Complexity and Postmodernism

Paul Cilliers (Routledge, 1998)

In a special issue devoted to books on complexity and management, Complexity and Postmodernism is slightly out of place as, although social structure does feature as an undercurrent, Paul Cilliers explicitly addresses management only in two brief paragraphs (pp. 110-11). The book focuses instead on complex systems and their relationship to theoretical approaches loosely bundled under the rubric “postmodernism,” arguing in particular that postmodernism displays a greater inherent sensitivity to complexity than does the analytical approach characteristic of the scientific method. Yet this focus may still hold some interest for those concerned with the central theme of this special issue. Indeed, I believe that the material does hint at useful conclusions about management. The conclusions are likely to differ starkly, however, from those that the author might prefer.

The first chapter is a précis of sorts of much of what follows, together with an attempt to outline the principal features of complex systems. Observing that a complex system “is not constituted merely by the sum of its components, but also by the intricate relationships between these components” (p. 2, emphasis original), Cilliers suggests that such systems reveal a “fundamental flaw” (p. 2) in the analytic method, which he describes as being all about conceptually dividing things into simpler units and then putting them together again. But if the analytic method is taken from the very start to preclude analysis of systems in terms of simple units displaying intricate interrelationships, then I am at a loss to name even a single scientist who practices this “analytic method”. Perhaps we are all closet “technologists,” since in place of “analytic” science Cilliers credits “technology”—in the form of powerful computer simulations—with making the study of complex systems possible at all.

Subsequent chapters continue with the critique of analysis as a subsidiary project, while the principal aim is to build up a picture of a postmodern alternative through a discussion intertwined with and informed by ideas about complex systems. This begins with introductions to connectionism and poststructuralism, and continues with discussions of representation and self-organization. Sandwiched in between is a chapter on John Searle’s now-famous Chinese Room argument about the possibility of computer intelligence, where Cilliers observes that premises adopted by Searle are incompatible with Derrida’s picture of language. Actual argument about either position is, however, conspicuously absent. The concluding chapter, which shares the book’s title, is largely a literature review of selected work—from both postmodernism and the study of complexity— which may be well positioned to support bridges linking the two fields.

While I cannot pretend any significant expertise in the ideas of Saussure and Derrida that form the core of Cilliers’ treatment of poststructuralism, that material is very well presented. Cilliers manages to communicate a poststructuralist picture of language as a network of meaning without slipping into the normal postmodern discourse and rhetoric that can be so impenetrable to those outside the field. Other areas of the book, however, left me less enthusiastic.

A principal distraction of the connectionism chapter, for instance, is its focus on material that is almost exactly 10 years out of date, with the bulk of the discussion limited to backpropagation. And Cilliers’ critique of Fodor and Pylyshyn’s famous attack on connectionism completely misses the central issues of systematicity and constituent structure. The chapter’s concluding section suggests that, “many phenomena ... simply cannot be understood properly in terms of deterministic, rule-based or statistical processes” (p. 35), citing quantum mechanics as a specific example. Yet critical readers will ask: what are the two central processes of quantum theory, if not deterministic and rule based on the one hand (unitary evolution) and statistical on the other (state vector reduction)?

The book’s treatment of self-organization rests mainly on metaphor, with occasional links to actual technical work appearing only in expository settings and often burdened by inappropriate background or interpretation. For instance, Cilliers reiterates his affinity for Hebb’s 50-year- old rule of synaptic modification in the context of Gerald Edelman’s work on neuronal group selection, yet Edelman himself explicitly predicts that purely Hebbian learning will never be found to occur in human brains. (Elsewhere, Cilliers links Steve Grossberg to Hebbian learning, yet it was Grossberg who explicitly identified the rule’s failure in terms of the noise-saturation dilemma and who continues to develop myriad architectures that have long since superseded Hebb.) Worse, Cilliers’ picture of self-organizing systems is strangely narrow, requiring for instance an explicit capacity for adaptation supported by memory storage.

Each researcher is likely to harbor differing opinions on how other authors’ work might have helped or hindered Cilliers’ project, but for my part I cannot help but think that he has missed out by passing over certain other authors. Nancy Cartwright, for instance, might have brought some sophistication to the view of modeling advocated in the first chapter, while the notion of supervenience so widely used in “analytic” philosophy might have helped readers pin down exactly what Cilliers means by emergent properties. Other perspectives on networks of meaning, perhaps from Quine or Saul Kripke, would have been helpful to those new to the field of postmodernism. Together with Greg Chaitin’s algorithmic information theory, introduced briefly—and immediately rejected—as a tool for quantifying complexity, Charles Bennett’s notion of logical depth surely could have been a great help to Cilliers in lending formal substance and clarity to frequent assertions about the complexity of systems and their models.

These considerable nitpickings aside, those with an interest in complexity and management may nonetheless find something here worth thinking about. I won’t be alone in recognizing Cilliers’ postmodern view of the “agonistics of the network” as a startlingly appropriate description of decision making and other behavior within many large organizations. The picture of behavior motivated not by scientific analysis or rational argument, but by “narrative knowledge”—which grounds itself in the “network” of the local social group and which “cannot be the subject of argumentation and proof’ (p. 129, Cilliers approvingly paraphrasing Jean-Frangois Lyotard)—rings all too true.

