Let us take "man", whom most of us would agree is a self-organizing system. A man is any member of a well-specified set of men. But this set can be well-specified (that is, specified in a way that meets common approval) in a vast number of ways, according to an observer's objctive. Man, for example, may be specified anatomically (two legs, head, and so on), or alternatively as a decision maker which influences and is influenced by his circle of acquaintances. Each specification is equally valid and entails criteria of similarity. The point is, there are objectives for which neither the first specification (and the criteria it entails), nor the second (and the criteria it entails) are sufficient. In conversation, when trying to control a man, to persuade him to do something, how do I define him? Manifestly I do not, at least, I continually change my specification in such a way that he appears to me as a self-organizing system. (ATC 47)

The recently automated candy factory in Figure 15 (ii) illustrates an industrial application of adaptive control. An indication of sucessful activity -- a reward variable X, is derived from one of two sources (only one at once) namely -- (Mode I). An output meter (1) which measures quality and quantity of candy (according to a predetermined criterion), or (Mode II) a manager (2) who developes a preference for certain states of the plant upon a diversity of evidence, sampling the candy, watching his material bills, and altercating with customers who object to the sweetmeat.

In either case values of X are conveyed to clerk (3). The engineer (4) (having the status of overall controller) knows that the factory can be run by some possibly changing controller, because a limited number of taps used to be turned and a limited number of measures used to be made by men (5) before the place was automated. But the men (5) are disgrutled and will not disclose their arts. So (4) has to experiment by changing the parameters of a versatile sub-controller (which is equivalent to selecting different sub-controllers from a box) (6). For each setting of the parameters (7), the clerk records a value of X in a table (8) and these records are averaged and guide the engineer who wishes to maximize the average value of X. Finally (9) and (10) represent the imperfections which disturb any real control system. (ATC 63-64)

A steady state control system is a device that interacts with a subject who is performing a task in order to maintain his proficiency at a near constant level by posing problems at an appropriately slected level of difficulty, n. If p increases n is made to increase and vice versa so that p approximates a stipulated constant value X. The controller has the (ideal) effect of clamping the subject in a working region where the problems posed are neither so difficult as to prove incomprehensible nor so easy that there is insufficient relevant variation to occupy his attention. By prior hypothesis, either circumstance would lead to a state of affairs in which the subject would be bound to focus his attention upon problems deemed irrelevant by the external observer (a subject can neither assimilate problems that he fails to comprehend as problems, nor can he stop solving problems; it is only a matter of whether he solves relevant problems). Viewed thus, the steady state controller is an instrument that captures the subject's attention or, more accurately, that permits him to honour an agreement to solve problems of a particular kind. (CCL 26-27)

The reference frame itself is a system. It satisfies a definition proposed by Cloin Cherry that a system is an "ensemble of attributes". But it has no predictive value. In order to show how it becomes of predictive value we shall first introduce a convention for representing U, L, called a "phase space". Secondly, we shall credit the observer with a special objective v, namely to make predictions about any behavior in U, L. In other words, to discover all he can about a given way of looking at the assembly. Although "special" v is shared by nearly all "scientific observers". Perhaps it is also true thatt we are impelled to adopt v by a belief in the underlying regularity of the world, and that this regularity will be apparent in the reference frame we have chosen.

Suppose the observer can unambiguously describe his attributes. If he can, his senses can be replaced by instruments which convert events from the assembly into numerically valued attribute variables (including, possibly, two valued variables which equals 1, if an attribute is present and equals 0 if it is absent), labelled x1, x2, . . . xm and displayed in a common modality (perhaps on dials or meters). In the simplest case, the observer knows very little about the assembly. It is a black box with m initially unrelated outputs. By the usual convention, we represent these outputs, the values of the x variables, as independent co-ordinates in a phase space. (CCL 26-27)

Home Cybeneticians Gordon Pask Exemplars

Symbolic Generalizations | Beliefs, Models, and Analogies | Values | Exemplars | Techniques | Guiding Questions