(*) The author is especially indebted to Johannes van der Zouwen of the Vrije Universiteit, Amsterdam, who has commented in great detail on a first draft of this paper, and with whom a paper on a related subject was recently completed (see ref. 34). Our intensive collaboration over nearly two decades in organizing sociocybernetics sections at different international cybernetics congresses has resulted in a number of co-edited books, and a certain amount of scientific symbiosis. Responsibility for the contents of the present paper, however, is all mine. Thanks are also due to Kitty Verrips of SISWO, who commented specifically on the potential usefulness of second-order cybernetics for sociological theorizing, and to the members of SISWO's Working Group on Sociocybernetics where this paper was first presented, especially to discussants Cor van Dijkum and Loet Leydesdorff.
First-order cybernetics originated in the 1940's 6, and indeed tried to steer observer-external systems. Although it had an interdisciplinary orientation, it might be called an engineering approach, and focussed on studying feedback loops and control systems, and on constructing intelligent machines7.
A few examples: Norbert Wiener8, often considered the father of cybernetics, was engaged in automating the operation of anti-aircraft batteries, which led to the construction of ILLIAC, the world's first computer. Shannon and Weaver9, working at Bell labs in the late 1940s, were confronted with the problem how to reduce noise in telephone lines, and developed information theory. MIT's Marvin Minsky10 constructed amongst others M. Speculatrix, a small robot that could find its way out of a dark room, dexterously moving around objects towards the light; he initiated the now flourishing field of Artificial Intelligence, or AI as it is known for short.
With its stress on efforts to steer especially technological devices, and developing all kinds of control systems, it is not amazing that first-order cybernetics was especially interested in negative feedback loops, rather than positive ones. When a negative feedback loop either naturally occurs, or is constructed, the performance or output of a system is compared with a preset goal, and corrective action is taken whenever there is a deviation from that goal. The thermostat of a central heating system may serve as an example: there is a feedback loop from the thermostat to the heater, whenever room temperature rises above a certain maximum, or falls below a certain minimum. It is noteworthy that even in this simple example, although it clearly is a control system, there is no specific controlling agent; control is dispersed through the system, and any part of it could be said to control the rest of the system.
As a result of the above, first-order cybernetics - with its engineering approach and the corresponding stress on constructing control systems, and with its predilection for negative rather than positive feedback phenomena - was interested primarily in homeostasis or equilibrium-maintenance, or at least in restoring a system's equilibrium whenever it was disturbed by external influences impinging on that system. As is still the case in much of science, environment mastery was an implicit goal, based on the Newtonian conception of an in principle orderly universe: admittedly complex, but knowable by means of a continuing and cumulative effort to discover its basic laws. Positive feedback loops, which cause morphogenesis rather than homeostasis, and are the motor behind change, were paid much less attention to. A simple example of a positive feedback loop is cumulative interest, or to put it more esoterically: "the devil shits on the big heap", recently formulated in economics11 as the law of increasing returns.
Early efforts to apply this homeostasis-oriented type of cybernetics or systems theory to the field of the social sciences, as for example those of Parsons12, met with the resistance of a social science community which by then had turned overwhelmingly liberal, and considered the systems or cybernetic approach to be not only conservative, but also too simplistic, mechanistic and linear to be really applicable to the world of human interaction.
Leaving in the middle for the moment to what extent this left-wing or liberal criticism is generally correct, one can certainly say that some applications of first-order cybernetics like system dynamics - a simulation procedure originally developed by Forrester and Meadows13 to simulate the behavior of systems with several feedback loops - have made remarkable inroads in the general scientific community. One need only think of the enormous popularity of the Club of Rome Report, even among laymen14, where systems modeling was applied to an extremely complicated problem area, with many interacting variables.
The liberal criticism is understandable - the stress of first-order cybernetics was indeed largely, though not exclusively on constructing mechanical control systems - but not quite correct. In the Club of Rome report, for example, positive feedbacks were certainly important, and the same goes for many other technological systems: one of the world's perhaps most impressive technological achievements, the atom bomb, would be unthinkable without positive feedbacks. As Van der Zouwen15 put it succinctly: without negative feedback loops the organism cannot maintain itself in its environment, and without positive feedback loops it has no chance to survive as a species in view of environmental changes to which it has to adapt by setting new goals.
3. The potential usefulness of the classical (first-order) cybernetic approach for sociology:
While the concepts and procedures of second-order cybernetics, to be discussed later, have some specific advantages for sociology, dealing as they are with the interactions of self-organizing, self-referential systems, and thus with the continuous emergence of new levels of complexity, the principles of first-order cybernetics can certainly also be fruitfully applied to the field of the social sciences.
3.1 System boundaries:
Cybernetics or systems theory - we use these terms interchangeably, as they pertain to virtually the same fields of inquiry - holds that one can mentally and arbitrarily carve out any part of the universe16 and call it a system. However, once one has thus delineated the boundaries of the system, one should keep to them, at least for a while. This follows from the so-called black box approach, named after the early metal boxes containing electrical wiring and circuitry, which were invariably painted black. The black box approach presupposes that the external observer can never really observe the system from within, but can only determine what goes in (the input) and what comes out (the output). From the differences between the two, inferences can then be made about the way the system works, depending of course on the mindset of the observer.
The way system boundaries are drawn is obviously observer-dependent, time-dependent and most importantly also problem-dependent. In other words: two observers may be inclined to draw slightly different boundaries when talking about the same problem; and the same observer may draw the boundaries of a system to be studied differently tomorrow than today. And finally, even when the boundaries are not drawn differently as a result of time- or observer-dependence, they may be drawn in a different way because a different problem needs to be studied. It is necessary to be fully aware of this when one has to determine what falls inside and what falls outside one's field of enquiry, and when one has to formulate a research design.
For example, one can define an individual as a system confined by his or her body. But when one is looking at that individual from a medical perspective, it may be relevant to include the input: all the food, drugs and alcohol that have been ingested lately, not to mention the output. When looking at that individual as a scientist, it may be relevant to include at the very least his word processor, if not the entire support structure of his institute or department. And when looking at his emotional problems, one may have to enlarge the system by including his family members, as is done in systems-oriented family therapy17. An extreme example of problem-dependence occurred at a cybernetic conference, where the speaker produced a saw, convinced everyone it was indeed a normal saw, and then started using it quite passably as a fiddle.
