© Copyright JASSS

  JASSS logo ----

Carl Henning Reschke (2001)

Evolutionary Perspectives on Simulations of Social Systems

Journal of Artificial Societies and Social Simulation vol. 4, no. 4

To cite articles published in the Journal of Artificial Societies and Social Simulation, please reference the above information and include paragraph numbers if necessary

Received: 13-Oct-01      Published: 31-Oct-01

* Abstract

The first part discusses modeling social systems from an evolutionary perspective. The main problems for evolutionary modeling, differentiation and structural change, possible approaches to these problems as well as hindrances to successful modeling are discussed. It is found that concepts of change processes need to be delimited more clearly, and combinatory emergence should be given more attention. Likewise theorizing on evolutionary processes needs to be broadened to be able to capture social phenomena. The results are used to evaluate simulation approaches for evolutionary processes in social systems: Classifier Systems, Genetic Algorithms, Evolution Strategies, Evolutionary Programming, Genetic Programming and simulations suites/programs are considered. For the latter some useful resources on the web are indicated.

evolution, combinatory emergence, social systems

* Introduction

Simulations of populations of many interacting, heterogeneous agents are interesting to analyze in terms of evolutionary models, since evolutionary theory builds on the competitive interaction of carriers of differing traits. In this paper I therefore want to draw together three lines of research: computer science and simulation, evolutionary research and social science.

To become more suitable for evolutionary treatments, we need to identify precisely the entities and range of applicability of evolutionary theories as well as mechanisms of evolutionary change in social science. The questions of entities and range of applicability have found a suitable 'base camp' in the work of Nelson and Winter (1982) and Hannan and Freeman (1989).

My discussion is based on the assumption that there is a general evolutionary theory applicable to social science as well as biology (see e.g. Hass 1970, Pantzar and Csanyi 1991). A (general) evolutionary theory, applicable to social systems, requires inclusion of a systemic perspective and mechanisms for self-referential, 'reflexive' processes, sensu Soros (e.g. Soros 2000), as well as processes of combining lines of tradition and/or artefacts. Feedback serves to 'control' change and achieving adaptation comparatively fast and combination of information and/or units seems necessary for the build-up of complexity.

Here, I want to focus on one part of the first issue, systemic feedback, and on one of the latter mechanisms, combinatory emergence, since I suspect they are presently the 'reverse salients' of developments.

* Theory and Concepts

Differentiation and Structural Change

If you look at von Neumann's (1945) model of balanced growth (which is a mathematical representation of an economy as 'network of entities' producing and consuming goods via 'activities') and think about it in terms of structural change, you become aware of the fact that it does not quite match into such a conception. Surely one could argue that the structure and strength of connections between entities changes and that a measure thereof could be used to represent (structural) change. One of the reasons the von Neumann model does not square with an evolutionary view of structural change is missing information on hierarchical organization of entities and subentities as well as criteria for separating different identities. The question that immediately pops up is: What constitutes change in such a model? How does change come about? In real world the answer would be: By slight changes in existing entities, recombination of existing (sub-)entities and possibly creation of entities de novo.

What we need is thus a clear separation between types of change and a good way to represent and model these types of change. This task is made difficult, but not impossible, by the differing conceptions of change processes in the relevant three fields (and even their subfields).

While the term differentiation is used in e.g. economics in the straightforward sense of increasingly differing characteristics of goods, structural change is a construct that refers to a change in the make-up of an economy in terms of percentage composition of (types of) production factors (capital, labor, and in newer accounts human capital/'knowledge'), or the structural composition of types of businesses/economic activities and their size.

In sociology, as far as I can make the various uses out, change, differentiation and evolution are often used interchangeably, denoting focus and school of researcher. Differentiation refers to the process of societies becoming more complex due to increasing specialization. Sometimes this is called social or structural change and sometimes evolution (see e. g. Levine 1995).

