Copyright Sociological Research Online, 1997


Byrne, D. (1997) 'Simulation - A Way Forward?'
Sociological Research Online, vol. 2, no. 2, <>

To cite articles published in Sociological Research Online, please reference the above information and include paragraph numbers if necessary

Received: 18/4/97      Accepted: 20/6/97      Published: 30/6/97

...the epistemological problem of nonlinear modeling can be crudely summarized as the dichotomy between engineering and science. As long as a representation is effective for a task, an engineer does not care what it implies about underlying mechanisms; to the scientist though the implication makes all the difference in the world. The engineer is certainly concerned with minimizing implementation cost ... but the scientist presumes, at least, to be focused on what the model means vis-a-vis natural laws. The engineering view of science is that it is mere data compression; scientists seem to be motivated by more than this. (J. P. Crutchfield, 1992: p. 68)


I have a problem with simulation, and I have a problem with having a problem with simulation. The problem of the first instance and the meta-problem have exactly the same origin in what has moved beyond an interest in the ideas which can be described by the words Chaos and Complexity and become a conviction that the analogies which constitute Chaos/Complexity theory are essential for a proper understanding of the world as it is. The problem of the first instance is ontological. If the world is essentially only describable in terms of chaotic/complex causation then it has to be understood as characterized by deterministic chaos. But:

Deterministic chaos is generated in the system model by destabilizing mechanisms such as positive feedback, and by nonlinear constraints such as human values. Systems specified in this way are very sensitive to initial conditions, with the result that they cannot be used to predict future system states from initial states. (Seror, 1994: p. 34)

In other words, if the world is chaotic and complex how can we ever set up the initial parameters of any simulation exercise with sufficient precision so that we can actually drive forward a simulation through time in a way which has any correspondence with what might really happen?

The meta-problem is methodological. Much of what constitutes the new science founded on chaos and complexity actually derives exactly from computer based simulation exercises which involve both the iconographic modeling of chaos/complex mathematical functions and the simulation of artificial eco-systems in a way which seems to demonstrate the possibility of emergent properties. In these abstract exercises which are an essential part of the chaos/complexity project, mathematics and programming come together in an experimental form. What was always seen as a deductive and nomothetic process with its foundations in logic uses the new information processing technology to become inductive and aesthetic/idiographic.

There seems to be a contradiction here. On the one hand Chaos/Complexity seems to correspond to Stephen J. Gould's seminal description of contingency:

I am not speaking of randomness ... but of a central principle of all history - contingency (original emphasis). A historical explanation does not rest on direct deductions from laws of nature, but on an unpredictable set of antecedent states, where any major change in any step of the sequence would have altered the final result. This final result is therefore dependent, or contingent, on everything that came before - the uneraseable and determining signature of history. (Gould, 1991: p. 283)

This certainly allows for the chaos/complex non-linear modeling of real historical process, but it suggests that any attempt at simulation as a way of seeing what might be as opposed to what has been is a pointless exercise. At the same time in the domains of experimental mathematics and artificial intelligence people are building both chaotically deterministic models and chaotically stochastic1 models, and specifying the behaviour of AI agents, and it is from the results of these exercises that much of the reinforcement of the chaos/complexity programme actually comes. The development of digital computing and of our capacity for using it to extend our cognitive range, seems to provide a new means of knowing the world which is fundamentally different from that of experimental control. The epistemological implication of this is that we are dealing with holistic systems, that reductionism and its apparatus of hypothesis testing no longer holds. For sociologists, the most important aspect of this is exactly emergence. Conte and Gilbert are (mostly) right when they say:

Social simulation studies provide an opportunity to fill the gap between empirical research and theoretical work while avoiding the individualist tendency of most mathematically-based approaches. In particular, social simulation provides not only for testing hypotheses, but also an observatory of social processes. It can, therefore, offer the basis of new efforts to devise categories of description and new analyses of social reality. In other words, social simulation can provide instruments for modeling sociality. (Conte and Gilbert, 1995: p. 5)

Hypotheses, I am going to suggest, have nothing to do with it. However, the possibility of handling robust chaos and the reality, a word used in the most deliberate and uninnocent sense, of emergent properties does mean that simulation may very well have a lot going for it. It may be one of the tools a reflexive social science has to offer to the processes of politics and administration. If that sounds like social engineering, well that is what it is, but it is a social engineering without the privilege of determined certainty and with the proper respect offered to both the knowledge and the intentions of people in general.

