interacting lower-level components, some theorists argue that social
systems do not produce higher-level organisation in this way (Sawyer 2002, 2003,
2004).
Sawyer’s development of the idea of non-reductive individualism, in which
certain properties of a society cannot be reduced to the actions of individual actors
in that society, does pose a fundamental problem for agent-based modellers. If
agent-based models proceed on the assumption that individual-level interactions
can produce the higher-level functions of a social order purely through those interactions,
then such models may be missing a vital piece of the explanatory puzzle
within social science. In this respect individual-based models need a theoretical
underpinning in which emergence is a valid means for the development of highlevel
social structures.
Schelling’s model focuses entirely on the actions of individual agents, and the
impact of their individual residential preferences on the racial make-up of a given
neighbourhood. In this sense the model does seek to demonstrate the emergence of
a higher-level conﬁguration from the actions of individuals, which according to this
view of social explanation is problematic.
7.9 The Final Hurdle: Social Explanation 141
However, Macy and Miller’s take on this social explanation problem (Macy and
Willer 2002) would allow for Schelling-type models, as the model focuses purely on
a situation for which there is no central coordination of the behaviour in question.
The agents in Schelling’s chequerboard world do act individually, but the results he
sought were not a higher-level social organisation or function, but instead merely
a particular conﬁguration of those individuals at the end of a simulation run. Even
in Sawyer’s less forgiving view, Schelling’s simulation does not strive to explain a
larger-scale structure that might be dubbed irreducible.
7.9.2 Implications for Luhmannian Social Explanation
Our earlier discussions of Niklas Luhmann’s ideas regarding the development of
the human social order presented a means for applying these ideas to agent-based
models designed to examine the fundamentals of society. By applying Luhmannian
ideas of extremely basic social interactions that lead to the formation of a higher
social structure, a modeller may be able to remove the theoretical assumptions often
grafted into a model through the design of restrictive constraints on agents within
that model.
However, our examinations of the difﬁculty of social explanation remain problematic
for a Luhmannian approach. If certain high-level aspects of human society
depend on functions which are irreducible to the properties of individuals within
that society, then even the most cleverly-designed agent-based model would be
unable to provide a complete explanation for the functioning of society. This places
a fundamental limit on the explanatory power of the Luhmannian approach, barring
us explaining the social order in its entirety.
Perhaps Sawyer’s comparisons with non-reductive materialism within the philosophy
of mind may provide a solution (Sawyer 2004). As he notes, ‘the science
of mind is autonomous from the science of neurons,’ alluding to the disconnect
between psychology and neuroscience (Sawyer 2002). Indeed, within the study of
mind there are conscious and unconscious phenomena which are irreducible in
any straightforward way to the functioning of the underlying neurons; after all,
psychologists still struggle to explain how neuronal ﬁrings lead to the subjective
experience of individual consciousness.
However, very few psychologists still adhere to the concept of dualism: the idea
that conscious experience is a separate entity from the physical brain itself. Sawyer
clearly does not, as is evident from his discussion of non-reductive materialism. In
that case, while we may not be able to draw a direct relation between conscious
phenomena and brain activity, the relation nonetheless exists; neurons are the cause
of conscious phenomena, merely in a way we cannot understand straightforwardly.
In the same sense, individuals will be the fundamental cause of large-scale social
phenomena, whether those phenomena display clearly evident relationships or not.
If we accept that a sufﬁciently advanced and appropriately constructed computer
may be able to achieve consciousness, then surely a similarly advanced social
142 7 Schelling’s Model: A Success for Simplicity
simulation could allow sophisticated structures to emerge as well? In either case, the
functioning of those individual units would be very difﬁcult to relate to the emerging
high-level phenomena but developing artiﬁcial (and hence de-constructable) models
of both these processes could provide a unique insight.
Thus, Sawyer may very well be correct, and understanding the relation between
high-level social structures or institutions to individuals in society may be difﬁcult,
or even impossible. But this is not to say that individuals, even in a model, could
not produce such phenomena; nor that study of such models would not produce any
insight. The Luhmannian modelling perspective aims to discover the roots of human
society, and if those individual roots produce irreducibly complex behaviour, those
models have surely functioned quite well indeed.
7.10 Summary and Conclusions
In this chapter we have examined one particular example of social simulation,
Schelling’s chequerboard model, in light of the various theoretical concerns raised
thus far. Schelling’s model was a marked success in the social sciences, producing
an endless stream of related works as the idea of social phenomena emerging from
the actions of individuals grew in the social sciences as it did in artiﬁcial life.
The simplicity and transparency of the model allowed it to have a stronger impact
than many more complex models. Along with being easily reproducible, Schelling’s
results were restricted to a very speciﬁc question: can individual housing preferences
drive the mysterious process of residential segregation? The answer, demonstrated
by the starkly-separated patterns of white and black squares so prominent in the
literature, appeared to be yes.
This approach, while falling within the remit of the theoretical frameworks laid
out by other social simulators, also provides an insight into new paths for producing
social theory. While the use of social simulations for empirical study is enticing, and
potentially both useful and lucrative, the methodological and theoretical difﬁculties
involved in such an approach are many and complex. In contrast, a simple model
with few inherent assumptions can offer a more basic and general description of the
origin of various social phenomena.
Schelling’s model also provides a view into the potential beneﬁts of the proposed
Luhmannian modelling approach for the social sciences. Like Luhmann, Schelling’s
model took very basic interactions as a basis for producing more complex social
behaviour; Luhmann takes a very similar approach, but on a much larger canvas. In
this respect Schelling shows that the Luhmannian approach could allow social theorists
to develop ideas regarding the social order that may have large ramiﬁcations
for social science as a whole.
References 143
Of course, the problem of social explanation still looms as large for Luhmann as
it does for Cederman or Axelrod. The debate over emergent phenomena in social
science is unlikely to subside, and as such the results of such simulations may
always be disputed on some level. However, even if we accept that some aspects of
a Luhmannian simulation may not provide complete explanatory power, the results
could still be revolutionary for social theory.
References
Axelrod, R., & Tesfatsion, L. (2005). A beginner’s guide to agent-based modelling in the social
sciences. In L. Tesfatsion & K. L. Judd (Eds.), Handbook of computational economics, vol. 2:
Agent-based computational economics, p. Appendix A.
Cederman, L. E. (2001). Agent-based modeling in political science. The Political Methodologist,
10, 16–22.
Clark, W. A. V. (1991). Residential preferences and neighborhood racial segregation: A test of the
Schelling segregation model. Demography, 28, 1–19.
Davis, J. A., & Smith, T. W. (1999). General social surveys, 1972-1998: Cumulative codebook.
Chicago: National Opinion Research Center.
Doran, J. E. (2000). Trajectories to complexity in artiﬁcial societies. In A. Kohler & G. Gumerman
(Eds.), Dynamics in human and primate societies. New York: Oxford University Press.
Epstein, J. M., & Axtell, R. L. (1996). Growing artiﬁcial societies: Social science from the bottom
up. Cambridge: MIT Press.
Freeman, L. (2000). Minority housing segregation: A test of three perspectives. Journal of Urban
Affairs, 22, 15–35.
Klüver, J., Stoica, C., & Schmidt, J. (2003). Formal models, social theory and computer
simulations: Some methodical reﬂections. Journal of Artiﬁcial Societies and Social Simulation,
6(2), 8.
Laitin, D. D. (1994). The tower of babel as a coordination game: Political linguistics in Ghana.
American Political Science Review, 88, 622–634.
Laitin, D. D. (1998). The cultural identities of a European state. Politics and Society, 25(3),
277–302.
Luhmann, N. (1995). Social systems. Stanford: Stanford University Press.
Macy, M. W., & Willer, R. (2002). From factors to actors: Computational sociology and agentbased
modeling. Annual Review of Sociology, 28, 143–166.
Massey, D. S., Condran, G. A., & Denton, N. (1985). The effects of residential segregation on
black social and economic well-being. Social Factors, 66(1), 29–56.
Sander, R. H. (1998). Housing segregation and housing integration: The diverging paths of Urban
America. Miami Law Review, 52, 977–(1010).
Sander, R., Schreiber, D., & Doherty, J. (2000). Empirically testing a computational model:
The example of housing segregation. In D. Sallach & T. Wolsko (Eds.), Proceedings of the
Workshop on Simulation of Social Agents: Architectures and Institutions (pp. 108–115).
Sawyer, R. K. (2002). Nonreductive individualism: Part I: Supervenience and wild disjunction.
Philosophy of the Social Sciences, 32, 537–561.
Sawyer, R. K. (2003). Artiﬁcial societies: Multi-agent systems and the micro-macro link in
sociological theory. Sociological Methods and Research, 31(3), 37–75.
Sawyer, R. K. (2004). Social explanation and computational simulation. Philosophical Explorations,
7(3), 219–231.
144 7 Schelling’s Model: A Success for Simplicity
Schelling, T. C. (1971). Dynamic models of segregation. Journal of Mathematical Sociology, 1,
143–186.
Schelling, T. C. (1978). Micromotives and macrobehavior. New York/London: W. W. Norton.
Silverman, E., & Bullock, S. (2004). Empiricism and artiﬁcial life. In J. Pollack, M. Bedau,
P. Husbands, T. Ikegami, & R. A. Watson, (Eds.), Artiﬁcial Life IX: Proceedings of the Ninth
International Conference on the Simulation and Synthesis of Living Systems. Cambridge: MIT
Press/Bradford Books.