Even when a problem has been more or less specified - for example the adaptation of ethnic minorities -one still has to decide how wide the system boundaries should be drawn in order to promise the most interesting research results. One might focus on the interactions between individuals and their environment if one wants to determine what Sennett18 called "the hidden injuries of class", between families and their environment if one wants to focus on defensive or adaptive joint strategies, or between the group as a whole and its environment if one wants to compare the adaptation problems of different ethnic groups.
3.2 Systems, subsystems and suprasystems:
Since one can arbitrarily draw the boundaries of a system when developing a research design, one can not only decide how one wants to define the system under consideration, bu one is also free to decide how one wants to define the subsystems - i.e., component parts that should be especially looked at - and the suprasystem(s) it forms part of. In the above case, for example, and obviously depending on one's research goals, one might select the ethnic family as the system under consideration, the family members as the subsystems, and the ethnic minority group and the nation in which it lives as the suprasystems. One obviously needs to be extremely careful how one defines such a hierarchically nested set of systems, as this will determine the kind of research results one will obtain.
3.3 Circular causality:
Many of us have still been educated to consider circular reasoning as being wrong: a mistake known in logic as the "circulus vitiosus". Something cannot cause itself; but cybernetics says it indeed can. One of the important contributions of first-order cybernetics has been to increase the awareness of ubiquitous circular processes, in technology, in nature, and in society. The circular causal cycle may be short - like A causes B and B causes A - or it may be long and cycle through the entire alfabet or more, in which case it will be harder to discover.
It may be interesting to speculate on what has caused the narrowing of this circular model of causal thinking during the cumulative buildup of the Newtonian-Laplacean world image, with its clockwork model of the universe, its mechanistic rather than organic bias, and its stress on linear causal chains unfolding through time, from past to future19. This resistance against circular causality may itself be part of a circular process where every success of the developing natural sciences since the 17th century strengthened the conviction that hard work could lay bare the implicit rules, the hidden clockwork, of the universe, which in turn attracted new scientists to engage in this prestigious area, whose research results then further strengthen the Newtonian image of the world.
For whatever reason, much of empirical sociology still follows the linear model: admittedly, multivariate analysis may demonstrate that a phenomenon has many different causes, and many different consequences, but this is still a far cry from concentrating on finding circular causal chains, whereby the phenomenon in question helps to create itself. It barely needs comment, of course, that such a concentration on unraveling circular causal chains would enormously complicate empirical research. The idea of circular causality has a certain intuitive attractiveness, and a logical one as well, but to prove its existence empirically is quite another matter.
3.4 Positive and negative feedback loops:
Both positive and negative feedback loops are examples of circular causality. They can either occur spontaneously, in nature as well as in society, or they can be engineered. As was mentioned before, negative feedback loops are of special interest to first-order cybernetics, since its purpose generally was to steer systems by keeping them on a certain course, rather than have them change direction, i.e., to let them fluctuate within a specified margin around an equilibrium. However, positive feedback loops were certainly already recognized in the engineering efforts of first-order cybernetics, but it was the originally biology-based second-order cybernetics that gave them special attention. Logically so, as they cause change rather than stability.
As an aside, it should be noted here that Magoroh Marayama20 spoke already in 1963 of the "second cybernetics" (not: second-order cybernetics), thus designating the cybernetics which concentrates on deviation-amplifying mutual causal systems, where positive rather than negative feedback, and morphogenesis rather than morphostasis are at issue.
Often, negative feedback loops will spontaneously emerge in human interaction when that interaction continues over a certain period of time. A famous example is formed by the well-known prisoner's dilemma which, when played over several cycles, changes from a non-zero-sum game to a zero-sum one: at first, both prisoners tend to betray one another to maximize their own profit. Rapoport21 discovered that both partners start empathizing with the other's position after a while, and then both converge to what he calls a tit-for-tat strategy: an honest move will be rewarded by an honest counter-move, and a dishonest one will be punished by a dishonest counter-move.
While a technique rather than a concept, one can certainly say that simulation, originally a technique of first-order cybernetics, is widely used now also in second-order cybernetics to study phenomena of emergence, and has become a much-used tool in the social sciences as well as in most other disciplines. With the increasing mass scale availability of high-speed computing equipment, even on PC's, it becomes possible to realistically simulate ever more complex problems, with the possibility to incorporate an increasing number of interacting variables in one's models. The obvious advantage of such simulations is that one can investigate the effects of changing some of the variables without actually changing them in reality, i.e., without engaging in policy action. Also, simulations with complex models allow one to discover latent consequences of certain intended actions, and to forecast the emergence and the effects of counter-intuitive behavior.
4. Modern (second-order) cybernetics:
Second-order cybernetics originated some thirty years later than first-order cybernetics, in the early 1970s. The term was coined by Heinz von Foerster22 in a paper for the 1970 meeting of the American Society for Cybernetics, entitled "Cybernetics of cybernetics". He defined first-order cybernetics as the cybernetics of observed systems, and second-order cybernetics as the cybernetics of observing systems.
Indeed, one of the main differences with first-order cybernetics is that second-order cybernetics explicitly includes the observer(s) in the systems to be studied. Moreover, it generally deals with living systems, and not with developing control systems for inanimate technological devices. These living systems range from simple cells all the way up the evolutionary scale to human beings; while the observers themselves are obviously also human beings. Thus, in contrast to the engineering approach of first-order cybernetics, most of second-order cybernetics could be said to have a mainly biological approach, or at the very least a biological basis. As Umpleby23 states, this difference has important consequences:
1) Living systems, no matter how primitive, have a "will" of their own. They exhibit what Maturana and Varela have termed autopoiesis or self-production: they not only reproduce, but also produce their own "spare parts" whenever necessary, generally utilizing elements from their environment. Living systems thus are organizationally closed, but informationally open.
