You are on page 1of 7

This article was downloaded by: [University of Sydney]

On: 04 September 2014, At: 09:09


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

International Journal of General Systems


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/ggen20

COMPLEXITY AS A SYSTEM PROPERTY†


a
ROBERT ROSEN
a
Department of Physiology and Biophysics , Dalhousie University , Halifax, Nova Scotia,
Canada
Published online: 06 Apr 2007.

To cite this article: ROBERT ROSEN (1977) COMPLEXITY AS A SYSTEM PROPERTY†, International Journal of General Systems,
3:4, 227-232

To link to this article: http://dx.doi.org/10.1080/03081077708934768

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the
Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Int. J. General Systems © Gordon and Breach Science Publishers Ltd.
1977, Vol. 3, pp. 227-232 Printed in Great Britain

COMPLEXITY AS A SYSTEM PROPERTYt


ROBERT ROSEN
Department afPhysiology and Biophysics, Dalhousie University, Halifax, Nova Scalia, Canada.
(Received May 15, 1975; illf1lialform July 20, 1975)

Complexity is generally viewed as an intrinsic property of certain kinds of systems, or at least, as a property of a specific
description of such systems. The view towards complexity taken in the present note is different; namely, that complexity
reflects the necessity for many distinct modes of description of a system. This in turn depends upon the number of ways we
can effectively interact with a system, and ultimately on the number of distinct subsystems which available observational
techniques make accessible to us. A number of important implications of this point of view, bearing particularly on system
analysis in biology, and on reductionism as a general analytic strategy, are pointed out. The relation between the concepts of
complexity, error and emergence is briefly explored.

INDEX TERMS Systems, complexity.


Downloaded by [University of Sydney] at 09:09 04 September 2014

INTRODUCTION Our views regarding the concept of "complexity"


have tended to be as richly varied as complexity
The early successes of science, particularly physics, itself. There have always been scientists uneasy with
arose from its ability to take apparently incom- the view that all complexities could be reduced or
prehensible situations, such as the motions of explained by simple laws, and not all of them can be
celestial bodies, and explain them on the basis of dismissed as "vitalists" or "holists." Bertrand Russ-
simple and immutable dynamical laws. It is im- ell remarks somewhere that if asteroids were much
possible to overestimate the impact which New- more common in the solar system than they in fact
Ionian mechanics had upon Western thought; if the are, the effects of their gravitational perturbations
whimsy of supernatural agents could be eliminated upon the planets would have made the general-
from the heavens, then surely any phenomenon, izations of Copernicus and Kepler, and ultimately
however complex it might appear, could likewise be the dynamics of Newton, impossible, and goes on 10
understood on the basis of laws as basically simple state that "thus we see how complexity can defeat
and universal as those of celestial mechanics. This science." Von Neumann t argued that there was a
conviction has informed all of modern science, and kind of "threshold" of complexity, below which the
has molded scientific development in all areas in world behaved with its familiar regularities, but
innumerable ways. above which entirely new modes of behavior
The notion that all complex processes could be appeared (such as self-reproduction, evolution and
understood in terms of underlying simple universal free will) which were sui generis, with no counter-
laws is indeed an article of faith, but one which has parts in systems of lesser complexity. We have all
been put increasingly to the test during the past heard it said that complex systems are coulltel'-
twenty-five years. Previously we have tended to intuitive, meaning that their behavior departs from
sidestep problems arising from complexity; we what "common-sense" (i.e., simple rules) suggests
could argue that these problems could be dealt with their behavior ought to be. To what extent are such
if necessary, but since there were many simpler and observations consistent with the faith that, because
more tractable problems at hand, we would concern we can explain celestial mechanics in terms of simple
ourselves with those. We are no longer able to do universal laws, we can explain any complexity in
this; social, biological and technological problems similar terms?
of the first magnitude now bombard us from all The present note in intended as a contribution to
sides, demanding imminent resolution, and these this question.
problems force us to confront directly the problems
of complexity.
2 THE MEASUREMENT PROBLEM
t'This paper was prepared with the support of NIH grants
#2ROI HD05136-04 and # I POI HD07328-01, and NASA One view of complexity, implicit in the remark of
Grant # NGR33015002. von Neumann cited above, is that it is a system
227
22H R. ROSEN

