Generated on 2011-09-28 14:05 GMT / Public Domain, Google-digitized / http://www.hathitrust.

org/access_use#pd-google

Copyright, 1948

By

THE MASSACHUSETTS INSTITUTE OF TECHNOLOGY

All Rights Reserved

This book or any part thereof must not

be reproduced in any form without the

written permission of the publisher.

THIRD PRINTING JANUARY, 1949

Generated on 2011-09-28 14:06 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

Printed in the United States of America

Generated on 2011-09-28 14:06 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

org/access_use#pd-google .hathitrust.Generated on 2011-09-28 14:07 GMT / Public Domain. Google-digitized / http://www.

Generated on 2011-09-28 14:07 GMT / Public Domain.hathitrust. Google-digitized / http://www.org/access_use#pd-google .

Generated on 2011-09-28 14:08 GMT / Public Domain.org/access_use#pd-google . Google-digitized / http://www.hathitrust.

hathitrust. Google-digitized / http://www.org/access_use#pd-google .Generated on 2011-09-28 14:08 GMT / Public Domain.

hathitrust. it was felt that it was essential to have someone present who could examine mathematical questions critically. There are fields of scientific work. he will regard the next subject as something belonging to his colleague three doors down the corridor. and in which important work has been triplicated or quadruplicated . Google-digitized / http://www. which have been explored from the different sides of pure mathematics. Moreover. in which every single notion receives a separate name from each group .8 CYBERNETICS method for a long time. the Mexicans. as we shall see in the body of this book. For many years Dr. electrical engineering. but there was a Gauss. A century ago there may have been no Leibniz. and the Russians — an inextricable tangle of exploration. statistics. science has been increasingly the task of specialists.org/access_use#pd-google will consider any interest in it on his own part as an unwarran table breach of privacy. and will know all its literature. Today there are few scholars who can call themselves mathematicians or physicists or biologists without restriction A man may be a topologist or an acoustician or a coleopterist. Since that time. a Faraday. Rosen- blueth's call to Mexico in 1944 and the general confusion of the war ended the series of meetings. and all its ramifications. and Generated on 2011-09-28 14:08 GMT / Public Domain. nomenclature. Since Leibniz there has perhaps been no man who has had a full command of all the intellectual activity of his day. and laws. These specialized fields are continually growing and invading new territory. more frequently than not. and neuro- physiology. He will be tilled with the jargon of his field. while still other important work is delayed by the unavailability in one field of results that may have already become classical in the next field. in fields which show a tendency to grow progressively narrower. Rosenblueth and I had shared the convic tion that the most fruitful areas for the growth of the sciences were those which had been neglected as a no-man's land be tween the various established fields. I thus became an active member of the group until Dr. and a Darwin. It is these boundary regions of science which offer the richest . The result is like what occurred when the Ore gon country was being invaded simultaneously by the United States settlers. but. the British. and had in fact been a participant in Josiah Royce's Harvard seminar on the subject in 1911-1913.

INTRODUCTION 9

opportunities to the qualified investigator. They are at the

same time the most refractory to the accepted techniques of

mass attack and the division of labor. If the difficulty of a

physiological problem is mathematical in essence, ten physio

logists ignorant of mathematics will get precisely as far as one

physiologist ignorant of mathematics, and no further. If a phy

siologist, who knows no mathematics, works together with a ma

thematician who knows no physiology, the one will be unable

to state his problem in terms that the other can manipulate, and

the second will be unable to put the answers in any form that

the first can understand. Dr. Rosenblueth has always insisted

that a proper exploration of these blank spaces on the map of

science could only be made by a team of scientists, each a specia

list in his own field, but each possessing a thoroughly sound and

trained acquaintance with the fields of his neighbors; all in the

habit of working together, of knowing one another's intellectual

customs, and of recognizing the significance of a colleague's new

suggestion before it has taken on a full formal expression. The

mathematician need not have the skill to conduct a physiological

experiment, but he must have the skill to understand one, to

criticize one, and to suggest one. The physiologist need not be

able to prove a certain mathematical theorem, but hemust be

able to grasp its physiological significance and to tell the mathe

Generated on 2011-09-28 14:08 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

matician for what he should look. We had dreamed for years

of an institution of independent scientists, working together in

one of these backwoods of science, not as subordinates of some

great executive officer, but joined by the desire, indeed by the

spiritual necessity, to understand the region as a whole, and to

lend one another the strength of that understanding.

We had agreed on these matters long before we had chosen

the field of our joint investigations and our respective parts in

them. The deciding factor in this new step was the war. 1 had

known for a considerable time that if a national emergency

should come, my function in it would be determined largely

by two things : my close contact with the program of compu

ting machines developed by Dr. Vannevar Bush, and my own

joint work with Dr. Yuk Wing Lee on the design of electric net

works. In fact, both proved important. In the summer of

1940,1 turned a large part of my attention to the development of

10 CYBERNETICS

computing machines for the solution of partial differential equa-

tia.u. I had long been interested in these, and had convinced

myself that their chief problem, as contrasted with the ordinary

differential equations so well treated by Dr. Bush on his differen

tial analyzer, was that of the representation of functions of more

than one variable. 1 had also become convinced that the pro

cess of scanning, as employed in television, gave the answer

to that question, and in fact that television was destined to

be more useful to engineering by the introduction of such new

techniques than as an independent industry.

It was clear that any scanning process must vastly increase

the number of data dealt with as compared with the number

of data in a problem of ordinary differential equations. To ac

complish reasonable results in a reasonable time, it thus became

necessary to push the speed of the elementary processes to

the maximum, and to avoid interrupting the stream of these

processes by steps of an essentially slower nature. It also

became necessary to perform the individual processes with so

high a degree of accuracy that the enormous repetition of the ele

mentary processes should not bring about a cumulative error

so great as to swamp all accuracy. Thus the following require

ments were suggested :

(1) That the central adding and multiplying apparatus of the

Generated on 2011-09-28 14:09 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

computing machine should be numerical, as in an ordinary ad

ding machine, rather than on a basis of measurement, as in the

Bush differential analyzer ;

(2) That these mechanisms, which are essentially switching de

vices, should depend on electronic tubes rather than on gears

or mechanical relays, in order to secure quicker action ;

(3) That, in accordance with the policy adopted in some

existing apparatus of the Bell Telephone Laboratories, it would

probably be more economical in apparatus to adopt the scale

of two for addition and multiplication, rather than the scale of ten;

(4) That the entire sequence of operations be laid out on

the machine itself so thai there should be no human interven

tion from the time the data were entered until the final results

should be taken off ; and that all logical decisions necessary for

this should be built into the machine itself;

INTRODUCTION 11

(5) That the machine contain an apparatus for the storage of

data which should record them quickly, hold them firmly until

erasure, read them quickly, erase them quickly, and then be

immediately available for the storage of new material.

These recommendations, together with tentative suggestions

for the means of realizing them, were sent in to Dr. Vannevar

Bush for their possible use in a war. Atthat stage of the prepara

tions for war, they did not seem to have sufficiently high priority

to make immediate work on them worth while. Nevertheless,

they all represent ideas which have been incorporated into the

modern ultra-rapid computing machine. These notions were

all very much in the spirit of the thought of the time, and 1 do

not for a moment wish to claim anything like the sole respon

sibility for their introduction. Nevertheless, they have proved

useful, and it is my hope that my memorandum had some effect

in popularizing them among engineers. At any rate, as we shall

see in the body of the book, they are all ideas which are of in

terest in connection with the study of the nervous system.

This work was thus laid on the table, and although it has

not proved to be fruitless, it led to no immediate project by

Dr. Rosenblueth and myself. Our actual collaboration resul

ted from another project which was likewise undertaken for the

purposes of the last war. At the beginning of the war, the Ger

Generated on 2011-09-28 14:09 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

man prestige in aviation and the defensive position of England

turned the attention of many scientists to the improvement

of anti-aircraft artillery. Even before the war it had become

clear that the speed of the airplane had rendered obsolete all

classical methods of the direction of fire, and that it was neces

sary to build into the control apparatus all the computations

necessary. These were rendered much more difficult by the fact

that, unlike all previously encountered targets, an airplane has

a velocity which is a very appreciable part of the velocity of

the missile used to bring it down. Accordingly, it is exceeding

ly important to shoot the missile, not at the target, but in such

a way that missile and target may come together in space at

some time in the future. We must hence find some method of

predicting the future position of the plane.

The simplest method is to extrapolate the present course of the

or in some other way take evasive action. it merely changes the acceleration of the plane. for example. an aviator under the strain of combat conditions is scarcely in a mood to engage in any very compli cated and untrammeled voluntary behavior. and this Generated on 2011-09-28 14:10 GMT / Public Domain. and are in fact realizable by apparatus which we can build. For one thing. except perhaps in the case of a very waste ful barrage fire. This has much to recommend it. All this made an investigation of the problem of the curvilinear prediction of flight worth while. other things are nol equal. first into change of velocity. However. and any too sudden deviation from his course will produce an acceleration that will render him unconscious. and may disintegrate the plane. and the pilot were to make the sort of intelligent use of his chances that we anticipate in a good poker player. by the time the first shell has burst. before it is finally effective. he has so much opportunity to modify his expected position before the arrival of a shell that we should not reckon the chances of hitting him to be very good. To predict the future of a curve is to carry out a certain operation on its past. the less time it has to accomplish a mission. stunt. On the other hand. Then too he can only control the plane by moving his control surfaces. Other things being equal. the less is its effective velocity. and the pilot will probably zig-zag. and is quite likely to follow out the pattern of activity in which he has been trained.hathitrust. but there are certain operators which bear it a certain resemblance. Moreover. he is in a plane going at an exceedingly high speed. a plane will fly as straight a course as possible. Even when it is fully deve loped.org/access_use#pd-google change of acceleration must be converted. and the longer it remains in a dangerous region. The more a plane doubles and curves in flight. The true prediction operator cannot be realized by any conslructible apparatus. If this action were completely at the disposal of the pilot.12 CYBERNETICS plane along a straight line. I suggesled to Professor Samuel Caldwell of the Mass achusetts Institute of Technology that these operators seemed . and the new regimen of flow that is established takes some small time to develop. Google-digitized / http://www. the pilot does not have a completely free chance to maneuver at his will. whether the results should prove favorable or unfavorable for the actual use of a control apparatus involving such curvilinear prediction. and then into change of position.

and acting as an essential part of it. the plane. In this second case. any friction or other delaying force which hampers the motion of the tiller will increase the admission of . I shall discuss this in considerable detail in the appropriate chapters. the difference between this pattern and the actually performed motion is used as a new input to cause the part regulated to move in such a way as to bring its motion closer to that given by the pattern. Thus the tiller turns so as to bring the other end of the valve-regulating offset amidships. Bigelow and 1 came to the conclusion that an extremely important factor in voluntary activity is what the control engineers term feed-back. using this as a ready- made model of the desired fire-control apparatus. We did so. but in the more usual case.org/access_use#pd-google Mr. It is enough to say here that when we desire a motion to follow a given pattern. I found myself engaged in a war project. and in the second.hathitrust. the forecasting of the future. the execution of a complicated pattern of computation . we should not avoid the discussion of the performance of certain human functions. Moreover. and he immediately suggested that we try them out on Dr. and it is desirable to know its performance character istics. In some fire-control apparatus. with results which will be discussed in the body of this book. It is essential to know their characteristics. it is true. their target. Julian H. is also humanly controlled. there is a human gun-pointer or a gun- trainer or both coupled into the fire-control system. Clearly. It will be seen that for the second time 1 had become engaged in the study of a mechanico-electrical system which was designed to usurp a specifically human function — in the first case. and in that way registers the angular position of the wheel as the angular position of the tiller. Bigelow and myself were partners in the investi gation of the theory of prediction and of the construction of apparatus to embody these theories. in which Mr.INTRODUCTION 13 worth trying. in order to incorporate them mathematically into the machines they control. At any rate. For example. Bush's differential analyzer. the original impulse to point comes in directly by radar. Google-digitized / http://www. Generated on 2011-09-28 14:10 GMT / Public Domain. one form of steering engine of a ship carries the reading of the wheel to an offset from the tiller which so regulates the valves of the steering engine as to move the tiller in such a way as to turn these valves off.

However. or to use a term now in vogue... and breaks down completely. On the contrary. If the proprio- ceptive sensations are wanting. It is a phenomenon which we understand very thoroughly from a quantitative point of view. To do this I have to move certain muscles. On the other hand. An ataxia of this type is familiar in the form of syphilis of the central nervous system known as tabes 1. there must be a report to the nervous system. 1945. our motion proceeds in such a way that we may say roughly Generated on 2011-09-28 14:10 GMT / Public Domain. Seruomechanixms. Van Nostrand. . under certain conditions of delay etc.hathitrust. we finda very precise discussion of feed-back. for all of us but a few expert anatomists. suppose that I pick up a lead-pencil. and even among the anatomists. Now. we are unable to perform the act of picking up the pencil. of the amount by which we have failed to pick the pencil up at each instant. L. and will be followed by a feed-back in the other direction which makes the rudder overshoot still more.org/access_use#pd-google that the amount by which the pencil is not yet picked up is decreased at each stage. a feed-back that is too brusque will make the rudder overshoot. what we will is to pick the pencil up. McColl. To perform an action in such a manner. proprioceptive. and the conditions under which it breaks down. and we do not replace them by a visual or other substitute. in such a way as to increase the torque tending to bring the tiller to the desired position. Once we have determined on this. This part of the action is not in full consciousness. the conditions under which it is advantageous. there are few if any who can perform the act by a conscious willing in succession of the con traction of each muscle concerned.14 CYBERNETICS steam to the valves on one side. but it is more generally kinaesthetic. In a book such as that by McColl '. Google-digitized / http://www. and will decrease it on the other. and find ourselves in a state of what is known as alaxia. until the steering- mechanism goes into a wild oscillation or hunting. If we have our eye on the pencil. conscious or unconscious. Thus the feed back system tends to make the performance of the steering engine relatively independent of the load. this report may be visual. at least in part. we do not know what these muscles are .

18-24 (1943). On the contrary. Ihe central nervous system no longer appears as a self-contained organ. pp. and goes into an uncontrollable oscillation ? Dr. Purpose & Teleology ». and we decided that if we could ever bring our plan for an interscientific insti tute to fruition. 10. « Behaviour. Bigelow and myself that the problems of control engineering and of communication engineering were I. Bigelow and myself approached Dr. Rosenblueth and I foresaw that this paper could only be a statement ot pro gram for a large body of experimental work. Rosenblueth. Is there any pathological condition in which the patient. an excessive feed-back is likely to be as serious a handicap to organized activity as a defective feed-back. Dr. . this topic would furnish an almost ideal center for our activity. We thus found a most significant confirmation of our hypo thesis concerning the nature of at least some voluntary acti vity. which we wrote up and published x. Rosenblueth immediately answered us tliat there is such a well- known condition. and that it is often associated with injury to the cerebellum. in trying to per form some voluntary act like picking up a pencil. Mr. where the kinaesthetic sense conveyed by the spinal nerves is more or less destroyed. This seemed to us to mark Generated on 2011-09-28 14:11 GMT / Public Domain.hathitrust. Rosenblueth with a very specific question. that it is called purpose tremor.org/access_use#pd-google a new step in the study of that part of neurophysiology which concerns not solely the elementary processes of nerves and synapses but the performance of the nervous system as an inte grated whole. It will be noted that our point of view considerably trans cended that current among neurophysiologists. emerging from the nervous system into the muscles. Google-digitized / http://www. and re-entering the nervous system through the sense organs. vol. However.INTRODUCTION 15 dorsalis. In view of this possibility. whether they be propriocep- tors or organs of the special senses. Philo- *ophy of Science. some of its most characteristic activities are explicable only as circular processes. On the communication engineering plane. receiving inputs from the senses and discharging into the mus cles. Wiener & Bigelow. The three of us felt that this new point of view merited a paper. overshoots the mark. it had already become clear to Mr.

hathitrust. and the longer it would be before such oscillations would die out. The better the apparatus was for smooth waves. and that the)' centered not around the technique of electrical engineering but around the much more fundamen tal notion of the message. Google-digitized / http://www. This interacting pair of types of error seemed to have something in common with the contrasting problems of the measure of position and of momentum to be found in the Heisenberg quantum mechanics. mechanical. and the choice of the particular apparatus to be used in a specific case Generated on 2011-09-28 14:11 GMT / Public Domain. Once we had this. whether this operator is realized by a scheme of mathematical computation. In this connection. we could translate the problem of optimum prediction to the determination of a specific opera- . of a roughly antagonistic nature. this refinement of behavior was always attained at the cost of an increasing sensitivity. as described according to his Principle of Uncertainty. The message is a discrete or continuous sequence of measurable events distri buted in time — precisely what is called a time-series by the statisticians.16 CYBERNETICS inseparable. While the prediction apparatus which we at first designed could be made to anticipate an extremely smooth curve to any desired degree of approximation. Thus the good prediction of a smooth wave seemed to require a more delicate and sensitive apparatus than the best possible prediction of a rough curve . or by a mechanical or electrical apparatus.org/access_use#pd-google was dependent on the statistical nature of the phenomenon to be predicted. or nervous means. it was not difficult to make what had originally seemed to be a difficulty in the the ory of prediction into what was actually an efficient tool for sol ving the problem of prediction.it became possible to derive an explicit expression for the mean square error of prediction by a given technique and fora given lead. Once we had clearly grasped that the solution of the problem of optimum prediction was only to be obtained by an appeal to the statistics of the time series to be predicted. Assuming the statistics of a time series. whether this should be transmitted by electrical. we found that the ideal prediction mechanisms which we had at first contemplated were beset by two types of error. the more it would be set into oscillation by small depar tures from smoothness. The prediction of the future of a message is done by some sort of operator on its past.

to achieve a physical realization of this solution by a constructible appa ratus. given its statistical nature. We thus have replaced in the design of wave filters processes which were formerly of an empirical and rather hapha zard nature by processes with a thorough scientific justification.org/access_use#pd-google a given lead. The notion of statistical mechanics has indeed been encroaching on every branch of science for more than a century. In doing this. . It oc curred to us that this was not an isolated case. The optimum design of this operator and of the apparatus by which it is realized depends on the statistical nature of the message and the noise. By reducing a problem of this sort to a minimization principle. and even further. the significance of the statistical element is immediately apparent. or the message modified by a given lag. at least one problem of engineering design took on a completely new aspect. Minimization problems of this type belong to a recognized branch of mathematics. the calcu lus of variations. With the aid of this technique. or the message under Generated on 2011-09-28 14:12 GMT / Public Domain. singly and jointly. we were able to obtain an expli cit best solution of the problem of predicting the future of a time series. but that there was a whole region of engineering work in which similar design pro blems could be solved by the methods of the calculus of varia tions. by an ope rator applied to the corrupted message. however. We often find a message contaminated by extraneous disturbances which we call background noise. a branch of statistical mechanics. Google-digitized / http://www. We attacked and solved other similar problems by the same methods. enginee ring design has been held to be an art rather than a science. We then face the problem of restoring the original message. we had established the subject on a far more scientific basis. We shall see that this dominance of statistical mechanics in modern physics has a very vital significance for the interpretation of the nature of time.hathitrust. In general. we have made of communication engineering design a statistical science. and this branch has a recognized technique. Once we had done this. Among these was the problem of the design of wave filters.INTRODUCTION 17 tor which should reduce into a minimum a specific positive quan tity dependent on this operator. In the case of communication engineering. The transmission of information is impossible save as a trans.

Fisher's motive in studying this subject is to be found in classical statistical theory. that of irritability. Generated on 2011-09-28 14:13 GMT / Public Domain. Google-digitized / http://www. Bull. then it may be sent most efficiently and with the least trouble by sending no message at all.3-14(1941). although a considerable part of my work was done before my attention was called to the work of the Russian school. Math. A. belongs to 1. Sec. A. . Sci. R.. Just as the amount of information in a system is a measure of its degree of organization. and that of the author in the problem of noise and message in electrical filters. in which the unit amount of information was that transmitted as a single de cision between equally probable alternatives. If only one contingency is to be trans mitted. « Interpolation uud Extrapolation von stationaren Zufal- ligen Folgen ». Such questions arise independently in the study of enzymes and other catalysts. we had to develop a statistical theory of the amount of information. and the one is simply the negative of the other. This point of view leads us to a number of considerations concerning the second law of ther modynamics. Let it be remarked parenthetically that some of my speculations in this direction attach themselves to the earlier work of Kolmogo- roff 1 in Russia. S.hathitrust. and their study is essential for the proper understanding of such fundamental phenomena of living matter as metabolism and reproduction. Acad.org/access_use#pd-google The notion of the amount of information attaches itself very naturally to a classical notion in statistical mechanics: that of entropy. so the entropy of a system is a measure of its degree of disorganization .18 CYBERNETICS mission of alternatives. V. that of Shannon in the problem of coding information. S. Dr. The telegraph and the telephone can perform their function only if the messages they transmit are continually varied in a manner not completely determined by their past. 5.. U. To cover this aspect of communication engineering. The third fundamental phenomenon of life. Fisher. and can only be designed effectively if the variation of these messages conforms to some sort of statis tical regularity. among them the statisti cian R. and the author. Shannon of the Bell Telephone Labora tories. This idea occurred at about the same time to several writers. and to a study of the possibility of the so-called Maxwell demons. Kolmogoroff.

or even of a single name for the field. whether in the machine or in the animal. which was published by Clerk Maxwell in 1868 2. 2. Maxwell. we wish to recognize that the first significant paper on feed-back mechanisms is an article on gover nors.hathitrust. Warren McCulloch. Eng. Rosenblueth at a meeting held in New York in 1942 under the auspices of the Josiah Macy foundation. On the other hand. we were seriously hampered by the lack of unity of the literature concer ning these problems. Among those present at that meeting was Dr. Press. and Wiener were disseminated by Dr. First the ideas of the joint paper of Bigelow. and as happens so often to scientists. we have come to the conclusion that all the existing terminology has too heavy a bias to one side or another to serve the future development of the field as well as it should . What is Life "! Cambridge U. the group of scientists about Dr.. Roy. of the Medical School of the University of Illinois. Proc. . In choosing this term. We have decided to call the entire field of control and communication theory. control. whether in the machine or in living tissue. and falls under the group of ideas we have just been discussing1.. we shall find it convenient to use in refer ring to earlier epochs of the development of the field. Google-digitized / http://www. Rosenblueth. J. who had already been in touch with 1. by the name Cybernetics. the development of the subject went ahead on in several fronts. Rosenblueth and myself had already become aware of the essential unity of the set of problems centering about communication. Thus as far back as four years ago.org/access_use#pd-google ship are indeed one of the earliest and best developed forms of feed-back mechanisms. 1868. Although the term cybernetics does not date further back than the summer of 1947. March 5. which we form from the Greek xugepv/lTY^ or steers man. and devoted to problems of central inhibition in the nervous system. From 1942 or thereabouts. Schrodinger Erwin. Cambridge. (London).INTRODUCTION 19 the domain of communication theory. and that governor is derived from a Latin corruption of xugspwiT/is. C. 1945. and statistical mechanics. we have been forced to coin at least one artificial neo-Greek expression to fill the gap. and by the absence of any common termi nology. Soc. We also wish to refer to the fact that the steering engines of a Generated on 2011-09-28 14:13 GMT / Public Domain. After much con sideration.

20 CYBERNETICS Dr. From these are descended the mathematical notation and the symbolic logic of the present day. that the development of a mathematico-logical theory is subject to the same sort of restrictions as those that limit the performance of a computing machine. both to the nominalists like Hilbert and to the intuitionists like Weyl. as in the case of mathematical induction. where we prove a theorem depending on a parameter n for n = 0. through a refe rence to the concept of infinity which can itself be stated in finite terms. and also prove that the case n + 1 follows from the case /7. If I were to choose a patron saint for cybernetics out of the history of science. and owe much to his . Moreover. may make an appeal to the notion of infinity. but this Generated on 2011-09-28 14:14 GMT / Public Domain. the reasoning machine. and who was interested in the study of the organization of the cortex of the brain. just as the calculus of arithmetic lends itself to a mechanization progres sing through the abacus and the desk computing machine to the ultra-rapid computing machines of the present day. Rosenblueth and myself. it has become quite evident. it is even pos sible to interpret in this way the paradoxes of Cantor and of Russell. even though they may appear to be otherwise. These symbols.org/access_use#pd-google appeal is one which we can sum up in a finite number of stages. At this point there enters an element which occurs repeatedly in the history of cybernetics — the influence of mathematical logic. As we shall see later. so the calculus ratiocinator of Leibniz contains the germs of the machina ratiocinalrix. I should have to choose Leibniz. Indeed. thus establishing the theorem for all positive values of n. was interested in the construction of computing machines in the metal. Google-digitized / http://www. A mathematical proof which we can follow is one which can be written in a finite number of symbols. It is therefore not in the least surprising that the same intellectual impulse which has led to the development of mathematical logic has at the same time led to the ideal or actual mechanization of processes of thought. The philosophy of Leibniz centers about two closely related concepts — that of a universal symbolism and that of a calculus of reasoning. Leibniz himself. like his predecessor Pascal. in fact. Now. the rules of operation of our deductive mechanism must be finite in number. I am myself a former student of Russell.hathitrust. In short.

Google-digitized / http://www. I met Dr.hathitrust. the consideration of nets containing cycles.INTRODUCTION 21 influence. Journal of The London Mathematical Society. W. W..org/access_use#pd-google cerning the union of nerve fibres by synapses into systems with given over-all properties. Lettvin of the Boston City Hospital. and of synaptic and other delays1. and Pitts. which are very far from being closed energetically. Math. J. He had been a student of Car- nap at Chicago. Ro- 1. . Pitts had the good fortune to fall under McCulloch's influ ence. Mr. II. Shannon took for his doctor's thesis at the Massa chusetts Institute of Technology the application of the techniques of the classical Boolean algebra of classes to the study of switching systems in electrical engineering. Independently of Shannon. Turing. who was very much interested in matters con cerning nervous mechanisms. served the British government during the war as a worker in electronics. Me Culloch. Turing. He was a close friend of Mr. Another young migrant from the field of mathematical logic to cybernetics is Walter Pitts. 5 115-133 (1943).42. 2. and is now in charge of the program which the National Physi cal Laboratory at Teddington has undertaken for the develop ment of computing machines of the modern type. In the summer of 1943. although it may seem to some of us that they are too dominated by problems of energy and poten tial and the methods of classical physics to do the best possible work in the study of systems like the nervous system. Pitts to come out to Boston. S. who is perhaps first among those who have studied the logical possibi lities of the machine as an intellectual experiment. « A logical calculus of the ideas immanent in nervous activity ». Dr. they had used the technique of mathematical logic for the dis cussion of what were after all switching problems. and had also been in contact with Professor Rashevsky and his school of biophysicists. Bull..230-265(1936). although they are certainly suggested by the ideas of Turing : the use of the time as a parameter. and to make the acquaintance of Dr. and the two began to work quite early on problems con Generated on 2011-09-28 14:14 GMT / Public Domain.. He induced Mr. Biophys. 2.. and made me acquainted with his work. Pitts2. They added elements which were not prominent in Shannon's earlier work. Let it be remarked in passing that this group has contributed much to directing the attention of the mathematically minded to the possibilities of the biological sciences.