2) As a result, living systems are inherently more difficult to steer; their interactions with their environment are more difficult if not impossible to forecast more than a few moves ahead. Thus, second-order cybernetics has become realistic about the possibilities for steering, and has concentrated more on understanding the evolution of biological and social complexity than on controlling it.
3) Given this, it is understandable that second-order cybernetics is more interested in morphogenesis and positive feedback loops, than in homeostasis and negative feedback loops.
4) Although first-order cybernetics certainly included important biologists - like Von Bertalanffy24, one of the founders of General System Theory - the impetus for second-order cybernetics came largely from biology and neurophysiology, and in a later stage also from epistemology. This is not to say that biology does not profitably use first-order cybernetic concepts: homeostasis, for example, remains an important concept in biology to explain different processes like hormonal balance, maintenance of temperature, etc. However, many biological phenomena that have to do with growth, change and emergence demand an explanation in terms of second-order cybernetics. Maturana, for example, considers knowledge to be a biological phenomenon.25 Attention was thus focused on the observer, and the biological basis of perception and knowledge acquisition processes. In epistemology, second-order cyberneticians became interested in the nature of knowledge, language, cognition and communication.
5) It was thus logical that the concept of self-reference was developed and stressed, especially biological and linguistic self-reference, A satisfying theory of biology should account for the existence of theories of biology; likewise, an adequate theory of cognition should give an understanding of understanding. The view of language changed from language as a string of symbols representing external "reality" to language as actions for coordinating actions. Umpleby16) gives the example of "performative utterances" like "I now pronounce you husband and wife", where the social status of two people is tranformed, while this transformation is described at the same time.
6) Summing up: in second-order cybernetics, the system - whether an individual or a group - is defined as having the ability to reflect on its own operations on the environment, and even on itself. These operations generate variety in the environment, or in itself, which can reflexively be recognized as being due to systemic variation, which makes them recursive: observations can be observed, communications can be communicated, etc.
Apart from von Foerster, several other authors haven given concise definitions of the differences between first-order and second-order cybernetics. These differences refer to, respectively, the purpose of a model vs. the purpose of the modeler (Pask), controlled systems vs. autonomous systems (Varela), the interaction among variables in a system vs. interaction between the observer and the system observed (Umpleby), and theories of social systems vs. theories of the interaction between ideas and society (Umpleby). The latter difference seems to reflect the respective approaches of Parsons as a first-order systems theorist interested in system stability and system maintenance, and Luhmann as a second-order cybernetician interested more in change and morphogenesis.
4.1 Second-order cybernetics and the philosophy of science:
While second-order cybernetics agrees with many of the tenets of the mainstream Newtonian philosophy of science - like the need to distinguish science fom non-science, e.g., by means of Popper's falsifiability criterion, the principle of verification through experimentation, the procedure of refuting conjectures by trial and error, etc. - it goes against it in some important respects:26
1) It disagrees with the idea that observations are independent of the characteristics of the observer. Von Glasersfeld27 developed the philosophy of constructivism as an alternative to the realism of mainstream philosophy of science, i.e., the idea that every individual constructs his or her own reality to fit personal experience28. The knowledge built up this way fits, but does not match the world of experience. It is considered an advantage of this approach that it supposedly increases tolerance, by leading to what De Bono 29 prefers to call "proto-truth", with all the relativistic implications this entails, rather than truth in any absolute sense. Perhaps the best illustration of such an inbuilt sense of relativity is presented in the Mel Brooks movie "History of the World", where Moses comes down Mt. Sinai carrying three topheavy tablets with the fifteen Commandments, stumbles and drops the last tablet to smithereens, then relaxes, shrugs, and says: "Well, ten left".
2) In classical philosophy of science, theories do not affect the phenomena they describe; it would be preposterous to assume that the Second Law of Thermodynamics would speed up the decay of the universe. But in second-order cybernetics, interaction between social theories and social systems is taken for granted30. Economic theories do change economic systems, and often entire societies, as any East-European in the audience can testify. And they are often developed precisely because the theorists do want to change social systems.
3) A core point of disagreement, however, at least with some extreme proponents of second-order cybernetics, is that they reject the necessity, claimed amongst others by Popper, of the unity of method. The methods used for the physical sciences cannot be used for the biological and social sciences, if only because they are self-organizing, self-referential and autopoietic. However, mainstream second-order cyberneticians have a more moderate viewpoint, i.e., that there is at least some unity of method across the sciences.
Clearly, second-order cybernetics strongly disagrees with the still neo-positivist, Newtonian mainstream philosophy of science in the abovementioned respects31, although this disagreement is certainly not the prerogative of second-order cybernetics: Norbert Wiener, for example, devoted a fascinating chapter32 to the difference between Newtonian and Bergsonian time. The changes suggested by second-order cybernetics amount to a scientific revolution in Kuhnian terms; but, as Umpleby33 suggests, the time has come perhaps to return to a period of normal science. The way to do this is by stressing the correspondence principle: i.e. the new theory, second-order cybernetics, should reduce to the old theory, the mainstream philosophy of science, to which it corresponds for those cases in which the old theory is known to hold. In other words, a new and previously neglected dimension is added.
5. The usefulness of second-order cybernetic concepts for sociology
Before dealing with some of the main concepts of second-order cybernetics - self-organization, self-reference, self-steering, autocatalysis, and autopoiesis - it is interesting to note that Norbert Wiener, the father of cybernetics, was quite ambivalent himself on the applicability of cybernetics to the social sciences and to society.34 On the one hand, Wiener was thoroughly convinced that the behavior of humans, animals and machines could be explained by making use of the same cybernetic principles: communication, control of entropy through learning by means of feedback, etc. This is evident already from the titles of his two best-known books: "The Human Use of Human Beings" 35 and "Cybernetics: or Control and Communication in the Animal and the Machine"36. On the other hand, Wiener was quite pessimistic about the applicability of cybernetics to social systems, and this for at least two reasons.
1) Social science data usually exemplify short statistical time series, affected by varying environmental conditions, while one would ideally need long runs under invariant conditions.
2) Wiener moreover considered the social sciences as the discipline where the coupling between observer and observed is hardest to minimize in both directions: apart from the obvious disadvantages of observer dependence, the researcher also inevitably influences the subjects of his research, and can sometimes even act as a catalyst in processes of sudden change: how many strikes have not broken out just after the researchers measuring job satisfaction left the premises?