property of the same character as any other system such that the values obtained by interacting the
property, and that it can in fact be measured. Thus a system with any measuring instrument are definite
good place to begin in a study of complexity is with a functions of those in the minimal set, then we say
study of general system properties and their that the minimal set defines a set of state variables
measurement. We shall here consider only those for the system. These state variables have the
system properties which can be perceived as property that any information pertaining to the
pertaining to the system at an instant of time, and system in question at an instant of time are in
shall exclude those properties which are manifested principle known when the values of the state
only over time (indeed, it is one of the tenets of variables at that instant are known. In this way, we
dynamical theory that these temporal properties can identify the instantaneous states of our system
can be predicted from a knowledge of system with sets of numerical variates, and the slate space of
properties at a particular instant). the system with a manifold. It should be noted that,
Let us suppose' that wc can represent our as a corollary of this definition, if two instantaneous
interactions with any given system in a particular states of a system are different at all (i.e., give
canonical form: that any such interaction can be different readings on some measuring instrument
replaced by a series of interactions' with special M) then they already must give ditTerent readings on
Downloaded by [University of Sydney] at 09:09 04 September 2014

known systems which we call measuring instru- at least one of the measuring instruments which
mellts. A system interacting with a measuring define the state variables. It should also be noted
instrumcnt will result in a modification of the that there is nothing unique about a set of state
measuring instrument culminating in a numerical variables,
readout. Thus any observable property of the In practice, of course, we do not have at our
system in which wc are interested can be trans- disposal all of the possible measuring instruments,
formed into a set of numerical readings on some but only a certain subset of them. If we treated that
appropriate measuring instruments; and con- subset as if it were the total set, we could proceed to
vcrscly, any set of numerical readings on the define a set of state variables relative to that subset,
measuring instruments refers to an observable and obtain thereby a description of the in-
property (orjeature) ofthc system being observed. stantaneous states of our system. Such a description
Thus, if we allow our system to interact with a would be a relative description, and would in
measuring instrument M 1 at an instant of time, we general be quite different from the "absolute"
obtain a certain amount of information about what description involving all possible observable quan-
our system is like at that instant. This information tities. Moreover, two such relative descriptions,
takcs the form of a number, which we assign as the each defined by a difTerent subset of the set of all
value of the particular observable quantity meas- possible observable quantities, would not only ditTer
ured by M I' If we wish. more information about from the "absolute" description, but would also
what the system is like, the only way to obtain it is to differ from each other. We have previously con-
interact the system with another measuring instru- sidered this problem of the relativity of system
ment M 2 at the same instant, obtaining a second descriptions (Rosen ').
number, t he value obtained by the observable M 2 al To give some idea of the dimensions of the above
that instant. If we wish still more information, we concepts, let us consider the situation in which a
must interact the system with a third measuring system of interest is specified by a finitenumber of
instrument M), obtaining a third number, and so state variables Xl"'" x•. Let us make the simplest
on. possible dynamical assumption about the system's
Complete information regarding the system at an temporal behavior; namely, that the rate at which
instant of time requires the values obtained by all each of the state variables is changing in a state is
possible measuring instruments upon interaction determined only by that state. We then have, locally,
with the system at that instant (we leave aside the familiar dynamical equations
problems of quantum uncertainty; these do not
change the character of the argument). It turns out,
i= 1,.. "n. (1 )
in general, that this information is highly re-
dundant; the values obtained on some of the
measuring instruments are functions of the values governing the change of state with time, and each
obtaincd on others. If we can find a set ofmeasuring initial state of the system generally determines an
instruments which is (a) of minimal size, and (b) entire system trajectory, a curve parameterized by
COMPLEXITY AS A SYSTEM PROPERTY 229

time winding through the Euclidean II-dimcnsional not an intrinsic property of systems, but rather
space which represents the space of states of the arises from the number of ways in which we are able
system. Now if F is any observable quantity, then by (or desire) to interact with a system. Indeed, from
definition F must be a function of the state variables; what we said in the preceding system, allY system
F = F(x 1"'" x n ) inherits a dynamics from the can be made to appear complex in this sense. But a
dynamical relations which govern the Xi' and this moment's reflection will reveal that this is quite in
can be computed if that dynamics is known. But if accord with our ordinary experience. A system like a
we look at F directly with the measuring instrument stone typically is regarded as simple, because we
which measures it, and do not look at the Xi' we will interact with it in only a few ways. For a geologist,
merely see a temporally varying quantity, with no who multiplies the number of ways in which he
underlying substructure. interacts with a stone, such a system can appear
If G is another such observable quantity, it may infinitely complex. Thus complexity is indeed a
turn out that the dynamical relations governing F function of the number of ways available to interact
and G may be put into the form with a system.
Another peculiarity of the definition is the
dF =cI> (F G) following. We said that the appearance of com-
dt v .
Downloaded by [University of Sydney] at 09:09 04 September 2014