Silverman, E., & Bryden, J. (2007). From artiﬁcial societies to new social science theory. In
F. Almeida e Costa, L. M. Rocha, E. Costa, I. Harvey, & A. Coutinho (Eds.), Advances
in Artiﬁcial Life, Proceedings of the 9th European Conference, ECAL 2007 (pp. 645–654).
Berlin/Heidelberg: Springer.
Taylor, P. (2000). Socio-ecological webs and sites of sociality: Levins’ strategy of model-building
revisited. Biology and Philosophy, 15(2):197–210.
Vickerstaff, R. J., & Di Paolo, E. A. (2005) Evolving neural models of path integration. Journal of
Experimental Biology, 208, 3349–3366.
Zhang, J. (2004). Residential segregation in an all-integrationist world. Journal of Economic
Behavior and Organization, 54(4), 533–550.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
Chapter 8
Conclusions
8.1 Overview
The previous chapter focused on demonstrating our developed modelling
frameworks in the context of one particular case study: Schelling’s well-known
residential segregation model (Schelling 1978). Despite this model’s inherent
simplicity, the results were seen as signiﬁcant within social science. Our analysis
of the methodological and theoretical underpinnings of Schelling’s model provided
some insight into how his simple model of societal micromotives became so
inﬂuential.
However, Schelling need not be the only example of a successful computational
modelling endeavour. While Schelling does fare well when viewed in the context
of a number of different modelling frameworks, there are other examples of
computational research which can provide useful results through varying means.
By revisiting our central bird-migration example, and viewing each of our
developed modelling frameworks from the ﬁrst two sections of this text in the
light of our analysis of Schelling, we can put together a more comprehensive set
of ideas regarding the limitations of computational modelling. The effect of such
ideas on substantive modelling works are also important to discuss; with these
methodological frameworks in place, research utilising computational modelling
will necessarily need to adapt to these restrictions.
8.2 Lessons from Alife: Backstory and Empiricism
8.2.1 Backstory in Alife
Our analysis of Alife in Chap. 3 focused ﬁrst on the distinction between two
proposed varieties of artiﬁciality: Artiﬁcial1
, a man-made example of something
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_8
145
146 8 Conclusions
natural; and Artiﬁcial2
, something made to resemble something else. This
distinction proved most important in the contrast between strong Alife and weak
Alife: strong Alife seeks to create systems that are Artiﬁcial1
in nature, while weak
Alife seeks only Artiﬁcial2
systems.
Along with the drive to create digital examples of life, members of the Alife
community have sought to use their simulations as a means for providing new
empirical data points useful for studying life. Such an empirical goal is far from
trivial, and requires a cohesive theoretical backstory to provide a basis for allowing
that data to be used as such. As described in Sect. 3.6, our PSS Hypothesis for Life
provides one example of such a perspective: if one accepts that life is an information
ecology, then a suitably-programmed computer can also instantiate such a system.
However, a backstory of this nature requires additional philosophical baggage.
While a researcher may utilise this PSS Hypothesis to justify investigations into a
digital living system, that system still exists only in a virtual form. The researcher
becomes a digital ecologist, studying the output of the simulation for its own sake.
8.2.2 Schelling’s Avoidance of the Issue
Our analysis of Schelling provides a means for escaping these philosophical
conundrums. Rather than proposing a model which is an Artiﬁcial1
instantiation
of human society in order to explore residential segregation, he produces a simple
model which only illustrates an example of his concept of micromotives and their
effect upon society (Schelling 1978).
A version of our bird-migration model could take advantage of simplicity in a
similar way. If we constructed a model which presented each bird as only a simple
entity in a grid-world, with simple rules which drives the movements of those
birds, we may be able to present an example of how singular micromotives could
drive birds to shift from one location to another (say by moving according to food
distribution or other factors that may contribute to our bird agents’ well-being under
the given rule set). In such a formulation the question of Artiﬁcial1
versus Artiﬁcial2
is unimportant; the model is obviously Artiﬁcial2
in nature, and no claim is made to
be creating real, digital bird equivalents. The model simply seeks to show the impact
of Schelling-esque micromotives in bird migration.
However, while avoiding the problem of artiﬁciality can be advantageous to the
modeller, the question of relating that model to empirical data still remains. Without
claiming that the agents within our bird model are instantiations of digital life, we
cannot be said to collect empirical data on those agents. Might we remain in danger
of constructing an artiﬁcial world which we proceed to study as a separate entity
from the real systems of interest?
8.3 The Lure of Artiﬁcial Worlds 147
8.3 The Lure of Artiﬁcial Worlds
8.3.1 Levinsian Modelling
Our examination of the more pragmatic concerns of modellers through the lens
of population biology provided some additional considerations for the simulation
researcher. Levins’ (1966) developed a framework in which three dimensions of
generality, realism and precision, important to any given model of a natural system,
must be balanced effectively to produce a useful model. He posits that a modeller
can focus on two of these modelling dimensions at a time, but only at the expense
of the third; this leads him to describe three possible varieties of models, which we
denoted L1, L2 and L3 (see Table 4.1).
As noted in Chap. 4, however, applying this framework to certain computational
models can be difﬁcult. If our bird migration model uses a highly-simpliﬁed set of
evolving agents to represent birds, and places those agents in a simpliﬁed world with
abstracted environmental elements to affect those agents, how might we characterise
that model under Levins’ framework? Certainly precision cannot apply, as there is
no attempt in this formulation to match these agents with any particular real-world
species. Nor does realism seem to apply, as the highly-abstracted nature of the model
divorces it from the natural systems it seeks to model. Can the model be referred to
as simply general in character, or even then is the model seeking insights which may
end up generalising in a different fashion than in other varieties of models?
8.3.2 Levins and Artiﬁcial Worlds
This difﬁculty leads us to the question alluded to in the previous section: do we
risk becoming mired in the study of artiﬁcial worlds for their own sake? Certainly
our proposed simulation above would suffer from a lack of available empirical
data to use in the model. For example, by using evolving agents to represent
the birds we can model the progression of this abstracted species over time, but
empirically-collected data following a species through its evolution is not available
to provide guidance for this aspect of the model. The modeller can proclaim a
serious advantage in being able to model something which is impossible to study
empirically through only a relatively modest investment of programming time and
processing time, but likewise, the lack of relevance to biology becomes ever more
acute (once again, see Webb 2009 for a discussion of this issue in relation to Beer’s
CTRNN model).
In a certain way, however, this lack of empirical relevance is an attractive
feature for the simulation researcher. Replacing the complexities of the real world
with simpliﬁed abstractions in a wholly artiﬁcial world not only makes model
148 8 Conclusions
construction potentially easier, but even avoids the practical difﬁculties often
associated with traditional empirical methods. As a consequence, our bird model
need not be tied down by Levins’ pragmatic modelling concerns, and balancing his
three dimensions of model-building suddenly appears much easier.
Of course, as discussed in Chap. 4, such artiﬁcial worlds create further methodological
difﬁculties of their own. While such a model may appear to avoid Levins’
concerns and thus produce more complete models of phenomena, conﬁning that
model to an artiﬁcial world creates a strong separation from the empirical data that
can inform other varieties of models. The modeller is thus in danger of studying the
model as a separate entity from the systems on which it is based.
Schelling avoids this difﬁculty by positioning his model as a means to illustrate
the importance of micromotives in social behaviour (Schelling 1971, 1978). While
his model does relate to a real system, and it does take place in a highly idealised
artiﬁcial world, within this context the model does not need a strong relationship to
empirical data. Schelling instead strives for transparency maximising tractability by
creating an abstract, easily computable model. The question of relating the model
to the real system thus becomes simpliﬁed: can the model illustrate the potential for
individual micromotives to exert a great inﬂuence on a society? The answer, for most
social scientists, appears to be yes, despite the model’s artiﬁciality and simplicity.
8.4 Modelling in Alife: Thoughts and Conclusions
8.4.1 Lessons from Schelling
As seen in the analysis above, there are a multitude of considerations relevant
to the ALife modeller. From crafting a suitable theoretical backstory to avoiding
the difﬁculties of artiﬁcial worlds, methodological problems are hard to avoid
completely. Schelling provides some insight into how to approach these issues.
The simplicity of the model allows for a coherent theoretical backstory, focusing
only on the possible effects of micro-motives on the larger system. Meanwhile, the
model’s transparency maintains tractability, though this brings with it a high level
of artiﬁciality in the model.
As we see with Schelling, however, this artiﬁciality is not necessarily the
problem. The larger concern is the intent with which the model is constructed.
A strong ALife practitioner who seeks to create digital life needs to demonstrate
Artiﬁcial1
, and in this case he would presumably require a much higher degree of
complexity than in Schelling’s model; even with the PSS Hypothesis for Life as a
backstory, a grid-world of simplistic homogeneous agents could hardly be said to
compose an information ecology.
8.4 Modelling in Alife: Thoughts and Conclusions 149
8.4.2 The Power of Simplicity
The circumstance of being driven away from simplicity toward complexity in this
search for Artiﬁcial1
also creates a much more difﬁcult pragmatic situation for the
modeller. As noted in Chap. 7, simple models like Schelling permit the modeller to
spend far more time theorising than tweaking. For our bird researcher, a situation in
which a few simple lines of encode provide the agent’s behaviour is preferable to one
in which each agent contains complex neural network elements, for example. In the
ﬁrst case, the researcher can write the code and run the simulation quickly, then take
the necessary time to examine results, run alternative versions of the simulation, and
so forth. In the second case, the researcher could spend far more time tweaking the
individual parameters of the agent structures: what sort of sensors might each agent
have? How do the neural networks control movement and receive input from the
environment? How many neurons are there, and what type are they? The questions
become many and varied for the researcher using complex agents, and each day
spent tweaking those agent parameters is one less day spent pondering the results of
the simulation.
