Robert Rosen on the implications of Anticipatory Systems (excerpt)

To augment the current discussion, I thought it might be most useful to simply include some of my father's words and thoughts from his book, "Anticipatory Systems". To my knowledge, Robert Rosen was the first to suggest that there IS such a thing as an "anticipatory system" in nature. The aspect of them that set off the controversy were the properties of TIME, which are revealed by the behavior of biological systems, properties that contradict the accepted notions of how time works and what it is. Organisms have behaviors that, my father said, are due to the "temporal spanning" (as he put it) that these systems have as a built-in capability, which he further said was a perfectly rigorously scientific concept if you discard the false presumptions and preconceptions that have insinuated themselves into science over the past several hundred years.

This excerpt is entirely prose and has none of the mathematical examples that he often employed while illustrating the concepts he was developing and explaining. Therefore, this is a much easier read! Anyone who wishes to is welcome to contact me for clarification, both on list and off. I hope people find these ideas of value; there are more suggestions of directions for scientific study here than my daughter's university could house...

[On a humorous note: I used to tease my father often about the fact that he wrote his books in a plural form, using the word "we" instead of "I"... He asked, with a twinkle in his eyes, if I had a problem with him using the "Royal WE"? to which I replied, "Oh! Well, the Royal We would be all right. I thought you were using the Schizoid We." (To be fair, he is actually using the Inclusive We.)]

"The point of departure for our entire development was the recognition that most of the behavior we observe in the biological realm, if indeed not all of the behavior which we consider as characteristically biological, is of an anticipatory rather than a reactive character. In fact, if it were necessary to try to characterise in a few words the difference between living organisms and inorganic systems, such a characterization would not involve the presence of DNA, or any other purely structural attributes; but rather that organisms constitute the class of systems which can behave in an antipicipatory fashion. That is to say, organisms comprise those systems which can make predictive models (of themselves, and of their environments) and use these models to direct their present actions.

We saw very early that the behaviors of systems which employ predictive models are vastly different from those which do not. At the most fundamental level, anticipatory systems appear to violate those principles as causality which have dominated science for thousands of years. It is for this reason that the study of anticipatory systems per se has been excluded routinely from science, and that therefore we have had to content ourselves with simulations of their behavior, constructed in purely reactive terms. Restriction to such simulations has limited us to the study of the mechanisms whereby biological actions are effected, thereby distorting our approach to biological processes, and in fact precluding from the outset any basic understanding of how these processes actually work.

Once we recognized that anticipatory behavior is the general rule in biological systems, and that it depends essentially on the presence of predictive models, our attention was inexorably drawn to the nature of the modelling relation itself. What, indeed, is a model? The better part of our development was concerned with attempting to clarify this question. In its most general terms, we found that a modelling relation between systems is established through an encoding of qualities pertaining to one of them into corresponding qualities of the other, in such a way that the linkages between these qualities are preserved. In science, we generally attempt to encode qualities of natural systems into purely formal (ie mathematical) ones, in such a way that the rules of inference of the mathematical system correspond to causal relations, and particularly dynamical relations, in the natural system. We found that the same natural system generally admits many models, depending on which of its qualities are thus encoded; indeed, the complexity of a natural system is perceived through the number of distinct modelling relations into which it can enter. Conversely, we found that many different natural systems can admit the same model; this provided the basis for the fundamental concept of analogy between systems, and the use of analogy as a powerful scientific tool. We sought to illustrate these ideas with many examples, drawn from the widest possible variety of scientific disciplines; not only to show their universality, but also to demonstrate that the concept of a model is not something exotic or unusual, but rather of the broadest currency imaginable.

The raw material for the construction of modelling relations is, and must be, the result of observation. In the broadest sense, observation provides the means by which the qualities of natural systems are defined and represented. As we developed it, observation involves the dynamical interaction of natural systems, and the employment of the change of state induced in one of the interacting systems as a label for the value of some corresponding quality of the other system. Thus, we stressed that the making of observations, which is generally considered the hallmark of empirical or environmental science, already involves the concept of a model in an essential way, and thus theoretical science (ie the study of models) is simply a kind of extension of the process of observation itself.