Analogies? Engineering vs Reductionist Science

I find it useful to think about simulation first in analogue form. Some of the most delightful and evocative2 objects I am familiar with are the ships' models which form part of the display of the Newcastle Museum of Engineering and Science. These are delightful in their detail but they were not made as art objects. Rather, they were produced as literal analogue models to be tested out in the giant tank at the British Ship Research Establishment in Wallsend. Naval Architects could not design a ship on the basis of classical deterministic mechanics alone. They had to use analogue models and test them under analogue wave conditions, precisely because the interaction of ship's hull and water could be complex in the form of turbulence. Exactly the same rationale lies behind the testing of aircraft models in wind tunnels. It is rather important to realize that a computer based simulation, although founded in digital processes, remains essentially an analogue in just the same way as a physical model being made to work in a tank.

What are analogues? They are essentially a kind of analogy. Here I find the discussion of the forms of metaphor developed by Khalil (1996: pp. 4 - 7) very helpful. Khalil distinguishes superficial, heterologous (or analogous), homologous and unificational metaphors ' the criterion of the kind of resemblance which a metatheoretical statement is supposed to inform' (Khalil, 1996: p.4). Superficial metaphors are simply similes. Khalil's distinction between heterologous and homologous metaphor is founded on the contrast between similarity of analytical function (his example is the wings of a bat and the wings of a butterfly which do the same thing but have very different origins) on the one hand, and common context or origin (the forelimbs of the bat and the mouse) on the other.3 Unificational metaphors ' similarities when they arise from the same common law' (Khalil, 1996: p. 6).

The important distinction here is between analogy (heterologous metaphor) and unificational metaphor. This is the issue identified by Crutchfield in the epigraph to this piece. Engineers have traditionally tested their models by the criterion of whether or not they work in the solution of problems of achievement - an engineering problem is how to get something done. Scientists in the tradition of Newton and Descartes have sought to establish fundamental laws which will explain why things are as they are. Simulations are fine for the first process and deny the validity of the second. In other words simulations cannot and must not be regarded as equivalent in a homologous sense to experiments. They are not even analogues and should not be seen as similes. There has been a good deal of understandable confusion about this. Gilbert and Doran endorsed the possibility of ' scientists experimenting with societies (of computational agents) rather than having to rely on social events' (Gilbert and Doran, 1994: p. vii). That might be regarded as simply a popular, if perhaps confusing, use of the term 'experimenting' but Gilbert has (see above) used the idea of hypothesis testing in relation to simulation.4 That won't do.

The reason why it won't do is because in a chaotic and complex world non-linearity means that no general and invariant law can hold. There is no point in establishing laws which are universal because in chaotic/complex systems only local solutions are valid. This is not a license for anti-rational postmodernism. Local solutions can be established and matter a great deal, but it is precisely the evolution of nonlinear processes which can be explored by simulation and it may that we can come to local, but nonetheless important solutions to problems.

Nota bene, this is not an abandonment of the idea of universalist meta-laws. Indeed the notion that the sort of system which is of interest to us, dissipative far from equilibrium systems (see Prigogine and Stengers, 1984; Reed and Harvey, 1992; Reed and Harvey, 1996), represents a general type all of which are heterologous with each other, all of which can be understood as working in the same way, is exactly the foundation on which simulation must stand. Simulations are interesting and useful because they involve the creation of modelled systems which are analogous in a fundamental way with the social systems which are our concern as sociologists. The way in which emergence is identifiable in simulations and seems the only way of sensibly resolving micro/macro, structure/action relations illustrates that very well. There still remains the issue of whether we can model closely enough to achieve the kind of predictive power of the old analogue engineering models, which is the issue of robust chaos, but as analogies simulations have very real possibilities.