As seen above, Schelling’s model provides one example of a means to avoid
these pragmatic issues. His model is of such simplicity that writing a version of his
simulation takes only a few lines of code compared to more complex simulations.
However, clearly other types of models can maintain similar simplicity; Beer,
for example, touts his CTRNN-based agents as displaying ‘minimally-cognitive
behaviour’ (Beer 2003a,b). Of course, Beer’s analysis of that minimally-cognitive
behaviour is extremely detailed and time-consuming, and may indicate that such
analysis is impractical for agents of that type even of such relative simplicity.
Nonetheless, Beer does demonstrate that relatively simple and analysable neural
models are not outside the realm of possibility.
8.4.3 The Scope of Models
With all of these points in mind, we see that Schelling-type models are hardly
the only permissible variety under these frameworks; in fact, a large number of
models may display appropriate theoretical back-stories while remaining tractable.
Schelling does, however, illuminate the central concerns tied to these modelling
frameworks: the importance of theoretical backstory, artiﬁciality, tractability and
simplicity, and the scope of the model.
The ﬁnal element, scope, is an important one to note. Schelling succeeds not only
by having a cogent backstory, using its artiﬁciality appropriately, and remaining
tractable, but also by limiting its approach: the model aims only to illustrate the
importance of micro-motives in a social system, not produce empirical data. In the
same way, if our bird migration researcher chose to model the movements of the
birds from place to place simply for the purpose of realistic mimicry of their travels,
150 8 Conclusions
as in Reynolds’ ﬂocking boids (Reynolds 1987), he could do so with little theoretical
baggage. If he then chose to declare these mimicked movements as instances of real
ﬂocking, or as producing relevant empirical data, suddenly far more justiﬁcation is
required.
8.5 The Difﬁculties of Social Simulation
8.5.1 Social Simulation and the Backstory
While our earlier analysis of ALife provided some valuable insight into the limitations
of agent-based modelling techniques, particularly in contrast to traditional
mathematical models, these reﬁned frameworks developed in those chapters do
not translate simply to social simulation approaches. Within a ﬁeld such as social
science, the philosophical considerations we addressed in those frameworks grow
even more troublesome. Imagine that we have created our bird migration model, and
the in-depth construction of our programmed agents allows those agents to begin to
communicate and even form social structures of a sort. If our stated goal is to use this
model to investigate properties of human societies by providing a new, digital source
of empirical data, we must not only accept the PSS Hypothesis (presuming that only
living things may create a true society), but also related points. The researcher must
be prepared to accept that these digital beings, alive but unrelated in a conventional
sense to natural life, can create a real societal structure.
In addition, this societal structure will be further removed from real, natural
societies by its dependence on a form of ‘life-as-it-could-be.’ In that case, even if we
accept that this virtual community of birds can create a society of their own through
their interactions, how might we be able to relate that behaviour to the development
of real-world societies? If we accept Sawyer’s concerns regarding the non-reductive
individualist character of human society (Sawyer 2002, 2003, 2004), are we not
placing ourselves even further from a possible social explanation by basing theories
upon this digital instantiation of society? Perhaps the non-reductive characteristics
of our digital society differ completely from those displayed in human society. Once
again, we would be stuck studying the simulation for its own sake.
8.5.2 Social Simulation and Theory-Dependence
As discussed in Chap. 3, the ﬁeld of ALife may be considered fundamentally theorydependent:
the structure and function of a given simulation is directly connected to
the theoretical assumptions made to create that model. The framework discussed in
that chapter chose to deal with this issue by noting the inherent theory-dependence
of empirical science, and presenting ideas regarding the importance of theoretical
back-stories to any empirically-motivated endeavour.
8.6 Schelling’s Approach 151
With social simulation, however, an additional layer is added to this theorydependent
aspect of modelling. Not only do the agents and environment of the
simulation present problems of theory-dependence, but also the additional aspects of
social communication and behaviour that allows the model to address issues relevant
to society.
Further, these additional layers of complexity are not easily subdivided into a
hierarchy or other framework which may ease the construction of a computational
model (Klüver et al. 2003). The interdependencies between individual action and
societal effects means that abstract models will lose a great deal of detail in
these missing elements, and conversely that highly-detailed models which address
these complexities will veer toward intractability. In either case, theory-dependence
becomes a greater issue: abstract models will require strong assumptions to remove
these non-hierarchical complexities; and complex models will require incorporating
ideas regarding the interaction of individuals and society.
8.5.3 Social Simulation and Explanation
Even if one does manage to construct a tractable social simulation with reasonable
theoretical justiﬁcation, as noted by Sawyer (2004), social explanation via social
simulation is a difﬁcult prospect. While the potential for agent-based models to
demonstrate the emergence of social complexities is a possible beneﬁt to social
science, whether or not those models can provide a coherent explanation of social
behaviour is unclear.
Sawyer argues that societies, like the human mind, display qualities which
are irreducible to simple interactions of individuals within that society; despite
our knowledge of neuroscience, the higher-level study of mental phenomena (i.e.,
psychology) is still required to understand the human mind. Similarly, Sawyer
argues that human society displays a non-reductive individualism, in which social
explanation cannot be complete without addressing the irreducible effects of higherlevel
social structures. The variation of the bird migration example given in
Sect. 5.9.2 gives one example of this phenomenon.
8.6 Schelling’s Approach
8.6.1 Schelling’s Methodological and Theoretical Stance
Schelling’s modelling approach, as in our analysis of ALife, provides some
important insights into addressing the difﬁculties of social simulation. Once again,
Schelling avoids some of the difﬁculties facing other social simulations by virtue
152 8 Conclusions
of the residential segregation model’s simplicity. As a C1 model in Cederman’s
framework (see Table 5.1 for a summary), Schelling avoids the methodological
difﬁculties of using complex agent structures (as in C2), or seeking profound
emergent behaviours (C3). Likewise, problems of theory-dependence are minimised
by using only simple agents with simple rules, with no attempt to address larger
social structure. As a consequence, Schelling’s methodology remains inﬂuential
today with researchers who seek simple models of social phenomena (Pancs and
Vriend 2007; Benito 2007).
Schelling further strengthens this approach through his own theoretical framework
regarding the best use of models. He posits that general models of behaviour
can produce general insights, and that within a given discipline some such insights
may not be evident from empirically-collected data. This view seems vindicated by
the surprising result of his segregation model, the impact of which was immediate
and lasting within social science.
8.6.2 Difﬁculties in Social Explanation
However, while Schelling’s model does address a number of concerns relevant to
social simulation, the problem of social explanation still presents a difﬁculty. As
noted above, his model takes no interest in the presence or inﬂuence of higher levels
of social structure; he is concerned only with the actions of individuals. Within the
study of residential segregation, his result is likely to be only a partial explanation
simply for that reason: housing reform, tax incentives, and other measures designed
to inﬂuence racial integration in the housing sector are likely to have an impact
as well. Schelling of course does not strive for such a complete explanation, as
discussed in Chap. 7; however, those who do seek an explanation of residential
segregation could try to use Schelling’s model as a starting point.
Unfortunately, Sawyer’s perspective argues that avoiding these higher-level
elements in an agent-based model and hoping for them to emerge may be fruitless
as well (Sawyer 2004). Even if one were to add additional capabilities and complexities
to Schelling’s model, in the hope of allowing for more complex emergent
behaviours, an explanation derived from such a model would lack the contributions
of higher-level, irreducibly-complex social institutions. As in the birdsong example
in Sect. 5.9.2, there is an argument that such elements must be incorporated to
produce a complete picture of the development social behaviours. The issue of
how to progress from Schelling’s initially successful modelling framework to deeper
social explanation thus remains a complex one.
At least the social scientist can take solace in the fact that such concerns are not
alien to other modelling disciplines, as in Bedau’s discussion of ‘weak emergence’
as a means for avoiding the difﬁculty of the effect of downward causation in natural
systems from higher-order elements (Bedau 1997). Unfortunately, if anything such
difﬁculties are more acute in social simulation than in biologically orientated
8.7 Luhmannian Modelling 153
simulation, as the additional elements of social institutions, mass communication,
and other distinctly societal factors add additional layers of unknowns into an
already difﬁcult theoretical situation.
8.6.3 Schelling and Theory-Dependence
As noted above, the issue of theory-dependence looms large within social simulation,
and even for Schelling’s simple and abstract model these problems seem to
remain. Schelling’s model is constructed as a singular, vitally important assumption:
if we presume that individuals choose their preferred housing based on the racial
makeup of their neighbourhood, then segregation will result.
Schelling, and likely other theorists, would argue that such an approach is
commendable: Schelling was using his model to test a hypothesis, which is an
acceptable role for models. By introducing a highly tractable, transparent model to
illustrate the potential import of these factors in residential segregation, he was able
to present a new perspective on potential individual choices that can lead to such
undesirable social outcomes. However, such an approach becomes difﬁcult when
the goal is not hypothesis-testing, but the development of new social theory.
8.7 Luhmannian Modelling
8.7.1 Luhmann and Theory-Dependence
Luhmann’s inﬂuential treatises on the development of social order (Luhmann
1995) provide a perspective which illuminates potential methods to use social
simulation to develop social theory. His ideas regarding the low-level basis for
human communication, and the subsequent development of social order, seems a
natural partner for the agent-based modelling methodology.