In dealing with long-term processes of social (or biological for that matter) systems we have to separate between the development of characteristics and that of structures involved in evolutionary processes more precisely than it has been done in the past. In evolutionary biology it usually is argued that increasing differentiation, helped by barriers to reproduction, leads to separation of hereditary lineages and evolutionary change (see e.g. Ridley 1997 for an overview). This view is only partially correct, since differentiation alone has not much to do with structural change, if understood as change in complexity, it is difficult to grasp how this should affect the structural composition of this entity.

Competitive pressure might explain an increased likelihood for differentiation. If differentiation is understood as the increasing separation of entities in terms of their characteristics, where does that leave structural change? The 'traditional' explanation in development biology is to invoke a hierarchical organization of entities, which is accompanied by a (increasing) hierarchical organization of control structures and temporal changes of developmental events (see Riedl 1975 for an overview). This is a fair argument. But why do control structures change at some times and do not change most of the other times? One possible answer is given by an systemic model of evolutionary processes (Riedl 1975).

If one takes an engineering or morphological view on systems, say on a social system like an economy, the characteristics of and (structural) relationships between subsystems can be used to separate differentiation from structural change. In the case of an economy this might be the production processes and their vertical or horizontal integration in businesses. Differentiation is the increase in difference between entities without major structural change of the system e.g. the construction of cars since the 1930s, or to take an extreme example the marketing differentiation between rather 'homogenous' goods such as beer or cigarettes. Evolutionary change thus is used in practice to refer to two types of changes - the effects of differentiation and the effects of structural change. The evolutionary model of the so-called Neo-Darwinian synthesis is suited to deal with differentiating change, but needs elaboration to account for hierarchical control structures and systemic feedback effects from 'selection results' to be suitable for 'evolutionary characterizations' of social systems involving structural change. If this type of model offers useful insights for evolutionary processes in biology it might be welcome in that discipline too.

Evolutionary change has also been associated with the phenomenon of emergence. This concept might be used to further disentangle the conceptual melee on evolutionary change and differentiation. The Neo-Darwinian synthesis model does not properly take into account what I want to call 'combinatory emergence' in its one-hierarchy-level form. Combinatory emergence denotes the combination of formerly independent entities into new systems with new - emergent - capabilities, functions or characteristics. Combinatory emergence is what happens in symbiosis in biology and inventive combination of ideas in social science. This type of synthetic emergence has to be differentiated from what is often understood in social science by emergence: the interaction of independent entities leads to the emergence of new patterns of aggregate behavior, such as Adam Smith's 'invisible hand', or to take a computer simulation example, patterns in Conway's game of life.

Table 1: Change Processes in Evolving Systems

mutualisminteractive emergencecombinatory emergence1

1 Thanks to Alesh Kubik for pointing out Peter Cariani's earlier use of the term in his Ph.D. dissertation (Cariani 1989), as well as Peter Cariani for discussions about his concepts.

Table 1 shows a conceptual separation of terms and assumed driving forces: competitive interaction leads to differentiation of two types of entities into separate niches. Interactive emergence is the result of the interaction of independent entities with mutual interests in the way described by Adam Smith's term of the invisible hand or sociologists' use of the term emergence. Combinatory emergence by combination of systems that fulfill in some way useful functions for each other is possibly the result of changes in hierarchical control structures. Combination of entities competing with each other under limited resources into one system should lead to the destabilization and possibly demise of one of the entities or the combined system.

In the following sections I want to postpone the treatment of combinatory emergence and focus on the systemic characterization of evolutionary process and its relevance for social systems instead. Combinatory emergence will be taken up again in the part of the paper on simulation.

In the last two sections of this part, I shall make some remarks on the suitability of evolutionary computation methods for its simulation. Systemic treatment of evolutionary processes seems presently more amenable to agent programming efforts than that of combinatory emergence.

Delimiting Evolutionary Social Systems

An evolutionary conceptualization of social systems is useful only for long-term processes, where agents, artifacts, or concepts can undergo substantial change. These entities must be connected by some line of tradition, where some of these entities in a certain period influence the make-up of entities of a following period directly or indirectly. This would allow us to speak of generations even when the precise temporal relationships of individual entities are overlapping and/or if they are muddled.