Realism/Emergence - Micro/Macro

Gilbert indicates the significance of emergence in relation to simulation in these terms:

The description of complex systems suggests that a candidate criterion is that it should not be possible to derive analytically the global emergent behaviour solely from a consideration of the property of agents. In other words emergent behaviour is that which cannot be predicted from knowledge of the properties of agents, except as a result of simulation. (Gilbert, 1995: p. 150)

He then goes on to draw back from the implications of this statement by remarking that this means that emergence is always going to be a transient property because it is always possible that in future we will find some analytical solution and that what is essentially interesting is the relationship between the micro and macro properties of systems. In my view the original statement is correct and stands as an assertion of the essential character of simulation as an investigative tool. The retreat/qualification is associated for Gilbert with an endorsement of Giddens' conception of structuration. It is precisely the elision of levels central to structuration theory which means that it is not adequate as a way of conceptualizing a complex world in which non-linear emergence is a fundemental property of social systems.

Archer has recently made this very plain. Her explicit rejection of structuration as an analytical strategy if founded on a commitment to '... social realism's ontology in which "structures" and "agents" belong to different emergent strata of social reality ... (This) avoids reducing one to the other or eliding the two' (Archer, 1996: p. 679). Reed and Harvey (1992; 1996) have demonstrated the compatability of social realism as a philosophical ontology with chaos/complexity as a scientific ontology, but we may be able to take this further into the domain of methodology by a consideration of simultion in realist terms. Archer makes a point about the ontological/epistemological position of realism which can take us in this direction:

Instead of a one-dimensional reality coming to us through hard data supplied by the senses, to speak of emergence implies a stratified (original emphasis) social world including non-observable entities, where reductive talk about its ultimate constituents makes no sense, given that relational properties pertaining to each stratum are real, that it is nonsense to discuss whether something (like water) is more real than something else (like hydrogen and oxygen), and that regress as a means of determining 'ultimate constituents' is of no help in this respect and an unnecessary distraction in social or any other type of theorizing. Reed and Harvey (1996: p. 686)

I want to agree with the anti- reductionist assertion of emergence in this passage and disagree radically with its proscription of empiricism and regress. Of course the term empirical is ambiguous here. If it means the developed and precise meta-theoretical position of empiricism, then there is no quarrel. However, it seems to me to have at least possibly the much more general sense of actual enquiry into the world, the equivalent of general induction and that is particularly important if regress is precluded as a method of enquiry.

Regress is not a synonym of analysis, although in the passage quoted Archer seems to be using it as such. Interestingly, in statistics 'analysis' has to be added to ensure that regression is an analytical programme - the full term for what is often shortened to regression is regression analysis. In other words going back over is not necessarily a process of breaking things up into parts.

One of the most interesting of Prigogine's assertions is that what we need is not a science of what is but a science of becoming. If we are dealing with a world characterized by emergent properties then what we want to be able to describe is the way in which those properties emerged. This is not a process of analysis but it is a process of historical account, a regress, the kind of historical science validated by Gould in his description of 'the nature of history'. In this frame, simulation is becoming driven forward as analogue. It is of course progression. What we do is look at the progression backwards when it has happened. We examine the regression to see what happened and how. I am irresistibly reminded here of E.P. Thompson's definition of class as not consisting in a static relationship but in the noise, smell, sound and fury, not the machine but the way the machine works. Simulation certainly offers us the possibility of building little machines and seeing how they work. The question is can we really do this if we are dealing with chaos, with the inability in a deterministic system to specify with enough precision? There are two answers to this, one involving art and the other leaving open the possibility of science. Let us take them in turn.

Iconography: The Art of Simulation?

The mark of a good simulation is that it separates the essential from the incidental, cutting through what is deemed irrelevant detail to get at the heart of a problem. This involves making instinctual judgements about which details are crucial and which can be ignored (Johnson, 1996: p. 244).