In a sense, one might say these two objections are interrelated: to the extent that the observer tends to influence his subjects more than in the natural sciences, he thereby contributes to a disruption of the constancy of the conditions needed for longer statistical time series.37 This is of course not to deny that many examples can be found of the inverse: content analysis or non-participant observation does not influence the subjects of the research, while bombarding protons or vivisection clearly does.
The difference between first- and second-order cybernetics was described before, but it should be clear already from the main concepts: while the concepts of first-order cybernetics do not point specifically to either the system or its environment, the important concepts of second-order cybernetics all start with "self", if not in English, then in Greek ("autopoiesis"). While it may be interesting to speculate about the extent to which this "selfishness" is related to the increasingly rapid social change taking place since the late 1960s, this would fall outside the bounds of this paper.
Due to the increasing complexity of today's world, and the seeming intractability of many of its problems, contemporary scientific interest in many disciplines has centered on the emergent, self-organizing properties of certain complex aggregates. Especially developments in biology in this respect have stimulated second-order cybernetics, although first-order cybernetics was certainly also aware of self-organizing systems, but did not pursue this line of inquiry. Norbert Wiener, for example, mentions the synchronization of the behavior of fireflies38, while Ashby also stressed self-organization in his "Design for a Brain" 39.
Recent developments in cognitive science demonstrate the emergence of self-organization (itself a matter of emergence) as a core concept. Cognitive science can be viewed as the result of an interdisciplinary effort which includes neuroscience, AI, linguistics, philosophy, and cognitive psychology. Though it is barely three decades old, one can already distinguish two schools: cognitivism and connectionism.
Cognitivism and cognitive technology gave the impetus for the development of artificial intelligence (AI). Cognitivism conceived mind as a manipulation according to the rules of deductive logic, as localized and serial information processing of symbols which were supposed to represent an external reality.40 In this respect, it seems to be close to first-order cybernetics, although during the Macy conferences of the late fourties41 it was already hypothesized that brains have no central logical processor, and no firm rules or specific spots to store information, but rather operate on the principle of distributed intelligence.
In spite of this, cognitivism was the mainstream paradigm in cognitive science until well into the 1970s, when its drawbacks became more apparent:42
- At some stage, symbolic information processing encounters the so-called "Von Neumann bottleneck": it is based on rules that are sequentially applied, which obviously gives problems with large numbers of sequential operations, as in weather forecasting or image analysis.
- Moreover, symbolic information processing is localized, rather than distributed, and local malfunctioning of some of the symbols or rules therefore tends to result in overall malfunctioning, without the resilience towards disturbances which distributed processing offers.
Connectionism, in clear contradistinction to cognitivism, incorporates many of the views of second-order cybernetics. It is explicitly bottom-up rather than top-down: it eschews abstract symbolic descriptions, but assumes simple, "stupid", neuron-like components that develop cognitive capacities when appropriately connected. In other words: the intelligence is in the structure, in the connections made - hence the name connectionism - and not in the components. Thus, it stresses emergence and self-organization.
Hebb's rule, formulated in 1949, occupies a central place in connectionism. It states that learning is based on postulated changes in the brain as a result of correlated activity between neurons. If two neurons tend to be active together, their connection is strengthened; if not, then it is weakened. In this way history, which is always a history of transformations, is incorporated. Simulations of a simple network of neurons according to Hebb's rule demonstrate that pattern recognition is possible, after a learning phase in which some of the connections are strengthened or weakened, when the number of patterns presented is no more than about 15% of the participating neurons.43
The components of neural networks do not need an external processing unit which guides their operations; they work locally according to local rules, but since they are part of a network, global cooperation emerges when the states of all participating neurons reach a mutually satisfactory state. This passage from local rules to global coherence is the essence of self-organization - otherwise also denoted as emergent or global properties, network dynamics, nonlinear networks, complex systems, etc. Once one cares to look for it, emergence abounds. In different domains, like vortices and lasers, chemical oscillations, genetic networks, population genetics, immune networks ecology and geophysics, self-organizing networks give rise to new properties.
Varela describes simulation experiments with extremely simple cellular automata, whose components are arranged in a circle, can only have only two states ( 0 or 1), and change their states according to simple rules (i.e. the states of their two neighbouring components). These experiments have demonstrated that they can be divided into four classes or attractors:
- in simple attractors all components end up either being all active (1) or being all inactive (0);
- in somewhat more complex cyclical attractors spatial periodicities emerge: some components remain active, while others do not;
- a third type of attractor, also cyclical, runs through at least two cycles before returning to the same state;
- finally, for a few rules the resulting dynamics give rise to the chaotic attractors, studied especially in chaos theory, where one cannot detect any regularities either in space or time, although such systems may unexpectedly return to perfect order, as has been Prigogine's point of departure in this theory of complexity.
When these simple cellular automata are structurally coupled to an external world, for example by dropping them in a simulated "primal soup" of 0s and 1s, with the rule that each component takes over the state of the environmental element it encounters, then nothing happens with the first and fourth class of attractors: they go back to their homogeneous state, respectively remain in their chaotic state. The cyclical attractors, however, demonstrate an admittedly primitive kind of intelligent behavior, for which they were definitely not programmed: for example, some types react ony to double perturbations from the environment; others, the so-called odd-sequence recognizers - only to an uneven number of perturbations.
These experiments show that a simple system with a form of closure (the network's internal dynamical emergences), which is structurally coupled to an environment (replacement of each component by the state of the element it encounters) selects (or enacts) a domain of distinctions from a world of randomness which is relevant to its structure. On the basis of its autonomy the system selects or enacts a domain of significance, which involves some minimal interpretation. While admittedly a far cry from
describing anything like human intelligence, these experiments present a minimal example of how autonomous systems can draw significance from a random background, which may not be totally unlike the way humans draw significance from an after all neutral universe: i.e., by being autonomous, or having operational closure, and by being structurally coupled.