The differentiation between evolutionary and non-evolutionary systems should be made along the following lines:

If the characteristics of entities do not change, or stay in a variability range which is controlled by the factors internal to the entities (control structures, information carriers like genes in biological entities) but not a selecting influence, there is no evolution in the Darwinian sense of the term and 'interactionist' models should suffice to describe the history of the system.

If the characteristics of the entities change, the characteristics are under some kind of selection and generations of characteristics are connected by tradition, i.e. there exists some form of heritability, we can meaningfully speak of and borrow from evolutionary models. These models will sometimes be confined to the biological, Neo-Darwinian type though. More on that below.

If there is change through generations but no connection through some mechanism of tradition, it is debatable which type of model to apply. Obviously the question will engender lots of discussion.

Table 2: Delimiting Interaction and Evolution

evolution yesnono

Table 2 lists the possible combination of the three differentiating criteria I have mentioned. According to the usual understanding of evolution, tradition should be by far the most important criterion. It is conceivable that the process is evolutionary in nature, but that no selection takes place because space, resources, partners for interaction are freely available, or that no change takes place because the variability of the entities is completely exhausted. If such a 'masked' evolutionary process is taking place this is important for modeling the situation in case conditions can change so as to 'suddenly' enable evolution. It is obviously highly desirable to capture such influences on the sensitivity of the model results to prevent being caught on the wrong foot.

The Criticism of Evolutionary Social Models

While evolutionary models have found increasing application in the social sciences, there is also strong criticism. This criticism is justified insofar as evolutionary social models are more complex and require higher effort to build useful characterizations of reality. As pointed out above it can be desirable to guard against unexpected surprises. This is even more important for decision-makers whose reputation depends on the quality of their anticipation of the future. Politicians and managers therefore may have a larger interest in such type of models than scientists. Criticism against evolutionary models for social processes is often justified also on the grounds that it has largely been tried to just copy biological evolutionary models of the Neo-Darwinist school, which do not capture relevant characteristics of social systems.

These relevant characteristics are feedback effects from 'the environment' on the medium that carries the tradition. These feedback effects occur constantly in social systems but are usually rejected in evolutionary biology, with some notable exceptions in systemic evolutionary theory (e.g. Riedl 1975, Wagner 1983). Systemic evolutionary theory for that matter does not reject evolutionary biology, although these views also exist (where it is questionable whether these can still be counted to what is called accepted science).

This situation serves to complicate the whole discussion, since it turns it into a war of beliefs with (at least) two frontiers. In this context it should be considered that what is called the Neo-Darwinian synthesis emerged in the 1930s in a painful process of combining the paradigms of Darwinian evolutionary change and Mendelian genetics resulting in the predominant population ecology/genetics type of model. The main architects of the synthesis each had some remaining criticism of the resulting theory. One of these triggered 'lately' the debate on the existence of punctuated equilibria and the associated explanatory factors for evolutionary change. The Neo-Darwinian synthesis is based on the assumption of undirected, random mutation, while systemic evolutionists argue that change is not random but constrained, if not directed (in a passive sense) by the existing structural make-up of organisms and its limited freedom for change (for an overview of these issues see Riedl 1975, cf. also Szathmary and Maynard Smith 1995).

The simple way out of the dilemma would be to differentiate between mutation and resulting change, arguing that mutations are random and favor no direction, while positively selected mutations show indeed certain trends. As interested but not fully involved social scientist I can not say what has been done to discriminate experimentally between the possible combinations of factors and phenomena. Taking my scant knowledge on human thinking and the course of scientific history into account, I would think it highly likely that the assumption of random mutation and the association with a random change is an outflow of human 'economizing' on thought. On the basis of this argument the development and test of the systemic view of evolutionary processes would be highly desirable.

Depending on the situation, the feedback effects from the outcomes of the evolutionary process on the medium respectively the content of tradition can be and are indeed ignored in some but not all cases of evolutionary based social modeling such as (sociological) population ecology (Hannan and Freeman 1989).