Google-digitized / http://www. Mr. and the University of Pennsylvania were already con structing machines. the construction of computing machines had proved to be more essential for the war effort than the first opinion of Dr. At that time Mr. Pitts came to the Massachusetts Institute of Technology in the autumn of 1943. At this time. Pitts was already thoroughly acquainted with mathematical logic and neurophysiology. which more than one of us had already contemplated as the most satisfactory basis of computing machine design. Bush might have indicated.org/access_use#pd-google certain combination of output from other selected elements will or will not act as an adequate stimulus for the discharge of the next element. Aberdeen Proving Ground. He was very much interested when I showed him examples of modern vacuum tubes and explained to him that these were ideal means for realizing in the metal the equivalents of his neuronic circuits and systems. We welcomed him into our group. The problem of interpreting the nature and varieties of memory in the animal has its parallel in the problem of constructing artificial memories for the machine. Harvard. must represent almost an ideal model of the problems arising in the nervous system. in order to work with me. but not yet christened. which had by that time been fairly born. and he had not had much experience of the possibilities of electronics.22 CYBERNETICS senblueth and myself. Shannon's work. but had not had the chance to make very many engineering contacts. and to strengthen his mathematical background for the study of the new science of cybernetics. and the Institute of Advanced Studies at Princeton and the Massachusetts Institute of Technology were soon to enter the same field. and was progressing at several centers along lines not too different from those which my earlier report had indicated. and must have its precise analogue in the computing machine. he was not acquainted with Dr. The all-or-none character of the discharge of the neurons is precisely analogous to the single choice made in determining a digit on the binary scale.hathitrust. From that time. depending as it does on conse cutive switching devices. In particular. In this program there was a gradual progress from the mechanical assembly to the electrical . it became clear to us that the ultra-rapid computing machine. The synapse is nothing but a mechanism for determining whether a Generated on 2011-09-28 14:15 GMT / Public Domain.

in par ticular to Dr. At this stage of the proceedings Dr. It was impossi ble to have Dr. similarly the computing-machine design ers presented their methods and objectives. Dr. and mathematicians were all represented.org/access_use#pd-google ever. and this meeting took place at Princeton in the late winter of1943-1944. while Dr. how Generated on 2011-09-28 14:15 GMT / Public Domain. Warren Weaver had published a document. There was a continual going and coming of those interested in these fields. from humanly directed operation to automatically directed operation . Bush. Dr. and Dr. Everywhere we met with a sympathetic hearing. from the scale of ten to the scale of two. Dr. Bige- low and myself on predictors and wave fillers. We had an opportunity to communicate our ideas to our colleagues. and that some attempt should be made to achieve a common vocabulary. Lorente de No of the Rockefeller Institute represented the physiologists. von Neumann and myself felt it desirable to hold a joint meeting of all those interested in what we now call cybernetics. The physiologists gave a joint presentation of cybernetic problems from their point of view. von Neumann of the Insti tute for Advanced Studies. from the mechanical relay to the electrical relay. Aiken of Harvard. Goldstine of the Eniac and Edvac machines at the University of Pennsylvania. physiolo gists. Rosenblueth among us. covering the work of Mr. as he had just accepted an invitation to act as Head of the laboratories of physiology of the Instituto Nacional de Cardiologia in Mexico. and in short each new machine more than the last was in conformity with the memorandum I had sent Dr. At the end of the meeting. and the vocabulary of the engineers soon became contaminated with the terms of the neurophysiologist and the psychologist. Google-digitized / http://www. McCulloch and Dr. but Dr. it had become clear to all that there was a sub stantial common basis of ideas between the workers in the different fields. A considerable period before this. Goldstine was one of a group of several computing- machine designers who participated in the meeting. Aiken was unable to be present . first secret and later restricted. Engineers. It was found . and myself were the mathematicians.INTRODUCTION 23 assembly. the war research group conducted by Dr. Pitts. that people in each group could already use notions which had been better developed by the others.hathitrust. Mr. von Neumann.

and Physics (M. Math. I had already developed part of my work in a course of lectures in the mathematics department of M.org/access_use#pd-google ready written of my manuscript to Professor Doob of the Univer sity of Illinois. but circumstances not completely under my control have preven ted its prompt publication. As I have said. At the same time. W. I myself have had a long mathematical expository paper under way for seve ral years to put the work I have done on permanent record. He is giving a course on the new methods lor the design of wave filters and similar apparatus in the M. I. Google-digitized / http://www.. Y. T. electrical engineering department in the fall of 1947. /. I received an invi- 1. Daniell. Lee. and have been used by the government for smoothing purposes. 2. Dr. but the principles proved to be sound and practical. has returned from China. . and Physics (M. after a joint meeting at the American Mathematical Society and the Institute of Mathemati cal Statistics held in New York in the spring of 1947. T.Wallmann. J. I have passed on what I have al Generated on 2011-09-28 14:15 GMT / Public Domain. Lee. Finally. my old student and collaborator 2. 1947.24 CYBERNETICS that the conditions of anti-aircraft fire did not justify the design of special apparatus for curvilinear prediction. in the summer of 1945. and in several fields of rela ted work. I. T. Phillips. and has plans to work the material of these lectures up into a book. Math. and others 1 written to fill the gap. now out of print. Rosenblueth returned to Mexico about the beginning of 1944. I.). T. 1932. Dr. and devo ted to the study of stochastic processes from a point of view closely allied to cybernetics. In particular. W. It also saw my government document. Thus in one way or ano ther.hathitrust.). the end of the war saw the ideas of prediction theory and of the statistical approach to communication engineering already familiar to a large part of the statisticians and communication engineers of the United States and Great Britain. and in many other problems of an applied mathematical interest. and a considerable number of expository papers by Levinson. to be developed in his notation and according to his ideas as a book for the Mathematical Surveys series of the American Mathematical Society. I. Levinson. the type of integral equation to which the calculus of variations problem reduces itself has been shown to emerge in wave-guide problems. the out-of-print government document is to be reprinted. In the spring of 1945. Since then. Y.

the tonic. on the one hand. Dr.. under its director Dr. Chavez for his unquestion ing hospitality. Rosenblueth. the tonic spasm. Dr. N. and the Instituto Nacional de Cardiologia.org/access_use#pd-google heart. « Conduction of Impulses in Cardine Muscle ». Walter 13.. Google-digitized / http://www. 16 : 205 : 265 . and phasic contractions in epi lepsy. Oliver G. This invitation was reinforced by the Comision Instigadora y Coordinadora de la Investigacion Cientifica. Rosenblueth to the investigation of the Generated on 2011-09-28 14:16 GMT / Public Domain. Rolh lines of work were developed in a paper 1. Our investigation took two directions : the study of phenomena of conductivity and latency in uniform conducting media of two or more dimensions. Me*. and fibril lation of the heart. clonic. Rosenblueth on a visit which unfortunately proved to be his last. Arch. Rosen- blueth and I decided to continue a line of work which we had already discussed with Dr. Cannon. Wiener. who was also with Dr. and although in both cases our earlier results have shown the need of a considerable amount of revision and of supplementation. We felt that heart muscle represented an irritable tissue as useful for the investigation of conduction mecha nisms as nerve tissue. We were also deeply grateful to Dr. of whom I have already spoken. and the statistical study of the conduc ting properties of random nets of conducting fibres.hathitrust. published by us. A. Selfridge of the Massachusetts Institute of Technology. Manuel Sandoval Vallarta. we were grateful to have an opportunity to contribute to its principal purpose. Cardiol. extended me its hospitality. under the leadership of Dr. and furthermore. Ignacio Chavez. Intl. and on the other hand. the work on flutter is being revised by Mr. Rosen- blueth invited me to share some scientific research with him. beat. I stayed some ten weeks in Mexico at that time. The first led us to the rudiments of a theory of heart flutter. This work had to do with the relation between. the latter to a certain possible understanding of fibrillation. while the statis tical technique used in the study of heart-muscle nets has been 1.INTRODUCTION 25 tation from the Mexican Mathematical Society to participate in a meeting to be held in Guadalajara that June. that the anastomoses and decussations of the heart muscle fibres presented us with a simpler phenomenon than the problem of the nervous synapse. and while it has never been the policy of the Instituto to restrict Dr.

and to hold them together for two successive days in all-day series of informal papers. would prove to be of this nature. not exceeding some"twenty in number. of workers in various related fields. and anthropolo gists. until they had had the opportunity to thresh out their differences and to make progress in thinking along the same lines. Much of the psychology of the past has proved to be really nothing more than the physiology of the organs of special sense . Dr. and have co-opted into the group a number of leading psychologists.26 CYBERNETICS extended to the treatment of neuronal nets by Mr. or of the perceptual formation of universals.org/access_use#pd-google assembled in Princeton in 1944. now a Fellow of the John Simon Guggenheim Foundation. The -experimental work is being carried on by Dr. and he who studies the mind cannot tor- get the nervous system. He who studies the nervous system cannot forget the mind. sociologists. Dr. Rosenblueth with the aid of Dr. Waller Pitts. worked out most efficiently by Dr. we have anticipated that the problem of the perception of Ges- talt. These meetings have been conducted in the traditional Macy way. who organized them on behalf of the Founda tion.hathitrust. Frank Fremont-Smith. Rosenblueth and I presented some of our results. Google-digitized / http://www. The need of including psychologists had indeed been obvious from the beginning. From the beginning. and meals together. F. At the Guadalajara meeting of the Mexican Mathematical So ciety.and the whole weight of the body of ideas which cybernetics is introducing into psychology concerns the phy siology and anatomy of the highly specialized cortical areas con necting with these organs of special sense. What is the mechanism by which we recog- . but Drs. We had already come to the conclusion that our earlier plans of col laboration had shown themselves to be practicable. The idea has been to get together a group of modest size. McCulloch and Fre mont-Smith have rightly seen the psychological and sociological implications of the subject. The nucleus of our meetings has been the group that had Generated on 2011-09-28 14:16 GMT / Public Domain. discussions. In the spring of 1946. We were for tunate enough to have a chance to present our results to a larger audience. Garcia Ramos of the Instituto Nacional de Car- diologfa and the Mexican Army Medical School. McCulloch had made arrange ments with the Josiah Macy foundation for thefirst of a series of meetings to be held in New York and to be devoted to the pro blems of feed back.

Norlhrup was inte rested in assaying the philosophical significance of our work. although distinct from. For the similar problems of human organization. His very important joint book on games with Dr. We also enlarged the group to contain more engineers and ma thematicians such as Bigelow and Savage . This does not purport to be a complete list of our group. represents a most interes ting study of social organization from the point of view of methods closely related to. M.INTRODUCTION 27 nize a square as a square. Morgenstern of the Institute of Advanced Studies was our adviser in the significant field of social organization belonging to economic theory. C. it is manifest that the im portance of information and communication as mechanisms of organization proceeds beyond the individual into the community. Lewiri and others represented the Generated on 2011-09-28 14:29 GMT / Public Domain. the late Dr. by the way. and Dr. As to sociology and anthropology. Ericsson of New York. irrespective of its position. and that before the next full meeting. Dr.hathitrust.moreneuroanatomists and neurophysiologists such as von Bonin and Lloyd. held in the spring of 1946. the subject matter of cybernetics. Google-digitized / http://www. S. Our first meeting. and its orientation ? To assist us in such matters. F. von Neumann. On the one hand. and to inform them of whatever use might be made of our concepts for their assistance. and so on. and we were fortunate enough to have the aid of Dr. we should have a small meeting for the benefit of the less mathematically trained to explain to . and Dr. Bateson and Margaret Mead . we sought help from the anthropologists Drs. Schierda in this matter. its size.org/access_use#pd-google newer work on the theory of opinion sampling and the prac tice of opinion making. Kurt Lewin of the Massachusetts Institute of Technology. was largely devoted to didactic papers by those of us who had been present at the Princeton meeting and to a general assessment of the im portance of the field by all present. while Dr. It was the sense of the meeting that the ideas behind cybernetics were sufficiently important and interesting to those present to warra'nt a conti nuation of our meetings at intervals of six months. we had among us such psychologists as Professor Kliiver of the University of Chicago. it is completely impossible to understand social communities such as those of ants without a thorough investiga tion of their means of communication.

for example.hathitrust. the frequency of oscillation. This is not the place to discuss the full significance of our results. This time we decided to take a nervous problem directly from the topic of feed-back. This circuit is not even approximately a circuit of linear operators if we take as our base of linearity the number of impulses transmitted by the efferent nerve per second. In many cases strychnine was used to increase the reflex responses.org/access_use#pd-google or electrical system exhibiting the same pattern of hunting. the methods of McColl's book on servo- mechanisms. This corresponds to the fact that the form of the envelope of . paying attention to the physiological condition of the cat. the base-level of the oscillation. which we are now repeating and preparing to write up for publication. and that it is much more nearly determined by the constants of the closed arc efferent-nerve- muscle-kinaesthetic-end-body-afferent-nerve-central-synapse- efferent-nervethan by anything else. We worked chiefly with cats. Google-digitized / http://www. first decerebrated under ether anesthesia and later made spinal by a thoracic transaction of the cord. and the quadriceps extensor femoris as the muscle to study. and to see what we could do with it experimentally. the following statements are either established or very probable : that the frequency ofclonic oscillation is much less sensitive to changes of the loading con ditions than we had expected . We also used an oscillograph to record the simultaneous electrical changes in the muscle itself. In the summer of 1946. and its amplitude. We observed this pattern of contraction. We cut the attachment of the muscle. fixed it to a lever under known tension. which is called clonus in the language of the physiologist. and recorded its contrac tions isometrically or isotonically. We chcse the cat as our experimental animal. The muscle was loaded to the point where a tap would set it into a periodic pattern of contraction. These we tried to analyze as we should analyze a mechanical Generated on 2011-09-28 14:30 GMT / Public Domain. However. I returned to Mexico with the support of the Rockefeller Foundation and thehospitality of the Instituto Nacional de Cardiologia to continue the collaboration between Dr. Rosenblueth and myself.28 CYBERNETICS them in as simple language as possible the nature of the mathe matical concepts involved. the load on the muscle. We employed. but seems to become much more nearly so if we replace the number of impulses by its logarithm.

We obtained theoretical oscillations of about 13. this agreement is excellent. the form of the curve of stimulation must be sinusoidal in all except a set of cases of zero probability. Unpublished articles on Clonus from the Instituto Nacional <le Cardiologia. and an even slower change in amplitude. and with data obtained from the conduction of single pulses through Generated on 2011-09-28 14:30 GMT / Public Domain.hathitrust.9 per second. even though this law is only a first approximation. using the technique already developed by the servo- engineers for the determination of the frequencies of hunting oscillations in feed-back systems which have broken down. The frequency of clonus is not the only important phenomenon which we may observe : there is also a relatively slow change in basal tension.INTRODUCTION 29 stimulation of the efferent nerve is not nearly sinusoidal. The most striking point is that on this logarithmic basis. in cases where the observed oscillations varied between frequen cies of 7 and 30. Furthermore. suffi- 1. and a partial inhibition means a multiplication by a small quantity. Google-digitized / http://www. and the outgoing fibre is only stimulated if the number of incoming impulses in a small summation-time exceeds a certain threshold. This approximate logarithmicity of the synapse mechanism is certainly allied to the approximate logarithmicity of the Weber-Fechnerlaw of sensation intensity. Under the circumstances.org/access_use#pd-google the various elements of the neuromuscular arc. However. but that the logarithm of this curve is much more nearly sinusoidal. but generally remained within a range varying somewhere between 12 and 17. Mexico. . Again. while in a linear oscillating system with constant energy level. If this threshold is low enough in comparison with the full number of incoming synapses. the synapse is a coin cidence-recorder. a complete inhibition means a multiplication by zero. the synaptic mechanism serves to multiply probabilities. These phenomena are certainly by no means linear. and that it can be even an approximately linear link is only possible in a logarithmic system. For example. It is these notions of inhibition and facilitation which have been used 1 in the dis cussion of the reflex arc. we were able to obtain very fair approximations to the actual periods of clonic vibration. the notions of facilitation and inhibition are much more nearly multiplicative than additive in nature.

org/access_use#pd-google a larger public. These were the main results presented by Dr. While this work has not yet been com pleted it is clear that it is both possible and promising. We felt that the time was now ripe for us to go jointly to them — that is. and to Dr. and fully convinced of the general practicability of work in this direction. in charge of the department of medical sciences — to establish the basis of a long-time scientific collabo ration. Robert Morison. Rosenblueth and myself at the Macy meeting held in the autumn of 1946. Google-digitized / http://www. to Dr. in order to carry on our program at a more leisurely and . and as though over each part of the oscillation the system behaved as it would if its parameters were those belonging to it at the time. the amplification of impulses in this arc is variable in one and perhaps in more points. Rosenblueth a grant for the equipment of a new laboratory building at the Instituto Nacional de Cardiologia. we felt nevertheless that the time of our collaboration had been too brief. and by many other causes. This variable amplification may be affected by the general level of central activity. This confirmation — which natu rally might amount to a refutation — we are now seeking in the summer and autumn of 1947. by decerebration.30 CYBERNETICS ciently slow changes in the constants of a linear oscillating sys tem may be treated to a flrst approximation as though they were infinitely slow. This is the method known in other branches of physics as that of secular perturba tions. in charge of the department of physical sciences. multi-neuron processes which run much higher in the central nervous system than the spinal chain primarily respon sible for the timing of clonus. by the use of strychnine or of anesthetics. The Rockefeller Foundation had already given Dr. Warren Weaver.hathitrust. and that our work had been done under too much pressure to make it desirable to publish without further experimental confirmation. and that some part of this amplification may be affected by slow. It may be used to study the problems of base-level and amplitude of clonus. While we were pleased with our results. and in a meeting of the New York Academy of Sciences held at the same time for the purpose of diffusing the notions of cybernetics over Generated on 2011-09-28 14:30 GMT / Public Domain. There is a strong suggestion that though the timing of the main arc in clonus proves it to be a two-neuron arc.

In this we were enthusiastically backed by our respective institutions. This is a definite analogue of the problem of the percep tion of form. Dr. the Instituto Nacional de Cardiologfa. for people wishing to go into this new field.org/access_use#pd-google work of considerable cybernetic importance. Dr. McCulloch had been given the problem of designing an apparatus to enable the blind to read the printed page by ear. to the very difficult problem of devising. The production of variable tones by type through the agency of a photocell is an old story. McCulloch's device involved a selective reading of the type-imprint for a set of different magnifica tions. Dr. The plan finally adopted was for five years. of Gesialt. Dean of Sciences. and medical techniques. psychological.hathitrust. and can be effected by any number of methods . Such a selective reading can be perfomed automatically as a scanning process. McCulloch and Mr. and engineering background. Rosenblueth would spend six months of the intervening years at the Institute. was the chief representative of the Massachusetts Institute of Technology during these negotiations. Google-digitized / http://www. whatever the size. it became clear that the laboratory center of the joint activity should be at the Instituto. during which I should spend six months of every other year at the Instituto. The time at the Instituto is to be devoted to the obtaining and elucidation of experimental data pertaining to cybernetics. and above all. In the spring of 1947. was a device which I had already suggested at one . George Harrison. the difficult point is to make the pattern of the sound substantially the same when the pattern of the letters is given. physical. while the intermediate years are to be devoted to more theoretical research. During the negotiations.INTRODUCTION 31 healthy pace. and the proper acquaintance with biological. Dr. both in order to avoid the duplication of laboratory equipment and to further the very real interest the Rockefeller Foundation has shown in the establishment of scientific centers in Latin America. while Dr. which allows us to recognize a square as a square through a large number of changes of size and of orientation. a scheme of training which will secure for them both the necessary mathematical. to allow a compari son between a figure and a given standard figure of fixed but different size. This scanning. Pitts did a piece of Generated on 2011-09-28 14:31 GMT / Public Domain. while Dr. Ignacio Chdvez spoke for his institution.

« Is this a diagram of the fourth layer of the visual cortex of the brain ? » Acting on this suggestion. with the assistance of Mr. and this is the approximate period of the so-called « alpha rhythm » of the brain. especially at Manchester and at the National Physical Laboratories at Teddinglon. Finally. and had a very good chance to discuss the work that Professor F.hathitrust. chiefly as a guest of my old friend Professor J. I had an excellent chance to meet most of those doing work on ultra-rapid com puting machines. Finally. Bartlett and his staff were doing on the human element in control processes involving such an element. In the spring of 1947. A diagram of the apparatus by which the selective reading was done came to the attention of Dr. Pitts. This was presented in the spring of 1947 both at the Macy meeting and at a meeting of the New York Academy of Sciences. and above all to talk over the fundamental ideas of cybernetics with Mr. which corresponds to what we call the « time of sweep » in ordinary television.org/access_use#pd-google harmonic analysis. I received an invitation to participate in a mathematical conference in Nancy on problems arising from Generated on 2011-09-28 14:31 GMT / Public Domain. B. S. H. von Bonin. Google-digitized / http://www. and Professors Haldane. and on my voyage there and back spent a total of three weeks in England. this scanning process involves a certain periodic time. These yield a time of the order of a tenth of a second fora complete performance of the cycle of operations. and to be important in the process of form perception. Levy and Bernal certainly . I found the inte rest in cybernetics about as great and well informed in England as in the United States. though of course limited by the smaller funds available. who immediately asked. C.32 CYBERNETICS of the Macy meetings. and in this theory the operation of scanning over a set of transformations plays an important part. and the engineering work excellent. There are various anatomic clues to this time in the length of the chain of consecutive synapses necessary to run around one cycle of performance. I found much interest and understanding of its possibility in many quarters. on quite other evidence. has already been conjectured to be of visual origin. Turing atTedding- ton. produced a theory tying up the anatomy and the physiology of the visual cortex. Dr. the alpha rhythm. I also visited the Psychological Laboratory at Cambridge. McCulloch. Haldane. I accepted.

in which circular processes of a feed back nature play an important part. that it is bound together by a system of communication .hathitrust. Loeve. I did not find. My colleague. Professor G. one of the directions of work which the realm of ideas of the Macy meetings has suggested concerns the importance of the notion and the technique of communica tion in the social system. introduced me to M. On this basis. One event during my French visit is particularly worth while noting here. Gre gory Bateson and Margaret Mead have urged me. that as much progress had been made in unifying the subject and in pulling the various threads of research together as we had made at home in the States. In France. Indeed. and in the more specific field of eco nomics . both in the general fields of anthro pology and of sociology. and that it has a dyna mics. Blanc-Lapierre and M. have been done in Mexico. 1 am particularly glad to receive his invitation as M Freymann is a Mexican. which we have already mentioned. I found also a considerable interest in the subject on the part of mathematicians. the Hungarian biochemist. of the firm of Her mann et Cie. I had discussed that subject in Boston before my depar ture with Professor Szent-Gyorgi. enters into this range of ideas. the meeting at Nancy on harmonic analysis con tained a number of papers uniting statistical ideas and ideas from communication engineering in a manner wholelyin confor mity with the point of view of cybernetics. I. and he requested of me the present book. to devote a large part . physiologists. however. and the writing of the present book. in view of the intensely pressing nature of the sociological and economic pro blems of the present age of confusion. of von Neumann and Morgenstern on the theory of games. This is true.INTRODUCTION 33 regarded it as one of the most urgent problems on the agenda of science and scientific philosophy. as well as a good Generated on 2011-09-28 14:31 GMT / Public Domain. Freymann. de Santillana of M. It is certainly true that the social system is an organization like the individual. Google-digitized / http://www. Here 1 must mention especially the names of M. T . particu larly with regard to its thermodynamic aspects in so far as they touch the more general problem of the nature of life itself.org/access_use#pd-google deal of the research leading up to it. Drs. and physical chemists. As I have already hinted. and had found his ideas concordant with my own. and the very important work.

Neither is there any important point in run ning statistics of the incidents of venereal disease in a single table which covers both the period before and that after the introduc tion of salvarsan.34 CYBERNETICS of my energies to the discussion of this side of cybernetics. Moreover. and economic quantities. but the runs of statistics on which they are based are excessively short. the advantage of long runs of statistics under widely varying conditions is specious and spu rious. Thus the human sciences are very poor testing-grounds for a new mathematical technique : as poor as the statistical mechanics of a gas would be to a being of the order of size of a molecule. unless for the specific purpose of studying the effectiveness of this drug. For a good statistic of society. There is no great use in lumping under one head the economics of steel industry before and after the introduction of the Bessemer process. unless the lens is made of a material so homogeneous that the delay of light in different parts of the lens conforms to the proper designed amount by less than a small part of a wavelength. anthro pological. 1 can share neither their feeling that this field has the first claim on my attention. Much as I sympathize with their sense of the urgency of the situation. The effective aperture of a lens is not appreciably increased by Generated on 2011-09-28 14:31 GMT / Public Domain. nor in com paring the statistics of rubber production before and after the burgeoning of the automobile industry and the cultivation of Hevea in Malaya.hathitrust. just as for a good resolution of light we need a lens with a large aperture. to whom the fluctuations which we ignore from a larger standpoint would be precisely the matters of greatest interest. nor their hopefulness that sufficient progress can be registered in this direction to have an appreciable therapeutic effect in the present diseases of society.is so great that it is no field for a newcomer who has not yet had the bulk of experience which goes to make up the expert. Similarly.org/access_use#pd-google augmenting its nominal aperture. the element of the judgment of the expert in determining the estimates to be made of sociological. in the absence of reasonably safe routine numerical techniques. the main quantities affecting society are not only statistical. To begin with. I may remark parenthetically . and much as I hope that they and other competent workers will take up problems of this sort which I shall dis cuss in a later chapter of this book. Google-digitized / http://www. we need long runs under essentially constant conditions.

The first Iwo losses are what the artificial-limbmaker now tries to replace. unless it is applied by a statistician by whom the main elements of the dynamics of the situation are either explicitly known or implicitly felt. The third has so far been beyond his scope. Google-digitized / http://www.org/access_use#pd-google support of the missing segment or its value as mechanical exten sion of the stump. which . The loss of a segment of limb implies not only the loss of the purely passive Generated on 2011-09-28 14:32 GMT / Public Domain. Here the instrument suggested by McCulloch lakes over quite explicitly some of the functions not only of the eye but of the visual cortex. and this interferes with his sureness of step on an irregular terrain. In the case of a simple peg-leg this is not important : the rod that replaces the missing limb has no degrees of freedom of its own. does not inspire me with any confidence. but in which this hope must wait on further developments. There is a manifest possibility of doing something similar in the case of artificial limbs. This is not the case with the articulated limb with a mo bile knee and ankle.hathitrust. thrown ahead by the patient with the aid of his remaining musculature.INTRODUCTION 35 that the modern apparatus of the theory of small samples. and the loss of the contractile power of its/ muscles. once it goes heyond the determination of its own specially defined parameters and becomes a method for positive statistical infe rence in new cases. There does not seem to be any insuperable difficulty in equipping the artificial joints and the sole of Ihe artificial foot with strain or pressure gauges. He has no adequate report of their position and motion. but implies as well the loss of all cutaneous and kinars- thetic sensations originating in it. I have just spoken of a field in which my expectations of cybernetics are definitely tempered by an understanding of the limitations of the data which we may hope to obtain. There are two other fields where 1 ultimately hope to accomplish something practical with the aid of cybernetic ideas. One of these is the matter of prostheses for lost or paralyzed limbs. in the construc tion of an instrument !o enable the blind to read print by hea ring. As we have seen in the discussion of Gestalt. the ideas of com munication engineering have already been applied by McCulloch to the problem of I he replacement of lost senses. and the kinaesthetic mechanism of the slump is fully adequate to report its own position and velo city.

they should receive one in the immediate future. The automatic factory. on intact areas of skin. Google-digitized / http://www. where the figure of the manikin familiar to all readers of books of neurology shows that the sensory loss in an amputation of the thumb alone is considerably greater than the sensory loss even in a hip-joint amputation. for example. but leaves the ataxia. With the aid of strain-gauges or similar agencies to read the performance of these motor organs and to reporl. but might very well be. it had occurred to me that we were here in the presence of another social potentiality of unheard-of im portance for good and for evil. the as sembly-line without human agents. It has long been clear to me that the modern ultra-rapid computing machine was in principle an ideal central nervous system toan apparatus for automatic control. are only so far ahead of us as is limited by our willingness to put such a degree of effort into their engineering as was spent. What we have said about the leg should apply with even more force to the arm. but up to now I have not been able to accom plish much. which should enable him to step out with a much surer gait. to the central control system as an arti ficial kinaesthetic sense. With the use of proper receptors. respectively. in the deve- . say through vibrators. and the performance of motors or solenoids. Long before Nagasaki and the public awareness of the atomic bomb. to « feed back ». Let me now come to another point whichlb elieve to merit at tention. In case they have not yet received a thorough practical consideration.hathitrust. much of this ataxia should disappear as well.org/access_use#pd-google output need not be in the form of numbers or diagrams. and that its input and Generated on 2011-09-28 14:32 GMT / Public Domain. The present artificial limb removes some of the paralysis caused by the amputation. the readings of artificial sense-organs such as photo-electric cells or thermometers. nor whether they have been tried out and found technically impracticable. we are already in a position to construct artificial machines of almost any degree of elaborateness of performance.36 CYBERNETICS are to register electrically or otherwise. I do not know whether the same ideas have already emanated from other sources. and the patient should be able to learn reflexes such as those we all use in driving a car. I have made an attempt to report these considerations to the proper authorities.

