Order this book
Department of Sociology and Social Policy, University of Durham, Durham, DH1 3JT.
I have now read through this book three times. Every reading was enjoyable and informative. I make these comments at the beginning of this review because when I saw this book announced, I dreaded reading it. The title suggested that the reader might be in for a trudge through a turgid and unintelligible assertion of the absolute relativism of knowledge with the general postmodernist programme reinforced by a turn to chaos and complexity. That is pretty well exactly what the text is not. It is clearly, indeed beautifully, written and although it seeks to reconcile poststructuralist perspectives and complexity, Cilliers is adamant in dismissing the notion that such a reconciliation provides a license for absolute relativism. This is an important book with a substantial argument to make. It is full of good things. At the same time there are important and suggestive absences in it, absences which are of very considerable significance for the general project of 'simulating society'.
In this review I propose to go through the text section by section, identifying Cilliers' themes and beginning to argue and/or agree with him as the exposition unfolds. At the end of this process I want to make some general remarks about the significance of the book in relation to simulation of the social. There I will note the resonance of much that Cilliers says with recent claims about the desirability of a relational sociology.
Cilliers begins in his preface, an important part of this text, and in his first chapter by defining complexity and locating our capacity to approach complex systems. His definition is quite usual in that he describes complexity in terms of emergence. However, he goes further. First he places very considerable emphasis on the specificity of complex systems, whilst still allowing for the possibility of a general programme of understanding and for the practice of modelling:
'The most obvious conclusion drawn from this perspective is that there is no over-arching theory of complexity that allows us to ignore the contingent aspects of complex systems. If something really is complex, it cannot by adequately described by means of a simple theory. Engaging with complexity entails engaging with specific complex systems. Despite this we can, at a very basic level, make general remarks concerning the conditions for complex behaviour and the dynamics of complex systems. Furthermore, I suggest that complex systems can be modelled.' (p. ix)
It is important to recognize immediately that this insistence on the significance of contingency is radically different from the approach of those such as Holland (1998) who want to understand emergence in terms of the potential of formulated rules and specific dynamic equations. For Cilliers, emergence arises both from the complexity of internal interactions in systems and from the interactions of systems with their external environment. Indeed, given the open character of complex systems, the construction of boundaries between the system and its environment is essentially a product of the process of observation. (p. 5)
Second, whilst not entirely dismissing any relationship between chaos and complexity, Cilliers is sceptical about this as a fundamental assumption. He considers that the complexity programme is in no way dependent on chaotic dynamics as a source, making the point that the mathematical programme of chaos deals with the non-linear interaction of relatively small numbers of equations in contrast to the huge number of interacting components in any real complex system. Cilliers claims that we can manage our understanding of phase shifts, sharp transitions between different states of a system, by using the idea of self-organised criticality rather than the metaphorical apparatus of chaos. This is an important point and one which has very much clarified my thinking but I would be very reluctant to lose the ideas of bifurcation and attractor states which derive from the chaos account. I will return to this point later with particular reference to the idea of control parameters.
The basis of our capacity to address complex systems, says Cilliers, lies in the capacities of electronic computers which essentially extend the cognitive range of science. However, they do so in a way which is absolutely antithetical to the reductionist programme which has underpinned science as a practice since Newton. He puts it like this:
'At the heart of the matter ... our technologies have become more powerful than our theories. ... We can do with technology what we cannot do with science.' (pp. 1-2)
Here plainly Cilliers is using science in the sense of constructed general theory, although he does not always stick to this usage. Later in Chapter One, the object of science is defined in terms of prediction in contrast to the object of philosophy which is cast in terms of understanding. In the latter usage, Cilliers is treating science as the equivalent of engineering as defined by Crutchfield:
' ... the epistemological problem of nonlinear modelling 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--vis natural laws. The engineering view of science is that it is mere data compression; scientists seem to be motivated by more than this.' (Crutchfield 1992, p. 68)
This approach to representation is not merely something Cilliers accepts. He argues that for complex systems, it is all we can achieve, although he does not necessarily think we aspire to predictive power.
For Cilliers, representation is always distributed. We cannot generate algorithms to describe significant natural complex systems. We might however be able to create models which work in the same way. His theory of representation is, I think, essentially one of heterologous analogy as that term is defined by Khalil (1996). It is not a theory of strong representation, a term which I take to be equivalent to Khalil's conception of unificational metaphor, similarities arising from the operation of the same basic law. Representation is distributed and based on connectionism - the core argument of the second chapter of this book. Cilliers' distributed representation using primarily (but not exclusively) neural networks, can represent because such network systems are to be understood in terms of the interactions among elements and not through an underlying generative law. They are themselves part of the general class of complex systems. I will note here that Cilliers' identifies evolutionary algorithms as another component of the distributed representational armoury, which is interesting given that Holland (who developed this approach) is insisting on rule bases for emergence.