This approach demonstrates the fundamental limitation of Schelling’s methodology
alluded to in the previous section. While the residential segregation model can,
and does, provide a useful test of a hypothesis which demonstrates the importance
of individual behaviour in human society, the overall import of that factor is
unaddressed by such a model, given that it is conﬁned to a singular behavioural
assumption related to a singular social problem. We may build a Schelling-type bird
migration model which demonstrates the importance of certain individual-based
factors in driving migration behaviour, but the deeper question remains open: how
do these behaviours arise in the ﬁrst instance?
For the social scientist, these are questions that must be addressed in order to
develop a deeper understanding of the origin and development of human society.
While we can imagine innumerable scenarios in which a Schelling-type model may
154 8 Conclusions
illuminate certain singular aspects of social issues and problems, the simplicity
and related theory-dependence of the approach limits our ability to investigate the
fundamental beginnings of society and communicative behaviour.
8.7.2 Luhmann and New Social Theory
Luhmannian modelling can address this larger goal of social simulation, clearly
a larger goal of the research community (Cederman 2001; Axelrod 1997; Epstein
1999), by removing these elements of theory-dependence. Any model constructed
based upon the functioning of human society will incorporate a fundamental
theoretical bias, and remove the possibility of investigating the factors that lead our
society developing in that way initially.
Doran’s perspective regarding the undeﬁned nature of computational agents in
social science (Doran 2000) ties in closely with our developed Luhmannian view.
Doran argues that social simulations should begin below the level of agent, allowing
for the emergence of agent structures that interact without pre-existing theoretical
biases affecting those interactions. In both cases, the removal of theory-dependence
from the model is paramount to its success in providing insight for the social
scientist into the origin of society.
8.7.3 Luhmann and Artiﬁcial Worlds
Recalling the earlier discussion of Levins (1966, 1968) and the lure of artiﬁcial
worlds for the modeller, this Luhmannian approach seems to fall squarely within
this realm. After all, a simulated environment, not based upon empirical data, which
includes abstracted virtual agents is already quite separated from the traditional
modes of empirical study. A simulation in which most elements of existing theory
and data are removed to study the fundamentals of society seems even further
away from the real world; one could easily imagine such a model producing quite
unusual agents with idiosyncratic interactions. Once again we return to the prospect
of studying a model for its own sake, removed from the natural world.
Where the Luhmannian approach is unique in this regard is the way in which
this separation from the real world is vital to the model. The search for a
fundamental social theory requires the removal of pre-existing theory-dependent
elements in order to produce models which illuminate the importance of pre-societal
communications and interactions. In essence, the Luhmannian approach utilises an
artiﬁcial world to illuminate factors in the real world that we may miss, simply
by virtue of our pre-existing biases derived from forming our models and theories
within a society.
8.8 Future Directions for Social Simulation 155
8.8 Future Directions for Social Simulation
8.8.1 Luhmannian Modelling as a Way Forward
Clearly the Luhmannian modelling approach displays promise for those in search of
new social theory. Without the removal of more theoretical bias from future social
simulations, issues of theory-dependence will continue to provide difﬁculty for the
social scientist who hopes to use these methodologies. Likewise, this approach
offers a wide scope for developing new ideas regarding the origin of society,
something not provided by Schelling-type methods.
Perhaps more importantly, this perspective brings us closer to the larger goals
expressed by proponents of social simulation: a means for understanding the
detailed workings of human society. From Levins’ enthusiasm for L3 models
(Levins 1966) to Cederman’s for C3 models (Cederman 2001), researchers in
various ﬁelds continue to seek to explain higher-level behaviour through the interactions
of component parts following simple rules. For the social scientist, Luhmann
reduces social interaction to its simplest components, giving the community access
to a new and potentially stimulating view of the earliest beginnings of human
society.
8.8.2 What Luhmann is Missing: Non-reductive Individualism
Despite these promising elements of the Luhmannian approach outlined thus far,
the problem of social explanation remains. Sawyer’s perspective of non-reductive
individualism in social science (Sawyer 2002, 2003, 2004) implies that even an
elegant portrayal of the origins of society may not be able to allow for the emergence
of complex social structures. Given that some of these structures are irreducible to
the actions of component individuals in the society in question, there may be great
uncertainty as to whether a given set of Luhmannian rules for a simulation may be
able to produce such complexity.
Perhaps, then, our comparison with the study of the human mind is more apt
that initially thought. As Sawyer notes, there is both a study of mind and a study
of neurons (Sawyer 2004); likewise, such an approach is an option for social
simulation. Analogous to the study of neurons, Luhmannian approaches can probe
the low-level interactions of individuals in a society through the use of models.
Then, analogous to the study of mind and mental phenomena, other models may
probe the inﬂuence of higher-level organisation upon those low-level agents. A
combination of these approaches could provide insight into social science that, as
Schelling describes, may be unattainable by other means (Schelling 1978).
For example, Luhmann’s discussion of the function of social order (Luhmann
1995) includes an in-depth discussion of the major elements of the modern social
institutions which pervade most human societies. This in turn inspired a model in
156 8 Conclusions
which each of those institutions was modelled very simply, as a monolithic inﬂuence
on a society of agents, with each institution affecting one element of overall agent
behaviour (Fleischmann 2005). Thus, this model attempts to capture the ‘science
of mind’ level of societal interaction, while also incorporating lower-level agent
behaviours. Such a model shows great promise, as Sawyer’s objections and the
difﬁculties of strong emergence hold much less weight if these downward causation
effects can be harnessed appropriately in a model.
8.8.3 Integrating Lessons from ALife Modelling
As our analysis of Schelling shows, the frameworks in the earlier chapters that
underwrite certain varieties of ALife models can be brought to bear on social
simulations as well. The issue of theory-dependence has proved vital enough to the
success of social simulation that the latter chapters of this text aimed to develop
a modelling framework which removes that difﬁculty. Similarly, the pragmatic
concerns of generality, realism, precision and tractability will remain important
even in a Luhmannian approach which aims to develop social theory; our modiﬁed
Levinsian framework provides a useful guide to the limiting factors present in all
models of natural phenomena.
A larger question in both ALife and social simulation is illustrated neatly by
Schelling’s model and related theoretical justiﬁcations: how does the intent and
scope of a model affect its usability, either empirically or theoretically? Schelling
shows how a simple model, intended to demonstrate the importance of a singular
factor in a singular problem, can have wide-ranging effects on related theory.
Despite using agents that followed only a single rule, Schelling’s demonstration
of the potential impact of individual micro-motives on the segregation problem lead
to a great deal of research investigating the impact of such individual factors on all
varieties of social problems.
Likewise, our analysis of ALife demonstrated the importance of theoretical
backstory in driving the acceptance of a model as a contribution to empirical data
within a discipline. Empirical science is full of such back-stories, but they are
implicit: the trans-cranial magnetic stimulation researcher believes tacitly that such
methods are analogous to producing real lesions in a patient’s brain. In contrast,
the agent-based modeller deals with artiﬁcially-generated data that is not simply
produced in a natural source through a different means, but is produced entirely
artiﬁcially, in a constructed world. For this reason, the theoretical backstory for
agent-based models must be explicit, as otherwise a connection between the model
and related data from the natural world becomes difﬁcult to establish.
8.9 Conclusions and Future Directions 157
8.8.4 Using the Power of Simplicity
As discussed throughout all three parts of this text, the issue of complexity in
models has numerous ramiﬁcations. For the researcher, highly complex models are
difﬁcult to develop successfully; there are a number of choices to be made at the
implementation level. For example, an evolutionary model requires a number of
parameters to govern the evolution and reproduction of its agents. How should the
agents replicate? Should crossover and sexual reproduction be used? What sort of
mutation rates might be necessary, and will test runs of the simulation illuminate
the best rate of mutation with which to produce interesting results? Additional
complexities such as neural network elements or detailed agent physiologies require
even more numerous questions at the implementation level.
Similarly, as discussed in relation to the Levins framework, greater complexity
leads to greater difﬁculties in tractability (in the context used here, a greater
difﬁculty in producing substantive analysis, as well as computability). Levins
discussed the possibility of producing models of such complexity that they exceed
the cognitive limitations of the scientist studying them (Levins 1968). For the
mathematical modeller, a model which captures every possible factor in a migrating
bird population could end up consisting of hundreds of linked partial differential
equations, a mathematical morass of such complexity that analysis becomes fruitless.
For the computational modeller, a model of similar character could produce
highly-detailed agents with individual physiologies and complex neural structures
that allow for remarkably rich behaviour; yet, such a scenario is a nightmare for
the analyst, for whom divining the function and impact of these complex internal
structures takes incredible amounts of time even for the simplest neural networks
(see Chap. 4 for discussion in relation to Beer’s model).
Thus, we must take inspiration from Schelling in this respect. Greater ease
in implementation and analysis are two enormous advantages of simpler models.
Particularly in the case of social theory, where the potential complexities when
studying human society in simulated form are vast, simple and elegant models
which illuminate crucial aspects of social systems are the most likely to produce
substantive insights.
8.9 Conclusions and Future Directions
8.9.1 Forming a Framework
This text has sought to investigate in-depth both the theoretical and methodological
concerns facing researchers who utilise agent-based modelling techniques. By
examining these issues, and developing frameworks to understand the limitations of
such models, future directions for substantive modelling research can be identiﬁed.
158 8 Conclusions
Our early analysis of ALife in the beginning of the ﬁrst section of this text
demonstrated the importance of theoretical justiﬁcation in the model-building
process, as well as the potential theoretical pitfalls of artiﬁciality in simulations.