The implications of these developments in cognitive science for sociology are potentially interesting because of the possible analogies that may exist with the many cases of self-organization in human societies:
1) autonomous systems, though in this case simulated on a computer whose coupling with the environment is specified by input-output relations, and thus by an outside source, give meaning to their interactions on the basis of their own history, rather than on the basis of the intentions of the programmer - or should one say a manipulating environment?
2) neural networks, and possibly other networks like human networks as well, produce emergent phenomena as a result of both simultaneous processes (the emergent pattern itself arises as a whole) and sequential ones (participating components have to engage in back-and-forth activity to produce the pattern.44
The phenomenon of self-reference is assumed to be typical of human beings, both on the individual and the group level, although recent work with apes seems to open up the possibility that they too may have some degree of self-reference. Nevertheless, self-reference - at least in the sense used here - is not a concept in first-order cybernetics, which - as Norbert Wiener so explicitly stressed - concerns itself with the commonalities between man, animals and machines, rather than with the differences between them.
Three meanings of self-reference may be distinguished in this respect:
1) the "neutral" meaning, which is used also and especially in first-order cybernetics, and is also applicable to non-biological systems, where "self-referencing control" indicates that any changes in the state of a system are dependent upon the state of that system at a previous moment, like birth rate being dependent upon population size;
2) the "biological" meaning, where senses and a memory are the minimum requirements, and where a self-referential system can be defined as a system that contains information and knowledge about itself, that is, its own state, structure, and processes; like, e.g., human beings45;
3) the "stronger" second-order cybernetics meaning used here, where the system - whether an individual or a social system - collects information about its own functioning, which in turn can influence that functioning; minimal requirements in this case are self-observation, self-reflection and some degree of freedom of action.
One of the main characteristics of social systems, distinguishing it from many other systems, is indeed their potential for self-referentiality in the latter sense. This means not only that the knowledge accumulated by the system about itself in turn affects both the structure and the operation of that system, but it also implies that in self-referential systems like social systems, feedback loops exist between parts of external reality on the one hand, and models and theories about these parts of reality on the other hand.
Concretely, whenever social scientists systematically accumulate new knowledge about the structure and functions of their society, or about subgroups within that society, and when they subsequently make that knowledge known, through their publications or sometimes even through the mass media - in principle also to those to whom that knowledge pertains - the consequence often is that such knowledge will be invalidated, because the research subjects may react to this knowledge in such a way that the analyses or forecasts made by the social scientists are falsified. In this respect, social systems are different from many other systems, including (most?) biological ones. There is a clearly two-sided relationship between selfknowledge of the system on the one hand, and the behavior and structure of that system on the other hand.
Biological systems, like social systems, admittedly do show goal-oriented behavior of actors, self-organization, self-reproduction, adaptation and learning. But it is only psychological and social systems that arrive systematically, by means of experiment and reflection, at knowledge about their own structure and operating procedures, with the obvious aim to improve these. This holds true on the micro-level of the individual, as in psychoanalysis or other self-referential activities, and on the macro-level of world society, as in planning international trade or optimal distribution of available resources.
For social scientists, the consequences of self-referentiality are interesting not only for gaining an insight in the functioning of social systems, but also for the methodology and epistemology used to study them. There is a paradox here: the accumulation of knowledge often leads to a utilization of that knowledge -both by the social scientists and the objects of their research - which may change the validity of that knowledge.46, 47
The usual examples of self-referential behavior in social science consist of self-fulfilling and self-defeating prophecies. Henshel46), for example, has studied serial self-fulfilling prophecies, where the accuracy of earlier predictions, themselves influenced by the self-fulfilling mechanism, impacts upon the accuracy of the subsequent predictions. In much of empirical social science research, however, self-referential behavior does not loom large - which is rather amazing in view of its supposedly being an essential characteristic of individual human functioning. However, if one does not see UFO's, this neither means they are not there, nor that one is blind. In this case the research methodology used may be an issue: survey research, where people are asked what they think or feel, offers little opportunity to bring out self-referential behavior, while depth interviews, which concentrate more on the "why" than the "what" of people's opinions have a better chance to elicit self-referential remarks in this respect.
The futility of large-scale and detailed planning efforts has led to the increasing realization that both individuals and organizations are to a large extent self-steering. After all, perfect planning would imply perfect knowledge of the future, which in turn would imply a totally deterministic universe in which planning would not make any difference. While recognizing the usefulness of efforts to steer societies, a cost-benefit analysis, especially in the case of intensive steering efforts, will often turn out to be negative: intensive steering implies intensive social change, i.e. a long and uncertainty-increasing time period over which such change takes place, and also an increased chance for changing planning preferences and for conflicts between different emerging planning paradigms during such a period. Nevertheless, given a few human cognitive predispositions, there unfortunately seems to exist a bias for oversteering rather than understeering.48
A historical overview of planning efforts concludes that, in spite of intensified theorizing and energetic attempts to create a thoroughly planned society during the last two centuries - the different answers given so far regarding the possibility of planning cancel each other out. There is even no consensus about a formal definition, though usually planning is seen as more comprehensive, detailed, direct, imperative or expedient when compared with other steering activities that are not defined as planning. Increased knowledge about human (i.e. self-referential) systems often does not help us to improve our planning of such systems.
Aulin49 tried to answer two basic questions in this respect:
1) Should one opt for the "katascopic" or the "anascopic" view of society; in other words, should the behavior of individuals and groups be planned from the top down, in order for a society to survive in the long run, or should the insight of actors at every level, including the bottom one, be increased and therewith their competence to handle their environment more effectively and engage more succesfully in goal-seeking behavior?
2) What should be the role of science, especially the social sciences, in view of the above choice: should it try mainly to deliver useful knowledge for an improved steering of the behavior of social systems and individuals, or should it strive to improve the competence of actors at grass roots level, so that these actors can steer themselves and their own environment with better results?