Throughout the text Cilliers has two referent systems for his account of complexity. These are the mammalian brain and natural language. The connectionist account of representation, presented against the rule based approach of strong representation, is primarily developed in relation to the debates surrounding the possibility and nature of artificial intelligence. The arguments about language begin in earnest in Chapter Three where Cilliers presents an exposition of Saussure's model of language with its absolute insistence on the significance of meaning (as opposed to structure) and continues with a review of Derrida's 'deconstruction' of Sassure's account. The point of this exercise is that Cilliers can interpret Derrida's approach as one which identifies language itself as an open system far from equilibrium, with (and he cites Luhmann here) 'more possibilities than can be actualized'. (p. 42) Derrida's conception of the significance of différance is equated in systemic terms with Freud's account of the relationships among neurones as one of differences.
This is an interesting and persuasive approach. What Cilliers does is take two fundamental areas where scientific and philosophical debate intersect, the nature of language and the nature of mind, and argue for the objects of both intellectual projects as being far from equilibrium systems founded on distributed connections and working through the differences among the components of the systems. In other words, the account is absolutely relational. Explicitly this is a programme of analogy, but I think there is more to it than that. Certainly in implicit terms, and perhaps explicitly as well, Cilliers is saying that the two great basics of understanding, mind and language, can only be comprehended as complex far from equilibrium systems and that this is fundamental (a word with a different content from foundational) to our capacities for engaging with the world.
Following arguments seeing off John Searle in Chapter Four, Cilliers turns to the basic question of representation. There is a lot in this chapter, but what is most interesting for a review to be published in JASSS is Cilliers' use of Baudrillard's notion of the hyperreal and simulation. For Baudrillard, and it seems Cilliers (although I am less sure about this), the real is something that can be copied. Here, we come to one of the significant absences in this book. Cilliers does not engage at all with Bhaskar's critical realism in which the domain of the real is not something which can be copied but something which contingently generates the actual. Instead, he seems to endorse Baudrillard's dismissal of essential abstraction of the real - certainly Bhaskar's generative mechanisms would seem to have an essential character in this sense - turning in consequence to the idea of simulation as something which attempts to repeat the real. (p. 84) This point is illustrated by reference to the way in which a neural network can be simulated on a digital computer, but I wonder if the illustration does in fact support the argument. Certainly we can consider a simulated neural network as being the same kind of thing in terms of system character as a constructed neural network, but both do share a reality exactly founded on their essential connectionism - an argument for a non-reductionist essentialism.
This turn to hyperreal simulation has its problems because of the emphasis on meaning itself as an emergent property in the process of interpretation, instead of some causal process inherent in the system being observed. For example, following Derrida, Cilliers considers that the significance of images in our culture is the source of the general tendency in science '... to fall prey to a metaphysics of presence.' (p. 82) This would seem to imply a rejection of the iconological approach (see Reed and Harvey 1996) and yet the turn to interpretation through images, and moving images at that, seems a fundamental part of simulation's project for dealing with our inability to reduce complex systems to sets of differential equations. Simulation is to a considerable extent an exercise in representing the real, again in a way which is explicitly not reductionist/analytical but which nonetheless asserts that the representation is an analogue of the thing being represented. Really this is the crucial ambiguity in Cillier's text, and one which he himself keeps coming back to. At one level, the postmodernist account seems to deny the possibility of any representation but Cilliers is always arguing that a distributed holistic analogy is possible.
'It bears repetition that an argument against representation is not anti-scientific at all. It is merely an argument against a particular scientific strategy that assumes complexity can be reduced to specific features and then represented in a machine. Instead it is an argument for the appreciation of the nature of complexity, something that can perhaps be 'repeated' in a machine, should the machine itself be complex enough to cope with the distributed character of complexity.' (p. 86)
And yet as Cilliers puts it, in a way which any sociologist must agree with:
'... a certain theory of representation implies a certain theory of meaning - and meaning is what we live by ...' (p. 88)
Chapter Six deals with 'self organisation in complex systems'. Again there is much that is interesting and important here, but this discussion contains (an appropriately postmodern mode of expression) two of the crucial absences in the text. The first is that there is no sense of the nested character of complex systems. Words and neurones can perhaps be properly considered as atomic nodes. In other words, we don't have to think about them as complex systems in their own right, although a neurone is certainly internally a complex system with potential implications if cancerous for the whole distributed system in which it is a node. However, the social is always nested. Individual selves, which may not be much after Derrida, but they are something, are themselves complex systems with their own evolutionary trajectories. Even the individual, the macro-actor, may have effects on social systems if powerful enough. Certainly collectivities of individuals, collective actors with their own emergent properties, form part of a social order. If we think about a purely structural representation of this nesting, the discussions of localities within regions within a global economic and cultural system, then we can see the issue. The urban is one of the few domains of significant social simulation and work on simulating urban processes demonstrates absolutely the need to conceptualise systems as nested (Batty 1995, Batty and Xie 1997).