It is obvious that the decision whether to include systemic feedback effects or not depends on the 'grain', time-range, and size of potential feedback effects of the analysis/research we want to undertake. If we deal with an aggregated system description, where we do not know the variance of characteristics of entities, or if this variance is not meaningfully affected by feedback effects we can (have to) forget about them. Likewise we can disregard them, if we deal with a time window that is much shorter than the feedback needs to influence the aggregate system its constituent entities. For example, analyzing the effects of introducing a new competitive strategy in a given market-environment (or an ecological system for biologically inclined readers) can be analyzed in a framework of interaction if we want to deal with questions of competitive displacement and strategy adaptation (and as long as those are not thought to be hard-wired in some form). If we want to deal with changes in the make-up the population of strategies, where there is a limited ability to change a strategy 'endogeneously' and the strategy of one generation influences that of the next generation we should use an evolutionary model. Depending on the occurrence of feedback effects and the system adaptability we can differentiate further (see Table 3).

Table 3: Types of Systems

static systemadaptation/flexible system
interactionstatic types of interaction,
changes only in composition
systemic interaction, leading to changes in interacting components
evolutionevolutionary selection operating on random supply of variation, changing composition of types (like 'manna from heaven')evolutionary systemic feedback influencing supply of variation and speed of evolution, changes in types depend on history of system

The interest of the interactionist perspective is usually focusing on the emergence of aggregated patterns of behavior and its effect in larger systems. This perspective can be found in ecology, sociology and economics. Methodwise, a lot of complex systems research usually focuses also on this type of interaction and the resulting cases of emergent phenomena. From the preceding conceptual differentiation it is clear that the relationship to evolutionary mechanisms needs to be clarified more exactly. Processes of interaction are often called evolutionary and relevance for a higher level (aggregated, long-term) evolutionary process is claimed, which cannot necessarily be justified on the above typology. (The problem lies of course also in the tension between the term in a Darwinian framework and its traditional meaning of 'development').

Systemic Evolutionary Models

If we are interested in the composition of populations of entities such as firms with strategies, and how this composition changes over time we need an evolutionary model. This type of model has the added benefit that we can consider the influence of the environment on the composition as well. This allows us to cover a wider range of situations and do the sensitivity analyses that have been mentioned above. If we want to know how the system works internally, in the sense of how it generates variation and whether it adapts its rate of change to changing environmental conditions, we need to build a systemic evolutionary model.

The argument for a systemic character of evolutionary social models can also be made on the basis of the relevance of perception in social processes. Perception categories are influenced by cultural traditions and change on the basis of the change of the associated social system. I do not want to go deeply into a discussion of the various constructivist schools, suffice it to say that Schmidt's (1994) review of these makes clear that a) some form of 'constructivism' is a useful perspective for dealing with perception issues in social systems and b) a lot of work needs to be done in terms of delimiting terms, cases, typology and models, c) a major part of the criticism of this perspective is directed against a non-existent 'straw man'.

In social systems perception, and its medium of exchange - communication, is the prime transmission mechanism for the feedback effect from evolutionary selection results to the generation of the entities of the 'next generation', be that humans, mental concepts, institutions, organizations or artifacts. Humans observe their environment as well as the effects of their actions in this environment and tend to draw conclusions from these observations. These conclusions are usually governed by the perception categories of the social systems, e.g. what counts as success, how strongly one can influence one's success in life, and what is acceptable as a means to reach this 'success'.

From studies in the area of innovation management we know that perception matters in the success and failures of projects in organizations as well in the fate of whole organizations. IBM was able to produce the original PC because it managed to isolate the project team from the traditional operation, which would have called for in-house development of the design and components, if not barricaded the development of such a computer that deviated radically from the traditional perception of large, service-intensive mainframe computers. Apple management failed to perceive the power of either 'open-sourcing' their design or their operating system software which could have greatly improved Apple's position in the market, by signaling that one would not lock-in consumers into a two-stage monopoly. (A lesson which was in a different form used much earlier by Woolworth, who sold petroleum lamps at a loss to make profits on supplying petroleum to households.)

In social systems the perception feedback is more important than the implicit feedback effect (from successful mutations on next generations supply of variation as well as the morphological constraints on variation) that is the subject of biological evolutionary systemic theory.