i
plexity arises from the use of different subsets of the
(2)
set of all available measuring instruments, and the
corresponding different modes of system de-
scription which arise therefrom. But surely we could
In this situation, it will appear to us that the state put all the available subsets together, and obtain a
space of the system is two-dimensional, quite unified (if still relative) description. It turns out,
different from the "real" system described by (I). however, that the problem of reconciling such
That is, we have found a "subsystem," specified by alternate descriptions is a very deep problem,
the observable quantities F and G, with quite a closely related to ideas of structural stability." In
dilTerent description and set of properties than those general, merely juxtaposing alternate modes of
of the "real" or "total" system from which it was description does not in general lead to a more
obtained. There can be many such subsystems; comprehensive description which embraces all of
indeed, there exist dynamical systems (I) which are the components.
ulliversal in the sense that they contain "sub- Intuitively, how does one approach a complex
systems" which inherit allY prescribed dynamics system scientifically? One attempts to analyze the
whatsoever. 2 These subsystems are all present in the system into a family of simpler subsystems. By
original system, and which one we actually "see" is definition, a simpler subsystem is one which is
specified entirely by how we choose to interact with capable of fewer modes of interaction than the
the system (i.e., which measuring instruments we original system (at least, with respect to the
apply to the system). measuring instruments we are using). By charac-
terizing these simpler subsystems, we may hope to
3 ABSTRACTION, ANALYSIS, obtain an understanding of how these subsystems
REDUCTIONISM can interact with each other, and thereby, an
We are going to define a complex system as one with understanding of the original system from which the
which we can interact effectively in many different subsystems were isolated.
kinds of ways, each requiring a different mode of Insofar as an analytic procedure isolates a
system description. That is, a complex system is one subsystem from a system, it involves a process of
for which we have at our disposal alarge number of abstraction. In general, observable quantities pre-
subsets of measuring instruments, each of which sent in the original system are discarded in the
gives rise to a dilTerent mode of description of the extraction of the subsystem. These observable
system. Another way of saying this is that a complex quantities represent potential interactive capacities,
system is one which allows us to discern many and their removal means precisely that the sub-
subsystems (in the sense we used that term above), system is capable of fewer interactions than was the
depending entirely on how we choose to interact original system. This is indeed the sense in which it is
with the system. true that a subsystem is simpler than the system
The above definition seems peculiar in several from which it was taken, and in fact comprises a
ways. For one thing, it requires that complexity is general definition of subsystem, as was used
F
DO R. ROSEN

implicitly in the preceding section. It is the hope of the precise sense noted above, and these are the
an analytical procedure to specify subsystems which subsystems which are appropriate for analysis. This
each retain some subfraction of the interactive is true even though we cannot measure the U i
capacities of the original system, so that each directly, and are interested in the directly observable
interactive capability is captured in one of the quantities Xi'
subsystems of the analysis, and the total capabilities This example leads to the next important point.
of the original system can be recaptured by putting What is "natural" from the standpoint of easy
the subsystems back together appropriately. measurability, and hence easy isolability of sub-
It is clear that not every mode of system systems, can often be most' unnatural from the
decomposition or analysis into subsystems will have standpoint of effective analysis of complex systems.
such a property. Suppose we consider a simple In the above example, the "natural" subsystems are
linear dynamical system; e.g., a set of n chemicals the Ili> not the Xi' The mere ease of isolation of a
reacting by first-order kinetics. Such a system can be subsystem does not at all imply that the subsystem
described by the dynamical equations will be analytically useful. Thus in biology the
conspicuous properties of cells has Jed to the "cell
(3 ) theory," that every biological property is to be
Downloaded by [University of Sydney] at 09:09 04 September 2014

understood by means of analysis into constituent


cells. But the cell is often not the basic unit of
Wc might try to isolate subsystems consisting of the biological function, and the attempt to preserve the
individual reacting species. These are certainly cell as the analytical unit may actually render
.~impler than the system from which they are analysis impossible. And this even though the cell is
abstracted, but (at least at this kinetic level) they indeed simpler than the entire organism.
allow none of the properties of the original system to This type of "vested interest" in an analytical unit
be inferred from them. Thus, although the Xi pervades much of the science of complex systems.
represent the observable quantities we choose to To insist on the cell as the ultimate unit of biological
measure, and in which we are directly interested, organisms, or on individual organisms as the
they do not provide a useful mode of analysis of the ultimate unit of social structures, is often quite
system. inappropriate for the analysis of many properties of
Actually, of course, when confronted by a system the higher structures. This kind of "vested interest"
like (3), we proceed with a mathematical analysis. is often called reductionism, and is of course most
We introduce nell' state variables conspicuous in molecular biology. Its basic feature
n is not that it advocates analysis of complex systems
lIj=Uj(xJ, ...,x,,)= L fiijx
j= 1
j, into simpler subsystems (this is unavoidable in any
casejbut rather that it advocates, in advance, a set
of abstract subsystems which it posits to be the only
choosing the numbers fJij so that the matrix of rate
admissible units for analysis. But as we have seen,
constants (aij) assumes a particularly simple (dia-
the relativity of system descriptions, together with
gonal) form. If we can do this, then the dynamics of
the multiplicity of descriptions which provide the
the system, in the new state variables, can be written
very definition of complexity, preclude anyone class
dll,. I of subsystems, or any mode of analysis, from being
-=)~·ll· i= , ... ,11. (4 ) universally valid.
dt "
In short, there are two features which must be
These new state variables u, are usually regarded exhibited by a successful analysis of a complex
as convenient mathematical fictions. However, they system into subsystems:
arc by definition observable quantities of the system
1) The subsystems must be simpler than the
(i.e., functions of the state variables) and as such
system from which they were abstracted;
could be measured directly if we could interact with
2) The subsystems must allow the properties of
the system appropriately. The Ui have the property
interest in the original system to be reconstructed
that they decouple the dynamical equations, so that
from the properties of the subsystems.
instead of having a highly interactive (and hence
complex) system of n equations, we have a system of To these features, experimental science also appends
11 dccouplcd simple first-order equations which can a third: that the subsystems so abstracted must be
be directly solved. Each Il, specifies a subsystem, in "natural," in the sense that they are easily com-
COMPLEXITY AS A SYSTEM PROPERTY 23t