To answer these questions, Aulin followed a cybernetic line of reasoning that argues for non-hierarchical forms of steering. Ashby's Law of Requisite Variety indeed implies a Law of Requisite Hierarchy in the case where only the survival of the system is considered, i.e. if the regulatory ability of the regulators is assumed to remain constant. However, the need for hierarchy decreases if this regulatory ability itself improves - which is indeed the case in advanced industrial societies, with their well-developed productive forces and correspondingly advanced distribution apparatus (the market mechanism). Since human societies are not simply self-regulating systems, but self-steering systems aiming at an enlargement of their domain of self-steering, there is a possibility nowadays, at least in sufficiently advanced industrial societies, for a coexistence of societal governability with ever less control, centralized plannning and concentration of power.
As the recent breakdown of the Soviet Union and its gigantic planning apparatus demonstrates, this is not only a possibility, but even a necessity: when moving from a work-dominated society to an information-dominated one, less centralized planning is a prerequisite for the very simple reason that the intellectual processes dealing with information are self-steering - and not only self-regulating - and consequently cannot be steered from the outside by definition. In other words: there should be no excessive top-down planning, and science should help individuals in their self-steering efforts, and certainly should not get involved in the maintenance of hierarchical power systems.
Of course, this is not to deny that there is a type of system within a society that can indeed be planned, governed and steered, but this is mainly because such systems have been designed to be of this type in the first place, i.e. to exemplify the concept of the control paradigm of first-order cybernetics - although even in first-order cybernetics control does not necessarily imply hierarchy, as even the simple case of the thermostat mentioned before demonstrates. After all, in most developed countries, and even in many underdeveloped ones, the railways run on time, in spite of self-steering employees being involved. Most armies, though also replete with self-steering individuals, are still based on strict hierarchical control and nevertheless function reasonably well, although it has to be admitted that modern, technologically sophisticated and information-driven armies50 seem to thrive more on bottom-up initiative, while armies that explicitly incorporate such self-steering principles and bottom-up initiative in their training - like the Israel Defense Forces - are among the most successful. Thus, while there certainly still is a limited place for organizations that exemplify a more or less hierarchical control paradigm, modern, complex multi-group society in its entirety, conceptualized as a matrix in which such systems grow and thrive, can never be of this type.
If one investigates a certain system with a research methodology based on the control paradigm, the results are necessarily of a conservative nature; changes of the system as such are almost prevented by definition. According to De Zeeuw,51 a different methodological paradigm is needed if one wants to support social change of a fundamental nature and wants to prevent "post-solution" problems; such a paradigm is based on a multiple-actor design, does not strive towards isolation of the phenomena to be studied, and likewise does not demand a separation between a value-dependent and a value-independent part of the research outcomes.
5.4 Autocatalysis and cross-catalysis
Laszlo52 describes two varieties of (chemical) catalytic cycles: autocatalytic cycles, in which the product of a reaction catalyzes its own synthesis, and crosscatalytic cycles, where two different products or groups of products catalyze each other's synthesis. An example of a model of a cross-catalytic cycle, developed by Prigogine and colleagues, is the Brusselator:
(1) A X
(2) B + X Y + D
(3) 2X + Y 3X
(4) X E
With X and Y as intermediate molecules, there is an overall sequence in which A and B become D and E. Step 3 can be seen as autocatalysis, while steps 2 and 3 in combination describe cross-catalysis. Autocatalytic sets (A B C ...... Z A) bootstrap their own evolution, provided the complexity of interactions is rich enough; the system then changes from a subcritical to a supercritical state, and autocatalysis follows. Kaufmann53 even uses the concept to explain the origins of life from a "primal soup" of simple chemical elements as an inevitable production of order, rather than as a unique and extremely unlikely historical accident. Simple chemical laws coupled with the presence of a sufficient number of frequently interacting elements produce ever more complex elements, with new characteristics, that often turn out to be part of new catalytic processes at higher levels of molecular complexity - processes which in turn boost the emergence of still higher levels of complexity. Along similar lines, Swenson54 likewise maintains that - in spite of the Second Law of Thermodynamics - "the world is in the order production business".
The economist Arthur55, collaborating closely with Kaufmann at the Sante Fe Institute, applied the concept of autocatalytic sets to the economy: the economy too bootstraps its own evolution, as it grows more complex over time. Beyond a certain critical threshold, phase transitions occur; stagnant developing countries can enter the take-off stage when their economy has diversified sufficiently. Increased trade between two countries in a subcritical state can similarly produce a more complex and interwoven economy which becomes supercritical and explodes outward. Catalytic effects might also operate in phase transitions that are considered negative, where critical thresholds of violence are passed as in Northern Ireland or Bosnia.
Autopoiesis, or "self-production", is a concept introduced in the 1970s by the biologists Maturana and Varela,56 with the aim to differentiate the living from the non-living. An autopoietic system was defined as a network of interrelated component-producing processes such that the components in interaction generate the same network that produced them.
Although Maturana and Varela considered the concept applicable only in biology, and not in the social sciences, an interesting "theory transfer" was made by Luhmann.57 He defended the quite novel thesis here that, while social systems are self-organizing and self-reproducing systems, they do not consist of individuals or roles or even acts, as commonly conceptualized, but of communications. It should not be forgotten that the concept of autopoiesis was developed while studying living systems. When one tries to generalize the usages of this concept to make it also truly applicable to social systems, the biology-based theory of autopoiesis should therefore be expanded into a more general theory of self-referential autopoietic systems. It should be realized that social and psychic systems are based upon another type of autopoietic organization than living systems: namely on communication and consciousness, respectively, as modes of meaning-based reproduction.
While communications rather than actions are thus viewed as the elementary unit of social systems, the concept of action admittedly remains necessary to ascribe certain communications to certain actors. The chain of communications can thus be viewed as a chain of actions - which enables social systems to communicate about their own communications and to choose their new communications, i.e. to be active in an autopoietic way. Such a general theory of autopoiesis has important consequences for the epistemology of the social sciences: it draws a clear distinction between autopoiesis and observation, but also acknowledges that observing systems are themselves autopoietic systems, subject to the same conditions of autopoietic self-reproduction as the systems they are studying.
The theory of autopoiesis thus belongs to the class of global theories, i.e. theories that point to a collection of objects to which they themselves belong. Classical logic cannot really deal with this problem, and it will therefore be the task of a new systems-oriented epistemology to develop and combine two fundamental distinctions: between autopoiesis and observation, and between external and internal (self-)observation. Classical epistemology searches for the conditions under which external observers arrive at the same results, and does not deal with self-observation. Consequently, societies cannot be viewed, in this perspective, as either observing or observable. Within a society, all observations are by definition self-observations.
