The architecture of complexity

Herbert Simon
Carnegie Institute of Technology, USA
(with an introduction by Paul Cilliers, University of Stellenbosch)

Originally published as Simon, H. (1962). “The architecture of complexity,” Proceedings of the American Philosophical Society, ISSN 0003-049X, 106(6): 467-482. Reprinted with the kind permission of the American Philosophic Society. Special thanks goes to Mary McDonald. The original article can be downloaded from here.

What is inside and what is on top? Complex systems and hierarchies

In the days - about a decade ago - when a start was made to apply complexity theory to all sorts of real-world problems like social systems and organizations, the notion of ‘hierarchy’ came under pressure. A number of important insights were responsible for this, including the recognition of the importance of distributed representation, non-local causes, holism and the importance of relationships with two-way communication.

Another more philosophical reason why the notion of hierarchy was resisted had to do with the problem of reductionism. Crude forms of reductionism propose that the world, or systems within the world, are made up of levels arranged in a hierarchical format. Higher level phenomena could then be reduced to physical activity on lower levels. From this perspective the mind, for example, was nothing but the activities of neurons; neurons can be described chemically and chemistry can be reduced to physics. This view is clearly on oversimplification and the resistance to reductionism which followed included a resistance to the notion of hierarchy.

This resistance had specific effects on our thinking about complex systems. They were understood as consisting of components which were all equally important, interacting in a way which undermined the idea of ‘central control’. In this phase of complexity studies the influence of chaos theory was still quite strong, and together this resulted, in organizational theory at least, in the notion of ‘flat systems’. Organizations, for example, should be seen as things where the resources are distributed throughout the system. A hierarchical understanding of complex systems is just too rigid.

This was certainly an important phase in the development of complexity theory, but more recently it has become clear that this view is restricted in its own way. The main problem is that a view which underplays hierarchy also tends to underplay the fact that complex systems have structure. They are not homogenous things. As a matter of fact, it is clear that chaos in itself does not lead to complexity; that structure is an enabling precondition for complexity. The task now is to rethink the notion of structure without simply falling back into a crude form of reductionism.

In facing this task we can return to Herbert Simon’s seminal paper from the early 60s. As one reads it, it becomes clear that we could have saved ourselves a lot of trouble by taking Simon seriously. He argues with exceptional clarity for the unavoidability of hierarchies in complex systems. He shows how, from an evolutionary perspective, it is much more efficient for complex systems to be composed of sub-systems which are hierarchically organized. Hierarchy is not an accidental feature of complex systems, it is an essential one.

Of course, complex systems are not simply hierarchical systems, and Simon knows this. If they were simply hierarchical, they would be fully decomposable, and, as a result, easy to understand and model. Unfortunately they are not neatly nested like Russian dolls, there are cross-cutting connections. Simon holds the hope that those interactions which do not fit into the overall hierarchy are of less importance and that complex systems are what he calls “nearly decomposable.” If he is right, this would mean that our approximations in hierarchical terms would be close enough to the truth to enable a proper understanding.

I think that this assumption is a little optimistic. The cross-cutting connections are nonlinear, and it is therefore difficult to predict their effects in any general way1. Perhaps it is better to think of complex systems not as being nearly-decomposable, but as being decomposable and non-decomposable at the same time. These are issues to be worked out in more detail, but what is clear is that we still have to confront the notion of hierarchy in a serious way. In this confrontation Simon’s work will be indispensable. Even if the problem of hierarchy does not interest everyone, they should read Simon’s paper just for its eloquence and clarity, as well as for the wide range of issues he addresses with insight. I wish more academic papers were written like this!


1 I discuss this problem in a little more detail in Cilliers, P. (2001). “Boundaries, hierarchies and networks in complex systems,” International Journal of Innovation Management, ISSN 1363-9196, 5(2): 135-147.