Whether it is Newell and Simon’s PSS Hypothesis, or Silverman and Bullock’s
PSS Hypothesis for Life, computational models require a theoretical basis before
they can ﬁt into the conventional tapestry of empirical methods. In the case of
both of these frameworks, the implied empirical acceptability of models built on
such premises comes at the price of philosophical baggage for the modeller to bear.
Those who choose not to take such strong philosophical positions, as in weak ALife,
ﬁnd themselves in the situation of facing more difﬁcult theoretical problems despite
avoiding philosophical conundrums.
This baggage became increasingly evident during our examination of Levins’
modelling framework (Levins 1966). The modeller who seeks to produce useful
insights about natural systems must strike a difﬁcult balance between both Levins’
proposed three modelling dimensions and the additional aspect of tractability. This
fundamentally limits the ability of a researcher to develop models that capture the
complexities of natural systems; instead, they are left waiting for techniques that
may enhance tractability while still allowing a rough balance of the three Levinsian
dimensions.
Initially, our analysis of social simulation in the second section of this text
seemed even more problematic than for the ALife community. Social simulations
suffer the same methodological and theoretical complexities of ALife simulations,
but with the added problem of increased layers of complexity through the addition
of social considerations. Even assuming that such problems could be surmounted,
the difﬁculty of providing complete social explanation remained.
However, the introduction of Luhmannian systems sociology introduced another
means of utilising simulation in social science. In the case of developing new social
theory, creating models that are closer to reality in fact poses a fundamental problem:
the issue of theory-dependence prevents the social scientist from producing
simulations that avoid biases drawn from our own societal experience. Artiﬁciality
thus becomes a desirable trait, bringing the social theorist away from pre-existing
theoretical biases in his models. The lure of artiﬁcial worlds in this case is thus
practical: rather than attempting to avoid the inherent difﬁculties of modelling
illuminated by Levins, and instead losing explanatory capacity, the Luhmannian
modeller avoids theory-dependence and gains the ability to generate new elements
of social theory.
In essence, the root of the issue comes back to artiﬁciality and intent. A modeller
who wishes to create an Artiﬁcial1
instance of life or mind can simply look to
the PSS Hypotheses. One who wishes to mimic a natural behaviour as accurately
as possible, without claiming any theoretical revelations as a result, can simply
declare his work Artiﬁcial2
. A weak ALife researcher, or a similar perspective
from other disciplines, can apply his Artiﬁcial2
simulation to the examination
of a natural system or an element of human society by balancing his model’s
dimensions within the modiﬁed Levins framework, and ensuring both tractability
and a reasonable theoretical backstory (as in Schelling and his focus on general
insight and transparency in modelling).
8.9 Conclusions and Future Directions 159
For those who wish to apply their models to the development of new social
theory, the issue of artiﬁciality becomes more nuanced. The modeller in this case
does not seek an Artiﬁcial1
instantiation of a natural phenomenon, nor does he seek
an Artiﬁcial2
imitation of something natural. Instead, the modeller seeks to develop
a simulation in which pre-existing elements of the natural system are removed, along
with pre-existing theoretical positions on that system, in order to develop theories
independent of bias.
8.9.2 Messages for the Modeller
Now that we have examined a number of modelling frameworks in detail, analysed
and compared each in Chap. 6, and applied them to Schelling in Chap. 7, we have
been able to ascertain the most fruitful directions for future models in the social
sciences to take. However, on a more general level, the models proposed here will
not be the entirety of social simulation. Indeed, a great variety of different types of
models will continue to develop in this ﬁeld, and while Luhmannian modelling may
hold great promise for those interested in formulating new social theory and in using
simulation by playing to its greatest strengths, this is not to say that other types of
models in social science must take a lesser role.
For the social science modeller interested in producing different varieties of
models, this argument has brought forth a number of philosophical and methodological
frameworks which can provide insight into making those models more
relevant to empirical social science. Our discussion of modelling frameworks in
Alife illuminated the importance of creating a theoretical backstory; since modelling
is inherently a theory-dependent approach, an understanding and elucidation of the
theoretical underpinnings of a given model is vital for creating a basis upon which
to understand that model. Merely stating that the model is somehow reminiscent of
a real-world incidence of some social phenomenon is not enough to create a useful
justiﬁcation.
Our subsequent discussion of modelling in social science saw these frameworks
applied to a new area of enquiry, and described the various theoretical problems
at issue in the ﬁeld which have a signiﬁcant impact for the model-builder. While
the idea of a balancing act inherent in creating a model between complexity and
analysability is nothing new, bringing the Levins framework into the discussion
in social science helps to illuminate more speciﬁc pragmatic modelling concerns
that are important in computational models. The in-depth discussion of problems
of explanation in social science demonstrates how the complexities of modelling
human society create additional difﬁculties for the modeller, adding to the problems
of strong emergence in the ﬁeld of Alife. The modeller interested in social science
must take care to understand the limitations of the computational approach, and to
avoid incorporating too many elements which may lead to unanalysable complexity.
Models offer some new research directions to the social scientist, but also new
160 8 Conclusions
methodological difﬁculties; models cannot solve all of these problems and must
be deployed appropriately in conjunction with empirical data and useful theoretical
background.
Finally, we applied the frameworks discussed in relation to both Alife and social
science simulation to one particular example, that being Schelling’s residential
segregation model. Schelling provides the best means for demonstrating the tensions
between model implementation and model design within social science. The
model’s inherent simplicity limits the conclusions which one can draw from its
results; yet, that same complexity allows for greater communicability and impact
among the social science community. Replicating and extending Schelling’s model
is simple in comparison to many computational models, and as a consequence a
greater number of the social science community could join in the conversation
spurred by the creation and dissemination of the model. In this respect, Schelling’s
limitation of his model’s scope to a simple demonstration of the possible impact of
a single individual factor on a complex social problem was a big part of its success;
the modeller would do well to remember this point as a reinforcement that models
need not encompass every element of a given problem to create intriguing insight
and new directions for empiricists and theorists.
8.9.3 The Tension Between Theory-Dependence
and Theoretical Backstory
A central element running throughout the three major sections of this text is the
discussion of theory-dependence in models. Early on in the Alife discussion this
element became important, and the possible theoretical biases on display in a given
model become a serious potential difﬁculty in applying that model to the ﬁeld in
question. The proposed solution to this issue has been the creation of a salient
theoretical backstory, one which describes the reasons which make a given model
acceptable as contributing in some important way to the discourse of the ﬁeld
to which it is applied. By doing so, the researcher can justify their model and
its conclusions as important and not simply relevant only to those interested in
mathematical curiosities; as discussed in Chap. 3, the Alife researcher might do so
by contending that their model is an instance of an information ecology and thus an
instance of empirical data-collection despite its inherently artiﬁcial nature.
However, such theoretical backstory can also limit the applicability of a model.
Too strong of a theoretical framework, or one which does not hold under closer
scrutiny, can result in a model which suffers from greater theory-dependence,
rather than less. The tension here then is the delicate balance between providing a
theoretical context for a model, and having that model stuck too deeply in theoretical
elements that limit its usefulness to the broader ﬁeld. There is no easy solution to
this tension, as these pages have demonstrated. However, the modelling frameworks
discussed herein provide guidelines to avoid such pitfalls. A careful examination
8.9 Conclusions and Future Directions 161
of the theoretical and pragmatic circumstances surrounding a model can help the
modeller to avoid both theoretical pitfalls that may make his model’s applicability
suspect (as in Webb’s discussion of Beer Webb 2009), and to avoid those elements of
model construction and implementation that may lead to great difﬁculties in analysis
and the presentation of useful results.
8.9.4 Putting Agent-Based Modelling for the Social
Sciences into Practice
Having investigated the use of agent-based models in Alife and social science, we
are more aware of the issues facing modellers in these areas and have developed
frameworks to help us navigate these complexities. As modellers seeking the most
effective application of agent-based models to the social sciences, our work is not
yet done, however. General principles for effective modelling are of great value,
but within each discipline and sub-discipline of the social sciences there are further
complexities to face before we can put these ideas into practice.
Creating individualised modelling frameworks for all the major disciplines of
the social sciences is obviously beyond the scope of this text. In order to situate our
modelling ideas properly in an individual discipline requires in-depth knowledge of
its history, scope and research aims. We would need to analyse the methods of each,
determining how agent-based models can enhance researchers’ efforts in that ﬁeld.
Finally, in order to be truly convincing to those looking to branch out into agentbased
approaches, we would need to present some worked examples that illustrate
the potential of these methods to contribute useful knowledge.
In Part III, we will build upon the foundations presented thus far and offer
an example of the development of a model-based approach to a social science
discipline – in this case demography, the study of populations. Demography is a
ﬁeld with a long history, and close ties to empirical population data. We will present
a framework for model-based demography, a considered approach to demographic
simulation that takes into account the needs of the discipline and its practitioners.
In order to develop a model-based demography, we will start by analysing the
methodological development of the ﬁeld, from its earliest beginnings to the bleeding
edge. Demography, given its nature and close ties to real-world data and policy,
tends toward the social simulation approach to social science modelling. As a
consequence, our analysis will focus particularly on the demographic approach to
data and how simulation can maintain the ﬁeld’s focus on real-world populations.
We will show that demography not only can live up to demographers’ expectations
in this respect, but it can even enhance their ability to gain new insight about
the processes and functions underlying demographic change by helping us to
address demography’s core epistemological challenges of uncertainty, aggregation
and complexity.