Google-digitized / http://www.hathitrust.org/access_use#pd-google .Generated on 2011-09-28 14:32 GMT / Public Domain.

as I have said. and for those who have this training.38 CYBERNETICS being of mediocre attainments or less has nothing to sell that it is worth anyone s money to buy. are the unions receptive to such people. sociological. if the best comes to the best. of course.. that the labor unions and the labor movement are in the hands ot a highly limited personnel. not very comfortable. The answer. We have contributed to the initia tion of a new science which. Further than these individuals neither I nor any of them was able to go. and from them I received a very intelligent and sympathetic hearing. and otherwise — who knows ? I thus felt it my duty to pass on my information and understanding of the position to those who have an active interest in the conditions and the future of labor — that is to the labor unions. thoroughly well trained in the specialized problems of shop-stewardship and disputes concer ning wages and conditions of work. To arrive at this society we need a good deal of planning and a good deal of struggle -- which. The best we can do is to see that a large public understands . technical.org/access_use#pd-google the exacting life of an administrator without any opportunity for a broader training . (). They belong to the age. and this is the world of Belsen and Hiroshima. a union career is not generally inviting . to say the least.hathitrust. may be on the plane of ideas. nor. I. as it had been my previous obser vation and information. Google-digitized / http://www. both in the United Stales and in England. embraces tech nical developments with great possibilities for good and for evil. It was their opinion. We do not even have the choice of suppressing these new technical deve lopments. Those of us who have contributed to the new science of cybernetics thus stand in a moral position which is. and totally unprepared to enter into the larger political. I did manage to make contact with one or two persons high up in the C. We can only hand it over into the world that exists about us. quite naturally. is to have a society based on human values other than buying or selling. and the most any of us can do by suppression is to put the development of the subject into the hands of the most irresponsible and most venal of our engineers. The reasons for this are easy enough to see : the labor union official generally comes from the exacting life of a workman into Generated on 2011-09-28 14:32 GMT / Public Domain. and economic questions which concern the very existence of labor.

Ciu<lad de Mexico. Mr.INTRODUCTION 39 the trend and the bearing of the present work. Generated on 2011-09-28 14:33 GMT / Public Domain. As we have seen. in the hands of the most unscrupulous). and to confine our personal efforts to those fields. 1 write in 1947. 194. Frederic Webster for aid in correcting the manuscript and preparing the material for publication. there are those who hope that the good of a better understanding of man and society which is offered by this new field of work may anticipate and outweigh the incidental contribution we are making to the concentration of power (which is always concentrated. Instituto Nacional de Cardiologia. Google-digitized / http://www. Mr.7.hathitrust. and I am compelled to say that it is a very slight hope. Georges Dube and Mr. by its very conditions of existence. Oliver Selfridge. Walter Pitts. The author wishes to express his gratitude to Mr. such as physiology and psychology.org/access_use#pd-google November. . most remote from war and exploitation.

that not one of the whole great number be lacking ». It goes : « Weisst wie viele Sterne stehen In dem blauen Hiinmelszelt ? Weisst. The more familiar astronomical phenomena can be predicted for many centuries. This little song is an interesting theme for the philosopher and the historian of science. and if a human Durchmusteriing of the stars — as we call these catalogues — . Astronomy is the oldest of the sciences. In the first place. In English this says : « Knowest thou how many stars stand in the blue tent of heaven ? Knowest thou how many clouds pass far over the whole world ? The Lord God hath counted them. To go back to the poem. while a precise prediction of x tomorrow's weather is generally not easy and in many places Generated on 2011-09-28 14:33 GMT / Public Domain. a star is a definite object. but which in almost every other respect offer an extreme contrast.CHAPTER I NEWTONIAN AND BERGSONIAN TIME There is a little hymn or song familiar to every German child. the answer to the first question is that. in that it puts side by side two sciences which have the one similarity of dealing with the heav ens above us. we do know how many stars there are.hathitrust. Google-digitized / http://www. while meteorology is among the youngest to begin to deserve the name. apart from minor uncertainties concerning some of the double and variable stars.org/access_use#pd-google very crude indeed. within limits. emi nently suitable for counting and cataloguing . wie viele Wolken gehen Weit hinuber alle Welt ? Gott der Herr hat sie geziihlet Dass ihm auch nicht eines fehlet Von der ganzen grossen Zahl ».

The pattern for all events in the solar system was the revolution of a wheel or a series of wheels. Cirro- cumulus ». and in any . Sky 38 % overcast. and Newton. defined as an object with a quasi-permanent identity . It is the astronomy of the solar system which is that chiefly associated with the names of Copernicus. Google-digitized / http://www. was originally concerned rather with the solar system than with the world of the fixed stars. Kepler. January 17.1950. or he might patiently explain that in all the lan guage of meteorology. whether in the form of the Ptolemaic theory of epicycles or the Copernican theory of orbits. It is indeed an ideally simple science. as pursued for example by Ghandrashekhta. he might laugh in your face. and would at most represent an extremely transitory state. There is of course a branch of astronomy which deals with what may be called cosmic meteorology : the study of galaxies and nebulae and star-clusters and their statistics. he neither possesses the facilities to count them. A topologically inclined me teorologist might perhaps define a cloud as a connected region of space in which the density of the part of the water content in the solid or liquid state exceeds a certain amount. apart from its purely classificatoiy. Galileo. there is no such thing as a cloud. even as far back as the Babylonians. It was realized that time itself could better be measured by the motion of the stars in their courses than in any other way.hathitrust. Even before the exis tence of any adequate dynamical theory. there is nothing too repugnant to us in the idea of a divine Durch- musterung going much further. hut this definition would not be of the slightest value to anyone. and which was the wet-nurse of modern physics. What really concerns the meteorologist is some such statistical state ment as. nor is he in fact interested in counting them. On the other hand. and is something Generated on 2011-09-28 14:33 GMT / Public Domain. and that if there were. but this is a very young branch of astronomy. ((Boston. it was realized that eclipses occurred in regular pre dictable cycles.org/access_use#pd-google outside the tradition of classical astronomy.NEWTONIAN AND BERGSONIAN TIME 41 stops short for stars less intense than a certain magnitude. if you were to ask the meteorologist to give you a similar Durchmuslerung of the clouds. Durchmusterung aspects. younger than meteorology itself. This tradition. extending backwards and forwards over time.

42 CYBERNETICS such theory. or where they are not. and the very high order coup ling effects are completely negligible. The planets move under conditions more favorable to the isolation of a certain limited set of forces than those of any physical experiment we can set up in the laboratory. and so on indefinitely. it would still be a possible picture of pla nets conforming to the Newtonian mechanics. when all this was reduced by Newton to a formal set of postu lates and a closed mechanics. and coupled wi»h one another in a sufficiently loose way that the second order coupling effects do not change the general aspect of the picture we observe. The music of the spheres is a palindrome. the planets are either very nearly rigid bodies . the fundamental laws of this mechanics were unaltered by the transformation of the time- variable t into its negative. lightning preceding instead of following the changes of cloud which usually precede it. if we were to take a motion-picture photograph of the turbulence of the clouds in a thunderhead. the future after a fashion repeats Ihe past. speed ed up to show a percepiible picture of activity. the meteorological system is one involving a vast number of approximately equal particles. On the other hand. greatly diverse in size. their internal forces are at any rate of a relatively slight . Compared with the elastic and plastic deformations they suffer. There is no differ ence save of initial positions and directions between the motion of an orrery turned forwards and one run in reverse. Compared with the distances between them. and reverse it. turbulence growing coarser in tex ture. We should see down-dralts where we expect up-drafts. Google-digitized / http://www. What is the difference between the astronomical smd the meteorological situation which brings about all these differences. while the astronomical system of Ihe solar universe contains only a relatively small number of particles. Thus if we were to take a motion-picture of the planets.org/access_use#pd-google meteorological time ? In the first place. Finally. and even the sun. and the book of astro nomy reads the same backwards as forwards. the planets.hathitrust. are very nearly points. some of them very closely coupled to one another. and in particular the difference between the apparent reversibi lity of astronomical time and the apparent irreversibility of Generated on 2011-09-28 14:33 GMT / Public Domain. it would look altogether wrong. and were to run the film backward.

are all terms referring not to one single physical situation. and if this record were actually made. If I set up a physical experiment. « temperature ». together with certain a priori assumptions. is easy and pre cise in principle. On the other hand. over the set of possible atmospheres. I bring the system I am considering from the past into the present in such a way that I fix certain quantities.hathitrust. because the questions to which they are answers are asymmetrical. questions of probability and prediction lead to answers asymmetrical as between past and future. they would not give a bil lionth part of the data necessary to characterize the actual state of Iheatmosphere from a Newtonian point of view. all that we can predict at any future time is a probability distribu tion of the constants of the system. and at most. and the computation of their future and past positions. capable of giving as a probability distribution. and in their mutual attraction. The positions. The terms « cloud ».. Google-digitized / http://www. The space in which they move is almost perfectly free from impeding matter . and have a reasonable right to assume that certain other quantities have known statistical distributions. and even this predictability fades out with the increase of time. we should have nothing but an impenetrable mass of figures wliich would need a radical reinterpretation before it could be of any service to us. the num ber of particles concerned is so enormous that an accurate record of their initial positions and velocities is utterly impos sible . their masses may be very nearly considered to lie at I heir centres and to be constant. The departure of the law of gravity from the inverse square law is most minute. in which time is perfectly reversible. and masses of the bodies of the solar system are extremely well known at any time.NEWTONIAN AND BERGSONIAN TIME 43 significance where the relative motion of their centres is con cerned. a measure. 1 then . and their future positions and velocities computed. in meteorology. Now. even in a Newtonian system. « turbulence ». Using the New tonian laws. while not easy in detail. velocities. or any other system of causal laws whatever. They would Generated on 2011-09-28 14:33 GMT / Public Domain. etc.org/access_use#pd-google only give certain constants consistent with an infinity of diffe rent atmospheres. If all the readings of all the meteorological stations on earth were simultaneously laken. but to a distribution of possible situations of which only one actual case is realized.

and not too obvious. In order to do so. while if there are any stars whose evolution is in the . Why then does the unidirectional thermodynamics which is based on experimental terrestrial observations stand us in such 'good stead in astrophysics ? The answer is interesting. However. This is not a process which I can reverse. and in which there seems to be no unidirectionalness in the nature of our experiment. in which we are observing remote heavenly bodies in a single observation. for a system starting from an unknown position to end up in any tightly defined statistical range is so rare an occurrence that we may regard it as a miracle. Google-digitized / http://www. In the perception of incoming light. and we cannot base our experimental technique on aw aiting and counting miracles.org/access_use#pd-google of light. and all our answers to these questions are equally conditioned by it. and we wrap our plates in black paper to prevent halation. and perceived by us. All our questions are conditioned by this asymmetry. We condition these for the reception of images by putting them in a state of insulation for some time past : we dark-condition the eye to avoid after-images. we are directed in time.44 CYBERNETICS observe the statistical distribution of results after a given time. We can perceive incoming light. we end up with the eye or a photographic plate. or at least the perception of outgoing light is not achieved by an experiment as simple and direct as that of incoming light. and our relation to the future is different from our relation to the past. of rays or particles emerging from the observed object. would end up within certain statistical limits. we can see those stars radiating to us and to the wbole world . It is clear that only such an eye and only such plates are any use to us: if we were given to pre-images we might as well be blind. This being the case. but can not perceive outgoing light . it would be necessary to pick out a fair distribution of systems which.hathitrust. without intervention on our part. In short. photography would be a very difficult art indeed. Our observations of the stars are through the agency Generated on 2011-09-28 14:34 GMT / Public Domain. and if we had to put our plates in black paper after we use them and develop them before using. A very interesting astronomical question concerning the direction of time comes up in connection with Ihe time of astrophysics. and find out what the antecedent conditions were a given time ago.

To such a being all communication with us would be impossible. Thus the part of the universe which we see must have its past-future relations. they will attract radiation from the whole heavens. Even astronomy. is nearly surrounded by oceans. but this is not quite Hie case. These antecedents would already be in our experience. and known as the theory of tidal evolution. and the water on the other side is less strongly attracted. and even this attraction from us will not be perceptible in any way. Within any world w th which we can communicate. Any signal he might send would reach us with a logical stream of consequents from his point of view. as far as the emission of radiation is concerned. but explainable by natural laws — by which Generated on 2011-09-28 14:34 GMT / Public Domain.org/access_use#pd-google that square would cease to exist. Our counterpart would have exactly similar ideas concerning us. The very fact that we see a star means that its thermodynamics is like our own.NEWTONIAN AND BERGSONIAN TIME 45 reverse direction. as we have seen. If he drew us a square. contains a cosmic meteoro logy. and would have served to us as the natural explanation of his signal. but most are rather nearer to meteorology than to astronomy. Its meaning would seem to be as fortuitous as the faces we read into mountains and cliffs. It contains as well that extremely interesting field studied by Sir George Darwin. The earth. The drawing of the square would appear to us as a catastrophe — sudden indeed. for example. The water nearer the moon than the center of the earth is more strongly attracted to the moon than the solid part of the earth. Indeed. Google-digitized / http://www. we should see the remains of his figure as its precursors. in view of the fact that we already know our own past but not our future. We have said that we can treat the relative movements of the sun and the planets as the movements of rigid bodies. it is a very interesting intellectual experiment to make the fantasy of an intelligent being whose time should run the other way to our own. without presupposing an intelligent being to have sent it. and it would seem to be the curious crystallization — always perfectly explainable — of these remains. antecedents from ours. To return to the contrast between Newtonian astronomy and meteorology : most sciences lie in an intermediate position.hathitrust. the direction of time is uniform. This relatively slight effect pulls the water into two . concordant with our own.

Generated on 2011-09-28 14:34 GMT / Public Domain. However. Thus even gravitational astronomy involves frictional proces ses that run down. the building up of tissues.hathitrust. Indeed. dissipative forces of a character much like the forces met in meteorology. In a perfectly liquid sphere. though in anything like historic times. oceanography may be called the meteorology of the hydrosphere rather than of the atmos phere. and the moon always presents nearly the same face to the earth. nor is anabolism. These friclional forces drag the moon back in its course about the earth. their breaking down. In ages past it has seriously moditied the face of the solar system. when the moon contained some liquid or gas or plastic material which could give under the earth's attraction. and need a statistical treatment. It consequently lags behind the position of the moon. They lend to bring the lengths of the month and of the day ever closer to one another. and the forces producing this are largely turbulent. They would consequently have a pull on the moon which would not greatly influence the angular position of the moon in the heavens. one under the moon and one opposite lo the moon. and accelerate the rotation of the earth forward. but may be observed to some degree throughout all gravitating systems. these hills could follow the moon around the earth with no great dispersal of energy. the tidal wave they produce on the earth gets tangled up and delayed on coasts and in shallow seas such as the Behring Sea and the Irish Sea. the exact reverse of catabolism. this modification is slight compared with the « rigid body » motion of the planets of the solar system.org/access_use#pd-google and in so giving could dissipate large amounts of energy.46 CYBERNETICS hills. Birth is not the exact reverse of death. and conse quently would remain almost precisely under the moon and opposite to the moon. Indeed. nor does the union of the germ-cells to form the fertilized ovum. There is not a single science which conforms precisely to the strict Newtonian pattern. The individual is an arrow pointed through time in one C . The division of cells does not follow a pattern symmetrical in time. the day of the moon is the month. This phenomenon of tidal evolution is not confined to the earth and the moon. The biological sciences certainly have their full share of one-way phenomena. Google-digitized / http://www. It has been suggested that this is Ihe result of an ancient tidal evolution.

Several of these.NEWTONIAN AND BERGSONIAN TIME 47 way. The work of Mendel has given us a far more precise and dis continuous view of heredity than that held by Darwin. and a posterior appendage which will catch the water . Generated on 2011-09-28 14:35 GMT / Public Domain. Charles Darwin's son. either from the point of view of the indivi dual or of the race. Darwin's principle still holds today. We have studied Ihe fine anatomy of the chromosome. from the simple to the complex.org/access_use#pd-google Darwinian evolution is thus a mechanism by which a more or less fortuitous variability is combined into a rather definite pat tern. from the time of de Vries on. and it is no accident that the problem of discovering its mecha nisms was carried ahead through the same great step by two men working at about the same time : Charles Darwin and Alfred Wallace. have made the statistical study of Mendelianism an effective tool for the study of evolution. such as Haldane. and if it is dependent for its food on the pursuit ot swift prey. This step was the realization that a mere for tuitous variation of the individuals of a species might be carved into the form of a more or less one-directional or few-directional progress for each line by the varying degrees of viabilily of the several variations. will swim better with a fusiform shape. or mammal. The record of paleontology indicates a definite long-time trend. An aquatic animal. The list of modern geneticists is long and distinguished.hathitrust. A mutant dog without legs will certainly starve. Google-digitized / http://www. while a long thin lizard that has developed the mechanism of crawling on its ribs may have a better chance for survival i* it has clean lines and is freed from the impeding projections of limbs. By the middle of the last century this trend had become apparent to all scientists with an honestly open mind. nor the choice of the . has com pletely altered our conception of the statistical basis of mutation. Neither the connection of the idea of the son with that of the father. whether fish. lizard. and the race is equally directed from the past into the future. while the notion of mutation. and have localized the gene on it. We have already spoken of the tidal evolution of Sir George Darwin. powerful body muscles. interrupted and complicated though it might be. though we have a much better knowledge of the mechanism on which it depends. its chances of survival may depend on its assuming this form.

and in which our conclusions concern not all such systems but an overwhelming majority of them. The third of the dynasty of Darwins. The succession of names Maxwell-Boltzmann-Gibbs represents a progressive reduction of thermodynamics to statistical mechanics : lhat is. but which was at odds with the whole remainder of physics .48 CYBERNETICS name. but in which the complete collection of data for the present and the past is not sufficient to predict the future more than statistically. Sir Charles. The ether showed much less power to absorb radiations of high fre Generated on 2011-09-28 14:35 GMT / Public Domain. not with a single dynamical system. in which the statistical Newtonian dynamics of (iibbs is replaced by a statis tical theory very similar to that of Newton and Gibbs for large- scale phenomena. It is thus not too much to say that not only the Newtonian astronomy. and of the molecules of the water. In tidal evolution as well as in the origin of species. we have a mechanism by means of which a fortuitous variability. « evolution ». This fact may be fortuitous. Google-digitized / http://www. The synthesis is the statistical theory discovered by Heisenberg in 1925. Planck gave a quasi-atomic theory of radiation — the quantum theory — which accounted satisfactorily enough Tor these phenomena. has become a picture of the average results of a statistical . but even the Newtonian phy sics. is fortuitous. The theory of tidal evolu tion is quite definitely an astronomical application of the elder Darwin. particularly where it concerned radiation. and Nils Bohr followed this up with a similarly ad hoc theory of the atom. About the year 1900 it became apparent that there was something seriously wrong with ther modynamics. Thus Newton and Planck-Bohr formed respectively the thesis and antithesis of a Hegelian antinomy. is con verted by a dynamical process into a pattern of development which reads in one direction.org/access_use#pd-google quency — as shown by the law of Planck — than any existing mechanization of radiation theory had allowed. a reduction of the phenomena concerning heat and temperature to phenomena in which a Newtonian mechanics is applied to a situation in which we deal. but to a statistical distribution of dynamical systems. that of the random motions of the waves in a tidal sea. but it nevertheless represents an even further inva sion of Newtonian ideas by ideas of statistics. is one of the authorities on modern quantum mechanics.hathitrust.

and hence an account of an evolutionary process. the wall has been erected to surround so wide a compass that both matter and life find them selves inside it. It is true that the matter of the newer phy sics is not the matter of Newton. The realization that the Newtonian physics was not the proper frame for biology was perhaps the central point in the old controversy between vitalism and mechanism. time has had its philosophical echoes. The civil engineers of ancient days were land-surveyors. A watch is nothing but a pocket orrery. and to convert the commerce of the great oceans from a thing of chance and adventure to a regular understood business. the steam-engine. they are effects to be over come. and the irreversible time of evolution and biology. astronomers. so that the resulting motion of the hands maybe as perio dic and regular as possible. From the Newcomen engine . but it is someting quite as remote from the anthropomorphising desires of the vitalists. In the end. The chief technical result of this engineering after the model of Huyghens and Newton was the age of navigation.NEWTONIAN AND BERGSONIAN TIME 49 situation. the vita- list proved too much. although this was complicated by the desire to conserve in some form or other at least the shadows of the soul and of God against the inroads of materialism.and Tyche is as relentless a mistress as Ananke. Generated on 2011-09-28 14:35 GMT / Public Domain. reversible time to a Gibbs- ian. Berg- son emphasized the difference between the reversible time of physics. To the merchant succeeded the manufacturer. the craftsmen made their tools in the image of the heavens. those of the seventeenth and early eighteenth centuries were clockmakers and grinders of lenses. Google-digitized / http://www. Instead of building a wall between the claims of life and those of physics. The thought of every age is reflected in its technique.org/access_use#pd-google and navigators .hathitrust. in which for the first time it was possible to compute longitudes with a respectable precision. moving by necessity as do the celestial spheres. and to the chronometer. irreversible. in which there is always something new. in which nothing new happens. as we have seen. As in ancient times. It is the enginee ring of the mercantilists. The chance of the quantum-theoretician is not the ethical free dom of the Augustinian. and if friction and the dissipation of energy play a role in it. This transition from a Newtonian.

a science in which time is eminently irreversible. There is in electrical engineering a split which is known in Germany as the split between the tech nique of strong currents and the technique of weak currents. Actually. to be reproduced as the tap of a telegraph receiver at the other end . Ihe theory of the conservation of energy and the later statistical explana tion of the Cfrnot Principle or second law of thermodynamics or principle ot the degradation of energy — that principle which makes the maximum efficiency obtainable by a steam-engine depend on the working temperatures of the boiler and the con denser — all these have fused thermodynamics and the Newto nian dynamics into the statistical and the non-statistical aspects of the same science. It is this split which separates the age just past from that in which we are now living. and the first telegraphers. received as the angular position of the rudder. Thus communication engineering began with Gauss. Google-digitized / http://www. and Joule. and the physics of Newton has been supplemented by that of Rumford. Heat has been converted info usable energy of rotation and translation. or it may be a sound transmitted and received through the apparatus of a telephone . it was perhaps Heaviside who did the . commu nication engineering can deal with currents of any size whatever. Carnot. the present time is the age of communication and control. Thermodynamics makes its appearance. and which we know as the distinction between power and commu Generated on 2011-09-28 14:36 GMT / Public Domain. or it may be the turn of a ship's wheel. what distinguishes it from power engineering is that its main interest is not economy of energy but the accurate reproduction of a signal. and from the eighties on. Wheatstone.hathitrust. This signal may be the tap of a key. It received its first reasonably scienti fic treatment at the hands of Lord Kelvin. and although the ear lier stages of this science seem to represent a region of thought almost without contact with the Newtonian dynamics.org/access_use#pd-google nication engineering.50 CYBERNETICS almost to the present time. If the seventeenth and early eighteenth centuries are the age of clocks. the central field of engineering has been the study of prime movers. and the later eighteenth and the nineteenth centuries constitute the age of steam-engines. and the movement of engines powerful enough to swing massive gun turrets. after the failure of the first transatlantic cable in the middle of the last century .

This is done to avoid questioning the orthodox Chris tian attitude that animals have no souls to be saved or damned.hathitrust. the present automaton opens doors by means of photocells.NEWTONIAN AND BERGSONIAN TIME 51 most to bring it into a modern shape. Just how these living automata function is something that Des cartes. that figure of clay into which the Rabbi of Prague breathed in life with the blasphemy of the Ineffable Name of God. burning some combustible fuel instead of the glycogen of the human muscles. To begin with. we have the bizarre and sinister concept of the Golem. Finally. In the time of Newton. is one which Descartes does discuss. In the days of magic. both in sensation and in will. never discusses. This idea has played a very genuine and important role in the early history of modern philosophy. although we are rather prone to ignore it.org/access_use#pd-google Neither the Greek nor the magical automaton lies along the main lines of the direction of development of the modern machine. Google-digitized / http://www. The discovery of radar and its use in the second world war. together with the exigencies of the control of anti-aircraft tire. This desire to produce and to study automata has always been expressed in terms of the living technique of the age. which was certainly never so actively pur sued in the past as it is at the present day. or computes the solution of a differential equation. the automaton becomes the clockwork music box. the important allied question of the mode of coupling of the human soul. have brought to the field a large number of well-trained mathematicians and physicists. It is far different with the clockwork automaton. Generated on 2011-09-28 14:36 GMT / Public Domain. Descartes considers the lower animals as au tomata. so far as I know. He places this coupling in the one . the ability of the artificer to produce a working simulacrum of a living organism has always intrigued people. the automaton is a glorified heat engine. However. although in a very unsatisfactory manner. with the little effigies pirouetting stiffly on top. or points guns to the place at which a radar beam picks up an airplane. nor do they seem to have had much of an influence on serious philosophic thought. In the nineteenth century. At every stage of technique since Daedalus or Hero of Alexan dria. with its material environ ment. The wonders of the automatic computing machine belong to the same realm of ideas.

he replaces the pair of corresponding elements. In Spinoza. As to the nature of this coupling — whether or not it represents a direct action of mind on matter and of matter on mind — he is none too clear. This is the situation from which Leibniz starts. it is entirely natural to attribute the correspondence between our will and the effects it seems to produce in the external world to a similar divine intervention.org/access_use#pd-google self-contained attributes of God . by a continuum of corresponding elements : the monads. Each of them lives in its own closed universe.hathitrust. While these are conceived after the pattern of the soul. and gives little or no attention to the mechanism of this correspondence. This is the path followed by the Occa- sionalists. Geulincx and Malebranche. but he attributes the validity of human experience in its action on the outside world to the goodness and honesty of God. the pineal gland. or He is an active participant. Google-digitized / http://www. in which case it is hard to see how Descartes'explanation really explains anything.52 CYBERNETICS median part of the brain known to him. He probably does regard it as a direct action in both ways . they correspond one to the other through the pre-established harmony of God. who is in many ways the continuator of this school. but Leibniz is as dynamically minded as Spinoza is geometrically minded First. but closed though they are. Thus the causal chain of material pheno mena is paralleled by a causal chain starting with the act of God. with a perfect causal chain from the creation or from minus infinity in time to the indefinitely remote future . in which case it is hard to see how the guarantee given by His honesty can be anything but an active participation in the act of sensation. they include many instances which do not rise to the degree of selfconsciousness of full souls. but Spinoza is not dynamically minded. the doctrine of Occa sionalism assumes the more reasonable form of asserting that the correspondence between mind and matter is that of two Generated on 2011-09-28 14:36 GMT / Public Domain. by which He produces in us the experiences corresponding to a given material situation. mind and matter. The role attributed to God in this matter is unstable. which have so been . Leibniz compares them to clocks. Either God is entirely passive. and which form part of that world which Descartes would have attributed to matter. Once this is assumed.

this fact is pushed into a corner and glibly explained by a contrast between the chemical energy of the living organism and the thermal energy of the heat engine. As he says. Google-digitized / http://www. and proteins. into carbon dioxide. we are coming to realize that the body is very far from a conservative system. The living organism is above all a heat engine. or rather more self-contained than. In the nineteenth century. The conservation and the degradation of energy are the ruling principles of the day. as is natu ral in a disciple of Huyghens.org/access_use#pd-google ratures of animal muscle attract attention as opposed to the high working temperatures of a heat engine of similar efficiency. The monad is a Newtonian solar system writ small. The engineering of the body is a branch of power engineering Even today. this reflec tion does not consist in a transfer of the causal chain from one to another. the passively dancing figures on top of a music-box. and the chief of these is that of potential. they have no windows. almost all of which . but this is due to the miraculously perfect workmanship of the Creator. and if the low working tempe Generated on 2011-09-28 14:36 GMT / Public Domain. fats. and urea.hathitrust. and the whole trend of thought of such biophysicists as Rashevsky and his school bears witness to its continued potency. All the fundamental notions are those associated with energy. Unlike humanly made clocks. he constructs after the model of clockwork. burning glucose or glycogen or starch. water. Thus Leibniz considers a world of automata which. the animals and plants of the materialist. and that its. conservative physiologists . They are actually as self-contained as. The electronic tube has shown us that a system with an outside source of energy. this is the predominating point of view of the more classically minded. The apparent organization of the world we see is something between a figment and a miracle.NEWTONIAN AND BERGSONIAN TIME 53 wound up as to keep time together from the creation for all eternity. are studied from a very different aspect. It is the metabolic balance which is the center of attention. the automata which are humanly constructed. nor are they effectively influenced by it. They have no real influence on the out side world. and those other natural automata. Today. Though the monads reflect one another.component parts work in an environment where the available power is much less limited than we have taken it to be. they do not drift into asynchronism .

their metabolism. and so on. hydrogen-ion-potential recorders. whose function is to recombine the incoming impressions into such form as to produce a desired type of response in the effectors. which may be said to taste . They comprise photo-electric cells and other receptors for light. we deal with automata effectively coupled to the external world. Moreover. receiving their own short Hertzian waves . the atoms of the nervous complex of our body. Finally. These correspond among other things to the kinaesthetic organs and other proprioceptors of the human system. as long as the automaton is running.54 CYBERNETICS is wasted. pressure-gauges of various sorts. The information fed into this central control system will very often contain information concerning the functioning of the effectors themselves. and its cardinal notions are those of message. amount of disturbance or « noise » — a term taken over from the telephone engineer — quantity of information. The effectors may be Generated on 2011-09-28 14:37 GMT / Public Domain. This is the analogue of memory. but may be delayed or stored so as to become available at some future time. The organs by which impressions are received are the equivalents of the human and animal sense-organs. and that the bookkeeping which is most essential to describe their function is not one of energy. microphones and so on. do their work under much the same conditions as vacuum tubes. its very rules of operation are susceptible to some change on the basis of the data which . but also by a flow of impressions. the newer study of automata. whether in the metal or in the flesh. especially if it is worked at a low energy level.hathitrust.org/access_use#pd-google electrical motors or solenoids or heating-coils or other instru ments of very diverse sorts. the information received by the automaton need not be used at once. etc. thermometers. In such a theory. and of the actions of outgoing messages. for we too have organs which record the positon of a joint or the rate of contraction of a muscle. is a branch of communication engineering. of incoming messages. may be a very effective agency for performing desired operations. coding technique. Betweenthe receptor or sense-organ and the effector stands an intermediate set of elements. radar systems. with their relatively small power supplied from outside by the circulation . Google-digitized / http://www. We are beginning to see that such important elements as the neurons. not merely by their energy flow. In short.

