Cybernetyczna teoria układów samodzielnych (‘A Cybernetic Theory of Autonomous Systems’, 1966) — an overview

Preface (pp. 7–10)

The preface explains why the theory has the shape it has. In its final version Mazur no longer sets the organism side by side with the machine; instead he studies a general concept and derives theorems from it. The price is having to build the concepts of thinking, linkages and motivation anew, and moving character to the end of the exposition. Earlier versions, developed over more than two decades (the author dates the beginning of the work to 1942), bore working titles that described them as a theory of character; what they had in common was the conviction that the psyche can be described in the same language in which an engineer describes regulation. Wiener's cybernetics helped the author to place his subject among the sciences and to make his vocabulary more precise, although cybernetics itself developed in a different direction. The human examples in the book are only illustrations: the conclusions apply to every entity that satisfies the assumptions of the theory, whether it is an animal, a human being or a device.

Concepts introduced in the chapter: no dictionary definitions come from this part of the book.

Chapter 1. Introduction (pp. 11–17)

Part I. Basic concepts (Chapters 1–4).

The chapter establishes the method of the whole book. Instead of reasoning by analogy — which first requires the similarity to be proved and does not guarantee that agreement in a few features will carry over to the rest — Mazur applies a ‘method of generalisation’: he introduces a broader concept, covering organisms and machines alike, and from its properties deduces what they have in common. Such an entity is the autonomous system, and its character is the set of its control properties.

The framework is an idea of cybernetics as a single science of control, not divided into technical, biological and social cybernetics, and not excluding goals and motivation, which are precisely the subject of the book. Most of the space goes to matters of vocabulary: the author does not adopt the terms of psychology, because they describe symptoms rather than causes, and a few of them, such as emotion or tolerance, he keeps only as names for concepts defined anew. The calculations in the book operate with physical quantities, but contain no units.

Concepts introduced in the chapter: none in the dictionary — the concepts of this chapter have definitions there from elsewhere (links in the text).

Chapter 2. Couplings (pp. 18–34)

The thesis of the chapter concerns the scope of the concept: Mazur rejects the technical division into ‘closed-loop’ and ‘open-loop’ control, because a loop exists in the latter too — only its missing link is the person who designed the device. The apparatus this thesis needs is modest — simple coupling and feedback coupling, and objects acting affirmatively or negatively — and a simple calculation yields only two kinds of feedback, positive and negative, with convergent and divergent variants and with oscillation.

The second idea of the chapter concerns self-equalising processes, in which potential differences disappear by themselves. Hence the picture of control as a chain: every counteraction to spontaneous equalisation — control's main adversary — is itself such an equalisation, set off at another point, and itself needs a further link. The illustrations are simple, taken from physics, from control engineering and from everyday experience.

Concepts introduced in the chapter: no dictionary definitions come from this part of the book.

Chapter 3. Information (pp. 35–46)

The chapter does two things: it gives the concept of information a precise definition, since until then it had been described only in words, and it adds a physical treatment to it. The physical treatment is required by the very nature of control: what counts in it is which reaction will follow and what flow of energy leads to it, and a measure in bits does not settle this. The author will develop both threads in Jakościowa teoria informacji (1970).

The chapter's conclusion is a physical one: in a set of connected paths the task of the code is performed by the distribution of the power flow, determined by conductances and potentials, and it is this that returns in the chapters on correlation. What matters most for the rest of the book is the equating of two vocabularies: stimulus, reaction, receptor, effector and control path are, according to the author, the same things that the language of information names differently — code, image, original, message, information (all of which have their own entries in the site's dictionary) — described only from another side. Along the way Mazur distinguishes the transfer of information (the messages change, the information remains) from its processing (the information changes too). The concepts are explained with examples from measurement, cartography, geometry, telegraphy and language.

Concepts introduced in the chapter (2): effector, receptor.