The other absence, essentially a deliberate deletion by Cilliers, is the idea of control parameters. He says:
'In our analysis of complex systems ... we must avoid the trap of trying to find master keys. Because of the mechanisms by which complex systems structure themselves, single principles provide inadequate descriptions. We should rather be sensitive to complex and self-organizing interactions and appreciate the play of patterns that perpetually transforms the system itself as well as the environment in which it operates.' (p. 107)
Yet complex systems do change state on the basis of the transformation of key parameter values. I have argued (Byrne 1997) that the polarised form of cities is something that can be understood in terms of the transformation of parameters to do with economic engagement and power. It seems to me that abandoning the idea of control parameters, however complex may be the interactive processes through which the control of these parameters are expressed, is also an abandonment of the possibility of effective agency based on understanding.
This tension runs through Cilliers' final chapter which deals with 'Complexity and Postmodernism'. I think this is fascinating. Cilliers argues that the postmodern programme cannot be understood in purely relativist terms but is instead to be understood as an assertion of the local validity of accounts rather than their universal validity. I have absolutely no problem with this but it is an unusual version of the postmodernist position, although Cilliers marshals a good number of citations, particularly from Derrida, in support of this positions. The chapter includes some interesting ideas on understanding society in connectionist and complex terms (although here the absence of a sense of nested systems again is a weakness), a fascinating discussion of language and some important preliminary statements about the ethical implications of the general argument. The element which I want to focus on is Cilliers discussion of the nature of the scientific project. Again, there is a great deal to agree with and some things I would want to argue about. Certainly the notion that the complexity programme is fundamentally antithetical to disciplinary boundaries in science is absolutely correct. However, I think Cilliers is hung up on, and moreover knows that he is hung up on, the distinction between the postmodernist programme as a general deconstruction of science and the possibility of using complexity as the basis of a coherent and acceptable postmodern science. The first is the usual story. The second is the interesting one. There is a lot in this book to help us get going with that project. In an afterword called 'Understanding Complexity', Cilliers puts his finger right on the fundamental issue. As he says, his work could be accused of the performative fallacy - attempting to do what he claims is not possible: generating a theory which insists on radical contingency but claims to be generally valid. Well yes, that is what he is doing but it is not fallacious. If we think of complexity, of his distributed connectionism, not as a law but as a metalaw, here understood as a general statement about the complex character of dissipative far from equilibrium systems, then we have something which allows for the local but is general. This is entirely compatible with the realist account. Postmodernists in general don't like realism, but for my money Cilliers ends up as an implicit realist.
So what does this all have to say to those interested in simulating societies? Well let me try to explain that by citation of resonances. Emirbayer argues (in his 'Manifesto for a Relational Sociology') that:
The imageries most often employed in speaking of transactions are accordingly those of complex joint activity in which it makes no sense to envision constituent elements apart from the flows within which they are involved (and vice versa).' (Emirbayer 1997, p. 289)
The implication for me of Cilliers' arguments is that simulation is indeed a necessary part of the sociological project - he resonates absolutely with Abbott who remarks:
'If I may use another forbidden word, we will have to employ simulation. Game theory will not get us very far because it is ignorant, except in the most general terms, of a serious concern with structure and with complex temporal effects. But simulations may help us understand the limits and possibilities of certain kinds of interactional fields, and that would be profoundly sociological knowledge.' (Abbott 1998, pp. 176-177, with original emphasis)
This is what the social world is like - it has to be understood and explored in relational terms. Of course Cilliers proscribes some approaches to simulation, essentially all hard representations, and prescribes others, those based on distributed connectionism. He has persuaded me, not that I needed much persuading. Let me conclude. I liked this book and learned from it. In this long review I have tried to give a reasonable representation of the book, based on a sampling strategy. Of course in the nature of the thing, appropriately enough, it needs to be taken as a whole.
ABBOTT A. 1998. The Causal Revolution, Sociological Methods and Research, 27:148-181.
BATTY M. 1995. Cities and Complexity: Implications for Modelling Sustainability. In J. Brotchie, M. Batty, E. Blakely, P. Hall and P. Newton, editors, Cities in Competition, Longman Australia, Sydney, 469-486.
BATTY M. and Y. Xie 1997. Possible Urban Automata, Environment and Planning B - Planning and Design, 24:275-292.
BYRNE D. 1997. Chaotic Places or Complex Places: Cities in a Postindustrial Era. In S. Westwood and J. Williams, editors, Imagining Cities, Routledge, London, 50-72.
CRUTCHFIELD J. P. 1992. Knowledge and Meaning: Chaos and Complexity. In L. Lam and V. Naroditsky, editors, Modelling Complex Phenomena, Springer-Verlag, New York, NY, 66-101.
EMIRBAYER M. 1997. Manifesto for a Relational Sociology, American Journal of Sociology, 103:281-317.
HOLLAND J. H. 1998. Emergence: From Chaos to Order, Addison-Wesley, Reading, MA.
KHALIL E. L. 1996. Social Theory and Naturalism. In E. L. Khalil and K. E. Boulding, editors, Evolution, Complexity and Order, Routledge, London, 1-39.
REED M. and D. L. Harvey. 1996. Social Science as the Study of Complex Systems. In L. D. Kiel and E. Elliott, editors, Chaos Theory in the Social Sciences, University of Michigan Press, Ann Arbor, MI, 295-324.
Return to Contents of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 1999