We are scarcely ever interested in the performance of a communication-engineering machine for a single input. and the need of handling the . It is clear of course that the relation input-output is a consecutive one in time. automatic gyro-compass ship- steering systems. the very old steam-engine governor belongs among them — but the great mechanization of the second world war brought them into their own. and the equivalent of a nervous system to integrate Generated on 2011-09-28 14:37 GMT / Public Domain.org/access_use#pd-google the transfer of information from the one to the other.extremely dangerous energy ot the atom will probably bring them to a still higher point of development. and this means a statistically satisfactory performance for the class of input which it is statistically expected to receive.NEWTONIAN AND BERGSONIAN TIME 55 have passed through its receptors in the past. and this is not unlike the process of learning. ultra-rapid computing machines. The machines of which we are now speaking are not the dream of the sensationalist. What is perhaps not so clear is that the theory of the sensitive automata is a statistical one. To function adequately it must give a satisfactory performance for a whole class of inputs. effectors. nor the hope of some future time. The relation of these mechanisms to time demands careful study. self-propelled missiles — especially such as seek their target — anti-aircraft fire-control systems. They had begun to be used long before the war — indeed. We shall study this in . They already exist as thermostats.hathitrust. Thus its theory belongs to the Gibbsian statistical mechanics rather than to the classical Newtonian mechanics. automatically controlled oil-cracking stills. and involves a deQnite past-future order. and the present age is as •truly the age of servo-mechanisms as the nineteenth century was the age of the steam engine or the eighteenth century the age of the clock. and the like. To sum up : the many automata of the present age are coupled to the outside world both for the reception of impressions and for the performance of actions. It is scarcely a miracle that they can be subsumed under one theory with the mechanisms of physiology. They lend themselves very well to description in physiological terms. Scarcely a month passes but a new book appears on these so-called control mechanisms or servo-mechanisms. Google-digitized / http://www. They contain sense-organs.

Thus the modern automaton exists in the same sort of Bergso- nian time as the living organism. the new mechanics is fully as mechanistic as the old. but as we have said this victory is a complete defeat. and « materialism » has come to be but little more than a loose synonym for « mechanism ». . Whether we should call the new point of view materialistic is largely a question of words : the ascendancy of matter charac terizes a phase of nineteenth-century physics far more than the present age. Google-digitized / http://www. Vitalism has won to the extent that even mechanisms correspond to the time-structure of vitalism .org/access_use#pd-google of badly posed questions.hathitrust. for from every point of view which has the slightest relation to morality or religion.56 CYBERNETICS much more detail in the chapter devoted to the theory of com munication. the whole mechanist-vitalist controversy has been relegated to the limbo Generated on 2011-09-28 14:37 GMT / Public Domain. and hence there is no reason inBergson's considerations why the essential mode of functioning of the living organism should not be the same as that of the automaton of this type. In fact.

hathitrust.org/access_use#pd-google . Google-digitized / http://www.Generated on 2011-09-28 14:37 GMT / Public Domain.

Google-digitized / http://www. Thus an event of probability one. « Yes ». this will determine in a completely Newtonian way the distribution of the momenta and positions for any future time. and so on. or certain other characteristics with pro bability zero. a third member. The distinction between these two cases involves rather subtle considerations concerning the nature of sets of instances. and some of these will have the character of assertions that the future system will have certain characteristics with probability one. was never a very subtle one. and the expression Generated on 2011-09-28 14:38 GMT / Public Domain. Probabilities one and zero are notions which include complete certainty and complete impossibility. a third. If we consider all the distinct decimal . Thus we cannot sum probabilities in all conceivable cases. and Gibbs. forming a sequence of contingencies in which every term has a definite position given by a positive integer. and so on — each of which has a known probability. in each specific case I must actually hit some specific point. to get a probability of the total event — for the sum of any number of zeros is zero — while we can sum 1hem if there is a first. is the reso lution of a complex contingency into an infinite sequence of more special contingencies — a first. that of my hitting some point. may be made up of an assemblage of instances of probability zero.hathitrust. and indeed. although it is not impossible that I hit it. Is it possible for a class to be infinite. such as that of the positive integers? This problem was solved towards the end of the last century by Georg Cantor. although a very powerful mathematician. which form an infinite sequence. a second. but include much more as well. and yet essentially different in multiplicity from another infinite class.58 CYBERNETICS of the system. Nevertheless. the chance that I hit any specific point on the target will generally be zero. a second. and the answer is.org/access_use#pd-google of the probability of this larger contingency as the sum of the probabilities of the more special contingencies. and Gibbs is nowhere clearly aware of it. although it is used implicitly. It will then be possible to make statements about these distributions. If I shoot at a target with a bullet of the dimensions of a point. which is an event of probability zero. one of the processes which is used in the tech nique of the Gibbsian statistical mechanics.

The whole theory depends on the properties of the average of a series. dependent on the time alone. multiplied by a given weighting function.org/access_use#pd-google of the product of the function to be represented. and to the then moot question of the generality of the sets of motions of a linear system which can be synthesised out of the simple vibrations of the system — out of those vibrations. is A. in terms of the average of an individual term. and zero from A to 1. and that the Gibbsian theory does not involve con tradictions. had also solved that of Gibbs. it is known that they cannot be arranged in one.GROUPS AND STATISTICAL MECHANICS 59 fractions. Thus a single function is expressed as the sum of a series. The service of Lebesgue to the Gibbs theory is to show that the implicit requirements of statistical mechanics concerning contingencies of probability zero and the addition of the probabilities of contingencies can actually be met. all the terminating decimal fractions can be so arranged. This is the reason why Lebesgue. This goes back to the eighteenth-century physics of waves and vibrations. In these series. Lebesgue's work. the coefficients are expressed as averages Generated on 2011-09-28 14:38 GMT / Public Domain. for which the passing of time simply multiplies the deviations of the system from equilibrium by a quantity. with N degrees of freedom. The particular distributionsdiscussedby Gibbs have themselves a dynamical interpretation. and not on position. lying between zero and one. terminating or non-terminating. If we consider a certain very general sort of conservative dynamical system.hathitrust. in other words. In other words. but on what looks like a very different theory. Notice that the average of a quantity which is one over an interval from zero to A. the theory of trigonometric series. the theory needed for the average of a series is very close to the theorjr needed for an ade quate discussion of probabilities compounded from an infinite sequence of cases. three order — although. if it is known to lie between 0 and 1. in solving his own problem. Thus the distinction demanded by the Gibbs statistical mechanics is not on the face of it an impossible one. was not directly based on the needs of statistical mechanics. Google-digitized / http://www. positive or negative. strangely enough. however. and may be regarded as the probability that the random point should lie in the interval from 0 to A. two. we find that its position and velocity coordinates .

However. (2). which are unchanged as the system develops. and there are no forces attached to fixed positions and fixed orientations in space. Cand Ulam. and the moment of momentum of the system as a whole. and regular enough to be subject to the system of integration based on Lebesgue measure.org/access_use#pd-google momentum. If all the bodies in the system act only on one another. and the moment of Generated on 2011-09-28 14:38 GMT / Public Domain. Google-digitized / http://www. N of which are called the generalized position coordinates and N the gene ralized momenta. the continual change of the boundary of the region does not change its 2N-dimensional volume. it is known that systems in which another invariant quantity exists. These determine a 2N-dimensional space defining a 2N-dimensional volume . there is one other numerically valued entity which also remains constant : the energy. Ann. Oxtoby J. They are not difficult to eliminate. For example if our family is that of concentric spheres. and if we take any region of this space.60 CYBERNETICS may be reduced to a special set of 2 N coordinates. dependent ontheinitial coordinates and momenta of a dynamical system. Bolh of these are vectors : the momentum. In highly specialized systems. momentum. there may be other quantities. there are two other expressions which also remain constant. the notion of volume generates a system of measure of the type of Lebesgue. are very rare indeed in a quite precise sense l. . 874-920 (1941). not determined by the energy. just as measure in space will determine area on a two-dimensional surface out of a family of two-dimen sional surfaces. so that the system is replaced by a system with fewer degrees of freedom. and in the conservative dynamical systems which are transformed in such a way as to keep this measure constant. and in the space of the remaining coordinates. In systems without other inva riant quantities. the measure determined by the position and momentum coordinates will itself determine a sort of sub-measure. and let the points flow with the course of time. the momentum. « Measure-Preserving Homeomorphisms and Metri cal transivity ». for sets not so simply defined as these regions. which changes every set of 2N coordinates into a new set depending on the elapsed time. of Math.hathitrust. and total moment of momentum. then the volume between two concentric spheres close 1. In this system of measure. 42. In general. we can fix the coordinates corresponding to energy.

Erg. We shall call this measure phase measure. and total moment of momentum are determined. Hopf. « Ergodentheorie ». only useful to show to young mathematicians the needs and possibilities of rigor. However. If. In the notion that these two types of average were related. would have nothing to do with it till his dying day. or as we can make it by a change in scale. For another fifteen years. It was not until about 1930 when a group of mathematicians — Koopman. von Neumann. and averages taken with respect to it phase averages. he was utterly and hopelessly wrong. Let us thentakethis new measure on a region in phase space for which energy. Osgood 1.F. it was a museum curiosity. any quantity varying in time may also have a time average. — finally established the 1. one. d. o In Gibbs' statistical mechanics.02) ^x j f V(0 dt.GROUPS AND STATISTICAL MECHANICS 61 together. For this he was scarcely to blame.. it will itself be invariant. Let the total measure of this restricted region be constant. (1936). its time average for the past will be : (2.hathitrust. and let us suppose that there are no other measurable invariant quantities in the system. /"(/) depends on /. and in the method by which he tried to show this relation. the same. both time averages and space averages occur. . As our measure has been obtained from a measure invariant in time. E. A mathematician as distinguished asW. Google-digitized / http://www. Gibbs Generated on 2011-09-28 14:38 GMT / Public Domain. It was a brilliant idea of Gibbs to try to show that these two types of average were. 2. when normalized by taking as one the total volume of the region between the two spheres. the fame of the Lebesgue integral had just begun to penetrate to America. for example.org/access_use#pd-google was perfectly right . Nevertheless some of Osgood's early work represented an important step in the direction of the Lebesgue Integral. Birkhoff 2.01) and its time average for the future : (2. in some sense. total momentum. Math. will give in the limit a mea sure of area on the surface of a sphere. in a way invariant in time. Even at the time of his death.

We shall later see what these foundations were. we should be forced to await each new catastrophe in a state of perplexed passiveness. Now.hathitrust. These notions were certainly familiar to Gibbs. We have . Google-digitized / http://www. in the first place. we need a more precise analysis ofthe notion of invariant. and who is generally right. Gibbs is one of the scientists whose physico-mathematical acumen often outstrips his logic. as Plancherel. even if it is of infinite length. a system generally passes inde finitely near to every point in the region of phase-space deter mined by the known invariants. Like his contemporary Heaviside. The followers of Gibbs.org/access_use#pd-google zero elsewhere — which were most urgently needed to make sense out of Gibbs' theory. Besides the notions of average and of measure — the average over a Universe of a function one over a set to be measured and Generated on 2011-09-28 14:39 GMT / Public Domain. there is no significant case where that hypothesis is true. including at the end perhaps Gibbs himself. it is possible to maintain that he did not assess them at their full philosophical value. This hypothesis he called the ergodic hypothesis. There is no logical difficulty as to the truth of this : it is merely quite inadequate for the conclusions which Gibbs bases on it. and others have shown. from the Greekwords £?7<>v. saw this in a vague way. it is necessary that there exist phenomena which do not stand isolated. as well as the notion of transformation group. In a world ruled by a succession of miracles performed by an irrational God subject to sudden whims. « path ». as his study of vector analysis shows. and 6'coe. « work ». No diffe- rentiable path can cover an area in the plane.62 CYBERNETICS properfotmdations ofthe Gibbs statistical mechanics. in order to appreciate the real signifi cance of ergodic theory. It says nothing about the relative time which the system spends in the neighborhood of each point. Gibbs himself thought that in a system from which all the invariants had been removed as extra coordinates. in the study of ergodic theory. For the existence of any science. which merely asserts that in the course of time. while he is often unable to explain why and how he is right. almost all paths of points in phase-space passed through all coordinates in such a space. Nevertheless. and replaced this hypothesis by the quasi-ergodic hypothesis.

Generated on 2011-09-28 14:39 GMT / Public Domain. known as the product or resul tant BA.hathitrust. is a transformation. y into — z and z into — x. and z into — y.. Ideally. while B takes x into z. such that both AA"1 and A'*A are that very special transfor mation which we call I. but will have the property that every element is the result of transfor ming an element. The change in scale which occurs when we examine a preparation under the magnifying action of a microscope is likewise a trans formation. Sometimes. Google-digitized / http://www. there is a unique transformation A'1. and invariant. likewise subject to locomotor initiative of their own. y into — x. it should represent a property of the system discussed which remains the same under the flux of particular circumstances. Thus if A is the transformation which takes the coordi nates into the coordinate (/.where the mallets are flamingos.GROUPS AND STATISTICAL MECHANICS 63 a picture of such a world in the croquet game in Alice in Won- derland. We are thus led to the notions of transformation. A transformation of a system is some alteration in which each element goes into another. z into — x. which trans forms every element into itself. the identity transformation. the transformation A will not only carry every element of the system into an element. The modification of the solar system which occurs in the transition between time /A and time tz is a transformation of the sets of coordinates of the planets. The essence of an effective rule for a game or a useful law of physics is that it be stateable in advance. but not always. unpre dictable Queen of Hearts. then BA will take x into y. or subject our geometric axes to a rotation. and y is unchanged . playing-card soldiers.the balls. it is a property which is invariant to a set of transformations to which the system is subject. If AB and BA are the same. In this case we call A"1 the . we shall say that A and B are permutable. transformation-group. and the rules are the decrees of the testy. hedgehogs.org/access_use#pd-google The result of following a transformation A by a transforma tion B is another transformation. In this case. The similar change in their coordinates when we move their origin. and that it apply to more than one case. while AB will take x into z. the hoops. In the simplest case. which quietly unroll and go about their own business. Note that in general it depends on the order of A and B.andyinto — o.whilezis unchanged.

Then if Tx stands for the element resulting from x under the transformation T. and if f(x) is a function of abso lute value 1. it is called an invariant of the group.04) A(*) = A*/i(x). Google-digitized / http://www. and where the resultant of any two transformations belonging to the set itself belongs to the set. are non-Abelian groups. . with certain appropriate properties of continuity or integrability. Let the elements transformed by an Abelian group be the terms which we repre sent by x. in some such form as.hathitrust.«(T) fix). The set of all translations along a line.(To.) <.org/access_use#pd-google elements. It is clear that A is the inverse of A"1. The set of rotations about a point. If we can represent any function h(x) defined over the group as a linear combination of the characters of the group. If f(x) and g(x) are group characters. where any two trans formations of the group are permutable. It is an invariant of the group in a slightly generalized sense. and even more. such that: (2.64 .03) /. and let f(x) be a complex-valued function of these Generated on 2011-09-28 14:39 GMT / Public Domain. we shall say that f(x) is a character of the group. or in a plane. There exist certain sets of transformations where every trans formation belonging to the set has an inverse. as is [f(x)"\ ~l. If this quan tity is unchanged when each element is changed by the same transformation of the group. where <x(T) is a number of absolute value 1 depending only on T. and of all motions of a rigid body in space. CYBERNETICS inverse of A. and that the inverse of AB is B^A 1. Let us suppose that we have some quantity attached to all the elements transformed by a transformation group. it is a transformation group of the special sort known asAbelian. whatever that transformation may be. clearly f(x) g(x) is one also. There are many sorts of such group invariants. is a transformation group. These sets are known as transformation-groups. (2. likewise belonging to the set . that I is its own inverse. or in a three-dimensional space. of which two are especially impor tant for our purposes. The first are the so-called linear invariants.

we shall obtain : (2. and e'Xx involves only the number bet ween 0 and 2* which differs from x by an integral multiple of 2*.r) that a(T) does to f(x) in (2. differing from x + T by an integral multiple of 2*. and it may simi larly be seen that the constant 1 is a character.07) a(T + 2*) = a(T).06) /. x and T in (2. a(T)=eiVr. In this case. for a given character/". Multiplication by a group character thus generates a transformation group of the group characters themselves. we have the extra condition that Generated on 2011-09-28 14:39 GMT / Public Domain. the values of .org/access_use#pd-google (2.06) will still hold. If the original group is the translation group on the infinite line. then (2. and the character group will be the group of translations changing x into x + T.r + T) = a(T) /"(or). negative. The characters will be the functions e'*x. which means that x must be a real integer. which is satisfied if f(x) =e''*. The character group thus corresponds to the translations of the real integers.05) Thus if we can develop /i(x) in terms of a set of group charac ters.GROUPS AND STATISTICAL MECHANICS 65 where /fc' x) is a character of the group.03).(. so that the operator T changes x into x + T.05) are con fined to integer values. thus having the same structure as the original group. Google-digitized / http://www. we can develop h(Tx) for all T in terms of the characters. This will not be the case when the original group consists of the rotations about a circle. positive.08) e««* = l.03) be comes : (2. or zero. In any character group. which is known as the cha racter group of the original group. the operator T changes X into a number between 0 and 2*. (2. If on the other hand the original group is that of the translations of the integers. and while (2.hathitrust. and a/r(T) bears the same relation to/fc(. If now we put f(x) = e^x as before. Thus the character group is essentially the group of rotations about a circle. We have seen that the characters of a group generate other characters under multiplication and inversion.

and must be 0. 2™ J o Generated on 2011-09-28 14:40 GMT / Public Domain. the reader should consult the following references 1.12) f(x) = I °°a(X) e». if we can express h(x) as in (2. if there is any reasonable basis of taking an average of these values which is not affected by the trans formation of the group by the multiplication of each transforma tion by a fixed one of its transformation. In the case of the group of rotations on a circle. this gives us directly that if (2.66 CYBERNETICS «(T) are distributed in such a way that the distribution is not altered when they are all multiplied by <x(S). we shall have : (2.* dx. either <x(T) is always 1.org/access_use#pd-google and the result for translations along the infinite line is closelj related to the fact that if in an appropriate sense. there 1.hathitrust.llf lln— I l\.04). Wiener. Google-digitized / http://www. In other words.• i r2:t \&. From this it may be concluded that the average of the product of any character by its conjugate (which will also be a character) will have the value 1. .10) f(x) = 2 an *»*. Cambridge Univ. then in a certain sense (2. . and without a clear statement of their conditions of validity. That is. then : 10 11 \ n — — I flf\ »—inx //i. The Fourier Integral and Certain of its Applications.^l " UJL. or this average is invariant when multiplied by some number not 1.13) a(x) = J. For more pre cise statements of the theory. Besides the theory of the linear invariants of a group.09) Afc= Average (h(x) fk(x) ). and that the ave rage of the product of any character by the conjugate of another character will have the value 0. (2. for any element S in the group.C" f(x) 2n J These results have been stated here very roughly. Press (1932).

The ordinary ergodic theorems start with an ensemble E. as we have already seen. definable in terms of the structure of the group itself.. where — oo < x < oc and where (2. of Math. « Der Maassbegriff in derTheorie der Kontinuierlichen Gruppen ». The basis on which this has been accomplished is known as the ergodic theory.hathitrust. .14) Tx . Haar has proved that a certain rather wide class of groups does possess a uniquely determined invariant measure. The most important application of the theory of the metrical invariants of a group of transformations is to show the justification of that Intel. These are the systems of Lebesgue measure which do not undergo any change when the objects transformed by the group are permu ted by the operators of the group. 1. f(Tlx) is taken to be measurable in x and x simultaneously. pp. As such. £ ™~r ln=o -.org/access_use#pd-google Ergodic theory concerns itself with complex-valued functions f(x) of the elements x of E. due to Haar 1. 147-169 (1933). As we have seen every group itself is a collection of objects which are permuted by being multiplied by the operations of the group itself.16) /•„(*) =_L. that is. Google-digitized / http://www. (2) 34. (2. which we can take to be of measure 1. In this connection. f(x) is taken to be measurable in x. Ann.changeability of phase averages and time averages which. Gibbs tried in vain to establish. Haar. it may have an invariant measure. or by a group of measure-preserving transformations T*.15) J"\ f(x) The theorem then asserts that dx < oe.GROUPS AND STATISTICAL MECHANICS 67 is also the general theory of its metrical invariants. we should cite the interesting theory of group measure. (2. f(x) is taken to be of class L2. Generated on 2011-09-28 14:40 GMT / Public Domain. and if we are concerned with a continuous group of-transformations. II. transformed into itself by a measure-preserving transformation T. T11 = Tx+|i. In all cases. In the mean ergodic theorem of Koopman and von Neumann.

This is impossible Generated on 2011-09-28 14:40 GMT / Public Domain. (2. or the set of transformations Tx.hathitrust.17). except for a set of values of x measure 0. A very interesting case is the so-called ergodic or metrically transitive one.21) /*(*)- and (2.A(a. leaves invariant no set of points x which has a measure other than 1 or 0. o as the case may be. the set of values (for either ergodic theorem) for which f*(x) takes on a certain range of value is almost alwaj's either 1 or 0. in the sense that : (2. The value which f*(x) then assumes almost always. 17) /. we have . f(x) is taken to be of class L .) \2dx = 0.68 CYBERNETICS or (2. In the « almost everywhere » ergodic theorem of Birkhoff. which means that (2. o That is.org/access_use#pd-google unless f*(x) is almost always constant. In such a case. in the Koopman theorem. Google-digitized / http://www. in which the transformation T.20) / | f(x) | dx The functions /N(x) and fA(x) aredeflnedasin(2.22) /*(*)= exist.23) Cf(x) dx.A(x) = -^ j* * /(P*) rfx. The theorem then states that.19) \ f*(x) . converges in the mean to a limit f*(x) as N -*• oo or A -*• oo respectively.16) and (2. is (2.

and such theories as that of the homogeneous gas. von Neumann has shown under very general condi tions that they can be reduced to ergodic components. such that one or another exists at a given time. E can be separated into a finite or denumerable set of classes En and a continuum of classes E(y). In particular. These transformations are all ergodic. not a mixture of both.2G) measure (S) = | measure (S(y) ) dy +V measure (Sn). of zero measure. The spatial analogue of temporal equilibrium is spatial homoge neity. Similar results hold in the continuous case. The whole of ergodic theory. Where the transformation T or the transformation group T> is not ergodic. liquid. or solid depend on the application of three-dimensional ergodic theory. the whole theory of measure-preserving transfor Generated on 2011-09-28 14:40 GMT / Public Domain. . That is. and if S(y) is the intersection of S with E(y) and Sn with En. E J Eg ""En In other words. except for a set of values of a. Incidentally. This is an adequate justiiication for Gibbs' interchange of phase averages and time averages. The case of three dimensions is physically important. we have 0 <2"25) .org/access_use#pd-google mations can be reduced to the theory of ergodic transformations. then (2.it may be applied to the translation group in vj dimen sions. may be applied to groups of transformations more general than those isomorphic with the translation group on the line. let us remark in passing. a non-ergodic group of translation transfor mations in three dimensions appears as the set of translations of a mixture of distinct states.GROUPS AND STATISTICAL MECHANICS 69 <2'24> N ^oc NTT o and in the Birkhoff theorem. Google-digitized / http://www. which is invariant under T or T>./^ = ZJ n=0 except for a set of values of a: of zero measure or probability 0. such that a measure is established on each E„and E(y).hathitrust.

org/access_use#pd-google assumed. with a very fair approximation even though no temperature is precisely determined except in a state of equi librium and by methods involving this equilibrium. this means that if we take a state of other than maximum entropy. let us consider the dynamics of n particles in a bottle.hathitrust. The system will spend most of its time in a state near that of greatest entropy. Gibbsian statis tical mechanics may well be a fairly adequate model of what .70 CYBERNETICS One of the cardinal notions of statistical mechanics. for a small number of regions of the given volumes and at the given temperature Generated on 2011-09-28 14:40 GMT / Public Domain. We may still talk of local temperatures. and so the temperatures read by ordinary thermometers in living tissues are gross averages. in the sense that for most of the time. liven the more refined discussions of thermal engines. n-m in B is a maximum. the entropy almost always increases. and observe what happens to it. nearly n-n^ in B. n-m in B. The logarithm is the entropy of the distribution : m particles in A. This fineness is far greater than that of the space-and-time scale of an ordinary thermometer. For example. do not change these conditions too radically. and not the true temperatures of thermodynamics. and its physiology is certainly of a corresponding fineness of texture. in living matter. in which a gas is expanding in a more complicated manner than in a cylinder. In the ordinary thermodynamic problems of the heat engine. and it will have a certain probability measure. and states within the limits of practical discrimination. divided into two parts. where the probability of the combination m1 in A. we lose much of even this rough homogeneity. nearly int particles will be in A. is that of entropy. which also receives an application in the classical thermodynamics. and expresses the logarithm of their probability measure. However. The states for which we study the entropy are states involving maximum entropy for a given temperature and volume. we have characterized a region in phase-space. It is primarily a property of regions in phase- space. Google-digitized / http://www. particularly of thermal engines like the turbine. The structure of a protein tissue as shown by the electron microscope has an enormous definiteness and fineness of texture. we are dealing with conditions in which we have a rough thermal equilibrium in large regions like an engine cylinder. For systems with a large number of particles. A and B. If m particles are in A. and n m in B.

this is the Maxwell distribution.hathitrust. we seem to obtain a perpetual-motion machine Generated on 2011-09-28 14:41 GMT / Public Domain. For a perfect gas. and certainly does not mean what it appears to mean. the law of the increase of entropy applies to a completely isolated system. but does not apply to a non- isolated part of such a system. either an anthropo morphic demon or a minute mechanism. Nothing is easier than to deny the possibility of such beings or structures. but when a particle of less than ave rage velocity approaches from compartment A or a particle of greater than average velocity approaches from compartment B. it must receive information from approaching particles.GROUPS AND STATISTICAL MECHANICS 71 happens in the body . chemical. and do not try to demonstrate it. the gate is closed. When a particle of more than average velocity approaches the gate from compart ment A or a particle of less than average velocity approaches the gate from compartment B. Let this gas be contained in a rigid container with a wall across it. containing an opening spanned by a small gate. the picture suggested by the ordinary heat engine certainly is not. This produces an apparent decrease in entropy. Accordingly. A very important idea in statistical mechanics is that of the Maxwell demon. we shall miss an admirable opportunity to learn something about entropy and about possible physical. Now. they must involve a coupling of the demon and the gas. the gatekeeper opens the gate. For a Maxwell demon to act. Let us suppose a gas in which the particles are moving around with the distribution of velocities in statisti cal equilibrium for a given temperature. We shall actually find that Maxwell demons in the strictest sense cannot exist in a system in equilibrium. and the particle passes through. concerning their velocity and point of impact on the wall. the only entropy . The thermal efficiency of muscle action means next to nothing. In this way.org/access_use#pd-google of the second kind. Google-digitized / http://www. so that if the two compartments are now connected by a heat engine. Whether these impulses involve a transfer of energy or not. It is simpler to repel the question posed by the Maxwell demon than to answer it. operated by a gatekeeper. the concentration of particles of high velocity is increased in compartment B and is decreased in compartment A. and biological systems. but if we accept this from the beginning.

it is impossible to obtain any information giving the position or the momentum of a particle. There is no reason to suppose that metastable demons do not in fact exist. All catalysts are ultimately poisoned : they change rates of reaction. decreasing entropy. and a system in statistical equilibrium is in equilibrium both in matters concerning entropy and those concerning energy. it may well be that enzymes are metastable Maxwell demons. it ceases to act as a Maxwell demon. The information must be carried by some physical process. Can we find terms involving the demon as well which contribute to this total entropy ? Most certainly we can. there may be a quite appreciable interval of time before the demon is deconditioned. and the stable state of a living organism is to be dead. but not true equili- . exceeding a minimum dependent on the frequency of the light used for examination. The demon can only act on informa tion received. and is incapable of clear perceptions. and as Leibniz says of some of his monads. and this information. indeed. until it falls into « a certain vertigo ». say some form of radiation. In the long run. In fact. We may well regard living organisms. and not that of the gas alone. Thus all coupling is strictly a coupling involving energy. Certainly the enzyme and the living organism are alike metastable: the stable state of an enzyme is to be deconditioned. but by some other equivalent process. Google-digitized / http://www. perhaps not by the separation between fast and slow particles. under the quantum mechanics.hathitrust. the Maxwell demon is itself subject to a random motion corresponding to the temperature of its environment. represents a negative entropy. Nevertheless. without a positive effect on the energy of the particle examined. and this time may be so prolonged that we may speak of the active phase of the demon as metastable. However. much less the two together. and that the transfer of energy between particle and demon is for a considerable time far less significant than the transfer of information. in this light. It may very well be that this information is carried at a very low energy level.72 CYBERNETICS which concerns us is that of the system gas-demon. it Generated on 2011-09-28 14:41 GMT / Public Domain. as we shall see in the next chapter. The gas entropy is merely one term in the total entropy of the larger system. such as Man himself.org/access_use#pd-google receives a large number of small impressions.