Peak and Frame's introduction to chaos for non-mathematicians is called Chaos Under Control: The Art and Science of Complexity (1994). The control element will be taken up in the next section. Here I want to focus on the word 'art' which resonates very closely both with Johnson's description of modeling as kind of art craft and with the discussion of iconological modeling presented by Reed and Harvey (1996). We are dealing here with the two ends of the artistic spectrum, with art as craft and with high art, but the essential thing about both is that neither can be understood or learned by reduction to first principles. Reed and Harvey describe iconological modeling as representing:

... a radical epistemological break in the type of knowledge it provides to researchers. Iconological modeling is rooted in a pictorial method, in visual correspondences rather than in deductive reasoning. (Reed and Harvey, 1996: p.309)

Iconological modeling is a response to the actual impossibility of an analytical account of the nonlinear differential equations which generate strange attractors. It is a product of the development of computer graphics as a tool for presenting the course of the development of a nonlinear mathematical process. As Reed and Harvey point out it replaces the analytic with the visual. 'In iconological modeling the gaze is more important than deductive logic in grasping the evolution of a chaotic structure' (1996: p. 310). It is important to note that the validating of this kind of graphical representation of the world is inherently and essentially qualitative. When we note that there is no achieveable analytical solution to these non-linear equations, that they do not integrate, we are saying that the major tool of applied science since Newton is not applicable to them. At the farthest end of the mathematical programme the quantitative breaks down into the qualitative.

Iconological modeling seems to be inherently simulational once it acquires any kind of interpretive force. In other words once the parameters specified for the start of the driving of an equation with a deterministic chaos form are considered to stand for the values of real properties of a real system, then we have a simulation. Interestingly, the end product presented for interpretation in simulation in general is often iconographic, as is the case with the results of Gilbert's historical simulation of 'The Structure of Academic Science' in his Figure 1.5 The presentation of holistic figures, and even more of moving sequences of development, as account, instead of the specification of a series of reductionist equations, is of enormous significance.

Robust Chaos and Social Action

There still remains the issue of real prediction. The historical adequacy of simulation as a new and complex form of narrative is not disputable. Can simulation be the basis of the development of even general theoretical accounts which have any forward reach? Conte and Gilbert suggest that this might be possible:

The question here is not 'what has happened', ... or even 'what might have happened', but rather 'what are the sufficient conditions for a given result to be obtained?' While the first two questions are exclusively descriptive, the latter may have prescriptive consequences. It may provide hints about how to enhance or reinforce some social strategies by telling us, for example, under what conditions these strategies become stabilized. (Conte and Gilbert, 1995: p. 3)

I actually think that this position is compatible with Seror's, at first sight, dismissal of predictive power when we are dealing with chaotic systems which:

...are highly sensitive to initially specified conditions. The simulation is not, therefore, useful to predict future states from initial states of the system; the explanatory focus is on the dynamics of change from one node to another, such as the change from a chaotic to a non-chaotic model or pattern of behaviour. Such changes are the representation of the model's active structure which mimic the evolutionary behaviour of social systems. (Seror, 1994: p. 38)

The crucial point is that prediction and prescribing are not similes. We can't know what will happen regardless of our acts. We can know what might happen if we act in a certain way. In Mao Tse Tung's words 'we know the world by changing it'. Poincare discussed the possibility of control of robust chaos early this century and Peak and Frame (1994: pp. 233 - 239) discuss engineering applications inititally in terms of stabilizing by perturbation, but subsequently precisely in terms of the adaptive capacitities of chaos based systems. In this way of thinking, simulation is clearly a tool which helps us not know what will happen, but what can be made to happen.


I started the processes of reading for this article and writing it, quite genuinely not knowing what conclusion I would come to about the value of simulation. The conclusion arrived at is, I am afraid, local and unstable and liable to non-linear transformation, but for the moment it is that simulation has its uses. Actually I have no problem with the value of regress, of historical simulation. It seems to me to be a useful tool of interpretation which will be the more useful, the more it is iconographic. I want simulation to be a tool for prescription. As of now I think it is.