* Methods for Simulating Evolutionary Processes

In this section I investigate in how far existing methods for Evolutionary Computation are suited for modeling combinatory emergence and a systemic evolutionary situation. For an overview over the described methods see Banzhaf et al. (1998), Michalewicz (1992). They also contain the references to the original sources for the methods described below. The methods described below are modeled after or include some element conducive to evolutionary processes.

Some General Remarks

Among others, Brenner (1998) and Chattoe (1998) have discussed the use of Evolutionary algorithms (EA) for social processes - the first in a very specific comparison setting while the latter gives an exhaustive overview over the relevant issues. Their discussion makes two problems clear: firstly, although EA are often claimed to be modeled after biological processes, they are in fact modeled after a - formalized - conceptualization of biological evolutionary processes. Secondly, it is necessary to have a good understanding of theoretical biology and its models to contribute to a meaningful discussion of evolutionary social processes.

I want to add that the traditional neo-Darwinian conception is hold to be insufficient by - among others - 'systemic' evolutionary scholars. The first step alone implies some loss of information. The transfer to social systems is usually associated with a further loss of information.

What I am interested in here is neither the discussion of particular EA methods nor their suitability for a specific problem, but solely the discussion whether EA characteristics make them suitable for modeling the two characteristics of change - systemic feedback and combinatory emergence - that I discussed in the beginning of this paper. The focus, here, is the 'structure' of processes rather than mathematical formulation.

Genetic Algorithms

Genetic Algorithms (GA) are modeled after processes in DNA, the molecule which carries biological genetic information. Usually a (binary) string of 0s and 1s codes for the characteristics of a solution to a pre-defined problem. It is possible to include a weight for the expression of characteristics. GAs are usually employed in optimization problems from economics to engineering.

In principle GAs are also suited for the simulation of open-ended evolutionary processes, where the 'solution' captured in the characteristics of the GA tries to hit the moving target of an evolving eco-system. Practically this implies that several GAs - representing different species of competitors and possibly some prey - need to be let run against each other. This is the idea behind Holland's Echo.

The transfer to social systems is easily achieved. Each individual string then represents an individual agent, fulfilling roles of e.g. producers, consumers, regulators, workers.

The systemic feedback element has found little attention up to now. The basic idea would be that strings are not only selected according to their characteristics but also can change their characteristics due to interactions with other strings and/or other elements of their environment. This is the equivalent to individual learning processes. A basis for the simulation of combinatory emergence exists in hybridization of strings through crossover. One disadvantage of GAs in this context is that they are not organized hierarchically and do not code for 'solutions' of combined, compartmental systems.

The non-convergence of a population of strings to the (known) optimal solution is a major issue for applications in optimization. For simulating biological or social systems it is quite useful, because it allows to investigate whether and under which conditions an evolutionary process leads to optimal or non-optimal 'solutions'.

Evolution Strategies

Evolution Strategies (ES) also work on populations of binary strings. Historically one difference between GAs and ES is that strings in a GA population usually work on discrete vectors, while ES have been used heavily in numerical optimization and strings code for floating point vectors, which has performance implications for different types of problems in optimization.

There is also a difference with respect to the selection of individuals in successive generations of potential solutions. Both, GAs and ES usually operate with fixed successive population sizes. But GAs randomly select an intermediate population and apply crossover (recombination) after that. In GAs weaker solution individuals may 'get through' the selection filter but fitter solution individuals usually survive in the long run.

ES allow to control the size and combination (parent strings vs. recombined offspring) of the intermediate population from which selections are made. In ES consequently always only the fittest x individuals survive. The most important difference with respect to the requirements for modeling systemic feedback in evolution is that ES allow for variation in reproduction parameters (probability of mutation and of crossover/recombination) such that the evolutionary process can influence itself.

Cellular Automata

Cellular Automata (CA) are another simple computer implementation to simulate interacting agents. On a grid of cells cell-agents interact according to special rules with their neighbor cells. The rules have been usually pretty simple and the focus on the emergence of self-organization patterns in cell aggregations. CAs seem to be less suited for the simulation of complex evolving entities, since they represent a greatly abstracted evolutionary process and contain little structure that the process could use to build up complexity.