GROUPS AND STATISTICAL MECHANICS 73 brium. The next chapter is devoted to the statistical mechanics of time series. Google-digitized / http://www. . This is another field in which the conditions are very remote from those of the statistical mechanics of heat engines and which . I refer especially to the work of Kryloff and Bogoliouboff.hathitrust. rather than assumed in advance. I do not wish to close this chapter without indicating that ergodic theory is a considerably wider subject than we have indicated above. Nevertheless.org/access_use#pd-google happens in the living organism. catalysts and Man alike have sufficiently definite states of metastability to deserve the recognition of these states as relatively permanent conditions. There are certain modern developments of ergodic theory in which the measure to be kept invariant under a set of transformations is defined directly by the set itself.s thus very well suited to serve as a model of what Generated on 2011-09-28 14:41 GMT / Public Domain. and to some of the work of Hure- wicz and the Japanese school.

although the apparatus by means of which they are combined and modified must in general be very rapid in its action. or a sequence of numerical quantities. The temperature as recorded by a conti nuous recording thermometer . corresponding to the whole apparatus of computing machines and schedules.org/access_use#pd-google tories. AND COMMUNICATION There is a large class of phenomena in which what is observed is a numerical quantity.CHAPTER III TIME SERIES. or the complete set of meteorological data published from day to day by the Weather Bureau . of the statistical laboratory. continuous or discrete. and are well suited to a treatment employing hand computation or ordinary numerical tools such as slide-rules and computing machines. taken day by day.hathitrust. simple or multiple. What is not generally realized is that the rapidly changing sequences of voltages in a telephone line or a television circuit or a piece of radar apparatus belong just as truly to the field of statistics and time series. distributed in time. and in fact must be able to put out results pan passu with the very rapid alterations of input. just as it has into the automatic range-finders and gun-pointers of an anti-aircraft fire control system. INFORMATION. wave filters. frequency-modulating networks and their corresponding receivers — are all in essence quick-acting arithmetical devices. and the staff of computers. and for . are all time series. Google-digitized / http://www. or the closing quotations of a stock in the stock market. automatic sound — coding devices like the Vocoder of the Bell Telephone Labora Generated on 2011-09-28 14:41 GMT / Public Domain. These pieces of apparatus — telephone receivers. These time series are relatively slowly changing. The ingenuity needed in their use has been built into them in advance. Their study belongs to the more conventional parts ofstatistical theory.

. between heads and tails in the tossing of a coin.org/access_use#pd-google where bk is the first digit not equal to 0. What is this information.— a choice. bn . each has the value 0 or 1.01) .. 0x02 .*1 ... which may with uniform a priori probability lie any where in this range.alC^a. If then we ask for the amount of information in the perfectly precise measurement of a quantity known to lie between A and B. INFORMATION. The chain of operation has to work too fast to admit of any human links. then the number of choices made and the consequent amount of information is infinite.. are significant.. We shall call a single choice of this sort a decision. On ••• and we shall take this quantity as the precise formula for the amount of information and its definition.. at a2 ..... no measurement which we actually make is per formed with perfect precision. where al.. an = —al + —a2 + + — an+ .02) iog2 . preservation.hathitrust. must unitary forms of informa tion is the recording of a choice between two equally probable simple alternatives.1). a2. If the measurement has a uni formly distributed error lying over a range of length bl b2. Google-digitized / http://www. We may conceive this in the following way: we know o priori that a variable lies between 0 and 1. and how is it measured? One of the simplest.03) — logz measure of (a. AND COMMUNICATION 75 the same reasons. However.. and use of information. for example. Then the amount of informa tion we have from our a posteriori knowledge is (3. and possibly to a*. an . we shall see that if we put A =0 and B = 1...2•••" . Here (3...TIME SERIES.. and a posteriori that it lies on the interval (a. Generated on 2011-09-28 14:42 GMT / Public Domain. while all the later decisions are not. transmission. we shall see that all the decisions from at to OA_i. The number of decisions made is certainly not far from : (3. One and all. b) measure of (0.. b) inside (0. time series and the apparatus to deal with them whether in the computing laboratory or in the telephone circuit. one or the other of which is bound to happen . have to deal with the recording..1) . and represent the quantity in the binary scale by the infinite binary number .

1 = log . In particular. 00 so that the average logarithm of the breadth of the region under l\(x) may be considered as some sort of average of the height of the logarithm of the reciprocal of fl(x). ^—. Fisher for statistical problems.76 CYBERNETICS However. 3O The quantity we here define as amount of information is the negative of the quantity usually defined as entropy in similar situations. let us now consider a case where our apriori knowledge is that the probability that a certain quantity should lie between x and x+ dx isf^x) efcr.04) fxf\ (x) dx = 1. Here the author makes use of a personal communication of J.07) log. A. if fl(x) is a constant over (a.= log.1). It will be noted that we are here assuming the variable to have a fundamental equipartition : that is.05) y°° (log. and can be used to replace Fisher's definition in the technique of statistics. -J—— . or any other function of x. and is zero elsewhere.and the a posteriori probability isf2(x) dx. although it is a statistical Generated on 2011-09-28 14:42 GMT / Public Domain. ^^ /** h n 1 1 00 Using this to compare the information that a point is in the region (0. (3. Thus a reasonable mea sure 1 of the amount of information associated with the curve (3.06) J (lo82A(x))A(x) dx = -^-^ log.hathitrust.log. Since fl(x) is a probability density. b). Kx)) fl(x) dx. our results will not in general be the same if we replace x by x3. The definition here given is not the one given by R. we obtain for the measure of the difference. How much new information does our a posteriori probability give us ? This problem is essentially that of attaching a width to the regions under the curves y=fl(x) and y=/2(*).org/access_use#pd-google definition. 1. Google-digitized / http://www. . we shall have : (3. von Neumann. (3.

INFORMATION. t/ -ao —oo and (3. Google-digitized / http://www. How much information do we gain concerning u if we know that u + v = wt In this case. We assume the a priori distributions of it and v to . . An interesting problem is that of determining the information gained by fixing one or more variables in a problem. In the two-dimen sional case f(x.hathitrust.. while a variable v lies bel- Tttt 1 _^ ween the same two limits with a probability 7=^ 2* dx.082) /« ^.r+</:r with the probability V/2 . 0) log2 A<*.081) f dx I dy A(*.083) f dx f dy 9(.30 Generated on 2011-09-28 14:43 GMT / Public Domain. (3.TIME SERIES. For exam ple.r) *({/) = 0 . AND COMMUNICATION 77 The definition which we have given for the amount of infor mation is applicable when the variable x is replaced by a va riable ranging over two or more dimensions. tx ". t. 0)log. y) is a function such that : (3.e 2a dx. and the amount of information is : (3.084) d.r dyA(*.00 then <f(x) dx = j <[>({/) dy. let us suppose that a variable u lies between x and. where w is fixed. y)=0.V) .A(*.08) C dx (*. it is clear that u = w — v.org/access_use#pd-google rOO x»00 dx <f(x) logsj <f(x) + i dy <?( Iog2 and the amount of information from independent sources is additive.30 . »)• jJ Note that if is of the form and (3.

78 CYBERNETICS be independent. We can consider this result in the following light : let us treat u as a message and v as a noise. Then the information carried by a precise message in the absence of a noise is infinite. It is one-half the logarithm of the ratio of the sum of the mean squares of n and v. dx 2 log .09) e 2ae 2* = Cl e '"* 2«» ' _ x^ (x —») _ 0+6 where Cl and c2 are constants. In the presence of a noise.. and it becomes infinite as b goes to 0. The excess of information concerning x when we know w to be that which we have in advance is . . and that it is inde pendent of iv. Then the a posteriori distribution of u is pro portional to (3. (*) + Tl~.org/access_use#pd-google logarithm of a quantity which we may consider as a probability. We have said that amount of information. a+b o Note that this expression (3. Google-digitized / http://www.091) is positive. it has the properties we associate with an entropy. Ihe amount of information concerning u which a knowledge of u + v gives is large. is essentially a negative entropy. on the average. to the mean square of v. however. Let <f(x) and fy(x) be two probability densities . then — — is also a probability density. Then (3. this amount of information is finite. It is interesting to show that. and it approaches 0 very rapidly as the noise increases in intensity. They both disappear in the for mula for the gain in information given by the fixing of w. being the negative Generated on 2011-09-28 14:43 GMT / Public Domain.hathitrust.10) f *(*> + *(*> log + *(*:) . If v has only a small range of variation.

b) and is zero elsewhere. For example.. Google-digitized / http://www.. or if in a function of several variables we allow some of them to range unimpeded over their natural range of variability..TIME SERIES... Here we have a precise application of the second law of thermodynamics in communication engineering. our distribution function will be proportional to f(xlt. xn+ 1. Let us take as a new set of variables ylt .. . They consist in the fusion of regions of proba bility which were originally distinct. AND COMMUNICATION 79 This follows from the fact that (3.. b) (3. An interesting case is when we have a probability distribution with n-fold density f(x. as we should expect.. ym.... Thus .) is a probability density vanishing outside (a. How much information do we get by fixing these m variables ? First let them be fixed between the limits «/l. INFORMATION.. if 9(0.y*m<ym<y*m + dy*mand 0 outside. ym.11) log In other words. as we have seen. . Then over the new set of variables. the greater specifica- Generated on 2011-09-28 14:43 GMT / Public Domain. yl+ dg^.org/access_use#pd-google tion of an ambiguous situation. closely analogous to the processes which gain entropy. ym. will.... No operation on a message can gain information on the average. It will be seen that the processes which lose information are... b —a This follows from the fact that the logarithm curve is convex upward. on the average.....12) f *(x)\ogv(x)dx is a minimum when f(x)— over (a.. we lose information. . xn.. the overlap of the regions under <?(x) and reduces the maximum information belonging to <p(x) +- On the other hand.. Conversely.xn) over the region R given by jKy*<y* + (dy)*.. and never lose it.xn) over the variables (xlt.. xn). and where we have m n dependent variables J/l.hathitrust. generally gain information. if we replace the distribution of a certain variable by the distribution of a function of that variable which takes the same value for different arguments. ym + dym.

*'V" OO 00 Closely related to this problem is the generalization of that which we discussed in (3.141).xn—m) to be a generalized message ...>3?m/ y ^-j rf*-+ ^« ~oo —oo —oo 0*1.. the set (xn—m+i.80 CYBERNETICS the amount of information obtained by the specification of the y's will be : (3.Xn)...org/access_use#pd-google / dxn-m+i I dxnf(Xl. ••. where the I! is taken over all sets of points (xn—m+i.. and the y"s to be a generalized corrupted message.... in the case just discussed how much information do we have concerning the variables xl..13).xn) / dxl. y* I flit* I nit* it T* i" i Inn tit* '*• \ /*°° r ~— I MX.. .-.. I dxn f(xl... .....141) f(xl.. On this basis. . •••.rn = J xn—m+lf..tn^ 10^2/V1 !»• •.. xn—„. we see that we have given the solution of a generalization of the problem of (3.t l» • •.>.13) / dxl. *"«) cor responding to a given set of y*'s.. and the un-normalizedprobability density after fixing the y*'s is: (3. xn) to be a generalized noise .. alone : Here the « priori probability density of these variables is: (3... though it will be a bit lengthy.. Google-digitized / http://www... I UJ n /^.xn)log2f(xl. we may easily write down the solution to our problem..... i dxnf(x1.14) Generated on 2011-09-28 14:43 GMT / Public Domain...xn). If we take the set (xlt.hathitrust.

It enables us to evaluate different systems. and with a message restricted to a definite frequency range and a definite power output for this range. In the first place. the original message must be coded for the grestest compression in transmission. no means of trans mission of information is more efficient than amplitude modu lation. INFORMATION. We wish to ascertain how much information these observations give concerning the messages alone.hathitrust. AND COMMUNICATION 81 We have thus at least a formal solution of a generalization of the message-noise problem which we have already stated. This is a technical problem and not suitable to a detailed discussion here . This is a central problem of communication engineering. it can be shown that with the definition of information given here. In general. the Generated on 2011-09-28 14:43 GMT / Public Domain. certain remarks are in order. C. in the design of the « vocoder » system by the Bell Telephone Laboratories. such as amplitude modulation or frequency modulation or phase modulation. Shannon of those laboratories. the information transmitted by this means is not neces sarily in the form most suitable for reception by the ear or by any other given receptor. This problem has been attacked. as far as their efficiency in transmitting information is concerned.TIME SERIES. however.org/access_use#pd-google efficient use of amplitude modulation or any other form of modulation must be supplemented by the use of decoding devices adequate to transforming the received information into a form suitable for reception by human receptors or use by mechanical receptors. but . with a random* static» on the ether equidistributed in frequency as far as power is concerned. There is indeed one operation which seems to contradict this. We shall now discuss the way in which infor mation may be presented in a form homogeneous in time. Google-digitized / http://www. So much for the definition and technique of measuring information. Here the specific characteristics of the ear and of other receptors must be considered by employing a theory very similar to the one just developed. On the other hand. Let it be noted that most of the telephone and other communication devices are actually not attached to a particular origin in time. and the relevant general theory has been presented in a very satis factory form by Dr. at least in part. although other means may be as efficient. Similarly. A set of observations depends in an arbitrary way on a set of messages and noises with a known combined distribution.

The extra message. It follows from this that if <!>(/"(/) (is a « functional » of/(/) — that is. 0 —A There is even more here than this. The group satisfies the properties that (3.16) A. — 00 < VL< oo).15) T>(T> f(t) )=T(*+* /(/) (— 00 < X < 00 . the time-average of $ j f(t) \ . in its simplest form. it will be seen that the situation will come under our general statement. All the n formation it contains is conveyed in an arbitrarily short interval of time. the transformation group consisting of the operators T> which change /(/) into f(t + x) leaves the probabi lity of the ensemble invariant. not altered by the change of t to / +T throughout. which has a measure defined over itself and also invariant under the . belongs to a sub-set (which may be the whole set) which goes into itself under the same transformation. we are in a position to use the Birkhoff ergodic theorem quoted in the Generated on 2011-09-28 14:44 GMT / Public Domain. is thus a single function or a set of functions of the time. adds nothing to the rate at which the system is carrying information.15). That is. any element belonging to a system which goes into itself under a group of measure-preserving transformations such as (3. or as the statisticians call it.hathitrust. (3. a time series which is in statistical equilibrium. due to von Neumann. A message homogeneous in time.82 CYBERNETICS which really does not. except for a set of values of f(t) of zero probability. which states that except for a set of elements of zero probability. This. If however we regard the factor sin (at + b) as an extra message which is put into the apparatus. We have stated in the last chapter another theorem of ergodic character. a number depending upon the whole hi story of f(t) — and *f the average of f(t) over the whole ensemble is finite.or in symbols. which we call the carrier. Google-digitized / http://www. which forms one of an ensemble of such sets with a well-defined proba bility distribution. converts a message /"(/) into one of the form /(/) sin (at + b). and to come to the conclusion that.- exists. and thereafter nothing new is said.org/access_use#pd-google last chapter. This is the operation of modulation.

we may determine the distribution of A. we only need to know the past of almost any one time series of the class. or measure 0. in which we have several quantities varying simultaneously. Then we can determine the simultaneous distribution of (cr0. and which has the further property that any portion of this sub-set with measure preserved under the group of transformations either has the maximum measure of the sub set. however.. Generated on 2011-09-28 14:44 GMT / Public Domain. we can compute with probable error zero the entire set of statistical parameters of an ensemble in statistical equilibrium to which that time series belongs.16) is in almost all cases the average of $ I /"(/)) °ver all the space of functions /"(/) .given the entire history up to the present. For example.hathitrust. the so-called phase average. In other words. Here we ap- . a^. it is equally true. an are all given. A) from the past of almost any single time series.. by using a time average instead of aphase average. we shall find that the time average (3. Up to here.. if the set of f s is taken in its narrowest possible sense.. 2. . and use its appropriate measure. Now let A be some function of the values of t in the future : that is. Thus in the case of such an ensemble of functions /"(/). an.TIME SERIES. rather than a single varying quantity. ). AND COMMUNICATION 83 transformation. Google-digitized / http://www. Moreover. we have fully deter mined /"when we know the set of quantities : (3. if ac.org/access_use#pd-google We are now in a position to discuss various problems be longing to time series. for quite a wide class of functions f(t) ( — o o < / < oo )... of a time series known to belong to an ensemble in statistical equilibrium. except in a set of cases of zero probability we can deduce the average of any statistical parameter of the ensemble — indeed we can simultaneously deduce any countable set of such para meters of the ensemble — from the record of any one of the component time series. INFORMATION. We shall confine our attention to those cases where the entire past of a time series can be given in terms of a countable set of quantities.. 1. If we discard all elements except those of such a sub-set. we have formulated this for single time series . for arguments greater than 0.17) an = °e« /n /(/) dt n = (0. for multiple time series.. In particular.

If then we have given any adequate interpretation to the « best value » of any of these statistical parameters or sets of statistical parameters — in the sense. We can compute the amount of information concerning any statistical parameter or set of statistical parameters. and future. we may wish to know the dis tribution of a value of another component. present. We ask for the distribution of the values of the message at some given time. In particular. Another interesting situation is that of a multiple time series.84 CYBERNETICS peal to the known theorem of Nikodym on conditional proba bilities. We then ask for an operator on the past of the cor rupted message which will best give this true message. We can even compute the whole amount of information which a knowledge of the past will give us of the whole future beyond a certain point.org/access_use#pd-google tain an infinite amount of information. or any set of quantities depending both on the past and the future. We also know the statistical joint distribution of the message and the noise as time-series. and so on. of which we know the past. The distribution of any quantity involving more than these pasts can be studied by means very similar to those al ready suggested. The same theorem will assure us that this distribution. perhaps. at some point of time. or a set of values of other components. of a mean or a median or a mode — we can compute it from the known distribution. past. in which we know precisely onl ythe pasts of some of the com ponents. past. which fixing of the past will give us. will tend to a limit as n -*• oo and this limit will give us all the knowledge there is con cerning the distribution of any future quantity. using any desired statistical basis of this merit — mean square error or maximum error or mean absolute error. together with a noise. We may similarly determine the simultaneous distribution of values of any set of future quantities. in some . The general problem of the wave-filter belongs to this class. and our knowledge of the present will con Generated on 2011-09-28 14:44 GMT / Public Domain. Google-digitized / http://www. and obtain a prediction to meet any desired criterion of good ness of prediction. present or future. We have a message.hathitrust. combined in some way into a corrupted message. although when this point is the present we shall in general know the latter from the past. under very general circumstances. when the past is known. We can compute the merit of the predic tion.

among them Einstein.18) CXP Generated on 2011-09-28 14:45 GMT / Public Domain. . and the motions over successive times are completely uncorrelated.. the motion shows a curious kind of undifferentiability. (3. we can make the set of paths correspon ding to the different possible Brownian motions depend on a parameter a lying between 0 and 1. Pub. If we normalize the scale of the Brownian mo tion to fit the time scale. On this basis. Unless we go down in the time scale to intervals so small that the individual impacts of the particles on one another are discernible. « Fourier Transforms in the Complex Plane »... which is unambiguous. a). This conforms closely to the physical observations. Google-digitized / http://www. between xn and xn + docn at time /„..hathitrust. Finally. ^ tn the particles lie between xt and xL + dxl at time t^. where x depends on the time / and the parameter of distribution a.. Final Chapter. We may ask for a statistical estimate of some measure of the error of our knowledge of the message. American Math. There is one ensemble of time series which is particularly simple and central. then the proba bility that if 0 ^ t1 . A very interesting question is that of determining the average with respect to a of x(lla).org/access_use#pd-google / — xl — (x2—xj2 (xn—Xn-l)2 (2*)« M/3-1) .... almost all paths will be continuous and non-differentiable. and Wiener N. INFORMATION..x(tna). Perrin. and if we let x(f) equal 0 for /=0. impelled by the random impacts of the other particles in a state of thermal agitation. (tn-tn-i) On the basis of the probability system corresponding to this. AND COMMUNICATION 85 given statistical sense.TIME SERIES. in such a way that each path is a function x(t. E. and consider only one coordinate x of the motion. This will be: 1. The mean square motion in a given direction over a given time is proportional to the length of that time. R. and where ths probability that a path lies in a certain set S is the same as the measure of the set of values of a corresponding to paths in S.. This is the ensemble associated with the Brownian motion. C. and the author1. Paley.^ /2 ^ .. A. Soc. Smoluchow- ski. The theory has been developed by many writers. we may ask for the amount of information which we possess concerning the message. (1933). The Brownian motion is the motion of a particle in a gas. is. Col.

JJ /*" /•- under the assumption that 0 < /t <> .19) will become : JJ / 0 if any x/ is odd .M . Let us put : (3.i + ^.(/„ — /„-!)] 2 — _L / d5i.hathitrust. a)) '0 X(x(/g.2 + .a) a:(/2.20) 5l. .86 CYBERNETICS (3. «) )• . ...j is even. a) — x(tg—1. 5« where Xk....org/access_use#pd-google group of Xfc.(/.n = n./ terms into which x is separated) = 2 A/ 211 f da(x(ik.a) ... Then if we put 5« = 0 the expres sion of (3. a) . Google-digitized / http://www. 19) P rfa x(/i. $. /„. * /L if every \k. ^k. T1 / /- _XM = 2 A* H (numDer of ways of dividing Xfc)/ terms into pairs) x (// —//-i)—^- — 2 Afc (numbers of ways of dividing n terms into pairs ft whose elements both belong in the same Generated on 2011-09-28 14:45 GMT / Public Domain. ._!.. / d5. o = (2«) ~ T [/ t(/ .

All we need to do is to write our coor dinates in the square in the decimal form : Generated on 2011-09-28 14:45 GMT / Public Domain. al a2. px p2. into pairs .22) P da x(tlt a) x(t2. a) . AND COMMUNICATION 87 -Here the first S sums over j.2 are t *.. p) = x(—t. .. There is a well-known mathematical device to map a square on a line segment in such a way that area goes into length. p. of \i .24) a = . an. x(tn. In other words. we define : (3... Up to the present... a) x(tk..). a. INFORMATION. a) /^ 0 a x(tj.TIME SERIES. ..i of the elements to be selected from /& and tq are 'ii ^fr... p = .... It immediately results that : (3. a.Y) = Z(t...n numbers.. and the * over all the pairs in each partition.ai) by pairs. where a and phave independent uniform distributions over (0. a. and the n is taken over those pairs of values k and q . where X*.... p) where t runs over the whole real infinite line.25) 5(/.. and so on. Using this substitution. Google-digitized / http://www. when we know the averages of the products of x(tj. /« into distinct pairs.26) f K(0 . J and we obtain a mapping of this sort which is one-one for almost all points both in the line-segment and the square.ai) where / is positive. where the £ is taken overall partitions of /!.. and thus their entire statistical distribution.. the second. we shall obtain a distribution of 5(/. a). We now wish to define : (3.. we have considered Brownian motions x(t.. and to put : Y = • al Pl *2 P2 an Pn-.org/access_use#pd-google (3. over all the ways of dividing n terms in blocks respectively. If we put t.hathitrust. pn. we know the averages of all polynomials in these quantities. \k. p) (t < 0).1).

I du I dv uv exp ( — ^TC \/ Is ( It " •" S j */ */ = f t / u2 Generated on 2011-09-28 14:45 GMT / Public Domain.Lecons sur I'Integration.Y) W.—"ft CO u2 e 8'du L r^e 2 du = I s I. we have formally : (3. H. . V ~*T£ +s 1. 1894." K'(0 5(/.(/) 5(*.Y) dr (3.29) y 5(*. if s and / are of opposite signs.Y) dt = / K.Y) S(/.Y) f X( 0 — °° _oo ' ° K. 1928. but 5 is a very irregular function of /. K runs sufficiently rapidly to zero at ± ooand is a sufficiently smooth function. and s \ < \ t \ C\ QC\\ / flo «A fH -*\ flv / ft I «• I n\ ft I / I ™"V tin \O»d\}j i s^AjY/ s\t>YJ "Y — / ^\ " 1» / *^\ I * |»*/ "* 00 = .Y) </Y = 0 .org/access_use#pd-google 1 r° . Stieltjes. o while if they are of the same sign.r) dt. de Toulouse.27) K(/) d5(/.28) p dY f\(t) d5(/. v/2rc I s y * ' «n —. and does not make such a definition possible. r e(*.Y) = . Under these circumstances.. des Sc.(0 5(/. If. Lebesgue. it is reason able to put : (3. Annales de la Fac. Google-digitized / http://www.hathitrust.(s) ds I Ka'(0 dt 00 00 0 Now. however.Y) dt °K. Paris.88 CYBERNETICS The obvious thing would be to define this as aStieltjes 1 integral.

hathitrust.(s)ds / /K^(/)d/- JJ 00 t 00 s /" ' ( f \ fx ' ( f* \ + \ J /J \ l J 1 I •»— /ad(Kt{a)Ki(a)) = In particular. AND COMMUNICATION 89 Thus : C dt /"W (3. and Generated on 2011-09-28 14:45 GMT / Public Domain. Google-digitized / http://www.Y) 0 k=l —M x^OO T. Y) represents a very important ensemble of time series in the variable /.TIME SERIES.org/access_use#pd-google the product is over the pairs in each partition. The expression : (3. INFORMATION.31) =— /"" K.34) K(/ + T) d fr. Y) 0 —oo /oo K(a) Moreover.<s) ds y/KJ(0 dt— f K^(s) ds f'tK't(t) dt 00 00 + / K. 00 (3. depending on a parameter of distribution Y. — T*) ds. (3. Y) = ft.33) ydr£[ f K(/ + Tft)d^.. We . TB into pairs.32) f df f K(t + Tl) d5(/. where the sum is over all partitions of TJ... Y) / K(/+T2) d 5(/.