Chapter 4. Autonomous systems (pp. 47–59)

The chapter answers the question of how an autonomous system differs from other entities, and arranges systems in a hierarchy of increasingly rich kinds. At the top stands the autonomous system: it has its own homeostat, an organ that changes the structure of the system itself, and it is its own organiser. Its ‘own interest’ is to preserve its ability to control itself, and the state furthest from the dangerous limits is functional equilibrium (here the author refers to Cannon and Ashby). It is defended along two routes, in which the author sees the counterparts of prophylaxis and therapy: one runs through the correlator and reactions to the surroundings before damage arises, the other through the accumulator when damage already has to be repaired.

Lower in the hierarchy stand self-controlling systems, with their own correlator but without a homeostat; controllable systems, with their own accumulator but controlled from outside; and finally organised systems, which an organiser supplies with both information and energy. With controllable systems a power balance appears: total power divides into idle power, lost to the surroundings, and disposable power.

The whole hierarchy rests on the thesis that every entity is described by matter, energy and their arrangement, that is, structure, and that control means changing structure, which turns out to be information. The author illustrates the distinction between the three factors with examples of technical devices. Organisms are autonomous; autonomous technical devices the author calls autonoms, and the quantities describing the control properties of a system he calls character parameters.

Concepts introduced in the chapter (1): total power.

Chapter 5. Registration (pp. 60–66)

Part II. Informational processes (Chapters 5–10).

The chapter leads to the conclusion that the system never has the present itself at its disposal: even the freshest trace of a stimulus forms with a delay, so the difference between ‘now’ and ‘once’ comes down to the age of the record. These records are registrates, messages about the past produced in the correlator by registrators; the process by which they form is registration. Thanks to them the system makes use of experience, and a message it receives means anything to it at all — the meaning depends on what the recipient has registered before.

The final part of the chapter concerns the durability of traces. The requirements are opposite at the two ends of the path: the receptor should clear quickly, the registrate should last, though not indefinitely, because correlator capacity is limited. Records therefore have to fade gradually — deregistration — preferably the oldest ones, least consistent with the present state of the surroundings. The examples come from everyday situations, from measuring instruments and from recording devices.

Concepts introduced in the chapter (1): registrator.

Chapter 6. Estimation (pp. 67–68)

This short chapter describes the output of the correlator, that is, the reverse side of registration, and gives a precise sense to the term decision: it is the moment at which the quantity changed by the estimator crosses a threshold. The definition also covers the everyday word: only what can no longer be reversed, because the threshold has been crossed, is a decision; everything before that is merely approaching the threshold. Estimation itself settles which effector will move, although the energy for the reaction comes from the accumulator; the time from the beginning of the rise to the crossing of the threshold Mazur calls the decision time.

Concepts introduced in the chapter (1): estimator.

Chapter 7. Correlation (pp. 69–83)

The key new concept of the chapter is the reflection potential, which the system itself adds to correlation in order to strengthen desired stimuli and suppress undesired ones instead of depending on its surroundings. By correlation Mazur means the processes in a common medium in which registrators and estimators are connected in many ways; only such a medium makes it possible to respond to the same stimulus in different ways, because with paths independent of one another the reaction would always be the same.

The description is physical: correlational power is the product of potential and correlational conductance, and a decision is made once the decision potential is exceeded. The author explains the whole with an example from automatic control, set against human behaviour, and gives in a table the equivalent technical, physiological and cybernetic terms; at the same time he stipulates that the similarity of the device to a human being is only apparent, and returns to this in Chapter 10.

Concepts introduced in the chapter (7): correlational elements, correlational energy, estimation potential, correlational potential, registration potential, correlational medium, reflection potential.

Chapter 8. Registrates and correlates (pp. 84–104)

The guiding idea of the chapter: a single mechanism, linkage, gives rise to learning, to general concepts and to mistaken beliefs. When two stimuli act together, or one immediately after the other, the route between their registrators is strengthened, and from then on one of them can evoke the reaction proper to the other. A general concept, according to the author, forms like a sediment: out of many similar experiences what survives is what recurred in each of them, while details met only once fade. By the same mechanism the author also explains superstitions, in which an earlier event is taken for the cause of a later one.