It has always been the hope of the CA community that CA-simulations would evolve complex structures endogeneously, but that has not happened yet beyond dynamic interaction patterns. The first computer simulations of evolutionary processes by Barricelli (1957) can be interpreted to give some hope. They are similar to CA simulation and show the emergence of networks (eco-systems) of interacting cells as well as symbiogenetic phenomena.

Genetic Programming

Genetic Programming (GP) applies the idea of mutation and selection to (parts of) programs. The goal is to find programming solutions that are more efficient and different from what a human programmer would have written. Problematic are of course the interaction of recombined chunks of programs as well as syntax matters. The degree of complexity and dependence on programs structures and syntax diminishes the evolutionary degrees of freedom relative to GAs for example. This allows to model and analyze effects a such as path-dependence of 'evolutionary' histories and morphological/engineering burden of 'design-decisions'. On the other hand it may leave too little evolutionary potential to be suitable for the modeling of complex social processes.

Neural Networks

Neural Networks (NN) are modeled after the ideas about the connectionist organization of human brains at the middle of this century. Data inputs are 'interpreted' in the sense that they are recorded by specific elements and lead to weighted connections between elements ('nodes), of the neural network. The interaction of these elements produces an output, e.g. a classification of photos according to their content. The quality of this classification is then judged and fed back to the network which reacts by rearranging the weights and connections between elements.

This iterative process of teaching the NNs contains the systemic feedback in that the internal organization is changed. Still, the analogy in NNs is more of a developmental than an evolutionary - in the selectionist interpretation - character. Therefore NNs are useful to model the brain respectively learning process of an actor. The problem that occurs here is the trade-off between computation intensive technique and simulation resources.

Classifier Systems

Classifier systems (CS) are rule matching systems coupled to decision or action outputs, which are often implemented on the basis of GAs. They consist of detector, effector, messages system, a potentially hierarchical set of population of classifier rules ('if characteristic x1 and z4 then concept a3'), a credit assignment system for the quality of output to individual rules (usually the bucket brigade algorithm) and a genetic procedure for reproduction of classifier rules.

CS can be used to model the internal cognitive workings of an agents 'brain'. In this they are more fine-grained than NNs which operate as input-output black-box. The question is whether the increased detail justify increased computational requirements.


As discussed already, GA string coding of characteristics is potentially suitable for combinatory emergence. It would of course be highly wishful to extend possibilities towards modeling hierarchical building blocks in GAs that are the equivalent to biological genes or the as of yet 'undiscovered' memes or culturgens.

ESs are suitable for modeling the systemic feedback effects because they are build in already. The present state should be suitable to test hypotheses for simple genetic systems. An untackled problem in ES (like in GAs) is the modeling of complex compartmentalized structures (allowing to represent organisms, organizations), which is also partly due to unsolved research questions in these areas. CS allow for fine-grained modeling of individual mental and social learning processes.

Table 4: Evaluation of Evolutionary Computation

EC MethodSystemic FeedbackCombinatory Emergence
GA- usually noto hybridization/crossover
ESo reproduction parametero hybridization/crossover
CAo (systemic) interaction- (interaction of cell-rules)
GPo + recombination of program
NN+ learning of NN- (interaction of
CS+ credit assignment+ recombination of rules

- not suited; o partly suited; + well suited for the effect in the column heading

Simulations Suites for Social Systems

This section analyzes some simulation suites with respect to their suitability for the simulation of evolutionary process as defined at the beginning of this paper. The selection is totally idiosyncratic in that it is based on the suites/programs I have come across over the last years. The simulation suites covered in Table 4 are suited to the modeling of the evolutionary systemic feedback systems in varying degrees. Those focusing on interaction and with inbuilt restricted possibilities for the change of programmed simulation objects are obviously less suited in principle (GA-based Echo, and Swarm-based Evo offer some potential but require programming knowledge). The general, largely unsolved, problem is that of fusing an evolutionary programming method with object oriented programming in such a way that simulation entities can change and develop changing, evolving characteristics. Furthest down this road seems Eve which implements recombination of locomotion and food search strategies for simple animal-agents a la Genetic Programming.