In other words. Y) \«/ .hathitrust. Google-digitized / http://www.00 • which is the autocorrelation function of the statisticians with log Y. O then the transformation of Y into r preserves measure.38) will be the limit of Generated on 2011-09-28 14:46 GMT / Public Domain. Y) and of /(a + si. In other words. (3. if we consider the average of (3. Moreover. our time-series f(t. Y) is in statistical equilibrium..36) f(t + /„ Y) = f(t.org/access_use#pd-google (3.37) under these circumstances. Y) tends to the joint distribution of the first and the second set as a—>-oo.OO it will consist of precisely the terms in / K(/+«-T)dl5(/. Thus the statistics of distribution of /(/.37) K^— *)*«<*.. Y) and f(t + a.Y) together with a finite number of terms involving as factors powers of (3. and if this approaches 0 when a —>-ao .39) f K(«+ T)K(T)dT . and it can be shown in fact that if (3. Y).f(° + sm.35) *(T)= f K(s)K(s + T)da = j K(s+ 0 K(s + t + T) ds . f (t. By a more generally phrased but entirely similar argument. Y) are the same as the statistics of f(t+tlt Y) . it may be shown that the simultaneous distribution of /(/„ Y)-. f(tn.90 CYBERNETICS have just shown what amounts to the statement that all the moments and hence all the statistical parameters of this distri bution depend on the function : (3. In independent other words. Y) are asymptotically independent in their distributions as a —>• oo . any bounded measurable functional or quantity depending on the entire distribution of the values of the .

40) oo If now &\f(t.43) lim^r /°/•(/. Y) i dY = [y v i f«. Y) j is any integrable functional of f as a function of t. and indeed any denumerable set of statistical para meters.org/access_use#pd-google There are certain quantities dependent on a time series of this sort which have quite interesting properties.hathitrust. AND COMMUNICATION 91 function oft. Y) i dY ] '. except for a set of zero measure. (i C" K(/) d5(/.Y) . for such a time series. T-** 1 J we know <&(/) in almost every case. Y) into f(t + <J. which we may write in the form 3> \f(t. INFORMATION. it is interesting to know the average of : (3.44) exp. Y)jrfY o = Lim 4" T-Kao 1 for all values of Y. (3. and we have a complete statistical knowledge of the time series.Yl) dt.42) y* V !/•(/. when we know (3.TIME SERIES. and only takes on the values 0 or 1. we can almost always read off any statistical parameter of such a time series. It follows that if & \f(t.Yi) A'—'. Y) j is invariant under a translation of t. Google-digitized / http://www. /"(/. Actually. Y).Y) is metri cally transitive. Y)|. That is. Generated on 2011-09-28 14:46 GMT / Public Domain. from the past history of a single example. so that the transformation group of f(t. In particular. we shall have : yv i /</. then by the ergodic theorem. must have the property that (3.

45)..— / ^y It is a very interesting problem to try to build up a time series as general as possible from the simple Brownian motion series.46) f " dx exp f i fX K(t + T.3. x)Y J \ &J i a — oo Let us take K(/ltx). Generated on 2011-09-28 14:47 GMT / Public Domain. In such constructions. 5.Yl-cCMf) /**dxexp(i" f -f T. the example of the Fourier developments suggests that such expansions as (3. a new independent variable n and solve for xt obtaining (3. In particular.1« K(/1. x))« dt . Then as in (3.hathitrust.46).47) /^dre^C•.49) x .. let us investigate time series of the special form : (3. Google-digitized / http://www.44) are convenient building blocks for this purpose.Y)Y a — oo Let us suppose that we know 5(vr)as well as expression (3.i f"(K(t + r.92 CYBERNETICS Formally.48) /'dx exp I .45) n=0 /""(K(0 )• dt}m(2m -1) (2m-3) .1 «•/ / \m 1 /*" exp 1 /* . this may be written : (3. if /l > /2 (3.X) d5(T. we obtain : (3.org/access_use#pd-google If we now multiply by e 2 let s(tt — /x) = i« and let tt = tl.

55) K(/lt x) = H(G(x).-i. Then : (3.T.0('».1 f W— '. O]1 dt. \ & j ] From this. INFORMATION. such that (3. x). .48) becomes : (3. Then the function Generated on 2011-09-28 14:47 GMT / Public Domain. there is a known func tion F(u. and put : (3. x).53) pd\ exp ( . AND COMMUNICATION 93 Then (3. F is a known function. by a Fourier transformation.00 Since the left-hand side of this equation does not depend on /lt we may write it G(x). where it is also a known function.TIME SERIES.l*)))'*\ &y. Here. If we integrate this function with respect to y. x) and tj. /j. we can determine : (3. and (3. and we can invert it with respect to the first argument. That is. we determine (3.56) G(x) = y^dx exp (^-\ f [H(G(x).org/access_use#pd-google a —«> (3.50) *Q(tl' ^ exp ( .hathitrust. y). /O = G(x).57) exp (— ~ f [H(iU)J«dA = R(u) \&JI will be a known function.54) F(K(/ l. x))« rf/ ) = F(K(/l. lies between K(/1( a) and K(/i. and put : (3.f (K(t.58) ~ = R(G).51) as a function of n when y. ^/ \ •« •/ / 0 .52) ~Xdx exp - as a function of K(^. 6). Google-digitized / http://www. /O.

55).63) f1 dx exp ( — -i. for our operator is not changed if we add a constant to all values of A. and this is most important for the concrete . or (3.52).62) a = S(o) +const. It is easy to see that if a is finite. (3.52). by (3. A good deal of work remains to be done to discuss the problem of the inversion of the functions inverted when the results are not single valued.Y) )• JJ\JI /•i /»& / /•» \ Generated on 2011-09-28 14:47 GMT / Public Domain. we have determined K(/. X.61) G(a) = 0. and thus G as a function of X.x) d5(/.52) and the numbers a and b.org/access_use#pd-google 0 a —oo Thus. under certain circumstances which remain to be defini tely formulated.Y) as well. we have at least taken a first step towards the solution of the problem of reducing a large class of time series to a canonical form. Thus.hathitrust. Still. We can hence make it 0.X) in (3. we need only know b. it does not matter what value we give it. Google-digitized / http://www. or (3. if a time series may be written in the form (3. We have thus determined X as a function of G. and we know £(/. This constant will be given by: (3. we can determine the function K(/.59) AC. This can be determined.x)]» dt */ \ ** *s a —oo with (3. however. and b. and the general conditions of validity of the expansions concerned.64) / dr / dx exp I i / K(/. = S(G) + const. There is no extra difficulty if b = + oo and it is not hard to extend the reasoning to the case where a = — oo of course./""[K(/.60) x = j —— + const.X). except for an undetermined constant added to a. by a comparison of (3. To finish the determination of (3.94 CYBERNETICS That is.

Google-digitized / http://www. AND COMMUNICATION 95 formal application of the theories of prediction and of the measurement of information. This is the mixing property of Koopman.Y) as well as the times series which we are expanding in form (3. Let KD be so selected that this density is D.65) / K(/ + f almost always have a definite density. The theory to be developed here has already been sketched by the author. Generated on 2011-09-28 14:48 GMT / Public Domain. If K(/) is a sufficiently continuous function.1). Y) from —oo to oo 00 will be called Zn (D.TIME SERIES. Y). that if we take intervals of fixed length. the distributions of any functionals of the segments of the time series in these intervals approach independence as the intervals recede from each other 1. Kac. Then /« KD (t — T) d5 (T. or at least as the limit in some sense or other of time series deter mined by Brownian motions? We shall confine ourselves to time series with the property of metrical transitivity. Y)). but remote in time.46). it is possible to show that the zeros of (3. Then let (3. by a theorem of M. is a parameter of distribution of the time series. Now.org/access_use#pd-google and that this density can be made as great as we wish by a proper choice of K. as we have sketched them earlier in this chapter. T) = T(t — Zn (D. 1. in the numera tion of these zeros. —oo < n < oe. n. There is still one obvious limitation which we should remove from this approach to the theory of time series: the necessity which we are under of knowing 5(/. /i is only determined except for an additive constant integer. and with the even stronger property. vary ing uniformly over (0. n) be any time series in the continuous variable /. INFORMATION. which is the necessary and suffi cient ergo die assumption to justify statistical mechanics. The question is: under what circumstances can we represent a time series of known statistical parameters as determined by a Brownian motion .66) TD (/. let T(/. Of course. while M. .hathitrust.

for almost every value of n.. which Generated on 2011-09-28 14:48 GMT / Public Domain..68) K(s) = f employing a Fourier transformation. Let us now come to the prediction problem for time-series of the form (3.n.oo. It is therefore not inappropriate to try to express TD (t.. . /„ of x. Google-digitized / http://www. We see that the only independent statistical parameter of the time series is *(/). Y) is completely determined by /. Nevertheless.org/access_use#pd-google means that the only significant quantity connected with K(/) is (3. D and 5 (T.67) /"" K(s) K(s + /) ds. in the opinion of the author. 2.v) will approach the simulta neous distribution of T(/K. n. Let us put (3.96 CYBERNETICS where the Zn taken is the one just preceding t.69) — /rC°K(5)K(s + T)ds= /" . However. for a given D and a given n. f) (K = 1. rather than one which we can consider as already accomplished. either directly in the form (3. the simultaneous dis tribution of TD (/K. Then (3 . n. To know K(s) is to know /<•(«) and vice versa.34).52). non-linear filtering. Y). or in some way or another as a time series which has a distribution which is a limit (in the loose sense just given) of distributions of this form. offers the best hope for a rational.35) . It must be admitted that this is a program to be carried through in the future. It will be seen that for any finite set of values /15 /2. Among these problems are perhaps the most pressing facing communication enginee ring. and the theory of the dense gas and turbulence. as given by (3. Here of course K is real.. Y).. the evaluation of the transmission of information in non-linear situations. TD (/. it is the program which.hathitrust. [t. consistent treatment of the many problems associated with non linear prediction. Y) for the same t'Ks as D ->..

In order to make this restriction.71) F(«) = R log (*(«) ). (F(«) + i G(«). Y) = . then the determination of K(s) is equivalent to the determination of the imaginary part of log (7r(<o) ). changes cos fco into sin X« and sin X<a into — cos X<o. and the Cauchy principal value of (3. . /f(<o) and (/(•(o))-1 will be assumed to be of algebraic growth on the real axis. Thus | /r(«) | z is a known function. INFORMATION.TIME SERIES.F(u «)=—/ — F(u) 7t J II — will exist. is such that K(s).hathitrust. under very general conditions.72) GO .74) *(<a) = exp. Then (F(<o))2 will be even and at most logarithmically infinite. The transformation indicated by (3. Since however K(s) is real. which means that the real part of log | /f(<a) | is a known function.as defined in (3. vanish es for all negative arguments. Thus (3. (3. 1 /•. it can be shown that k(u>) is a function which. Thus F(<o) + t"G(<o) is a function of the form (3. known as Generated on 2011-09-28 14:48 GMT / Public Domain. Google-digitized / http://www.org/access_use#pd-google the Hilbert transformation. AND COMMUNICATION 97 Thus a knowledge of *(T) is tantamount to a knowledge of k(<») k( —«). This problem is not determinate unless we put some further restriction on k(a>).68).75) /•(/. If we now put (3. If we write (3.72).70) / whence k(«) = k( — «o). The type of restriction which we shall put is that log *(O) shall be analytic and of a sufficiently small rate of growth for <o in the upper half plane.73) /r° and satisfies the required conditions for log | k(u>) \ in the lower half plane.

d Nn(x).79) <K/) = lim /~K(/+T)dT / Qn «— *•• y y — < — T or if we write (as in (3.77) 5(T. and vice versa.68) ) : (3.82) lim ?„(-»)= We shall find this result useful in getting the operator of predic tion into a form concerning frequency rather than time. it can be shown that we may write l//r(<->) in the form (3. Thus the past and present of 5(/.Y) (or properly. of the « differential » dl(t. r) = lim / ds I Qn(t + a) f(a.QO (3.80) K(s) = /*°°/f(<o) e'-« d<o then : /« Generated on 2011-09-28 14:48 GMT / Public Domain.98 CYBERNETICS On the other hand. Y) da.org/access_use#pd-google — oo gn(<o) ei-»d« — oo Y(s) = /~%» e«-«d<o .f). and that this can be done in such a way that (3. J (3. Y) d«. Y)) determine the past and present of f(t. .hathitrust. r~ /~x n—>x J J — * Here the Qns must have the formal property that /CO /*°° _* In general.76) lim f e*. n—»•« J o where the Nns are properly determined . Google-digitized / http://www. we shall have : K(t + T) dT / Qn(T + «)/(«.81) V(«) = lim (2w) Thus : n — ^ oo (3.

)dr / — I — T = A (<o) eiu>" .org/access_use#pd-google (3. if A >0. we shall find out (somewhat as in (3.86) *A («) = — f K(/ + A) e-i»* dt : 2fiC J 0 and if we apply the operator of (3. It is the error of the best possible prediction of/f/+A. (3.81) that : .83) f(l + A. AND COMMUNICATION 99 Now. / tells us nothing. Its mean square value is : (3. and is entirely independent of the second term. INFORMATION.Y). Y) d«. It may be shown to have a Gaussian distribution with this mean square value. The best possible prediction itself is the last term of (3.85) to eitat. If we now put : (3.83).84) y ~'(K<< + A + T) )« dr = yW) )t dT — t-A 0 and this tells us all there is to know about it statistically. Google-digitized / http://www.TIME SERIES.hathitrust.87) *™f K(/-A + T. obtaining: Generated on 2011-09-28 14:48 GMT / Public Domain.85) =n™» f K (/ + A + T) fcj Qn (T + «) /•(«. (3.Y) - J-A =y K —I—A Here the first term of the last expression depends on a range of d5(vr) of which a knowledge of /"(<vr) for «<.

90) m(/) = f Q(T) d 5(/— T.88) A («) = (2*)3/2 qn (— = A:A = -T-: .00 .901) f Q(T + a) d 5(/ . o arid the mean square error of prediction is : (3.T.Q(T)) dr i«i •00 R(|/|+T)R(T)dT . 1 r* /** 2n/f(«) J J A — » This is then the frequency form of the best prediction operator. Y) + /" RWd 5(« .902) y~°[ Q(T) ]« rfr + y* " [ R(T) ]• dr.34) is very closely allied to the prediction problem.r.T. r). o and let the message be of the form : (3.903) <^(t) = I dt fldSn(t + r)n(*) K(|/|-f T)-Q(|/|+T))(K(T)—Q(T))dT ~ Q( l/l + *) ) (K(r) . . Google-digitized / http://www. 8).hathitrust.«(0 = f K(TJ d W .100 CYBERNETICS (3. Then the predictable part of the m(t + a) is clearly : (3. The problem of filtering in the case of time-series such as (3.89) m(/) + ./ e— '"> (*— A) d/ / /f(u) «'<" du. —00 —00 where Y and « are distributed independently over (0. Y). let us suppose that we know the following quantities 00 0 -ul Generated on 2011-09-28 14:49 GMT / Public Domain.1).org/access_use#pd-google (3.00 Furthermore. Let our message plus noise be of the form : (3.

904) «I'n(T) = f d S m( \ t \ + T) m(T) dT oo = /" Q( I / |+')Q^) rfr+ /* R( | / | i OO ./ R(s) e'" Generated on 2011-09-28 14:49 GMT / Public Domain.Y) /" — a. INFORMATION.TIME SERIES.«. .906) *u(«)= where (3.907) Thai is. AND COMMUNICATION 101 = f K( I / I + T) K(T) dT — C Q( | 1 1 + T) K(T) dt Q( 1*1 (3.hathitrust.905) <I>l2(T)= p df fl = /rl . («) = . Google-digitized / http://www.908) *n(<o) ds.org/access_use#pd-google 2n _/ — 30 (3../ K(s)e--ds 1 T"30 = — .00 (3.Y) Q(T) dT— *U(T) — ( The Fourier transforms of Ihese three quantities are respec tively : (3.

— a is known as the lead. This will give us : (3.913) Kk(»)J L.913) may always be constructed with as much accuracy as we like. is : (3.902) may be represented as the sum of the mean square filtering error for infinite lag : 1.102 CYBERNETICS and (3.hathitrust. They may be found elsewhere 1.89). Google-digitized / http://www. where for symmetry we write : "^(oj) = <J>l («).910) q („> - Hence : (3.909) 9(«) 7^) = *„(«) + <I«2l(«). if we write it on the frequency scale. We can now determine /r(<a) from (3. Here we put *(/) fort 11(/)+*Zt( /)+2 R(<*>lt(t) ).911) and thus the best determination of ni(t).74). Lee. The mean square filtering error (3. It may be either positive or negative. W. we see that the operator on m(l) + n(l) by which we obtain the « best » representation of m(t). We refer especially to recent papers by Dr. when it is nega tive. The apparatus corresponding to (3. and using an argument similar to the one by which we obtained (3.) / y A..88). The details of its construction are more for the spe cialist in electrical engineering than for the reader of this book.908) as we have defined /r(«) before on the basis of (3.org/access_use#pd-google electricians know as a wave. . Y. with the least mean square error is : _T.r This operator constitutes a characteristic operator of what Generated on 2011-09-28 14:49 GMT / Public Domain.(O>) x7 Combining this with (3. The quantity a is the Ing of the filter. filter.

916) 4>12(<o) = *„(«) = 0.org/access_use#pd-google rate of transmission of information. where Y and 8 are distributed independently.hathitrust./ <P. let us consider : (3. Another question which is interesting in the case of messages and noises derived from the Brownian motion is the matter of Generated on 2011-09-28 14:50 GMT / Public Domain. It will be seen that the mean square error of filtering is a mono- tonely decreasing function of lag.914) [ R(T)] 2 dr = *u(o) - = . Y). when (3. ./ ^(<o) + «>ll(e») + <l>21(<o) -da + and a part dependent on the lag : / d<o . AND COMMUNICATION 103 (3. how much information do we have concerning m(t) ? Note that we should heuristically expect that it would not be very different from the amount of information concerning (3.T.A) . N(0 d !•(/ ./ 2* y uv ' 2« y 1 /*"* /Trf 1 " [QWl'dr d<o l(«) + *•!(») + *„(«) + <I'M(«) S \ J_ I frtiH ^zsK 2« . Google-digitized / http://www. INFORMATION.917) n(t) 5(/ — T. Let us consider for simpli city the case where the message and the noise are incoherent : that is.TIME SERIES.Y). 8) .918) / M (T) d 5 (/ — T. In this case.1(<o) a<o -. Let us suppose we know m(t) + n(t) over (— A.

If now A -» oo.921) 2 ilog2 ~—m 2 / M(T)C * dTJ— A+/ N(T)C A dTitnt/Nfr). the ordinate of lower boundary is the logarithm of the intensity of the threshold of audible intensity — what we may call the logarithm of the intensity of the internal noise of . and that its mean square value is proportional to I/ (3./• — A and the time-density of communication of energy is this quantity divided by 2A. it has a close rela tion to the audiograms used to measure the amount of hearing and loss of hearing in a given individual. it depends. As will be seen.919) /~AM( where f and 8 have independent distributions.920) M(T)e .org/access_use#pd-google This is precisely the result which the author and Shannon have already obtained for the rate of transmission of information in this case. the total amount of information available con cerning M is (3. It can. As a matter of fact.918) has a Guassian distribution independent of all the other Fourier coefficients. not only on the width of the frequency band available for transmitting the message. Google-digitized / http://www. Here the abscissa is frequency.1 * *.Tent Thus by (3. but also on the noise level.104 CYBERNETICS which we have when we know all values of (3.hathitrust.09). (3. however be shown that the n th Fourrier coefficient of (3.921) approaches : 22)— / dulog2 fJ—caM(T) e»UT dr•+/« N(T) e'« dx »/» N(T) e«»T dT =022 Generated on 2011-09-28 14:50 GMT / Public Domain.

of course. First : the theory gives us the best possible design of predictors and of wave-filters in the case in which the messages and the noises represent the response of linear reso nators to Brownian motions. This will not be an absolute best possible design. However.hathitrust. the logarithm of the intensity of the greatest message the system is suited to handle. and the quadratic quantities such as (3.88). The key formulae are (3.org/access_use#pd-google importance in economics.922). we have to develop a number of functions simultaneously in terms of the frequency.TIME SERIES. there will generally be some non-linear apparatus which gives a performance still better than that of any linear apparatus. in such a way as to satisfy certain auxiliary conditions in the complex plane. in so far as this can be done with apparatus performing linear operations. with Ihe definitions necessary to interpret these. In this case. AND COMMUNICATION 105 the receiving system —and the upper boundary. the time series here have been simple time series. meteorology. There are a number of variants of this theory. (3.922). and it is these which are of greatest Generated on 2011-09-28 14:50 GMT / Public Domain.35) and the I /f(«) 2 of the arguments following (3. The multi-dimensional theory represents a complication of the one already given. but in much more general cases. in which a single numerical variable depends on the time. at least in part. Google-digitized / http://www.70) are replaced by arrays of pairs of quantities —that is. The area between them. taken from day to day. is then taken as a measure of the rale of transmission ot information with which the ear is competent to cope. There are also multiple time series. This is the theory of prediction. The problem of determining Ar(<o) in terms of k(<a)\2. The com plete weather map of the United States. together. especially as the multiplication of matrices is not a permutable operation.914) and (3. in which a number of such variables depend simultaneously on the time . but it will minimize the mean square error of prediction and filtering. and the like. The theory of messages depending linearly on the Brownian motion has many important variants. which is a quantity of the dimension of (3. There is another closely related theory which is a simplification of it. filter- . the problems involved in this multi-dimen sional theory have been solved. they represent a possible design for predictors and filters. constitutes such a time series. matrices. INFORMATION. However. by Krein and the author. Next. becomes much more difficult.

and it is metrically transitive.tR J oo (3.927) — log *(«) = 2 P« cos 1 °° 2 —00 .106 CYBERNETIC? ing.org/access_use#pd-google for almost all a. Let us put : (3. its autocorrelation coefficient will be : (3. The theory of discrete time series is simpler in many respects than the theory of the continuous series.1). Each term (in the mixing case) will be representable as a combination of the previous terms with a quantity indepen dent of all previous terms. distributed uniformly over (0. for instance. The quantity a is as before the parameter of distribution. where n runs over all integer values from — x to oc.926) *(») = 2 *n eima. The time-series is said to be in statistical equilibrium when the change of n to n -f v (v on integer). Google-digitized / http://www.923) *m -- and we shall have : 1N (3.925) *„ = —I $(<•>) o i r* J. Let : (3. and the sequence of these independent factors may be taken to replace the Brownian motion which is so important in the con tinuous case. to make them depend on a sequence of independent choices. If fn (a) is a time series in statistical equilibrium.924) ^^-^ooiTfTo i1N lim l Y or Generated on 2011-09-28 14:50 GMT / Public Domain.1). It is much easier. and may be taken to run uniformly over (0. and amount of information in discrete time series.hathitrust. Such a series is a sequence of funcions fn (*) of a parameter a.

WM e^t"" = 21 eiu>lp—"j) I k(u) «—«(«* du.org/access_use#pd-google (3. INFORMATION. k(u>) will be the boundary value on the unit circle of a function without zeros or singular ities inside the unit circle.^ J This is the analogue of (3. (3.934) 2 W It will clearly be the result of the way we have formed k(u>) that in a very general set of cases.931) £ /•„_. if «o is the angle.929) «(<•') = £(<•>). We shall have : (3.932) y.hathitrust.928) G(<o) = ~ + ^pn «"«". we shall find that : o (3.930) | k(u) | 2 = *(<o).f*k(u) e-iv du.88). 2re J —it then 40 o Generated on 2011-09-28 14:51 GMT / Public Domain.933) kp = -^. Let: (3. Then under very general conditions. AND COMMUNICATION 107 and let : 00 (3. we can put: . Google-digitized / http://www.(a) Wu. Let us note that if we put (3. 00 1 °° /** o 2* /f(u) n^. If now we put for the best linear prediction of /«(a) with a lead of u.TIME SERIES.

a). The transfer of the filtering problem from the continuous to the discrete case follows much the same lines of argument.941) } X2 e-i-("-°) /r"«I>1 2wfc(«) „ J Generated on 2011-09-28 14:51 GMT / Public Domain. Thus for a prediction one step ahead.%) 2 Wn **" = *"*".934) becomes : (3. we can solve the entire problem of linear prediction for discrete time series.108 CYBERNETICS Then (3. as mathematical procedures to enable statisticians to obtain the best results with statistically impure data. Google-digitized / http://www.. this will be the best prediction possible by any method if (3. The filters for discrete time series are usually not so much physically constructible devices to be used with an electric circuit.939) — k0 00 and by a process of step-by-step prediction.fl— *t2 9* ***V o\o/ or (3. Formula (3. instead of integrals on /.937) f) Wl. the rate of transfer of information by a discrete time- series of fhe form . the best value for /"«(a) is : (3.9. en». except that all integrals on u or u are from — Ttton instead of from — o o to oo and all sums on u are discrete sums.org/access_use#pd-google where all the terms receive the same definitions as in the con tinuous case.= e-*.hathitrust. Finally. if u = 1.( 1 ~~ S oo / v— 1 0\0 In particular. (3.938) Wn= — gx+i*. As in the continuous case.940 /k (a) = f K(/J — T) d5(T.913) for the frequency characteristic of the best tilter takes the form : (3.

hathitrust. but which in reality transfer the responsibility for its use to the working statistician.922). will be the precise analogue of (3. Meanwhile the statistical theorist is quite 1./. In every case. R. Google-digitized / http://www. INFORMATION. involves an extension of existing methods of sampling.943) f N(/j — when Y and 8 are independent. The author and others1 have made a beginning in this direction. New York. AND COMMUNICATION 109 (3.942) f M(n — T)d!. as our observation does not run indefinitely into the past. (3. 2.org/access_use#pd-google represents the power distribution of the message in frequency. The development of our theory beyond this point. . Wiener. Wiley. N(dT) eiu' The statistical theories we have here developed involve a full knowledge of the pasts of the time series we observe.f"du log2 2* y—K /*o ) C'"T dr / \ J— N(T) C'" I/ where over (—«. we have to be content with less. and Doob. Y) in the presence of a noise (3.TIME SERIES. forthcoming book. or the person who ultimately employs his results. or of those terminological tricks in the theory of likelihood2. N. either of Bayes' Law on the one hand. Sec writings of. which seem to avoid the necessity for the use of Bayes' law. n). as a practical statistical theory.944) J. (3. on the other. It involves all the complexities of the use. namely: /» M( » 2 (3.945) /« M(T) e'n- » and Generated on 2011-09-28 14:51 GMT / Public Domain. A. Fisher and J. von Neumann.946) that of the noise.

or short wave length. and in particular. In the complete Gibbsian theory it is still true that with a perfect determination of the multiple time series of the whole universe. this chapter should end with a discussion of modern quantum mechanics. In the Newtonian physics. we must observe it with light or electron-waves or similar means of high resolving power. The great contribution of Heisenberg to physics was the re placement of this still quasi-Newtonian world ofGibbsby one in which the time series can in no way be reduced to an assembly of determinate threads of development in time. except in a loose and approximate way. by the de termination of all positions and momenta at anyone moment.110 CYBERNETICS honestly able to say that he lias said nothing which is not per fectly rigorous and unimpeachable. non-observed coordinates and momenta that Ihe time series with which we actually work take on the sort of mixing properly with which we have become familiar in this chapter. the sequence of physical phenomena is completely determined by its past. in the case of time series derived from the Brownian motion.hathitrust. and to illuminate a body with high-frequency light means to subject it to a change in its momentum which increases with the frequency. it is low-frequency light that gives the minimum change in the momenta of the particles it illuminates. To observe the position of a system as precisely as possible.org/access_use#pd-google system. In quantum mechanics. are not simul taneously observable. the knowledge of all positions and momenta at any one moment would determine the entire future. but merely the distribution of possible futures of Ihe Generated on 2011-09-28 14:51 GMT / Public Domain. These represent the highest point of the invasion of modern physics by the theory of time series. the light has a particle action depending on its fre quency only. Google-digitized / http://www. It is only because these are ignored. The quantities which the classical physics demands for a knowledge of Ihe entire course of a system. Finally. However. which nevertheless is sufficiently precise for the needs of the classical physics over the range of precision where it has been shown experimentally to be applicable. the whole past of an individual system does not determine the future of that system in any absolute way. The conditions of the observation of a momentum and its corresponding position are incompatible. On the other hand. and this has not a sufficient resolving .

in the language of the quantum-theorist. However. in a tentative manner. and consequently Ihe theory of entropy. a situation where many of these chains are alike. AND COMMUNICATION 111 power to give a sharp indication of positions. the probability of this depends in the long run on the relative probability or measure of the two states. Nevertheless. This turns out to be especially high for states which can be transformed into themselves by a large number of transformations. Since however we now are dealing with time series with the mixing pro perly even when our dataareas completeas they can be. In general. INFORMATION. since the two states are equivalent. This suggests that in a and |l | \/ system inwhich various building-blocks may combine themselves Generated on 2011-09-28 14:51 GMT / Public Domain. have a high internal resonance. that this may be the way in which genes and viruses reproduce themselves . or a high quantum degeneracy. . ajid which is Ihe copy. for states which. It was suggested by Haldane. we find that our system has no absolute potential barriers. there is no set of observations con ceivable which can give us enough information about the past of a system to give us complete information as to its future. as when a mixture of amino-acids organizes itself into protein chains.hathitrust. and go through a stage of close associa tion with one another. as in the case of all ensemblesof time series.TIME SERIES. it is not possible to say in such a case with more than fragmentary accuracy. Intermediate frequencies of light give a blurred account both of positions and of momenta. Google-digitized / http://www. the theory of the amount of information which we have here developed is applicable. and that in the course of time. which of the two exam ples of a gene that has reproduced itself in this manner is the master pattern. and although he has not asserted this suggestion of his with anything like finality. I see no cause not to retain it as atentative hypothesis. As Haldane himself has pointed out. sharp individuality. may be more stable than one in which they are different. The benzene ring is an example of this.org/access_use#pd-google intimately in various ways. any state of the system can and will transform itself into any other state. as no single particle in quantum theory has a per fectly.