162 8 Conclusions
We will then examine some worked examples of agent-based modelling, with a
particular focus on two modelling projects which will be presented in detail. These
two models illustrate how agent-based models can work in tandem with traditional
statistical demography to build simulations that closely mirror the behaviour of the
real-world populations under study. Finally, we will discuss the status of modelbased
demography and its potential to contribute to the discipline, and theorise about
the implications of this effort at methodological advancement for other areas of
social science that are experimenting with a model-based approach.
References
Axelrod, R. (1997). The complexity of cooperation. Princeton: Princeton University Press.
Bedau, M. A. (1997). Weak emergence. In J. Tomberlin (Ed.), Philosophical perspectives: Mind,
causation, and world (Vol. 11, pp. 375–399). Oxford: Blackwell.
Beer, R. (2003a). Arches and stones in cognitive architecture. Adaptive Behavior, 11(4), 299–305.
Beer, R. (2003b). The dynamics of active categorical perception in an evolved model agent.
Adaptive Behavior, 11(4), 209–243.
Benito, J. M. (2007). Modelling segregation through cellular automata: A theoretical answer.
Working Papers Serie AD, Instituto Valencia de Investigaciones Economicas (pp. 2007–16).
Cederman, L. E. (2001). Agent-based modeling in political science. The Political Methodologist,
10, 16–22.
Doran, J. E. (2000). Trajectories to complexity in artiﬁcial societies. In A. Kohler & G. Gumerman
(Eds.), Dynamics in human and primate societies. New York: Oxford University Press.
Epstein, J. (1999). Agent-based computational models and generative social science. Complexity,
4(5), 41–60.
Fleischmann, A. (2005). A model for a simple Luhmann economy. Journal of Artiﬁcial Societies
and Social Simulation, 8(2), 4.
Klüver, J., Stoica, C., & Schmidt, J. (2003). Formal models, social theory and computer
simulations: Some methodical reﬂections. Journal of Artiﬁcial Societies and Social Simulation,
6(2), 8.
Levins, R. (1966). The strategy of model-building in population biology. American Scientist, 54,
421–431.
Levins, R. (1968). Evolution in changing environments. Princeton: Princeton University Press.
Luhmann, N. (1995). Social systems. Stanford: Stanford University Press.
Pancs, R., & Vriend, N. J. (2007). Schelling’s spatial proximity model of segregation revisited.
Journal of Public Economics, 91, 1–24.
Reynolds, C. W. (1987). Flocks, herds and schools: A distributed behavioural model. Computer
Graphics, 21(4), 25–34.
Sawyer, R. K. (2002). Nonreductive individualism: Part I: Supervenience and wild disjunction.
Philosophy of the Social Sciences, 32, 537–561.
Sawyer, R. K. (2003). Artiﬁcial societies: Multi-agent systems and the micro-macro link in
sociological theory. Sociological Methods and Research, 31(3), 37–75.
References 163
Sawyer, R. K. (2004). Social explanation and computational simulation. Philosophical Explorations,
7(3), 219–231.
Schelling, T. C. (1971). Dynamic models of segregation. Journal of Mathematical Sociology, 1,
143–186.
Schelling, T. C. (1978). Micromotives and macrobehavior. New York: W. W. Norton.
Webb, B. (2009). Animals versus animats: Or why not the real iguana? Adaptive Behavior, 17(4),
269–286. https://doi.org/10.1177/1059712309339867.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
Chapter 12
Conclusions
12.1 Model-Based Demography: The Story so Far
As we have seen from the examples presented in Part III thus far, the development
of model-based demography is advancing, though still in a relatively early stage.
Demography is a ﬁeld with a lengthy and successful history, and by the nature of
the research questions it poses, has always been linked very tightly with concepts
of statistics and probability (Courgeau 2012). Far from preventing progress in the
ﬁeld, demography has been inﬂuential since its earliest beginnings, proving vital
for the study of population change and the implementation of critical social policies
relating to its core processes of fertility, mortality and migration.
Perhaps as a consequence of this, the introduction of a new methodology into
a ﬁeld with such an extensive history is one that bears careful consideration. After
all, if statistical modelling of population data can still produce ﬁgures that journal
editors, policy-makers, and research funders ﬁnd useful and are happy to support, is
there any particular need for new methods?
As outlined in Chap. 9, demography has wrestled with this question before, and in
each instance has incorporated these new methods and used them to enhance its core
strengths. Each of the four methodological paradigms we outlined – period, cohort,
event-history and multilevel – developed in response to shortcomings in previous
methods that left certain demographic questions out of reach. New methods that
addressed these questions thus became part of the demographic toolbox, augmenting
the ﬁeld’s capabilities but not replacing the methods that came before.
In that respect, model-based demography answers a similar call. The epistemological
challenges posed by the problems of uncertainty, aggregation and
complexity lend themselves toward a model-based approach. The application of
simulation to demographic problems is not just in pursuit of novelty, but is a means
to an end, offering the power to investigate and better understand the complex,
multilevel interactions that drive population change. As with cohort, event-history,
and multilevel approaches previously, model-based demography will become a key
© The Author(s) 2018
E. Silverman, Methodological Investigations in Agent-Based Modelling,
Methodos Series 13, https://doi.org/10.1007/978-3-319-72408-9_12
225
226 12 Conclusions
tool the demographic toolbox for certain categories of research questions, and the
four previous paradigms will continue to be useful for other questions. Indeed, as
we have seen in the examples in Chaps. 10 and 11, simulation models and statistical
demography can work in combination to answer questions about complex social
processes.
12.2 The Practice of Model-Based Demography
The modelling studies presented in Part III provide detailed examples of the
application of simulation modelling methodologies in demographic research. We
should note that the studies presented here are far from a comprehensive survey
of the whole of model-based demography, though we have tried to present some
of the key inﬂuences on these studies and acknowledge their critical contribution
to the state of the ﬁeld today. Without the efforts of Axtell et al., Billari et al.,
and numerous others, the development of model-based demography would not have
proceeded so quickly and effectively (Axtell et al. 2002; Billari and Prskawetz 2003;
Billari et al. 2007).
The Wedding Ring example in particular serves as a useful illustration of
how simple models can add to demographic knowledge. Much like Schelling’s
residential segregation model (Schelling 1971), the Wedding Ring focuses on a
single complex question and seeks to model a possible answer with very little in
the way of data or complexity. In both cases, we are not able to offer speciﬁc
point predictions about the future state of residential segregation or coming trends
in partnership formation, but we are able to offer some new conclusions about
the social functions underlying those phenomena. With Schelling’s model, we can
suggest that individual racial preferences may play an unexpectedly large role in
the manifestation of residential segregation, and with the Wedding Ring we can
propose that social pressure on the unmarried from the married can produce similar
partnership formation patterns to those seen in the real world. Most fascinatingly,
neither of these models required the incorporation of even a scrap of real-world data.
These modelling approaches thus hew closer to systems sociology as opposed
to social simulation (Silverman and Bryden 2007, to appear). The outcomes of
these models help us to understand the social functions underlying a populationlevel
outcome seen in human society, but in neither case are we able to state
anything speciﬁc regarding a real-world example. This is a marked difference from
traditional demography, in which the overwhelming majority of research is applied
in nature.
The Wedding Doughnut (Bijak et al. 2013; Silverman et al. 2013a) acknowledges
this disconnect and attempts to resolve it, primarily through the integration of statistical
demography. The model is extended to increase the inﬂuence of spatiality, and
fertility and mortality, augmented by statistical demographic projections using realworld
data, become key elements of the model’s behaviour. The model remains a
proof-of-concept more than a focused policy tool, but these extensions demonstrate
12.3 Limitations of the Model-Based Approach 227
the capacity for simulation and traditional demography to work in tandem, and in
the process harness the strengths – and mollify the weaknesses – of each approach.
The addition of uncertainty quantiﬁcation in the form of Gaussian process
emulators further eases the transition from statistical model to computational. The
emulator provides key insight into the impact of simulation parameters, helping the
modeller to redress the balance between precision and tractability. In practical terms,
we can better understand the model’s behaviour and provide more incisive analysis
of the results. In pragmatic terms, modellers accustomed to formal mathematical
models may feel more comfortable working with computational models that are
less of a ‘black box’.
12.3 Limitations of the Model-Based Approach
While the development of model-based demography provides new means for
generating demographic knowledge, practitioners should remain mindful of the
limitations of simulation as set out in Parts I and II. As with any modelling
methodology, agent-based methods are best used to answer research questions that
would explicitly beneﬁt from modelling individual behaviours and interactions.
When modelling problems that are closely related to complex social factors, agentbased
models may provide a more suitable platform for explanatory aims.