We also pointed out in this connection that an act of observation is a quintessential act of abstraction; the observation of a single quality of a natural system is indeed the greatest kind of abstraction which can be made of that system. From this point of view, the development of theoretical science is an attempt to combine observations in such a way that our view of systems becomes less abstract than it could be if we were restricted to observation alone. Thus we stressed what should be a commonplace; that there is no antagonism between "theory" and "experiment"; it is unfortunately not a commonplace, because it has been obscured by the antagonism between "theorists" and "experimentalists".

Even though models are, in this sense, less abstract than observations, they are nevertheless abstractions. Thus it becomes of great importance to understand how different models of the same systems are interrelated. The crucial concept here was that of bifurcation. Indeed, a modelling relation can generally be formulated in mathematical terms as a conjugacy; the failure of a modelling relation, which is a logical independence between two modes of description, thus becomes exactly a bifurcation in the mathematical sense. We saw how this concept of bifurcation was related, on the one hand, to phenomena of emergence, and on the other hand, to the concept of error. As we formulated it, error is not a stochastic phenomenon, but rather indicates a discrepancy between the behavior of a natural system and the corresponding behavior of a particular model of that system.

Since the point of our development was the consideration of predictive models, it was necessary to describe the concept of time in some detail. We found that time itself is complex, in the sense that it admits many different kinds of encodings, and these encodings can themselves bifurcate from one another. Indeed, it is only within a class of systems which are themselves already similar in some non-temporal sense that a concept of time can be uniformly introduced which is valid for all the systems in the class. It was in fact the non-comparability of time scales generated by different kinds of dynamical processes which was at the heart of our discussion of selection and adaptation, which is the process by which predictive models are actually generated in biological systems.

First we may ask how best to study the "physiology" of an anticipatory system. That is, we suppose given a system which contains some predictive model, and uses that model to determine its present behavior. We do not ask how that model was generated, either ontogenetically or phylogenetically, but rather seek to understand the behavior of the system as a whole. Certain aspects of the behavior of such a system can be understood without knowing the specific nature of the model employed by the system, but follow from the general character of modelling relations. For instance, we know that the model, as an abstraction, must ultimately bifurcate from what it models; thus any such system is in a sense "spanned", and must undergo a characteristic form of senescence, as we have seen. We indicated how the character of this senescence can tell us something about the nature of the model, and how it is linked to the other qualities of the system.

However, for detailed understanding of such a system, we need to know specifically what the model is which the system is employing to generate its behavior. Thus the basic question arises: "How can we determine this model, from observations performed on the system itself?" This is the basic question underlying the "physiology" of anticipatory systems; it is one which has no counterpart in the theory of purely reactive systems, and raises a host of entirely new problems of both a theoretical and a practical character. Analogous problems which can be formulated in a formal context should be of great help in telling us how to approach this basic question; for instance, how do we need to observe a computer, or a Turing machine, in order to determine its program? It should be noted that the reactive paradigm itself, based on developments initiated in particle mechanics, presume that there are, in general, effective procedures for extracting system laws (ie linkages between ovservables, or equations of state) from appropriate observations of the observables themselves; we are now asking explicitly how (and indeed, whether) this can be effectively done in the context of anticipatory systems. Such questions have received relatively scant attention in the past; they now are seen to be of the essence.

Another kind of basic question we can ask about anticipatory systems concerns their "ontogeny"; the manner in which they are generated. We have suggested in the preceding chapters of the present section that the ontogeny of the models, and their role in control of anticipatory behavior, is the natural consequence of selection mechanisms; that these in turn involve the closely linked ideas of fitness, adaptation and a mechanism through which genome can act on phenotype. Here, of course, the terms "genome" and "phenotype" are to be understood in very general terms, they are ways of classifying the qualities which appear in an equation of state. Thus, the operation of a selection mechanism is a metaphor both for a biological evolution and for the structurally quite different phenomena usually associated with the notion of learning. It is quite clear that a study of the ontogenesis of anticipatory systems, governed by the requirement that adaptation (as measured by fitness) increases, can throw some light on the physiological problems articulated above. Of particular interest in this regard, as we sketched in the preceding chapter, is the circle of ideas relating the several subsystems of an adaptive system, and the manner in which a study of subsystems might bear upon the determination of an anticipatory model in the system as a whole. However, here too, the necessary ideas have only just begun to be formulated, and are themselves still in an essentially embryonic state.