Such considerations may be very important in Generated on 2011-09-28 14:52 GMT / Public Domain. and such a storoge certainly occurs in muscle contraction.112 CYBERNETICS This same phenomenon of resonance is known to be very frequently represented in living matter. Subs tances with high resonance very generally have an abnormal capacity for storing both energy and information. but even within the individuals of a species. Google-digitized / http://www. not only from species to species. .org/access_use#pd-google immunology. the same phenomena that are concerned in reproduc tion probably have something to do with the extraordinary specificity of the chemical substances found in a living organism.hathitrust. Szent-Gyorgi has suggested its importance in the construction of muscles. Again.

Give him a glass of water. However. there seems to be nothing wrong with him. This will be followed by an equally futile swing in the other direction.CHAPTER IV FEED-BACK AND OSCILLATION A patient comes into a neurological clinic. What is the matter with him ? Both of these patients are suffering from one form or another of what is known as ataxia. He starts each step with a kick. He walks with a peculiar uncertain gait. The first patient suffers from tabes dorsalis. He is not para lyzed. If blindfolded. with eyes downcast. throwing each leg in succession in front of him.hathitrust.org/access_use#pd-google maged or destroyed by the late sequelae of syphilis. and he will empty it in these swings before he is able to bring it to his mouth. on the ground and on his legs. and totters to the ground. but they are unable to organize their actions. and this by still a third swing back. Google-digitized / http://www. if they have not totally disap peared. offer him a cigarette. The incoming messages are blunted. and he can move his legs when he receives the order. Nevertheless. and he will swing his hand past it in trying to pick it up. he cannot stand up. until his motion becomes nothing but a futile and violent oscillation. he suffers under a severe disability. Their muscles are strong and healthy enough. The part of the spinal cord which ordinarily receives sensations has been da Generated on 2011-09-28 14:52 GMT / Public Domain. The receptors in the joints and tendons and muscles and the soles of his feet which ordinarily convey to him the position and state of motion of his legs send no messages which his central nervous system can pick up and transmit. and for information concerning his posture he is obliged to trust to his . What is the matter with him ? Another patient comes in. While he sits at rest in his chair.

which conveys to the signalman its actual states and performance.hathitrust. and that the readings of these mon itors be properly combined with the other information coming in from the sense organs to produce a properly proportioned output to the effectors. It seems likely that the cerebellum has some function of propor tioning the muscular response to the proprioceptive input. switch. However. It may be that the switches have frozen fast. and which regulate the setting of the switches. To avoid the dangers inherent in this contin gency. upon the reception of an order. This is the mechanical equi valent of the repeating of orders in the navy. it is not only essential that we possess good effectors. Something quite similar is the case in mechanical systems. to show that he has heard and understood it. The second patient has lost none of his proprioceptive sense. must repeat it back to his superior. Google-digitized / http://www. a tremor may be one of the results. In the jargon of the physiologist. and that what he has supposed to be the actual state of the switches and the signals — his effectors — does not correspond to the orders he has given. in the cerebellum. and he is suffering from what is known as a cerebellar tremor or purpose tremor. but that the performance of these effectors be properly monitored back to the central nervous system. according to a code by which every subordinate. It is true that the signalman is not altogether a free agent . is attached to a tell-tale back in the signal tower. and if this proportioning is disturbed.org/access_use#pd-google the weight of a load of snow has bent the signal arms. it does not do for him to assume blindly that the signals and the switches have followed his orders.114 CYBERNETICS eyes and the balancing organs of his inner ear. We thus see that for effective action on the outer world. Let us consider a signal tower on a railroad. His injury is elsewhere. or that Generated on 2011-09-28 14:52 GMT / Public Domain. or signal. that his switches . he has lost an important part of his proprio- ceptive or kinaesthetic sense. The signalman controls a number of levers which turn the semaphore signals on or off. It is on such repeated orders that the signalman must act. Notice that in this system there is a human link in the chain of the transmission and return of information : in what we shall from now on call the chain of feed-back. every effector.

or increases the flow of fuel oil. as in the case of the steering engines of a ship. We have thus examples of negative feed-backs to stabilize temperature and negative feed-backs to stabilize velocity. which actuates a member which serves to open the intake valves of the cylinder when the engine slows down and ihe balls fall. and that he is not free to choose some of the more disastrous combinations. This position is transmitted by other rods to a collar about the shaft. and to close them when the engine speeds up and the balls rise. There are however feed-back chains in which no human element intervenes.org/access_use#pd-google are kept down by their own weight or by a spring. There is a setting for the desired room temperature . and is thus negative. it consists of two balls attached to pendulum rods and swinging on opposite sides of a rotating shaft.FEED-BACK AND OSCILLATION 115 and signals are interlocked. and if the actual temperature of the house is below this. In this way the temperature of the house is kept approximately at a steady level. Note that the constancy of this level depends on the good design of the thermo stat. either mechanically or electrically. They Generated on 2011-09-28 14:53 GMT / Public Domain. Google-digitized / http://www. The ordinary thermostat by which we regulate the heating of a house is one of these. and that a badly designed thermostat may send the tem perature of the house into violent oscillations not unlike the motions of the man suffering from cerebellar tremor. and always act so as to bring the position . In the original form designed by Watt. the dampers are turned off. or the flow of fuel oil is slackened or interrupted. an apparatus is actuated which opens the damper. They thus assume a compromise position likewise dependent on the angular velocity. Another example of a purely mechanical feed-back system — the one originally treated by Clerk Maxwell — is that of the governor of a steam-engine. and they are swung upwards by a centrifugal action dependent on the angular velocity of the shaft. which are actuated by the angular difference between the position of the wheel and the position of the rudder. There are also negative feed-backs to stabilize position.hathitrust. Notice that the feed-back tends to oppose what the system is already doing. and brings the temperature of the house up to the desired level. If on the other hand the temperature of the house exceeds the desired level. which serves to regulate its velocity under varying conditions of load.

but not when s is greater than /. of its defective behavior and its breaking into oscillation when it is mishandled or over loaded. we also add outputs. we will. for an input /(/). In this book. too. . This reading is simply subtracted from the input- We wish to give a precise theory of the performance of such a piece of apparatus . There are pieces of apparatus. The information fed back to the control center tends to oppose the departure of the controlled from the controlling quantity. which delay their input by a fixed time.t). and indeed we generally do not know which muscles are to be moved to accomplish a given task. and in parti cular in the last chapter.org/access_use#pd-google the symbolism of mathematics is the appropriate language. We may combine several pieces of apparatus of this kind. Our motion is regulated by some measure of the amount by which it has not yet been accomplished. we have avoided mathematical symbolism and mathematical technique as far as possible. but it may depend in widely different ways on this departure. The simplest control systems are linear : the output of the effec tor is a linear expression in the input. and when we add inputs. The feed-back of voluntary activity is of this nature. and these yield us. to pick up a cigarette. The output is read by some apparatus equally linear. Let/"(/)bea function of the time/ where / runs from minus infi nity to infinity : that is. and. and which are only intelligible to the reader acquainted with mathematical symbolism by virtue of his ability to translate them into this symbolism. Google-digitized / http://www. where T is the fixed delay. The best compro mise we can make is to supplement the symbolism by an ample verbal explanation. in the rest of the present chapter. Here. an output f(t. At any time /.hathitrust. electrical and mechanical. We do not will the motions of certain muscles. say. the quantities/"(s) are accessible to us when s is less than or equal to /. in particular. we are dealing precisely with those matters for which Generated on 2011-09-28 14:53 GMT / Public Domain. and we can only avoid it by long periphrases which are scarcely intelligible to the layman. although we have been forced to compromise with them in various places.116 CYBERNETICS of the rudder into accord with that of the wheel. let /"(/) be a quantity assuming a nume rical value for each time t.

01) £ an /•(*-*„). For example. This however involves the knowledge of the future of f(t). Wecan multiply each of these outputs by fixed quantities. There are physical cases where this may occur.FEED-BACK AND OSCILLATION 117 yieldingus outputs f(t-ct). deter mined uniquely by its past. with an undetermined amplitude. positive or negative.03) /"" where a(t) does not effectively vanish for negative values of /.02) /"" a In this expression. or even oscillation building up to infinity. a dynamical system with no input may go into permanent oscillation. and to obtain f(t + a). i By increasing the number of delays Tn. and not from — oo to oo. Google-digitized / http://www. we may approximate as closely as we wish to an output of the form : (4. we may obtain an output : (4.. and with the aid of these. Otherwise we could use various practical devices to operate on this result. and f(t) may be a quantity. and it is not too difficult to devise automatic balancing devices and amplifiers to multiply a voltage by quantities which are negative. it means that we have no longer a true operator onf(t).f(t-tt).org/access_use#pd-google yield us an operator (4. where a is posi tive.f(t-tn). or are greater than one. When a physical process seems to Generated on 2011-09-28 14:53 GMT / Public Domain. it is important to realize that the fact that we are integrating from 0 to oc.hathitrust. and suitably adjusting the coefficient an.. we may use a potentiometer to multiply a voltage by a fixed positive number less than one. For example. like the coordinates of a street car which may turn off one way or the other at a switch. It is also not difficult to construct simple wiring diagrams of circuits by which we can add voltages continuously. is essential.. In such a case the future of the system is not determined by the . which is not determined by its past..

( 0 (2 < X).T) dT. and (4.05) g(t + «) = J*" O(T) f(t + a .08) where Generated on 2011-09-28 14:53 GMT / Public Domain. The operation by which we obtain (4. ( 1 (0<X<1).02). 0 The second property is expressed by the statement that if then (4. eyery operator on the past of f(t) which is linear and is invariant under a shift of the origin of time is either of the form (4. As we have seen before.org/access_use#pd-google (4.07) f™ a(t) g(t — T) dT = A f °° O(T) /\(/ — T) dT ! — T)dT.hathitrust. and (2) it is linear. For example.09) a (X) = — 1 (1 < X < 2) . or is a limit of a sequence of operators of that form. and we may in appearance find a formalism which suggests an operator dependent on the future.02) from f(t) has two important further properties : (1) it is independent of a shift of the origin of time . 0 (4.T) dT. Google-digitized / http://www. It may be shown that in an appropriate sense.04) g(t) = f O(T) f(t . /"'(/) is the result of an operator with these properties when applied to (/). The first property is ex pressed by the statement that if then (4. the functions ezt are a set of func tions .118 CYBERNETICS past.

x + iy.12) / a (T) e-zr dr = A (Z) o rx is said to be the representation of the operator (4. since (4.15) I Ate +/y) I <) / I O(T) 2 dr / e—^dft ( /"*°° /""°° }l/2 (J J ) 00 = j^y°°i«wi2^|l/2 0 This means that A (x + jy) is a bounded holomorphic func tion of a complex variable in every half-plane x >e > 0. if y > 0 and (4.14) / | a(T) I2 dT < oo. cr-fr.02). so that by the well-known Schwarz inequality concerning inte grals. If Z is taken as the complex quantity. . where x and y are real.02) as a function of frequency.FEED-BACK AND OSCILLATION 119 f(t) which are particularly important from the point of view of operator (4.16) u + iv = \(x+iy). o we have : Generated on 2011-09-28 14:54 GMT / Public Domain.hathitrust. Thus (4.10) ez(«-T) = eu .org/access_use#pd-google (4.11) ez< o f a (T) e-z* dT. this becomes.13) / O(T) e— °* eiy dT. Let us put: (4.02) becomes : (4. and the delay operator becomes merely a multiplier dependent on Z. Google-digitized / http://www. The expression (4. /-» (4. and is also a multiplication operator dependent on Z only. and that the function A(/'y) represents in a certain very definite sense the boundary values of such a function.

hathitrust.org/access_use#pd-google . Google-digitized / http://www.Generated on 2011-09-28 14:54 GMT / Public Domain.

which is the difference between the original input X.FEED-BACK AND OSCILLATION 121 Subtracter Input X =X. 2. This will be infinite when and only when A = — 1/x. The diagram (4.X Ay Motor operatorA Multiplier operator \ XAy Fio. and the output of the multiplier. (4. In this case.18) and Y = X — XAY. and as a matter of fact the . Thus: (4.X The operator produced by the whole feed-back mechanism is then A/(l + XA).21) u + i v = A/\ and ao will be an interior point of this when and only when —1/x is an interior point of (4.org/access_use#pd-google (4. a feed-back with a multiplier X will certainly produce something catastrophic.20) AY .hathitrust. Here the input of the motor is Y.19) so that the motor output is Y= X 1+ xA' +xA Generated on 2011-09-28 14:54 GMT / Public Domain. which multiplies the power output AY of the motor by the factor X.17).17) for this new operator will be (4. Google-digitized / http://www.

02). assuming that the same argument will apply to these. let (4.+<.l__ . We shall not only consider operations (4. The CT(T) corresponding to this operator is : (4. It is des cribed in the clockwise sense.. Google-digitized / http://www. In this case too the admissible feed-back is unlimited. I •> a * 4 i lo Generated on 2011-09-28 14:55 GMT / Public Domain. It is perhaps worth considering several operators A and the ranges of feed-back which are admissible under them. and center at (1/2. it maybe shown that there will be no difficulty.hathitrust. as y goes from — oo to oo. and the interior points are those which we should ordinarily consider interior.22) the curve (4. .26) a(t) = Again.25) u8 + v2 = u.27) A(z) = Then (4. a more elaborate discussion is necessary. the system may go into an oscillation of an amplitude which does not increase. Under most circumstances.17) is . . This is a circle with radius 1/2. as —I/A is always outside the circle.— ^ \^**** *•* ~~~ j .23) u + i v = td. If — 1/X is on the effective boun dary. If the operator A corresponds to the differential operator. A(y) does the same. 24. and the interior points are the points interior to the right half-plane. The point — 1/X is always an exterior point.122 CYBERNETICS catastrophe will be that the system will go into unrestrained and increasing oscillation.4. but also their limits. A(z) = z.org/access_use#pd-google which we may write : (4.17) is 1 + If i y' or (4.i '(i-*'*>• ' . and the feed back is stable.0). If (4.28) . and any amount of feed-back is possible. If on the other hand the point — 1/X is an exterior point.

hathitrust. * (4. £* 2t It can be shown that these two equations represent only one curve.34) p — = ± — That is.FEED-BACK AND OSCILLATION 123 and (4.. The interior of this curve will contain no point of the negative real axis.33) pcos9 = P'—™!L = p-.. a cardioid with vertex at the origin and cusp pointing to the right. if u = p cos 9.org/access_use#pd-google sible amplification is unlimited.* + £2J5L. the admis Let Generated on 2011-09-28 14:55 GMT / Public Domain.29) «=/" ~ This yields : (4.35) ?*/2 = — sin -^. this becomes (4. Google-digitized / http://www.36) a(0 = 1•+ k Let p and 9 be defined as in the last case. Here the operator «(/) is : ^4. (4..38) PV3 cos -J + i P1/3 sin -1 = . .30) u2 + v2 = —— or . and as in the previous case.37) . pi/2 = cos — . or 4 44 /A *A\ COS f .31) y= -.32) v~ ' ' 4(uz+ !>»)' In polar coordinates. Then (4. A (Z) (4. v = p sin 9.—~V (4. !*" Then (4.

org/access_use#pd-google . Google-digitized / http://www.hathitrust.Generated on 2011-09-28 14:55 GMT / Public Domain.

45) u + i v = — k i y3. each of which may be compensated as well as we wish by a single feed-back. It concerns the steering of a ship by a gyro compass. produces a turning moment which serves to change the course of the ship in such a way as to decrease the difference between the set course and the actual course. can not itself be so compensated. The angle between the course set by the quartermaster and that shown by the compass expresses itself in the turning of the rudder. by a single feed-back.17) (4. and the force turning the ship does not determine the acceleration. we have slightly oversimplified the steering problem.FEED-BACK AND OSCILLATION 125 feed-backs. no servo-mecha nism whatever will stabilize the system. we have an example of a complicated system which can be stabilized by two feed-backs. however. which in view of the headway of the ship. and may be regarded as the limit of the additive composition of three operators with first-degree denominators. In the important book of McColl. and the operation which we have to stabilize by the feed-back from the gyro-compass is kz9. In this account.hathitrust. We thus get for our curve (4. Google-digitized / http://www. which is the resultant of the composition of three operators 1/(1 + kz) in cascade. Hence the amount of turning of the ship is proportional with a negative factor to the third derivative of the deviation from the course.org/access_use#pd-google us note that the angular position of the rudder is roughly proportional to the turning-moment of the ship and thus to its angular acceleration. where k is positive. If this is done by a direct opening of the valves of one steering-engine and closing of the valves of the other in such a way lhat the turning-velocity of the rudder is proportional to the deviation of the ship from this course. The operator 1/(1 + kz)9 may also be written: (4>44) ~2F~d^~fTH. . Actually there is a certain amount of friction. but not by one. possible to compensate as closely as we wish for an operator 1/(1 + kz)3. It is not. and as the left-half plane is the interior region. It thus appears that a sum of different operators. let Generated on 2011-09-28 14:55 GMT / Public Domain.

and (4. but in this case. In the human body. the operator to be stabilized by feed-back is klzt + kz. The output is an addi tive vectorial combination of the outputs of all these joints. the . and the inside of the curve is to the left. if we open the valves wide enough.48) u2 = — /f.org/access_use#pd-google course. To achieve this we may employ another stage of feed-back. We have seen that in general. Google-digitized / http://www. but by the difference between this quantity and the angular position of the rudder.17) becomes: This curve may be written (4. and the curve is described from y = — oo to . not by the discrepancy between the actual and the desired Generated on 2011-09-28 14:55 GMT / Public Domain. Correspondingly. the motion of a hand or a finger involves a system with a large number of joints. If we regulate the position of the valves of the steering engine. This curve may be written : (4. If on the other hand the position of the rudder is proportional to the deviation of the course. the outside of the curve is to the left.47) u + i v = = — kj y3 — kty2. we have : f\ tR\ d2e rfe and (4'46) -dtr=c*-c2-dT (4. This double feed-back system of control is in fact the one usually adopted for the automatic steering of ships by means of the gyro-compass.hathitrust.50) u2 = — k tu. so does v. which slill cannot be stabilized by any feed-back. a complex additive system like this cannot be stabilized by a single feed-back. as y goes from — oo to oo. if o is the angular position of the ship and 9 that of the rudder with respect to the ship. we shall keep the angular posi tion of the rudder as nearly proportional to the ship's devia tion from true course as we wish.126 CYBERNETICS Instead. if we allow a large enough feed-back — that is. As y goes from — oo to oe. and unlimited amount of amplification is possible. u goes from oe to — oo. In this case.us./ = oo.

org/access_use#pd-google feed backs. It is the voluntary feed-back which shows a tendency to break down or become deranged in cases of cerebellar injury. If the ope rator after feed-back is . When feed-back is possible and stable. and many of these must have their origin in defects of parts of the nervous system situated very differently. Parkinsonianism is known not to have its origin in a diseased condition of the cerebellum. and indeed often seems to be greatly mitigated when he attempts to perform a specific task. Let us consider that the load changes the characteris tic A by dA. is very different in nature from the tremor of Parkinsonianism or paralysis agitans. which appears in its most typical form when the patient is at rest. One of the great tasks of physiological cybernetics is to disentangle and isolate loci of the different parts of this complex of voluntary and postural Generated on 2011-09-28 14:56 GMT / Public Domain. we shall have : B= CTA" dB "V* ' A y AT dA C B i+JL i + Jl A A+C Thus feed-back serves to diminish the dependence of the system . for the ensuing tremor does not appear unless the patient tries to perform a voluntary tasx. This purpose tremor. There are surgeons with Parkinsonianism who manage to operate quite efficiently.hathitrust. Examples of component reflexes of this sort are the scratch and the walking reflex.FEED-BACK AND OSCILLATION 127 voluntary feed-back by which we regulate the performance of a task through the observation of the amount by which it is not yet accomplished needs the backing up of other feed-backs. as we have already said. in which the patient can not pick up a glass of water without upsetting it. The fractional change will be dA/A. but to he associated with a pathological focus somewhere in the brain-stem. is to make performance less dependent on the load. Google-digitized / http://www. It is only one of the diseases of the postural feed-backs. and they are associated with the general maintenance of tone of the muscular system. its advantage. These we call postural feed-backs.

will be to increase the stability of the system for low frequencies. The existence of a periodic non-sinusoidal oscillation is always a suggestion at least that the variable observed is one in which the system is not linear. A linear oscillating system has certain very special properties which characterize its oscillations. One is that when it oscil lates.hathitrust. This will not even be true in the first of the cases we have discussed. and serves to stabilize it. studied from the point of view of feed-back. the system may be rendered linear again by a new choice of the independent variable. but in very few. There are many cases in which even this degree of stabilization is advantageous.org/access_use#pd-google form (4.54) A sin (B* + C) e™.17) lying furthest to the left on the negative u-axis The quantity y is of course of the nature of a frequency. This is determined by the value of y in the iy corresponding to the point of the boundary of the inside and outside regions of (4. if it is at all stable. Another very significant difference between linear and non linear oscillations is that in the first. for all frequencies for which This is to say that the entire boundary between interior and exterior points must lie inside the circle of radius C about the point — C. We have now come to the end of an elementary discussion of linear oscillations. the amplitude of oscillation is completely independent of the frequency . there is generally only one amplitude. while in the latter. Google-digitized / http://www. In some cases. it always can and very generally — in the absence of independent simultaneous oscillations — does oscillate in the Generated on 2011-09-28 14:56 GMT / Public Domain. The effect of a heavy negative feed-back. or at most a discrete set of amplitudes for which the system will oscillate at a given .128 CYBERNETICS on the characteristic of the motor. bul generally at the expense of its stability for some high frequencies. A very important question which arises in connection with oscillations due to an excessive amount of feed-back is that of the frequency of incipient oscillation.

and the level of oscillation is completely indeterminate. This is well illustrated by the study of what happens in an organ pipe. but for certain velocities of air-flow. this motion will actually increase the coupling of the proper modes of oscillation of the pipe with the energy input.hathitrust. There is indeed a theoretical steady-state flow of air across the lip of the pipe which does not interchange any energy with any of the modes of oscillation of the pipe. No question is asked about how the pipe came to oscillate. In the case we have discussed. In the first. Google-digitized / http://www. that is. Thus the level of the non linear oscillation is determined just as definitely as its frequency.org/access_use#pd-google The case we have examined is an example of what is known as a relaxation oscillation : a case.FEED-BACK AND OSCILLATION 129 frequency. these two quantities must be identical. nearly linear part of the system. B. but not in phase. and that there are in fact relaxation oscillations where the predominating frequency is not near the frequency of linear oscillation of any part of the system. An example is given by a stream of gas flowing into a chamber open to the air and in which a pilot light is burning : when the concentration of gas in the air reaches a certain critical value. the system is ready to explode under . The rate of energy input and the rate of energy output by thermal dissipation and otherwise have different laws of growth. leads to a solu tion periodic — or corresponding to some generalized notion of periodicity — in time. the oscillation of the organ pipe is consid ered as dissipating energy. this steady-state condition is unstable. Generated on 2011-09-28 14:56 GMT / Public Domain. as well as a discrete set of frequencies for which the system will oscillate. the organ pipe is treated as a conser vative system. and this energy is considered to have its origin in the stream of air across the lip of the pipe. The slightest chance deviation from it will introduce an energy input into one or more of the natural modes of linear oscillation of the pipe . the frequency of oscillation of the system is close to that of some loosely coupled. and a more precise non linear theory. has pointed out that this is not always the case. There are two theories of the organ pipe — a cruder linear theory. one of the chief authorities on relaxation oscillations. and up to a certain point. In the second theory. but to arrive at a steady state of oscillation. where a system of equations invariant under a translation in time. Van der Pol. and determinate in amplitude and frequency.

org/access_use#pd-google brium is reached.. the rate at which air seeps in and the products of combustion seep out. until a stale of equili Generated on 2011-09-28 14:56 GMT / Public Domain. let 1. and the theory of secularly perturbed systems plays a most important role in gravitational astronomy. as far as I know. In this case. For example. There is however a specially tractable case. when the system depends for its future behavior on its entire past behavior. There is not. in which the system differs only slightly from a linear system. and the time it takes for this to happen depends only on the rate of flow of the coal gas. The classical cases for the study of such oscillations are those in which the equations of the systems are of a differential nature. we may study the non-linear system as if it were a linear system with slowly varying parameters. Then as the oscillation of the system builds up. However. non-linear systems of equations are hard to solve. Google-digitized / http://www. . We can see quite clearly in such a system why the steady-state amplitude level may be just as determinate as the frequency. any comparable adequate study of the corresponding integral equations. Non-linear systems of relaxation oscillations have been stu died in some cases by methods developed by Hill and Poincare1. Systems which may be stu died this way are said to be perturbed secularly. H. the gain may be cut down. Les \lethodes Noiwelles Jans la Mecanique Celeste. In general. it is not hard to sketch out the form such a theory should take. Let one element in such a system be an amplifier whose gain decreases as some long-time average of the input of such a system increases. In this case the slight modification of the constants of the equation should lead to a slight. and therefore nearly linear. and the terms which distinguish it change so slowly that they may be considered substantially constant over a period of oscillation. It is quite possible that some of the physiological tremors may be treated somewhat roughly as secularly perturbed linear systems.hathitrust. Poincare.130 CYBERNETICS ignition by the pilot light. especially where these differential equations are of low order. modification of the equations of motion. especially when we are only looking for periodic solutions. and the percentage composition of an explosive mixture of coal gas and air.

FEED-BACK AND OSCILLATION 131 Op I f(t) \ be a function of t which results from a non-linear operation on f(t). Google-digitized / http://www.58) Sf(t) .56) /"(/) = 2 anei«^. and which is affected by a translation.00 and /"(/) + 8 /"(/) is also periodic. The relative usefulness of the two methods of . lies in the future.55) Op [ /•(/) I = 0 and we change the dynamics of the system. This work. In this case it is at least conceivable that by starting with a linear equation (non-homogeneous) and gradually shifting the constraints. We shall thus obtain an infinite system of linear non- homogeneous equations in 8an+an in 8 x. and x and this system of Generated on 2011-09-28 14:56 GMT / Public Domain. and has a performance relatively inde pendent of the characteristic and changes of characteristic of the effector used. Then the variation of Op \ f(t) \. They both serve to bring the complicated input-output relations of an effectorinio a form approaching a simple proportionality. does more than this. being of the form : 00 (4.org/access_use#pd-google equations may be solvable by the methods of Hill. however. 8 Op \ f(t) \ corresponding to a varia- tional change 8 f(t) in f(t) and a known change in the dynamics of the system. The feed-back system. the feed-back systems of control discussed in this chapter and the compensation systems discussed in the last are competitors. we obtain a linear non-homogeneous equation for 8 /*(/). though not in f(t). If we now know a solution /(/) of (4.hathitrust.57) f(t) + 8f(l) = 2 (a„ + 8a„)e««rA+sx)* . . If 00 (4.= ]£ 8ane*»t + E The linear equations for 8 f(n) will have all coefficients develop able into series in elX«'. as we have seen.--. we may arrive at a solution of a very general type of non-linear problem in relaxation oscillations. is linear but not homogeneous in 8/(/). — oo then OO CO (4. since f(t) can itself be developed in this form. To a certain extent.