12.3.1 Demographic Social Simulation
Revisiting the agent-based model of the Anasazi Axtell et al. (2002) provides an
example. The model was built using archaeological data, and provides a useful
platform for exploring how the Anasazi population eventually declined. However,
as pointed out by Janssen et al. (2009), one could argue that the model did not result
in a substantial change in the discourse around the Anasazi’s decline. The authors’
replication shows that, while the model results do closely mirror archaeological data
and provide sensible conclusions, the model does not provide much information
beyond a comparatively simple model based on the carrying capacity of the Long
House Valley itself:
Within model-based archaeology two approaches can be identiﬁed: (1) detailed data-driven
simulation models that mimic observed trends like population numbers, (2) stylized models
that are informed by empirical data but explore a broader domain of possible socialecological
systems. Which approach is the most appropriate depends on the research
question at hand. The fact that the Long House Valley abandonment can not be explained
by environmental factors is demonstrated by the original Artiﬁcial Anasazi, but it could
also be explained by calculating the carrying capacity of the valley. A more comprehensive
question like whether exchange networks increase the resilience of settlements in the US
south west may need to be addressed by a series of models, including stylized models that
simulate various possible landscapes. (Janssen 2009, para. 5.4)
228 12 Conclusions
In model-based archaeology, as in social sciences more broadly, the datadriven
social simulation approach and the abstract, theory-driven systems sociology
approaches are in evidence. In the case of the Anasazi simulation, one might argue
that the model may have been more powerful than was actually demanded by the
research question. We might propose as well that the model in this case served a
useful role by conﬁrming a result through a more detailed modelling approach, as
well as providing a useful example of model-based demography in practice.
The choice of whether to apply an agent-based model to a particular question is
not always going to be straightforward, and there is often the chance that a model
may be overkill for certain types of research questions. However, as in the case
of the Long House Valley model, the model can serve additional purposes beyond
testing a hypothesis, by providing a test case for a particular approach, generating
new questions that can be examined with a reﬁnement of the model, or by directing
future data collection.
12.3.2 Demographic Systems Sociology
The Wedding Ring and Wedding Doughnut provide a useful example of the more
abstract side of agent-based demography. These models are more generalised and
abstract in their approach, in contrast to the Anasazi model. The Wedding Ring
eschews empirical data entirely, building a model focused on testing a particular
theory about the inﬂuence of social pressure on partnership formation timing (Billari
et al. 2007). The approach is more in line with a systems sociology approach,
in which we are examining the impact of social factors on a population-level
phenomena without reliance on empirical data. The theoretical focus of the model
is acknowledged from the outset, and as a consequence the model provides both
a useful exploration of theory and an inﬂuential proof-of-concept for demographic
models with similar research aims.
The Wedding Doughnut (Bijak et al. 2013; Silverman et al. 2013a) takes the
foundations of the Wedding Ring and takes them in a more empirical direction.
Empirical data is used to generate the initial population and to drive the patterns
of mortality and fertility amongst the agents. A simple model of health status is
added to illustrate how simple models can still be relevant for the study of social
policy. The model does not fully make the leap into social simulation, however, as
the authors are not aiming for speciﬁc point predictions regarding social care need
or UK population change; instead, the model provides an example of the integration
of statistical demography and agent-based approaches.
In an empirically-driven discipline like demography, models like these stand out
as a more theoretically-driven approach. This can easily lead to misunderstandings,
as the results may seem to lack relevance to real-world population change, and too
ill-informed by population data to provide demographic insight. However, as noted
by Courgeau and Franck (2007), a demography which exists to operate only on
successive sets of data using identical methods is not a ﬁeld which is progressing
12.4 The Future of Model-Based Demography 229
as a scientiﬁc practice. Model-based demography can provide a means to expand
the theoretical innovation of demography, as illustrated by the Wedding Ring and
subsequent extensions.
In practice, the makers of such models should be mindful of their theoretical
backstories, and ensure that the assumptions underlying their construction are
clearly delineated from the start. Setting out the aims and purpose of a more abstract
model will ensure that the results are properly placed into context by readers, and
alleviate potential misunderstandings due to misapprehension of the model’s scope
and intended impact. This will also help to ensure that comparisons between models
and demographic approaches will be made on a like-for-like basis. Demographic
systems sociology models will not evaluate well when compared against statistical
models of a particular population, for example, given that the simulation in that case
is not aiming for theoretical relevance in the ﬁrst place – but if the intended scope
of the model is not laid out from the outset, that may not be clear and could lead to
a negative evaluation of the methodology by the community.
12.4 The Future of Model-Based Demography
Model-based demography clearly has potential as an approach to certain types of
demographic problems, both focused empirical questions and broader, theoretical
concerns. However, the unique characteristics of agent-based approaches in particular
suggest some particular avenues where this approach would be most fruitful.
The demographic extension of the Linked Lives model discussed in Chap. 11
(Noble et al. 2012; Silverman et al. 2013b) demonstrates the potential of agent-based
demographic models for the study of major social policy concerns. Demography
has a long history of empirical relevance, and is frequently used by policy-makers
to guide their decisions (Xie 2000). The Linked Lives model aims to leverage this
strength by combining statistical demographic elements with a detailed model of
the supply and demand of social care in an ageing UK society, a problem receiving
a great deal of focus in the UK political context at present.
The simulation is built around a simpliﬁed version of UK geography, in which
individual agents live, form partnerships, migrate, and provide care for loved ones.
The original Linked Lives model (Noble et al. 2012) focused on the implementation
and demonstration of the model as a useful platform for the examination of
the cost of social care; the subsequent demographic extension (Silverman et al.
2013b) incorporated UK census data and demographic projections of mortality and
fertility to enhance the realism of the model. This combination produces population
dynamics that accurately reﬂect demographic projections of the UK population.
More importantly, however, the extended model demonstrates how a model of
this type can provide unique insights into policy-relevant problems that beneﬁt both
from demographic expertise and the modelling of complex interactions facilitated
by an agent-based approach. For example, the model illustrates that a policy change
that might at ﬁrst blush seem largely irrelevant to the cost of social care – in
230 12 Conclusions
this case, increasing the retirement age – has a signiﬁcant impact. The presence
of retired carers actually accounts for a surprisingly large amount of the informal
social care being provided in the model, and as a consequence, keeping older
potential carers in work longer can backﬁre when the retirement age is raised too
high (Silverman et al. 2013b). This result anticipates the later analysis by Age UK,
which highlights the signiﬁcant cost savings to society provided by these selﬂess
older citizens (Age UK 2016). The use of Gaussian process emulators to conﬁrm
the impact of the retirement age parameter further increases our conﬁdence in this
result, allowing us to peer deeper into the workings of the model and determine key
parameters that may be of particular importance to policy-makers concerned with
social care.
While the model remains more of a proof-of-concept, and does not claim to
provide speciﬁc and solid predictions for the future of UK social care, it does
provide a useful exemplar for future excursions into policy-relevant model-based
demography. Despite the relative lack of data compared to data-rich microsimulations,
the model is able to provide signiﬁcant insight into the dynamics of social
care supply and demand. The incorporation of UK demographic data shows that
simulations can be linked closely with population data in a relatively straightforward
way. Finally, the use of uncertainty quantiﬁcation in the form of emulators allows
us to more thoroughly explore the simulation’s parameter space, and in the process
generate scenarios that help us examine possible futures under a wide variety of
possible policy shifts.
Thus we may imagine a future for model-based demography in which the
approach becomes a trusted tool for the study of empirical questions of population
change where social factors are of particular relevance, but also where it ﬂourishes
particularly when applied to systems sociological models driven by social theory,
and policy-relevant models aimed at the generation and exploration of scenarios.
The latter case offers another area of growth for demography, where the implementation
of models combining population data and complex agent behaviour allows
us to create ‘policy sandboxes’ where future population trends can be studied
under a variety of possible futures. Interacting with models of this type can help
policy-makers to spot potential spillover effects of policy changes before realworld
implementation, and assist them in the creation of evidence-based policy
informed by real-world population data and a scientiﬁc approach to the modelling
of populations.
12.5 Model-Based Demography as an Exemplar for Social
Science Modelling
In Parts I and II, we examined the methodological difﬁculties inherent in the use of
agent-based modelling for the social sciences. By bringing together methodological
analyses from Alife, social simulation, population biology, and political science,
12.5 Model-Based Demography as an Exemplar for Social Science Modelling 231
we established the importance of a theoretical backstory for any given modelling
enterprise. These backstories delineate the assumptions on which our models
operate, the intended scope of the model, and the level of artiﬁciality we ascribe
to the model and its results.
In practice, however, addressing these concerns in detail every time we develop
a model seems redundant at best, even a waste of time at worst. The practice of
modelling often requires an iterative approach, in which previous simulations are
extended in various ways, tested and at times discarded, and as a consequence each
instance of the model could approach each of these elements slightly differently,
even if the overall research aims stay largely identical.
12.5.1 The Advantages of Developing a Framework
In the case of model-based demography, we are able to alleviate this additional
explanatory burden somewhat by developing a widely applicable methodological
framework – a paradigm which seeks to justify the general practice of modelling
population change in this way from the outset. Under this methodological paradigm,
modelling is focused on a classical scientiﬁc approach, informed by data and tasked
with studying the social factors underlying the processes generating population
change. As model-based demographers we seek the integration of demography’s
greatest strengths – rich population data and a centuries-long history of statistical
expertise and innovation – with simulation’s ability to surpass some troublesome
epistemological limits of demographic knowledge (Courgeau et al. 2017). By
extending the concept of the statistical individual to the simulated individual,
we establish model-based demography as a descendant of the methodological
tradition of the discipline, and enable a generation of researchers to embrace a new
technique without overly troubling themselves with the ﬁner points of Artiﬁcial1
and Artiﬁcial2
.
The advantage of this kind of approach is signiﬁcant. Establishing the theoretical
backstory in advance as a methodological addition to the ﬁeld, or as a sub-ﬁeld,
allows us to approach each new model identifying as ‘model-based demography’
with a pre-existing knowledge of the likely scope and intent of that model. Where
models depart from the core concepts of model-based demography, this can be
established when documenting the model by making reference to this paradigm.
We are able to spend more time constructing and validating models, conﬁdent that
our intentions will be understood by the community at large without excessive
explanation.