The physical basis is that the flow of correlational power increases the conductance of its own route, convergently, up to a certain limit; Mazur calls the increase in conductance the registrate of a stimulus, and the corresponding distribution of power its correlate; when the increase concerns the route between registrators, he speaks of a linkage registrate and linkage correlate. In this language he describes sensation, representation, three kinds of learning, and creativity as new combinations of known elements. Separately he lists four physical processes — the fading of a registrate, its renewal by the same stimulus, the outflow of power from the area of the registrate (detraction) and the return of that power (retraction) — and shows that the everyday words ‘forgetting’ and ‘remembering’ each lump two of them together. The examples come from the everyday experience of children and adults, from medicine, from experiments on animals and from the history of superstitions.

Concepts introduced in the chapter: none in the dictionary — the concepts of this chapter have definitions there from elsewhere (links in the text).

Chapter 9. Motivation (pp. 105–118)

The central concept of the chapter is motivation: the readjustment of correlation so that it promotes reactions favourable to the system and inhibits unfavourable ones, carried out in the mutual interaction of the correlator and the homeostat, because in the author's view motivation springs from this cycle and not from either of the organs. Experience consolidates it in two directions — from then on the system speeds up a favourable reaction (offensive motivation) and does not allow an unfavourable reaction (defensive motivation) — but learning from one's own mistakes can be fatal. That is why Mazur understands instincts as innate motivations, and in passing turns round the charge made against thinking machines: it is the human being who carries a ready-made program.

The mechanism works in two directions: the influence of the homeostat on the correlator is reflection, the reverse influence is emotion, and the channel is the perturbation potential, whose rise and fall give the two kinds of emotion. The homeostat judges reactions by the current state of the correlator, because it does not know the future; disturbances inside the system it eases by itself, but when they are too strong and their source lies in the surroundings, only a reaction can remove it. The same potential makes it possible to react to a danger before it arrives; such vigilance has its cost, however, because in the process the system disturbs its own equilibrium. The last pages apply these concepts to behaviour in everyday life.

Concepts introduced in the chapter (14): favourable reaction, unfavourable reaction, weakening reflection, strengthening reflection, motivational correlates, motivational registrates, disturbance-relieving emotions, disturbing emotions, offensive correlates, offensive registrates, defensive correlates, counter-offensive motivation, defensive registrates, primary motivational registrates.

Chapter 10. Correlational circuits (pp. 119–129)

The conclusion of the chapter: in the reflection circuit there is no first cause of a reaction, which is why in an autonomous system the same stimuli can evoke different reactions. Along the way the chapter corrects the comparison made in Chapter 7: a self-controlling system knows only the reaction circuit, which links the correlator with the surroundings, so it resembles a human being only when the latter acts by reflex. What sets the human being apart is the reflection circuit, which links the correlator with the homeostat, and a circuit in which the two alternate. In the reflection circuit Mazur places thinking — views and convictions differ only in how durable its result is — and consciousness, which is bound up with the continuous persistence of this circuit.

The theme of choice is connected with the passage from the reflection circuit to the reaction circuit. A quandary, that is, a struggle between offensive motivation and defensive motivation, is resolved in the same way in a machine and in an organism, by the flow of energy — contrary to the view that only a human being resolves ‘non-formalised’ quandaries — and it is ended by the fastest correlate, not the best one; in this the author sees the source of mistakes made knowingly. The illustrations come mainly from everyday life, and also from literature, from games and from engineering.

Concepts introduced in the chapter (1): quandary resolution.

Chapter 11. Ageing and expansion of autonomous systems (pp. 130–137)

Part III. Energy processes (Chapters 11–13).

The chapter leads up to the dynamism coefficient, the ratio of the rate of expansion to the rate of ageing, from which Part IV will derive the typology of character. The author defines both quantities through the state of the material. Ageing has a narrow meaning here: the decline in the quality of material; in this sense a system ages even while its power is still rising. The calculation is simplified by the assumption that quality declines exponentially. The counterweight is expansion, that is, increasing the quantity of material with energy taken in from outside; it is controlled by the homeostat, which ties its rate to how quickly the material deteriorates.