patible with contemporary observational tech- When we say, for example, that a mutation
niques. Indeed, it is often argued that simplicity plus represents an "error" in the cellular coding system,
"naturality" in this sense will imply the requisite what we actually mean is something like the
reconstructibility. This, however, is in general false; following: DNA is capable of many interactions
indeed, it is precisely the subsystems which will besides those involved in its coding functions. Some
allow reconstructibility that are the "natural" units of these interactions can affect the coding functions.
of analysis in a given situation, however far removed When such an interaction occurs, there will be a
they may be from presently available observational deviation between what our simple model tells us
techniques. ought to be coded, and what actually is coded. This
deviation we call a mutation, and we say that the
4 COMPLEXITY, ERROR DNA has behaved erroneously.
AND EMERGENCE Likewise, when we say that a neuron, or a
communication channel is "noisy," we mean pre-
A subsystem abstracted from a complex system, in cisely that the real physical system is exhibiting
the sense described above, may be regarded as a interactions not comprehended in a simple model of
model of that system. The model shares certain that system, and that these interactions are affecting
Downloaded by [University of Sydney] at 09:09 04 September 2014

properties with the original system, but other the modelled activities. The typical way of treating
properties have been abstracted away. The philo- this is to superimpose a stochastic variability on the
sophy of analysis of complex systems posits that a model. In formal terms, this stochastic aspect is
sufficiently comprehensive family of such models intended to reintroduce into the model the effect of
will allow us to reconstruct the properties of interest all those degrees of freedom which were abstracted
of the original system (and here is should be stated out in constructing the model.
explicitly that different families of models may be In this view, error is the result of replacing
necessary to understand different sets of properties simplicity by complexity. Its inevitable correlate is
of the original system). loss of individual functions, which presuppose
Any such model, be it a real physical subsystem simple systems capable of only those interactions
(such as a molecular fraction removed from a cell) or which support the functions in question. It is in this
a mathematical abstraction (such as the functions 1/; sense that senescence of organisms, and the failure of
in (4) above) has simpler interactive capacities than mechanical artifacts and social institutions, can be
the real system does, by definition. The model is regarded as the result of "error"; they are uncon-
what the real system would behave like if it had no trollable deviations between the behavior of an
other interactive capabilities besides those retained abstract functional model, and the behavior of a real
in the model. In environments in which no system, with interactive capabilities not present in
interactions are possible other than those retained the model. "Error correction," which involves the
in the model, the real system and the model would intervention in the real system on the basis of yet
behave identically. another simple model, cannot in principle solve this
However, in any real environment, the real problem.
system will in general exhibit interactions besides On the other hand, such "error" is in fact the
those retained in the model. These interactions are driving force for biological evolution, and for the
in fact unpredictable from the model (since the emergence of new properties in interacting systems.
dynamical variables through which they occur have Emergence has always been a puzzle, precisely
been abstracted away to create the model). But since because its roots were sought in simple models,
all the dynamical variables are "linked" with each where the very basis for its occurrence in real
other through the dynamics which govern the real systems was abstracted away. From the viewpoint
system, these unpredictable interactions will sooner expounded in the present note, it is seen as the
or later affect those observable quantities preserved inevitable corollary of the greater richness of
in the model. The result will be a deviation between interactive capabilities in a real system than is
what the real system is actually doing, and what the present in any abstraction from that system.
simple model predicts it will be doing. This The above considerations represent a sketch of a
deviation is the basis for the concept of error in more general development, which will be presented
system theory. elsewhere.
Downloaded by [University of Sydney] at 09:09 04 September 2014

You might also like