One of the first efforts to apply the concepts of both autopoiesis and autocatalysis to the social sciences was made by Gierer.58 He demonstrated "empirically" - by computer simulation - that inequality can be explained as resulting from the cumulative interaction over time of the auto-catalytic, self-enhancing effects of certain initial advantages (e.g. generalized wealth, including education) with depletion of scarce resources. It then turns out that striking inequalities can be generated from nearly equal initial distributions, where slight initial advantages tend to be self-perpetuating within the boundary conditions of depleting resources.
6. The science of complexity: a convergence of paradigms
Since complex modern societies - as compared to simpler ones - are highly dynamic and interactive, and thus change at accelerated rates, they are generally in a far-from-equilibrium situation. According to Prigogine and Stengers59 - who distinguish systems in equilibrium, systems fluctuating near equilibrium through feedback, and systems far from equilibrium - non-linear relationships obtain in systems that are far from equilibrium, where relatively small inputs can trigger massive consequences. At such "revolutionary moments" or bifurcation points, chance influences, but does not take over from determinism and the direction of change is inherently impossible to predict: a desintegration into chaos, or a "spontaneous" leap to a higher level of order or organization - a so-called "dissipative structure", because it requires more energy to sustain it, compared with the simpler structure it replaces.
In stressing this possibility for self-organization, for "order out of chaos", Prigogine comes close to the concept of autopoiesis. In modern societies, the mechanistic and deterministic Newtonian world view - emphasizing stability, order, uniformity, equilibrium, and linear relationships between or within closed systems - is being replaced by a new paradigm. This new paradigm is more in line with today's accelerated social change, and stresses disorder, instability, diversity, disequilibrium, non-linear relationships between open systems, morphogenesis and temporality. Prigogine indeed calls it the science of complexity.60 It is exemplified amongst others by Prigogine himself, Maturana and Varela,47 Laszlo,43 and "second-order cybernetics" in general: i.e. the (non-mechanistic) study of open systems in interaction with their observers.
Social scientists, often still thinking in terms of linear causality, would be well-advised to really study Prigogine's theoretical approach and try out the explanatory powers of his conceptual vocabulary -fluctuations, feedback amplification, disssipative structures, bifurcations, (ir)reversibility, auto-and cross-catalysis, self-organization, etc. - on the phenomena they study. This holds true as well for the concepts and methods of second-order cybernetics in general, as discussed in the foregoing. However, it is already quite difficult to apply first-order cybernetics - which also fully recognizes non-linearities - to social science data sets, and it may seem virtually impossible to do the same with second-order cybernetics; we will come back later on the reasons why this is the case. But indeed, second-order cybernetics is a paradigm that does more justice to the constantly emerging novel complexities of ongoing human interaction, and does not postulate simplistic assumptions about the constancy of human behavior.
What name one gives to this paradigm, or rather this convergence of paradigms over the last two decades, is a matter of secondary importance. What is reassuring in this novel and therefore risky field of research is that there seems to be indeed a convergence of paradigms: all the blind men seem to have their hands on the same elephant. We have generally called this field here second-order cybernetics; but it might also be designated by other names like cognitive science, general systems theory, artificial intelligence, artificial life, or perhaps indeed most aptly the science of complexity. Its main points should be clear by now:
1) Complexity is in the software, not in the hardware; it is in the structure rather than in the elements making up the structure, in the way simple building blocks are organized as a result of simple laws, and not in the building blocks themselves.
2) The emergence of complexity is a bottom-up process, without any central controller leading it, rather than a top-down one; it is a matter of local units, acting according to local laws, producing new levels of complexity by interacting.
The interesting new field of AI (Artifical Life)61 demonstrates these points by means of computer simulation. The flocking behavior of birds, for example, has been simulated with amazing accuracy by computer "boids" following three simple rules: 1) maintain minimum distance from other "boids", 2) match velocities with other boids, 3) move towards the center of the mass of birds.
Artificial life is the opposite of conventional biology: it tries to understand life by means of synthesis rather than analysis. It assumes, as stated above, that life is not a property of matter, but of organization of matter. Living systems are viewed as machines, with one difference from other machines: that they are constructed from the bottom up. Complex behavior does not need to have complex roots. On the contrary, top-down systems are forever running into combinations of events they do not know how to handle. Lindenmayer62 simulated leafs of totally different plants by only slightly changing the bottom-up rules for their construction. There is no ghost in the machine: a population of simple elements following equally simple rules of interaction can behave in always surprising ways. The AL people are convinced that life is not like a computation, but that it is a computation.
3) What can easily be demonstrated in computer simulations of neural networks goes for human networks as well: the more densely they are interconnected, the less likely they are to cycle through a limited number of states, or to ever repeat the same state. The more interdependence grows, the less likely it becomes that history will ever repeat itself, and can therefore be more or less forecasted on the basis of previous experience.
Growing interdependence implies increasing communication. As Leydesdorff63 has stated, "Communication systems change by communicating information to related communication systems; co-variation among systems, if repeated over time, can lead to co-evolution (rather than evolution per se). Conditions for stabilization of higher-order systems are specifiable: segmentation, stratification, differentiation, reflection, and self-organization can be distinguished in terms of developmental stages of increasingly complex networks."
Departing from Luhmann's conception of society as consisting of communications, rather than actions of participating actors, and commenting on Giddens's structuration theory, Leydesdorff64 cogently argues that the mutually conditioning relationship between structure and action can best be empirically studied by using the model of parallel distributed processing as employed in Artificial Intelligence. "The network networks, and the actor acts", i.e., the network performs its own self-referential loops, independent of the specific actors involved.
4) It should be clear by now that we have not been talking merely about complex systems in isolation -which probably do not even exist - but about complex adaptive systems, interacting with an environment. They are everywhere one cares to look: brains, immune systems, ecologies, cells, developing embryos, but also sociocultural systems like political parties, economic systems, and even scientific communities. Holland,65 who was one of the first to simulate neuronal networks in 1951, mentions the following characteristics:66
1) They have many agents acting in parallel, and their control is highly dispersed, with any coherent behavior resulting from competition and cooperation amoong the agents themselves.