The background to the whole part is this: an autonomous system has to defend itself against time, because self-equalising processes dissipate the energy on which every structure depends. The quality and quantity of material, together with unit power, determine total power, that is, the greatest power the system can safely process; over the course of life it first rises and then declines.

Concepts introduced in the chapter (4): ageing curve, ageing coefficient, expansion curve, expansion coefficient.

Chapter 12. Idle power (pp. 138–151)

The chapter proves that an autonomous system cannot last for ever, and determines its existence time — for an organism, the moment of death — as the point at which what it spends merely on persisting becomes equal to its declining total power.

Two general conclusions follow from the calculation. Existence time depends only on ageing, unit power and lossiness, that is, the coefficient by which idle power (in an organism, basal metabolism) grows with the quantity of material; restraining growth does not change it, but leaves more power at the system's disposal. The extreme case, a system that does not expand at all, corresponds to machines; an organism cannot afford this, because it starts from a small germ.

Concepts introduced in the chapter: none in the dictionary — the concepts of this chapter have definitions there from elsewhere (links in the text).

Chapter 13. Working power (pp. 152–166)

The calculation in this chapter leads to the freedom coefficient (the dictionary gives a definition from 1976, framed differently): a measure of how much of the disposable power is left to the system to use as it chooses; at zero, behaviour is entirely forced. Along the way Mazur divides the power taken in into its components. Some of this power is absorbed by the very effort of obtaining it — this is working power, which together with idle power makes up safeguarding power; the surplus, coordination power, the system spends on improving its conditions, and better conditions and the surplus reinforce each other. The struggle for better conditions nevertheless has a defensive aim: it allows the system to live to the limit set by ageing, but does not push that limit back.

The intake of power has two limits: the resources of the surroundings and the endurance of the system itself. What the system actually takes in and processes Mazur calls internal power, and what the surroundings offer, external power; for organisms he also uses the names physiological power and sociological power. In poor surroundings the share of working power grows, until in the extreme case the system cannot sustain itself. The examples come from nature, from engineering and from social life.

Concepts introduced in the chapter (5): internal power, external power, safeguarding power, supply coefficient, coordination power.

Chapter 14. Character dynamism (pp. 167–186)

Part IV. Character (Chapters 14–17).

The first character parameter, character dynamism, carries the dynamism coefficient from Chapter 11 over to a logarithmic scale: it says whether the system expands faster or more slowly than it ages. The character classes distinguished on this scale — endodynamism, endostatism, statism, exostatism and exodynamism — differ above all in what the system's control is directed at: an endodynamic system accumulates external power and takes distant consequences into account, while an exodynamic one keeps a large reserve of power for the present moment and removes disturbances quickly.

The author stresses that extreme types are rare and that none is ‘better’, because each plays its own role in a community and has its own distortions; with age, dynamism shifts from exodynamism towards endodynamism. A considerable part of the chapter is taken up by a survey of the manifestations of dynamism in human behaviour, later gathered in an extensive table, and towards the end the author draws conclusions from it for upbringing.

Concepts introduced in the chapter (1): character classes.

Chapter 15. Character breadth (pp. 187–199)

The second parameter, character breadth, says how wide a range of stimuli a system of a given dynamism accepts. Its practical meaning is clearest in the case of coercion: forcing begins to work only where tolerance ends, and its effectiveness is measured by susceptibility, which is the greater, the less the system can draw from its surroundings. Tolerance depends only on the system and grows with age, whereas susceptibility depends on the surroundings; in the dictionary both have definitions from 1976, framed differently than in this book.

In energy terms breadth is the ratio of safeguarding power to total power: the more power is absorbed merely by keeping the system going, the less choosy it is. Great tolerance prolongs quandaries, little tolerance speeds up decisions. Combinations of the two components give four types, which the author illustrates with characters from literature, and the reactions to a sudden loss of resources and to sudden success show that after a temporary change tolerance returns to its previous value, while susceptibility settles at a new level — higher after a loss, lower after success.