The third circle of questions concerns the basic question of how we can hope to apply an understanding of anticipatory systems to develop a technology of anticipatory control. Such a technology would be of vast importance, in many vital areas. For instance, in a medical context, it is clear that many of the so-called metabolic diseases, which are as yet so imperfectly understood, can be thought of as derangements of anticipatory mechanisms. We have suggested that senescence [aging] can be regarded as a generalized maladaptation arising from a growing discrepancy between what the system's internal models are predicting, and what the system itself is actually doing. As we saw, the hallmark of this type of senescence is a kind of generalized maladaptation, without any localizable failure in specific subsystems. Likewise, hormonal control seems to be of an essentially anticipatory character; a hormone is thus to be regarded as a predictor, representing some anticipated future state of the organism. To approach endocrine mechanisms in this way is to cast an entirely new light on endocrine disorders, and points up once again the need to extract from an anticipatory system some information about the character of the models employed by the system. Exactly the same may be said about other vital physiological subsystems, such as the central nervous system and the immune system, which in some sense seem to work entirely off models. Hence a theory of anticipatory systems seems bound to find crucial applications in medicine.

Finally, of course, we must return to the circle of applications which were the initial point of departure for our entire development; namely, the management of our own societies. Here too, we feel that a deep understanding of anticipatory systems in general, and the character of the modelling relations which direct them, will be central. Furthermore, the ubiquitous character of anticipatory mechanisms in biology, and their emergence through selection mechanisms, provides for us a vast encyclopedia for how to solve complex problems of the type with which we are presently confronted (and also, equally usefully, of how not to solve them). This encyclopedia represents a natural biological resource to be harvested; a resource perhaps ultimately more important to our survival than the more tangible resources of food and energy. To learn to expolit this resource to the full involves an understanding of the metaphoric relations between biology and social systems, which we are only beginning to be able to grasp. Indeed, it was for this purpose that we went so deeply into the character of metaphorical relationships between systems in the above pages. The formal tools arising from these considerations represent a beginning, but still only a beginning, in these directions.

One other problem regarding anticipatory systems, unique to the human realm, may also be mentioned. The basic situation with which we have dealt so far involves the interaction of an anticipatory system with an environment that is non-anticipatory; ie is describable entirely within a reactive paradigm. We may, however, ask what happens when an anticipatory system must interact with an environment which is itself anticipatory. This is the situation embodied most graphically in the illustration below: [There are two mystics sitting at a table, playing chess together, but instead of consulting the chess board to decide what their next move should be, they are each gazing into crystal balls, hoping that IT will tell them....] The behavior of such systems is characteristic of human interactions. The closest approach to a theory of such interactions is found, of course, in the Theory of Games. But this theory is in many ways phenomenological and unsatisfactory; it is like a probabilistic theory of error, and awaits a more basic theory arising from fundamental principles. The development of such a theory of interacting anticipatory systems represents yet another direction for future research.

Finally, we come full circle to the ideas of Robert Hutchins, with which we started. His basic question, it will be recalled, was: "What ought we to do now?" The crucial word here is "ought"; a word which has traditionally been regarded as foreign to science. Indeed, if we stay entirely within a reactive paradigm, this word never arises. Perhaps this is the fundamental reason why Hutchins was so suspicious of scientists, and hence of science. In his view, the word "ought" was excluded as a matter of principle. However, in the study of anticipatory systems, we find that "ought" is of the essence; the character of a predictive model assumes almost an ethical character, even in a purely abstract context. We might even say that the models embodied in an anticipatory system are what comprise its individuality; what distinguish it uniquely from other systems. As we have seen, a change in these models is a change of identity. This is perhaps why, for human beings, the preservation of models becomes identical with the preservation of self. The identification of one's self with one's models explains, perhaps, why human beings are so often willing to die; ie to suffer biological extinction, rather than change their models, and why suicide is so often, and so paradoxically, an ultimate act of self-preservation. The study of anticipatory systems thus involves in an essential way the subjective notions of good and ill, as they manifest themselves in the models which shape our behavior. For in a profound sense, the study of models is the study of man; and if we can agree about our models, we can agree about everything else."