2) They have many levels of organization, with agents at one level serving as building blocks for the agents at the next higher level.
3) These building blocks are constantly rearranged as a result of what one might call either learning, experience, evolution, or adaptation.
4) They all anticipate the future to some degree, making "predictions" on the basis of "mental" models of their environment that act like computer subroutines under certain triggering conditions and then execute certain behaviors - no matter how simple, as in the case of bacteria.
5) They all have many niches they can exploit, whereby filling one niche often opens up new ones that can be filled; complex adaptive systems, in other words, always create new opportunities.
6) As a consequence, it is meaningless to talk about complex adaptive systems being in equilibrium: they can never get there, but are always in transition. If they would get there, they would be dead.
7) Likewise, the agents in complex adaptive systems cannot optimize their fitness, utility, etc.: the space of possibilities is simply too vast in an environment which is also complex and rapidly changing; they can at best improve on some dimensions, but never optimize.
7. Second-order cybernetics: a bridge too far?
If one accepts these criteria as being valid for complex adaptive systems, and realizes that the social sciences indeed mainly study those systems - self-organizing, self-referential, autopoietic, and thus with their own strategies and expectations, with intertwining processes of emergence and adaptation - then one is confronted with one of the core problems of sociology, economics, and other social sciences: how to make a science out of studying a bunch of imperfectly smart agents exploring their way into an essentially infinite space of possibilities which they - let alone the social scientists researching them -are not even fully aware of.
There is indeed quite a methodological problem here. It is already very difficult to apply the principles and methods (e.g., feedbacks and non-linearities) of first-order cybernetics to empirical social research, much more so than to sociological theory, and nearly impossible to incorporate a second-order cybernetics approach in one's research design. Indeed, as far as empirical research is concerned, second-order cybernetics may be a bridge too far, given the research methodology and the mathematics presently available.
Applying the principles of first-order cybernetics in empirical research already poses heavy demands on the data sets and the methods of analysis: every feedback (Xt Y Xt+1), every interaction between variables [Z (X Y)], and every non-linear equation (Y = cX2 + bX + a), let alone non-linear differential equation (Y' = cY2 + bY + a), demands extra parameters to be estimated, and quickly exhausts the information embedded in the data set. Admitting on top of that the second-order notions that the research subjects can change by investigating them, let alone being aware of the fact that these subjects may reorganize themselves on the basis of knowledge acquired by them during the research, exceeds the powers of analysis and imagination of even the most sophisticated methodologists: it equals the effort to solve an equation with at least three unknowns.
In the case of second-order cybernetics these problems indeed multiply: how does one obtain reliable data within such a framework, where nothing is constant and everything is on the move, let alone base policy-relevant decisions on such data? How can one still forecast developments when at best retrospective analysis of how a new level of complexity has emerged seems possible? Certainly, these are problems that are far from solved, and a lot of work lies ahead before hypotheses derivable from second-order cybernetics will be fully testable. Nevertheless, the opportunities offered by this paradigm to present a truly realistic analysis of the complex adaptive behavior of interacting groups of agents seems to good to pass up.
But the inherent problem remains: the more realistic - and therefore less parsimonious - a theory, the more complex it becomes, and the more difficult to test the hypotheses and subhypotheses derived from it which are used in collecting and interpreting the data. If one accepts that social systems have a high degree of complexity, cybernetic theories become more relevant and fitting, but less testable as they grow more complex themselves, as is the case with second-order cybernetics as compared to first-order cybernetics. There is certainly a challenge here, for theorists and methodologists alike.
For the time being, sociology should perhaps model itself more on meteorology than on the natural sciences, and force itself to give up the ambition to make accurate medium- and long-term predictions, except in delimited areas of research where complexity is still manageable or can be more or less contained. Ex post facto explanation of how things have come to be as they are is already difficult enough for social scientists nowadays. The best they may do at the turn of the millennium is to get a grip on the underlying laws of change, perhaps by a theory transfer from those subfields within biology where second-order cybernetics was developed, and consequently to further develop the theories, the non-linear mathematics and the simulation techniques required to investigate the growth of complexity of human society.
This might ultimately result in adequate and empirically falsifiable models of self-referential, self-steering and self-organizing actors on individual and supra-individual levels, interacting with each other in ever more intricate networks to develop new and unforeseen higher levels of complexity, with new actors engaging in new activities, speeding up the growth of complexity even more. The best one can do as sociologists under these circumstances seems to accept that there is not any one desirable and sustainable state for society - only near-continuous transition, often coupled with the impossibility to forecast even the near future - and that consequently one can engage at best in some degree of damage control, by pointing out the probability of future catastrophes to those who might be able to help averting them.
Unfortunately, interdisciplinary and international-comparative research centers to study complexity-related problems, though sorely needed, barely exist as yet, although the systemic rather than piecemeal approach they could provide is required by the sheer complexity and interdependence of present-day societal problems. This is partly due to a lack of political commitment to finance them on the part of politicians who have to be re-elected every four years, but partly it is also caused by the fact that most researchers are not yet educated to work in truly interdisciplinary teams that presuppose an open mind, and at least a reasonable knowledge of the other disciplines involved.
Ashby's "Law of Requisite Variety" states that only variety within a system can force down the variety due to the system's environment - at least if the system is to make fully sense of the latter and to be able to steer it. Both the international political system and the international scientific system still have a long way to go in this respect, and are barely organized yet as reasonably integrated systems that are aware of their policy options. Grass roots pressure from below, such as is now visible regarding environmental pollution problems in most of the Western "television-democracies", is probably required to force politicians and scientists alike to finally get their act together and to start tackling other complexity-related problems as well.
NOTES AND REFERENCES
Engineering, Materials Science and Technology Division
MCB University Press
60-62 Toller Lane, Bradford, West Yorkshire BD8 9BY, UK
Tel: 44 (0) 1274 777700
Fax: 44 (0) 1274 785200