1 The word 'stochastic' is often used in the statistics literature as a synonym for random. However, its etymology does not imply absolute indeterminacy. Rather given its origins in 'aiming' it is better understood as descriptive of processes which whilst containing a random element, nonetheless have a direction and an outcome. Stochastic processes are often described by reference to a drunkard's walk, but as Chesterton informed us, the rambling English drunkard may have gone to Birmingham by way of Beechey Head, but s/he got there in the end. This matters because of course there is really no such thing as stochastic chaos if the term stochastic is taken as equivalent to random. Indeed the idea of chaos, of Poincare's effect which is too small to measure but which matters a great deal and which we choose to call chance, denies randomness as a meaningful concept. I know that my version of stochastic seems purposeful; the system may wander but it gets there. I see no problem in this in any system where human intentionality matters, and I really am prepared to go for it in nature as well.

2 Evocative of Modernity alongside those other absolute products of modern design, railway engines.

3 Frankly I find this distinction to be of not very much import.

4 I don't want to get at Gilbert here, far from it. In fact his book on Analyzing Tabular Data (Gilbert, 1993) with its very important recognition of the way in which apparently inductive tests were really used in exploratory model building, has played an important part in my coming to the position I hold on these issues.

5 Gilbert's function is not a chaos function but the general principle holds.


ARCHER, M. (1996) 'Social Integration and System Integration: Developing the Distinction', Sociology, vol. 30, no. 4, pp. 679 - 699.

CONTE, R. and GILBERT, N. (1995) 'Introduction: Computer Simulation for Social Theory' in N. Gilbert and R. Conte (editors) Artificial Societies.London: UCL Press.

CRUTCHFIELD, J. P. (1992) 'Knowledge and Meaning: Chaos and Complexity' in L. Lam and V. Naroditsky (editors) Modelling Complex Phenomena. New York: Springer-Verlag.

GOULD, S. J. (1991) Wonderful Life: The Burgess Shale and the Nature of History. London: Penguin.

GILBERT, N. (1993) Analyzing Tabular Data: Loglinear and Logistic Models for the Social Researcher. London: UCL Press.

GILBERT, N. and DORAN, J. (editors) (1994) Simulating Societies. London: UCL Press.

GILBERT, N. and CONTE, R. (editors) (1995) Artificial Societies. London: UCL Press.

GILBERT, N. (1995) 'Emergence in Social Simulation' in N. Gilbert and R. Conte, (editors) Artificial Societies. London: UCL Press.

GILBERT, N. (1997) 'A Simulation of the Structure of Academic Science', Sociological Research Online, vol. 2, no. 2, <http://www.socresonl>

JOHNSON, G. (1996) Fire in the Mind. London: Viking.

KHALIL, E. L. (1996) 'Social Theory and Naturalism' in E. L. Khalil and K. E. Boulding (editors) Evolution, Complexity and Order. London: Routeldge.

PEAK, D. and FRAME, M. (1994) Chaos Under Control. New York: W. H. Freeman.

REED, M. and HARVEY, D. L. (1992) 'The New Science and the Old: Complexity and Realism in the Social Sciences', Journal for the Theory of Social Behaviour, vol. 22, pp. 356 - 79.

REED, M. and HARVEY, D. L. (1996) 'Social Science as the study of Complex Systems' in L. D. Kiel and E. Elliott (editors) Chaos Theory in the Social Sciences. Ann Arbor: University of Michigan Press.

PRIGOGINE, I. and STENGERS (1984) Order out of Chaos. New York: Bantam.

SEROR, A. C. (1994) 'Simulation of Complex Organizational Processes: A Review of Methods and their Epistemological Foundations' in N. Gilbert and J. Doran (editors) Simulating Societies. London: UCL Press.

Copyright Sociological Research Online, 1997