Concepts introduced in the chapter: none in the dictionary — the concepts of this chapter have definitions there from elsewhere (links in the text).

Chapter 16. Character level (pp. 200–207)

The third parameter, character level, is the number of linked elements of the correlator, and behaviour is determined by the current level, not the limit level: the innate possibilities of the correlator still have to be filled with linkages. The author connects the level with intelligence (for him the intelligence quotient is an indirect, statistical measurement), with aptitude and with talent, which he explains by a concentration of registrators of one kind.

The justification is a matter of calculation: if all routes conducted equally, the system would distinguish only a single bit, and the possible number of linkages grows with the square of the number of elements; the proportion of registrators to estimators is of secondary importance here. The starting point is the primary level — Mazur considers it more logical that an organism inherits a certain distribution of conductances than that it starts from a ‘blank slate’ — and the increase slows down as the elements not yet linked run out; the upper limit is set by the number of all elements. The increase is favoured by frequent, strong and desired stimuli and a fairly long period of learning, and harmed by anything that arouses disturbing emotions, while unused linkages fade away. The examples come from medicine, from games and collecting, from school and from computing technology.

Concepts introduced in the chapter (4): primary level, limit level, assimilability, relative stimulus frequency.

Chapter 17. Evolution of character (pp. 208–214)

The chapter ends with the concept of character crystallisation: a sudden transition from an apparent character to the real one, so named by analogy with crystals precipitating from a supersaturated solution. This is how Mazur explains genuine conversions and distinguishes them from sham ones, which arise from a temporary rise in susceptibility after a loss of resources and pass when the resources return. Crystallisation is possible because character is sometimes only apparent: when something that in fact serves the system has been linked with something unpleasant, the system unwittingly fights its own inclinations — typically as a result of an upbringing that cuts it off from other stimuli.

The source of such a distortion is imposed stimuli, coming from other systems; random stimuli and provoked stimuli, evoked by the system's own reactions, work the other way, revealing the real character ever more clearly. Dynamism itself changes under the influence of the surroundings only when scarcity threatens existence, and always towards endodynamism; changes in resources shift only breadth. These changes are described with a diagram of the three parameters and, after division into classes, with the character square, on which a person is represented by a point or a segment and a group by an area. The examples come from literature, from fairy tales, from extreme situations and from everyday life.

Concepts introduced in the chapter (7): character square, character diagram, imposed stimuli, random stimuli, provoked stimuli, real character, character crystallisation.

Conclusion (pp. 215–216)

The conclusion takes up the question of free will and answers it with the difference between the circuits: thanks to the reflection circuit, an autonomous system directs itself through its own homeostat, whereas the reactions of an automaton, confined to the reaction circuit, are imposed by the surroundings. In this account free will turns out to be a name for such self-direction and requires no exception to the laws of physics. The author nevertheless leaves the last word to specialists, and restricts the conclusions of the theory to entities that really are autonomous systems.

Concepts introduced in the chapter: no dictionary definitions come from this part of the book.

How to read it

Start with Chapters 1 and 4: the first explains the method and the vocabulary, the fourth sets out the kinds of systems. Chapters 2 and 3, Part III and Chapter 15 involve the most calculation — it is enough to follow the conclusions and the graphs. The heart of the book is Chapters 7–10; read them slowly and with the dictionary, because each concept rests on the previous one. Part IV is the most accessible; its descriptions of behaviour illustrate the argument of Parts II and III, and Mazur developed the theory of character in Cybernetyka i charakter (1976).

Concepts from this book in the dictionary

In total: 48 entries with 48 definitions from this book — 27 in the domain theory of character, 21 in the domain cybernetics.

The lists under each chapter include the entries whose dictionary definition comes from the given chapter. Many other concepts in this book, such as homeostat, correlator, emotion or tolerance, have dictionary definitions from Mazur's later works — links to them appear in the text. These later definitions are sometimes framed differently from those in this book.

Our own account, working version; definitions of concepts are in the dictionary, with their sources.

All summaries