The opposite extreme is offered by extremely simple 'methods' such as GAs and CAs, which can be used to model changing characteristics but miss the implicit environment such that the meaning/effect of characteristics must be pre-defined, which limits the suitability for open evolutionary processes.

My personal favorites for getting beyond this dilemma are SDML and SeSam in terms of large suites working with pre-defined sets of agent (structures), not least because they allow for the inclusion of cognitive processes and behaviors. The future will have to show which suite/method will be suited to modeling systemic evolutionary social systems. As to modeling combinatory emergence this seems to be more complex in the large suites again. Evo offers a starting point here, although I wish genomes could be combined more freely and/or entities with different functions could 'live' inside other entities. The nesting of swarms of entities is a dominant characteristic of Swarm after all.

Table 4 gives an overview about the technical characteristics of some simulation suites I have investigated. It is compiled from the point of view of a beginner who wants to build complex models but has little programming expertise. In the table I have just mentioned the easiest way to interact with the program under 'interface'. 'Own language' refers to the possibility to 'program' models in a way that is adapted to the suite and intended to be easier than learning a full blown language.

Usually you can change all these programs if you are an able programmer by reprogramming the simulation modules or building your own, since most packages let you access the source-code. But of course not everybody wants to do so or can initially. So the best strategy might be to start with a simple idiosyncratic model written in something like Pascal or FORTRAN to learn programming concepts and switch to the suite that fits best and learn more complex languages like Java, or C(++). Smalltalk is interesting as an alternative since it is supposed to be a user-friendly, easy to learn language. Unfortunately it is comparatively slow.

It is also comparatively easy to get acquainted with LSD, which offers the possibility to use difference equations as objects. The major drawback is that it is structured around Nelson and Winter type models. So extending existing modules to other model structures is a tough mental exercise.

Table 5: Simulation Suites (as of September 2001)

NameNewest VersionLanguageInterfacePlatformAimwebtested versioninstallation,
ease of use
knowledge level (beyond script/example manipulation)
Ascape1.9.1Javascript (-interface),
allsocial interactionhttp://www.brook.edu/es/dynamics/models/ascape/ReadMe.html1.3easy?,
faulty operation
Cormas2001Smalltalkdrop-down, own languageallspatial/social systems, interactionhttp://cormas.cirad.fr/1999, 2000medium,
Echo1.3beta2CprogrammingSolaris (Unix)ecology, interactionhttp://www.santafe.edu/projects/echo/echo.html1.3beta2relatively easy
Ecolab4.0Tcl/Tk,C++scripts, programmingUnixecology (economics?)http://parallel.acsu.unsw.edu.au/rks/ecolab.html3.4can be difficult,
EVO1.0.1Obj. Cown language, programmingUnix, Winecology, behaviour, economicshttp://www.omicrongroup.org1.0easyprogrammer, looks like easy start for Swarm
LSD2C++, script-languagevisual structuring, programmingWin, UnixNelson + Winter modelshttp://www.business.auc.dk/~mv/Lsd2.0/Intro20.html2easy,
beginner, useful for building up programming expertise
Madkit2.0.1Java drop-down, programmingallsocial/ecological interactionhttp://www.madkit.org1.4okprogrammer
Repast1.3.1Javadrop-down, programmingallhttp://repast.sourceforge.net/1.1?
SDML4.1Smalltalkmenu based, own language
allmental representation, (inter)actionhttp://www.cpm.mmu.ac.uk4.1easy,
high initial learning effort, easy
SeSam-drop-down, own languageApple,
Java announced
http://ki.informatik.uni-wuerzburg.de/sesam/-like above?
not evaluated
Swarm2.1.1Obj. Cscripts, programmingall, Unix very usefulrule-based interaction, CA typehttp://www.swarm.org1.3, 2.1.1ok, can be tedious (linux libraries)programmer
TNG-C++programmingWininteraction (trade-networks)http://www.econ.iastate.edu/tesfatsi/acecode.htmeasyprogrammer

* Conclusion

In the beginning of the paper I have laid out two of the wishful characteristics of evolutionary models of social systems, systemic feedback and combinatory emergence and pointed to some research in biology that aims to extend biological evolution theory in similar directions.