Additional complexities do come into play, however, when we reach the stage
of analysing the results of our complex demographic models. If the advantages of
model-based demography are to be truly realised, then methods which enable us to
understand the impact of model parameters on population-level phenomena should
continue to be reﬁned. Uncertainty quantiﬁcation methods like Gaussian process
emulators provide a useful starting point here, and if model-based demographers
232 12 Conclusions
begin to embrace these techniques then it is likely we will see continued reﬁnements
in the future as we begin to adapt them to the particular case of agent-based social
simulations.
However, the development of a backstory remains important when working
with demographic systems sociological models. Abstract models are simpler in
their construction, generally speaking, but are not necessarily simpler in their
implications, as we saw with Schelling’s residential segregation model (Schelling
1971, 1978). Establishing the scope and artiﬁciality of a model is signiﬁcant, as
simplistic models can easily be misconstrued as making overly ambitious claims
otherwise.
12.5.2 Model-Based Social Sciences
The example of model-based demography illustrates the advantages of developing
a methodological paradigm as a kind of collective theoretical backstory for an
approach to simulation with a speciﬁc discipline. The speciﬁc case of demography
somewhat lends itself to this way of doing things, however; demography boasts a
lengthy history and a notable ability to absorb and reﬁne a wide variety of statistical
approaches. In that context, establishing another methodological framework to
underwrite the use of simulation seems an appropriate way to situate simulation
as a tool worthy of the same respect as statistical modelling.
In other areas of social science, however, the range of methodologies in use
can be much wider. We see researchers gathering data qualitatively via interviews
or surveys, or others analysing texts or artwork, or studying the geographical
distribution of people, artefacts and customs. Many of these disparate methods can
provide useful knowledge that can be utilised in simulation (Gilbert and Troitzsch
2005), but this does not mean in turn that the simulations will be considered
trustworthy or appropriate tools to those same researchers.
In this context we cannot simply write variations of the model-based demography
framework as model-based social science and expect them to provide an appropriate
theoretical backstory for such a broad range of research questions and methods.
However, model-based demography does demonstrate a process which can be more
transferrable. Model-based demography as a framework addresses core questions
that are just as salient elsewhere:
1. What are the key questions asked by our discipline?
2. What is the main unit of analysis in our discipline?
3. What are the main epistemological limits within our discipline?
4. Which of these limits can be addressed in some way using simulation?
5. How can our analyses inform a simulation process?
Model-based demography suggests that simulation efforts in other areas of social
science would beneﬁt from a concerted effort to address these core questions, and
in the process situate the approach clearly within a disciplinary context. Doing so
12.6 A Manifesto for Social Science Modelling? 233
not only alleviates some of the difﬁculties outlined above, but it provides a common
backstory which also clariﬁes and communicates the aims of the work to others
outside the discipline. This in turn allows for easier collaboration between social
scientists and simulation practitioners, and eases the time-consuming process of
developing a common language in simulation collaborations, which can inhibit
progress signiﬁcantly in new simulation ventures. Collaboration across disciplines
also becomes easier, as each member of the collaboration would have a clear
statement in hand of the methodological aims and limitations of the work to come.
In a sense, perhaps, we would beneﬁt from developing modelling manifestos
of sorts. Rather than individual justiﬁcations of each model, establishing a united
front through which we can embark on journeys into simulation allows us to get on
with development and implementation using a common framework. Some models,
particularly those of a more abstract, systems-sociological bent, will need to pay
more attention to individual statements of scope and purpose, but given the more
theory-driven and explanatory nature of such models this is naturally part of such
an enterprise anyway. Developing such ‘manifestos’ will certainly spawn its own
protracted arguments and divisions, naturally – we are academics, after all. That
being said, with the splintering of so many disciplines into sub-disciplines and subsub-disciplines,
we might beneﬁt from occasional forays into self-reﬂection on the
goals and limitations of our work and how it relates to our colleagues elsewhere in
our own disciplines and neighbouring ones as well.
12.6 A Manifesto for Social Science Modelling?
By most measures this volume makes for a rather unwieldy manifesto for social
modelling – the word count is excessive; it covers far too many disciplines; and
many of the conclusions are highly malleable depending on the reader’s own
disciplinary background and research convictions. Fortunately, this volume is not
intended to ﬁll that role; specialists within the varied specialisms of social science
are far better equipped to handle the task of establishing approaches to modelling in
their particular context (see, e.g., Conte et al. 2012).
This volume set out to expose and discuss the challenges faced by simulation
modellers, starting from the earliest pioneers in simulation (Schelling, Langton,
and the rest) and moving toward the current growing interest in models of human
sociality in many different ﬂavours. By bringing together insights drawn from Alife,
population biology, social simulation, and demography, we are able to develop a
better understanding of the power and the limits of simulation when applied to
the social sciences. The development of model-based demography shows us how
simulation can be investigated, applied, and reﬁned for a particular social science
context.
Ultimately, the further advancement of social modelling will still require signiﬁcant
work, both theoretical and practical. Conversations will need to be started
between colleagues who hardly understand one another; conceptual chasms will
234 12 Conclusions
need to be bridged; and social scientists will need to work with programmers and
computer scientists who may have very different views of the world. Hopefully the
discussions brought forth in this volume might make those discussions somewhat
easier, the bridges a bit shorter and easier to construct, and the gaps in practical
and disciplinary knowledge between social scientists and computer scientists less
insurmountable. If it facilitates some heated debates over the writing of some
modelling manifestos, then so much the better.
References
Age UK. (2016). Invisible army of oldest carers saving state billions. http://www.ageuk.org.uk/
latest-news/archive/invisible-army-of-oldest-carers-saving-state-billions/. [Online; posted 18
May 2016].
Axtell, R. L., Epstein, J. M., Dean, J. S., Gumerman, G. J., Swedlund, A. C., Harbeger, J.,
Chakravarty, S., Hammond, R., Parker, J., & Parker, M. (2002). Population growth and collapse
in a multi–agent model of the Kayenta Anasazi in Long House Valley. Proceedings of the
National Academy of Sciences, 99(3), 7275–7279.
Bijak, J., Hilton, J., Silverman, E., & Cao, V. (2013). Reforging the wedding ring: Exploring a
semi-artiﬁcial model of population for the united kingdom with Gaussian process emulators.
Demographic Research, 29(27), 729–766.
Billari, F., & Prskawetz, A. (2003). Agent-based computational demography: Using simulation to
improve our understanding of demographic behaviour. Heidelberg: Physica Verlag.
Billari, F., Aparicio Diaz, B., Fent, T., & Prskawetz, A. (2007). The “Wedding Ring”: An agentbased
marriage model based on social interaction. Demographic Research, 17(3), 59–82.
Conte, R., Gilbert, N., Bonelli, G., Ciofﬁ-Revilla, C., Deffuant, G., Kertesz, V., Loreto, V., Moat,
S., Nadal, J. -P., Sanchez, A., Nowak, A., Flache, A., San Miguel, M., & Helbing, D. (2012).
Manifesto of computational social science. European Physical Journal Special Topics, 214,
325–346.
Courgeau, D. (2012). Probability and social science: Methodological relationships between the
two approaches (Methodos series, Vol. 10). Dordrecht: Springer.
Courgeau, D., & Franck, R. (2007). Demography, a fully formed science or a science in the making.
Population-E, 62(1), 39–46.
Courgeau, D., Bijak, J., Franck, R., & Silverman, E. (2017). Model-based demography: Towards
a research agenda. In A. Grow & J. Van Bavel (Eds.), Agent-based modelling and population
studies (Springer series on demographic methods and population analysis, pp. 29–51).
Berlin/Heidelberg: Springer.
Gilbert, N., & Troitzsch, K. G. (2005). Simulation for the social scientist. Maidenhead: Open
University Press.
Janssen, M. A. (2009). Understanding artiﬁcial Anasazi. Journal of Artiﬁcial Societies and Social
Simulation, 12(4), 13.
Noble, J., Silverman, E., Bijak, J., Rossiter, S., Evandrou, M., Bullock, S., Vlachantoni, A., &
Falkingham, J. (2012). Linked lives: The utility of an agent-based approach to modelling partnership
and household formation in the context of social care. In C. Laroque, J. Himmelspach,
R. Pasupathy, O. Rose, & J. M. Uhrmacher (Eds.), Proceedings of the 2012 winter simulation
conference. IEEE.
Schelling, T. C. (1971). Dynamic models of segregation. Journal of Mathematical Sociology, 1,
143–186.
Schelling, T. C. (1978). Micromotives and macrobehavior. New York: W.W. Norton.
References 235
Silverman, E., & Bryden, J. (2007, to appear). From artiﬁcial societies to new social science theory.
In Proceedings of ECAL 2007.
Silverman, E., Bijak, J., Hilton, J., Cao, V., & Noble, J. (2013a). When demography met social
simulation: A tale of two modelling approaches. Journal of Artiﬁcial Societies and Social
Simulation, 16(4), article 9.
Silverman, E., Hilton, J., Noble, J., & Bijak, J. (2013b). Simulating the cost of social care in an
ageing society. In W. Rekdalsbakken, R. T. Bye, & H. Zhang (Eds.), Proceedings of the 27th
European conference on modelling and simulation (pp. 689–695). Dudweiler: Digitaldruck
Pirrot.
Xie, Y. (2000). Demography: Past, present and future. Journal of the American Statistical
Association, 95(450), 670–673.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give appropriate
credit to the original author(s) and the source, provide a link to the Creative Commons license and
indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.