Yet if these organizations appear “postmodern,” I believe that this is not, contra Cilliers, because they are complex (which they are), but because they have unwittingly already adopted the course of action that he promotes: they have given up on the tough job of understanding and managing their complexity through science and analysis and have descended into “network” philosophy instead. The book has correctly identified the existence of a relationship between complexity and postmodernism, but I wonder if it mightn’t have put things back to front? I believe that headway will be made not by adopting a philosophy that considers “complexity science” an oxymoron, but by embracing the scientific tools of relevant fields and doing the hard work that it takes to understand complexity. These tools have a great deal to offer, and it is in becoming intimate with them and the constraints and general heuristics that they suggest that managers of complex organizations will find insight.


This book is a gift to those of us who enjoy exploring novel ideas, revisiting established truths and finding new perspectives on old issues. Its stated aim is to “stimulate transdisciplinary discussion on the subject of complexity” (p. 141) and Paul Cilliers has achieved this aim. His work is stimulating, his claims interesting and the consequences of his approach to the subject of complexity far reaching. He develops his arguments with remarkable clarity—his intellectual honesty and insight are impressive. This work offers an exciting experience and is a true pleasure to read.

Yet, I am deeply disturbed by the conclusions that naturally follow the reasoning that he presents in this book. In this review I will examine my discomfort in more detail and, hopefully, exercise a comparable degree of honesty and clarity.

The book is written in a style that carries the reader effortlessly, it seems, through a rigorous process of thinking about complex systems. Cilliers looks for approaches to the study of complexity that are more sensitive to features such as self-organization and learning, as well as the special relationship between complex systems and their environments. He also places these in a broader philosophical and scientific context. He says:

I argue that the traditional rule-based and analytical approaches to complex systems are flawed, and that insights from postmodern and post-structural theory can help us to find novel ways of looking at complexity.

... [T]hese insights can influence our models of complex systems. The suggestion is that ‘distributed’ methods of modeling share some of the characteristics of complex systems and that they therefore hold more promise than rule-based models, models which incorporate a strong theory of representation.” (p 25)

The argument in favor of connectionist models, or neural networks, to account for complexity rests on three of their main characteristics:

For the purposes of this review I will take up only the issue of representations, and this I will consider only in relation to the claims that Cilliers makes about the role of representation in modeling language.

The debate about representations and, in particular, the differences between distributed representations in connectionist models and symbolic representations of the classical models has been a heated one. It is succinctly summarized in the introductory chapters, and developed further in Chapter 5, “Problems with representation.” He states:

In a representational system, the representation and that which is being represented operate at different logical levels; they belong to different categories. This is not the case with the neural network. There is no difference in kind between the sensory traces entering the network and the traces that interact inside the network. In a certain sense we have the outside repeated, or reiterated, on the inside, thereby deconstructing the distinction between the outside and the inside. The gap between the two has collapsed. (p 83)

For those of us brought up in the formalist tradition, this is not necessarily an unquestionable advantage. The relationship between language and the world is not the only issue we are concerned with when we deal with the issues of representation. By this I mean that the relationship between the representation and the part of the world it purports to represent is only one part of the story. Representations also involve notions such as purpose, agency, understanding, and ultimately an awareness within complex system both of itself and of its environment. It is in this sense that the definition of distributed representations as presented in connectionist literature and defended in this book is lacking in coherence.

Taking language as an example, Cilliers explains, “even though rules may be useful to describe linguistic phenomena, explicit rules need not be employed when language is acquired or when it is used” (p 32).

He suggests an unusual but appealing link between the Saussurean view of language and the connectionist approaches. However, a key feature of the Saussurean lexicon is that it has boundaries, that it has structure. The changes in the relationship between lexical items when a new item is added are meaningful only in terms of this structure. A lexicon is a system consisting of a finite set of items that stand in relationships to other items. These relationships partly determine their individual meanings. When a new item is added to the set, the relationships change and the meanings of individual items become more general or more specialized.

Moreover, such lexicons “belong” to people, agents who use them to reflect their attitudes and beliefs, and who make deliberate and conscious choices, for example in describing a female as “the big woman” or “a tall lady.” They do so in expectation that others share their awareness of the world and of the complex social system to which they all belong. Are we here talking about several levels of complexity, or complexity within complexity? Are we trying to ignore the differences between a lexicon as a complex system and the complex system of social relationships that may be manifested through the use of such a lexicon coupled with informed human choices?

Connectionist models do not seem to share such firm boundaries, or, at the very least, their boundaries are not as significant as the boundaries of a given lexicon. I feel that these features of open systems such as Saussurean lexicons are equally important for our understanding of complexity. Perhaps it is not the pros and cons of connectionist approaches that we need to concentrate on, but “the spirit in which this book is offered: one of openness, provisionality and adventure” (p 142). And this is why Paul Cilliers’ book is a gift.