This discussion was followed by a review of existing techniques of evolutionary programming for implementing a model of systemic evolution. One of the major shortcomings of these methods is their intellectual wedding to Neo-Darwinian conceptions of (biological) evolutionary change. The last part surveyed existing simulation suites with respect to their suitability to model social systems and combinatory emergence.

Systemic feedback evolutionary models should be no major problem e.g. on the basis of ESs that are adapted to open-ended evolutionary processes. Problems exist with the integration of cognitive/mental processes due to the required computational resources. The chief bottleneck for the simulation of evolutionary social systems exists on the level of simulations suites in object oriented programming techniques for combinatory emergence.

The aim of this paper is thus to announce my suspicion that the fields of computer science, biology and sociology have to offer even more to each other than has been discussed until now. This suspicion fuels also the view that a successful 'integration' requires a certain shift of focus in all of them. If I am correct, the combination might trigger genuine progress for each of them.

* Acknowledgements

This article is based on a paper presented as poster at WEHIA 2001, Maastricht, June 7-9

* References

BANZHAF, W., NORDIN, P., KELLER, R.E. and FRANCONE, F. D. (1998): "Genetic Programming. An Introduction", San Francisco, CA: Morgan Kaufmann, and Heidelberg: dpunkt.

BARRICELLI, N.A. (1957): "Symbiogenetic Evolution Processes Realized by Artificial Methods", Methodos, vol. 9, pp. 143-182.

BRENNER, T. (1998): "Can Evolutionary Algorithms Describe Learning Processes?", Journal of Evolutionary Economics, vol. 8, pp. 271-283.

CARIANI, P. (1989): "On the Design of Devices with Emergent Semantic Functions", Dissertation, State University of New York at Binghampton.

CHATTOE, E. (1998): "Just How (Un)realistic are Evolutionary Algorithms as Representations of Social Processes?", Journal of Artificial Social Societies and Simulation, vol. 1, https://www.jasss.org/1/3/2.html

HANNAN, M. T. and FREEMAN, J. (1989): "Organizational Ecology", Cambridge, MA: Harvard University Press.

HASS, H. (1970): "Energon. Das verborgene Gemeinsame", Wien: Fritz Molden.

LEVINE, D. N. (1995): "Visions of the Sociological Tradition", Chicago: University of Chicago Press.

MICHALEWICZ, Z. (1992): "Genetic Algorithms + Data Structures = Evolution Programs", Berlin: Springer.

NELSON, R. R. and WINTER, S. G. (1982): "An Evolutionary Theory of Economic Change", Harvard University Press, Cambridge.

NEUMANN , J. VON (1945): "A Model of General Economic Equilibrium", Review of Economic Studies, vol. 13, pp. 1-9.

PANTZAR, M. and CSANYI, V. (1991): "The Replicative Model of the Evolution of the Business Organization", Journal of Social and Biological Structures, vol. 14, pp. 149-163.

RIDLEY, M. (ed) (1997): "Evolution", Oxford Readers, Oxford: Oxford University Press.

RIEDL, R. (1990 [1975]): "Die Ordnung des Lebendigen. Systembedingungen der Evolution", paperback ed., Munich: Piper.

SCHMIDT, S.J. (1994). "Kognitive Autonomie und soziale Orientierung. Konstruktivistische Bemerkungen zum Zusammenhang von Kognition, Kommunikation, Medien und Kultur", Frankfort/Main: Suhrkamp.

SOROS, G. (2000): "Open Society. Reforming Global Capitalism", London: Little, Brown and Co.

SZATHMARY, E. und MAYNARD SMITH, J. (1995): "The major evolutionary transitions", Nature, vol. 374, pp. 227-232.

WAGNER, G.P. (1983): "On the necessity of a systems theory of evolution and its population biologic foundation. Comments on Dr. Regelmann's Article", Acta Biotheoretica, vol. 32, pp. 223-226.


ButtonReturn to Contents of this issue

© Copyright Journal of Artificial Societies and Social Simulation, [YEAR]