There are various ways of doing this.org/access_use#pd-google example. for Generated on 2011-09-28 14:57 GMT / Public Domain. On the other hand. One of the most simple is that illustrated in the following diagram : Subtracter In this. It is natural to suppose that cases arise in which it is advantageous to combine the two methods. the effector has an essentially lagging characteristic. Another type of arrangement is the following: Subtracter FIG. and it is not easy to see how it can ordinarily be made to increase that level to an important extent.132 CYBERNETICS control thus depends on the constancy of the characteristic of the effector. This change will in general alter the maxi mum feed-back admissible. If. Output Here the compensator and effector are combined into one larger effector. except that Ihe compensator must be arranged to compensate what is in some sense the ave rage characteristic of the feed-back system. for the same feed-back level. 5. it will most definitely improve the performance of the system. Google-digitized / http://www. and no new point arises. the entire feed-back system may be regarded as a larger effector.hathitrust. .

We explore the amplitude-phase rela tions of the high-frequency output to the input in order to obtain the performance characteristics of the effector. is not difficult to schematize into a mechanical form. will tend to hurry up the action of the effector mechanism. On the basis of this.hathitrust.and may well be worth while employing in practice. We super impose on the incoming message a weak high-frequency input. on a knowledge of the performance characteristics of the system car-road. not enough to throw the car into a major skid. Google-digitized / http://www. but that between the position of the gun and the anticipated position of the target. Any system of anti-aircraft tire control must meet the same pro blem. we shall discover ourselves in a skid before we know it. The flow-chart of the system is much as in the following diagram : .FEED-BACK AND OSCILLATION 133 the compensator will be an anticipator or predictor. We thus give to the steering-wheel a succession of small fast impulses. Generated on 2011-09-28 14:57 GMT / Public Domain. but quite enough to report to our kinaesthetic sense whether the car is in danger of skidding. designed for its statistical ensemble of inputs.org/access_use#pd-google and we regulate our method of steering accordingly. We have a compensator for our effector. separated from the rest of the output by an appropriate filter. If we wait to find this out by the ordinary performance of the system. The conditions of stability and effectiveness of anticipa tory feed-backs need a more thorough discussion than they have yet received. we modify in the appropriate sense the character istics of the compensator. Our entire conduct of driving depends on a knowledge of the slipperiness of the road-surface : that is. which we may call control by infor mative feed-back. When we go duck-shooting. Another interesting variant of feed-back systems is found in the way in which we steer a car on an icy road. Feed-backs of this general type are certainly found in human and animal reflexes. and lake off the output of the effector a partial output of the same high frequency. the error which we try to minimize is not that between the position of the gun and the actual position of the target. Our feed-back. This method of control. and this compensator has a characteristic which may be varied from outside. which we may call an anticipatory feed-back.

Of course. the system has no tendency to go into oscillation. There are very many cases where the change of load is secular in this manner. and that. and hence of the effector. This informative feed-back and the examples we have given of feed-back with compensators are only particular cases of what is a very complicated theory. This will often be the case if the character of the load. A great group of cases in which some sort of feed-back is not only exemplified in physiological phenomena. this informative feed-back will only work well if Generated on 2011-09-28 14:57 GMT / Public Domain. Before we end this chapter. For example.hathitrust. oscillator)Compare"* determine character1. we must not forget another im portant physiological application of the principle of feed-back.134 CYBERNETICS FIG. or give a good indication of.org/access_use#pd-google the characteristics of the load at high frequencies are the same as. in comparison with the changes of the original input. contains a relatively small number of variable parameters. if the characteristic of the load changes slowly enough. and if the reading of the load condition is accurate.or to sffector istics|l High Pass H Filter%^Effectori The advantages of this type of feed-back are that the compen sator may be adjusted to give stability for every type of cons tant load . The whole field is undergoing a very rapid develop ment. 6. but is absolutely . its characteristics at low fre quencies. but this stiffness will not change appreciably in a few swings of the turret. It deserves much more attention in the near future. and a theory as yet imperfectly studied. the frictional load of a gun turret depends on the stiffness of the grease. Google-digitized / http://www. and this again on the temperature . in what we have called a secular manner. Original inputCompeControl compensatorisatorndetermining characteristicAdderF.

etc.org/access_use#pd-google nent damage in a small fraction of a second. These are what we know collectively as our homeostatic mechanism. There are very few changes in physiological homeostasis — not even cerebral anemia — that produce serious or perma Generated on 2011-09-28 14:57 GMT / Public Domain. is found in what is known as homeostasis. and a permanent variation of <ive degrees is scarcely consistent with life. are quite narrow. our heart rate and blood pres sure must neither be too high nor too low. many .FEED-BACK AND OSCILLATION 135 essential for the continuation of life. the nerve fibres reserved for the processes of homeostasis—the sym pathetic and parasympathetic systems — are often non-myeli- nated. Google-digitized / http://www. . automatic hydrogen- ion-concentration controls. Many of the messages of the homeostatic system are carried by non-nervous channels — the direct anastomosis of the muscular fibres of the heart. which would be adequate for a great chemical plant. governors.our sex cycle must conform to the racial needs of reproduction . can continue in the higher animals. our leucocytes and our chemical defenses against infection must be kept at adequate levels . our calcium metabolism must be such as neither to soften our bones nor to calcify our tissues . our inner economy must contain an assembly of thermostats. The typical effectors of homeostasis — smooth muscles and glands — are likewise slow in their action compared with striped muscles. Any complete text-book on cybernetics should contain a thorough detailed discussion of homeostatic processes. The conditions under which life. and except in the case of the heart muscle. Besides all these. the typical effectors of voluntary and postural activity. the carbon dioxide content of the blood. A variation of one half degree centigrade in the body tempera ture is generally a sign of illness. The waste products of the body must be excreted before they rise to toxic concentrations. In short. or chemical messengers such as the hormones. and the like. these too are generally slower modes of transmission than myelinated nerve fibres. Our homeostatic feed-backs have one general difference from our voluntary and our postural feed-backs: they tend to be slower. and so on. and are known to have a considerably slower rate of trans mission than the myelinated fibres. especially heal thy life. The osmotic pres sure of the blood and its hydrogen-ion concentration must be held within strict limits. Accordingly.hathitrust.

W.1. . .org/access_use#pd-google the Environment. Cannon.136 CYBERNETICS individual cases of which have been discussed in the literature with some detail1. However. and the theory of homeostatic processes involves rather too detailed a knowledge of general physiology to be in place here. The Fitness of Generated on 2011-09-28 14:58 GMT / Public Domain. Henderson. L. this book is rather an introduction to the subject than a compendious treatise.hathitrust. Google-digitized / http://www. 1. The Wisdom of the liodij ..

org/access_use#pd-google (5. since if n-j regions are accurately established. each marked less accurately. with a pointer of some sort moving over this. Now let us divide this information over two scales. More precisely. The cost of recording this information will be about : i Generated on 2011-09-28 14:58 GMT / Public Domain. both in money and in the effort of construction. If we wish to record a number with an accuracy of one part in n. and giving the result in nume rical form.01) (1( — 1} A. we must finish each part of the movement of the pointer with this degree of accuracy. the approximate cost will be : (5. the remaining region will also be determined accurately. and the cost will be of the form An. the pointer assumes the desired position within this accuracy. Google-digitized / http://www. If the information be divided among N scales. goes to the simple problem of recording numbers clearly and accurately.03) N — IA. for an amount of information Iog2 n. .hathitrust.02) 2\V2S — I/A. That is. we have to assure that in each region of the scale. where A is not too far from a constant. operating with numbers. The simplest mode of doing this seems to be on a uniform scale.CHAPTER V COMPUTING MACHINES AND THE NERVOUS SYSTEM Computing machines are essentially machines for recording numbers. A very considerable part of their cost. the cost of recording an amount of information I will be about (5.

Various papers. x = 0. and that 1 is not a significant value. and in which the probability of an imperfect knowledge as to which half of the scale contains the observation is made vanishingly small. Google-digitized / http://www. when (5. and the accuracy is determined by the sharpness with which the contingencies are distinguished. or N = oc . each divided into two equal parts. Journal of the Franklin Institute. where the data are represented by mea surements on some continuous scale.07) u=Uo+_Ul + -^"2+ ••• + 27U«+"" where every un is either 1 or 0. In other words. The best significant value for 2i/N js 2. That is.org/access_use#pd-google There exist at present two great types of computing machines : those. so that the accuracy of the machine is determined by the accuracy of construction of the scale . Let us remember that 2i/N must be an integer.06) This will occur when.04) 2 N — 1 = ^ 2 * log 2. or if we put: -L I -L (5. we represent our numbers in the binary system on a number of scales in which all that we know is that a certain quantity lies in one or the other of two equal portions of the scale.hathitrust. which are known as analogy machines. In other words. . the number of alternative 1. as we then have an infinite number of scales each containing no information. N should be as large as possible to give the lowest cost for the storage of information. we represent a number u in the form : (5. like the Bush differential analyzer 1. in which case we record our number on a number of independent scales. where the data are represented by a set of choices among a number of contingencies. and those.138 CYBERNETICS This will be a minimum when : (5.05) -log 2 = *. Generated on 2011-09-28 14:58 GMT / Public Domain. like the ordinary desk adding and multiplying machine. which we call numerical machines. 1930 on. and only when.

was already in use when the Hindus made the great discovery of the importance of the zero and the advan tage of a positional system of notation. the numerical machines are preferable. and must be as free as possible from human interference to the very end. in business offices. and in taking off the machine numbers which must be written in the same conventional form.COMPUTING MACHINES AND THE NERVOUS SYSTEM 139 contingencies presented at every choice. While . It is not the way that the larger and more automatic machines are best to be employed . as in any combination of chemical reactions. as employed in banks. any computing machine is used because machine methods are faster than hand methods. at the very beginning and the very end. and to perform all intermediate processes on the binary scale. in which the number of alternatives presented at each choice is two. and in many statistical laboratories. We see that for highly accurate work.org/access_use#pd-google human element from any elaborate chain of computation.hathitrust. Under these conditions. This is in fact the use of the ordinary desk computing machine. Our use of machines on the decimal scale is conditioned merely by the historical accident that the scale of ten. it is the slowest which gives the order of magnitude of the time constants of the entire system. and must combine contingencies in accordance with a systematic algorithm. it pays to have an instrument for the change of the scale of nota tion. in general. at any rate. based on our fingers and thumbs. and the number of choices given. and only to introduce it where it is absolutely unavoidable. Thus the computing machine must be a logical machine as well as an arithmetic machine. to be used initially and finally in the chain of computa tions . In any combined use of means of computation. but also all the rules for combining them. as far as possible. in the form of instructions covering every situation which may arise in the course of the computation. Google-digitized / http://www. It is worth retaining when a large part of the work done with the aid of the mfichine consists in transcribing on to the machine numbers in the conven tional decimal form. This means that not only must the numerical data be inserted at the beginning. and above all. to remove the Generated on 2011-09-28 14:58 GMT / Public Domain. The ideal computing machine must then have all its data inserted at the beginning. those numerical machines constructed on the binary scale. It is thus advantageous.

are in the form of a set of choices between two alternatives. The reasons for its superiority to other systems are of the same nature as Ihe reasons for the superiority of the binary arithmetic over other arithmetics.hathitrust. Google-digitized / http://www. Thus all the data. the simplest of these is known as the algebra of logic par excellence.org/access_use#pd-google 01 Thus multiplication is simply a method to determine a set of new digits when old digits are given. is based on the dichotomy. This algorithm. if A and B are both 1. put into the machine. as in the decimal. The addition of numbers of more than one digit follows similar but more complicated rules. if O is a negative and I a positive decision.140 CYBERNETICS there are many algorithms which might be used for combining contingencies. and otherwise with 0. or the Boolean algebra. A and B. Multiplication in the binary system.10) 000 IOI . may be reduced to the multiplication table and the addition of numbers. with the same table as the numerical multiplication of the (140) system. numerical or logical. and is otherwise 0. The second digit is 1 if A =£ B.08) 00 Generated on 2011-09-28 14:59 GMT / Public Domain. and the rules for multiplication for binary numbers take on the peculiarly simple form given by the table : ^J (5. On the logical side. namely : 00I (5. When I add two one-digit numbers. which transforms 1 into 0 and 0 into 1 . every operator can be derived from three : negation. with the table 0I (5. I obtain a two-digit number commencing with 1.09) 0I I 1I and logical multiplication. logical addition. and all the operations on the data take the form of making a set of new choices depend on a set of old choices. the choice between being in a class and outside it. the choice between yes and no. like the binary arithmetic.

as in the case of a solenoidal relay. The relays used in a computing machine maybe of very varied character. in the form of high-vacuum tubes. which are known to be capable of the work of a computation system. Google-digitized / http://www. they are either at rest . the relays assume each a position dictated by the positions of some or all the relays of the bank at a previous stage of operation. while the other is transitory. either in the form of gas-filled tubes. depending according to a fixed set of rules on the decisions already made. the structure of the machine is that of a bank of relays. say « on ». or when they « fire ». or they may be electro-mechanical. they go through a series of changes almost independent of the nature and intensity of the stimulus. and « off ». They may be purely mechanical. While they show rather complicated properties under the influence of electrical currents.org/access_use#pd-google case. These stages of operation may be definitely « clocked » from some central clock or clocks. capable each of two conditions. while at each stage. in their ordinary physiological action they conform very nearly to the « all-or- none » principle : that is. They may be purely electrical systems with two altern ative positions of equilibrium. or only one may be stable. In other words. contain elements which are ideally suited to act as relays.hathitrust. It is a noteworthy fact that the human and animal nervous systems. Always in the second Generated on 2011-09-28 14:59 GMT / Public Domain. and to avoid the clogging up of the system which will ensue if one of the relays does nothing but repeat itself indefinitely. we shall have more to say concerning this question of memory later. every contingency which may arise in the operation of the machine simply demands a new set of choices of contingencies I and 0. There is first an . or what is much more rapid. in which the armature will remain in one of two possible positions of equilibrium until an appropriate impulse pulls it to the other side. or the action of each relay may be held up until all the relays which should have acted earlier in the process have gone through all the steps called for. it will be desirable to have special apparatus to retain an impulse which is to act at some future time. However. These elements are the so-called neurons or nerve cells. and generally in the first case. The two possible states of a relay system may both be stable in the absence of outside interference.COMPUTING MACHINES AND THE NERVOUS SYSTEM 141 That is.

a function equally in demand for computing machines. as we have said. these vary in number from a very few to many hundred. to which there succeeds a refractory period. This is perhaps an over-simplification of the picture : the « threshold » may not depend simply on the number of synapses. the ability to preserve the results of past . is that some definite combinations of impulses on the incoming neurons having synaptic connections with a given neuron will cause it to fire. or at any rate is not capable of being stimulated by any normal.hathitrust.org/access_use#pd-google another with respect to the neuron into which they feed . which either completely prevent the firing of the outgoing neuron. For a given outgoing neuron. but on their « weight » and their geometrical relations to one Generated on 2011-09-28 14:59 GMT / Public Domain. This is not to say that there may not be other. Leaving aside those neurons which accept their messages from free endings or sen sory end-organs. during which the neuron is either incapable of being stimulated. physiological process. and Ihe number of incoming synapses which « fire » within a certain very short fusion interval of time exceeds a certain threshold. while others will not cause it to fire. and there is very convincing evidence that there exist synapses of a different nature. is that of memory. or at any rate raise its threshold with respect to stimulation at Ihe ordinary synapses. the so-called « inhibitory synapses ». non-neuronic influences. each neuron has its messages fed into it by other neurons at points of contact known as synapses. Google-digitized / http://www. At the end of this effective refractory period. If it is neither firing nor refractory.142 CYBERNETICS active phase. but may be stimulated again into activity. and. transmitted from one end to the other of Ihe neuron with a definite velocity. secular changes tending to vary that pattern of incoming impulses which is ade quate for firing. however. What is pretty clear. perhaps of a humoral nature. which produce slow. Thus the nerve may be taken to be a relay with essentially two states of activity : firing and repose. combined with the antecedent state of the outgoing neuron itself. A very important function of the nervous system. then the neuron will fire after a known. it is the state of the incoming impulses at the various synapses. which determines whether it will fire or not. the nerve remains inactive. fairly constant synaptic delay.

until this circuit is cleared by intervention from outside. or with a minimal. either with no reference to each other. be read quickly. and before the errors inherent in the instrument have blurred it too much. never even approximately clears out its past records. such as a multiplication. or even. There is much reason to believe that this happens in our brains during the retention of impulses which occurs over what is known as the specious pre sent. of the machine or the brain. but rather the analogue of a single run on such a machine. while it favors the . under normal circumstances. Let it be remarked parenthetically that an important difference between the way in which we use the brain and the machine is that the machine is intended for many successive runs. We shall see later that this remark has a deep significance in psychopathology and in psychiatry.COMPUTING MACHINES AND THE NERVOUS SYSTEM 143 operations for use in the future. and to con tribute to the basis of all its future behavior. or at least suggested for such a use.hathitrust. the permanent record. It will be seen that the uses of the memory are highly various. and in which the operating apparatus should then be released for further use. and be erased quickly. a very satisfactory me thod for constructing a short-time memory is to keep a sequence of impulses travelling around a closed circuit. The first condi tion tends to rule out delays produced by the transmission of light. in nfany cases. There are two conditions which are desirable in such a retentive apparatus : the impulse should be transmit ted in a medium in which it is not too difficult to achieve a considerable time-lag . Such a memory should record quickly. in which the intermediate results are of no value when once the process is completed. limited reference. This method has been imitated in several devices which have been used in computing machines. at least during a single run of the machine. by electric circuits. There is first the memory which is necessary for the carrying out of a current process. and it is improbable that any single mechanism can satisfy the demands of all of them. Google-digitized / http://www. while the brain in the course ofnature. and that it can be cleared between such runs. is not the complete analogue of the computing machine. Generated on 2011-09-28 14:59 GMT / Public Domain. there is the memory which is intended to be part of the files. On the other hand.org/access_use#pd-google To return to the problem of memory. the impulse should be reconstructed in a form as sharp as possible. Thus the brain.

What makes this more remarkable is that this apparatus was not used merely to preserve a single deci Generated on 2011-09-28 15:00 GMT / Public Domain. or the imperfectly insula ting surface of the dielectric itself.org/access_use#pd-google sion. A much better way is the use of a large number of condensers. In the electrical industry. Williams of the University of Man chester. and such vibra tions have actually been employed for this purpose in com puting machines. a single « yes » or « no ». the delay produced at every stage is relatively short . we must insert some where in the cycle a relay which does not serve to repeat the form of the incoming message. Their success is all the more remarkable : in a piece of apparatus designed by Mr. To avoid this. in which one plate is either a small piece of metal sputtered in to a dielectric. One of the simplest modes of storing information for a relatively short time is as the charge on a condenser . and when this is supplemented by a telegraph-type repeater. this works on the scanning principle. They are known as lelegraph-iype repealers. the deformation of the message is cumulative. The ordinary means of doing this in volve mechanical inertia. while one of the connectors . Google-digitized / http://www. and this is never consistent with very high speeds. This is done very easily in the nervous system. it becomes an ade quate method of storage. If electric circuits are used for delay purposes. The great difficulty of using them for memories of long duration is that they have to function without a flaw over an enormous number of consecutive cycles of ope ration. it is desirable to be able to switch successively and very rapidly from one condenser to another. where indeed all transmission is more or less of a trigger phenomenon. a second consideration comes into play .144 CYBERNETICS use of one form or anolher of elastic vibrations. or as in all pieces of linear apparatus. and have been used in connection with telegraph circuits. and very soon becomes intolerable. Like other forms of apparatus intended to retain a large num ber of decisions. To use to the best advantage the circuit facilities attached to such a storage system. a device of this sort with a unit delay of the order of a hundredth of a second has continued in successful operation for several hours.hathitrust. but a matter of thousands of decisions. pieces of apparatus for this purpose have long been known. but rather to trigger off a new message of prescribed form.

These last-named methods for storing information can hold a message for quite an appreciable time. where many of the substances which act as developers . Phosphorescence as well is a phenomenon associated with a high quantum degeneracy . photo graphy. They seem to depend on systems with a high degree of quan tum degeneracy . Very many of the methods of storage of information already considered have an important physical element in common.COMPUTING MACHINES AND THE NERVOUS SYSTEM 145 to these condensers is a pencil of cathode rays moved by the condensers and magnets of a sweep-circuit over a course like that of a plough in a ploughed field.org/access_use#pd-google planting of a new record. which has been reduced to a few seconds. and the same sort of effect makes its appearance in the photographic process. which are thus especially valuable for use in condensers for the storage of information. The Eastman people have been working on just these problems. which have largely eliminated the tendency of messages on this material to spread . if not for a period com parable with a human life-time. There are various elabora tions of this method. and (at present) the fact that a photogra phic record is not subject to rapid erasure and the rapid im Generated on 2011-09-28 15:00 GMT / Public Domain. It suffers from two grave disadvantages : the time needed for de velopment. or in other words. This is certainly true in the case of ferromagnetism. with a large number of modes of vibration of the same frequency. and above all. and is also true in the case of materials with an exceptionally high dielectric cons tant. Williams. ideal again from the point of view of the shortness of exposure needed to record an observation. slow. Google-digitized / http://www.hathitrust. phosphorescent substances . For more permanent re cords. there is a wide variety of alternatives among which we can choose. Leaving out such bulky. and unerasable methods as the use of punched cards and punched tape. Photography is indeed ideal for the permanence and detail of its records. and it is possible that by this time they have found the answer. we have magnetic tape. which indeed was employed in a some what different way by the Radio Corporation of America before it was used by Mr. together with its modern refinements. which do not seem to be neces sarily insoluble. but is still not small enough to make photography available for a short-time memory .

the neurons and the synapses are elements of this sort. our whole life is on the pattern of Balzac's Peau de Chagrin. by changes in the permeability of each synapse to messages. In re inserting this information into the system.hathitrust. Apparently it is adequately established that no neurons are formed in the brain after birth. and to make them effective in the outer world. One of the simplest ways to do this is to have as the storage ele ments which are changed parts which normally assist in the transmission of messages. We have seen in the case of photography and similar pro cesses that it is possible to store a message in the form of a permanent alteration of certain storage elements. This is a possible explanation for a sort of senes- . until life itself squanders our capital stock of power to live. it is necessary to cause these changes to affect the messages going through the system. and the very process of learning and remembering exhausts our powers of learning and remembering. and it is a plausible conjecture that the chief changes of thresholds in the memory process are increases. we find them associated with a third fundamental property of living matter : the ability to receive and organize impulses. or. It is possible.org/access_use#pd-google may be regarded as another way of saying the same thing. Google-digitized / http://www. and of such a nature that the change in their character due to storage affects the manner in which they will transport messages for the entire future. It is probably not an accident that here. what Generated on 2011-09-28 15:00 GMT / Public Domain. It may well be that this phenomenon does occur. in the absence of a better explanation of the phenomenon. and it is quite plausible that information is stored over long periods by changes in the thresholds of neurons. that the storage of information in the brain can actually occur in this way.146 CYBERNETICS seem to have a great deal of internal resonance. It is conceivable for such a storage to take place either by the opening of new paths or by the clo sure of old ones. though not certain. Quantum degeneracy appears to be associated with the ability to make small causes produce appreciable and stable effects. In the ner vous system. Many of us think. in a non-living environment. We have already seen in Chapter II that substances with high quantum degeneracy appear to be associated with many of the problems of metabolism and reproduction. that no new synapses are formed . If this is the case.

when it is engaged in that activity known as logical thinking. The real phenomenon of senescence. but — and this is the important fact — any logic which means anything to us can contain nothing which the human Generated on 2011-09-28 15:00 GMT / Public Domain. but these discussions and their accom panying proofs are not infinite in fact. According to this. as a logical machine. (2) Pn has been proved for n = 1.hathitrust. the study of logic must reduce to the study of the logical machine. and con sequently the brain.org/access_use#pd-google mind — and hence the human nervous system — is unable to encompass. P«+i is true . \. is much too complicated to be explained in this way alone. We have said before that the machina raiiocinatrix is nothing but the calculus ratiocinator of Leibniz with an en gine in it . . however. It may be said by some readers that this reduces logic to psy chology. it considers every statement as es sentially concerned with possible experiments or observable processes. It is by no means trivial to consider the light cast on logic by such machines. and that the two sciences are observably and demon- strably different. so it is inevitable that its present engineering development should cast a new light on logic. Loc. Here the chief work is that of Turing 1. It is true. No admissible proof involves more than a finite number of stages. This is true in the sense that many psycholo gical states and sequences of thought do not conform to the canons of logic. we devote much of mathematics to discussions involving the infinite. All logic is limited by the limitations of the human mind. Psychology contains much that is foreign to logic. (3) If Pn is true. Google-digitized / http://www. but this is only apparent. For example.COMPUTING MACHINES AND THE NERVOUS SYSTEM 147 cence. with all its non-removable limitations and imperfections. both natural and artificial. whether nervous or mechani cal. The science of today is operational : that is. We have already spoken of the computing machine. a prool by mathematical induction seems to involve an infinity of stages. it involves just the following stages : (1) Pn is a proposition involving the number /?. In fact. cit. and just as modern mathematical logic begins with this calculus. (4) Therefore Pn is true for every positive integer n.

this parameter being the time at which it is asserted. A machine to answer this question would give the suc cessive temporary answers. However. It may goon grinding through different stages with out ever coming to a stop. according to the character of the objects with which it concerns itself . it is equally certainly a member of itself. This occurs in the case Generated on 2011-09-28 15:00 GMT / Public Domain. However. a logical machine following definite rules need never come to a conclusion. Thus some very interesting situations arise. « yes ». and so on . or by going into a repetitive process like the end of a chess game in which there is a continuing cycle of perpetual check. The method by which we resolve the paradoxes is also to attach a parameter to each statement. such as we find in mathematical induction. it is certainly not a member of itself . the so-called type. this mathematical induction is a far different thing from com plete induction over an infinite set. and if it is not.148 CYBERNETICS It is true that somewhere in our logical assumptions there must be one which validates this argument. A proof represents a logical process which has come to a definitive conclusion in a finite number of stages. The same thing is true of the more reflned forms of mathematical induction. Bertrand Russell's solution of his own paradoxes was to affix to every statement a quantity. and would never come to equilibrium. — whether these are « things ». etc. classes of « things ».org/access_use#pd-google of some of Ihe paradoxes of Cantor and Russell. In both cases. which occur in certain mathematical disciplines. Is this class a member of itself ? If it is. such as transfinite induction. it may be impossible to prove Pn for all n. Let us consider the class of all classes which are not members of themselves. « no ». Google-digitized / http://www. This contingency is recognized in what is known as metamathematics. in which we may be able — with enough time and enough computational aids — to prove every single case of a theorem P«. in the simplest sense. but if there is no sys tematic way of subsuming these proofs under a single argument" independent of n. classes of classes of « things ». we introduce what . which serves to distinguish between what seems to be formally the same state ment. « yes ».hathitrust. the discipline so brilliantly developed by Godel and his school. « no ». either by describing a pattern of activity of continually increasing complexity.

as exemplified in the work of Pavlov. and cause and effect. the content of the mind was considered to be made up of certain entities known to Locke as ideas. and following Turing. By some sort of inner activity. This may well have been a survival of the scholastic emphasis on substances. Nevertheless. Indeed. the collector and classifier. the experimental embryologists of the present day. as free from influence on the ideas it contained as a clean blackboard is on the symbols which may be written on it.COMPUTING MACHINES AND THE NERVOUS SYSTEM 149 we may call a parameter of uniformization. Similarly. these ideas were sup posed to unite themselves into bundles. with so much of the world to explore. Of these principles perhaps the most significant was contiguity : ideas or impressions which had often occurred together in time or in space were supposed to have acquired the ability of evoking one ano ther. but the idea of a dyna mics had not yet filtered through from physics to the biological and psychological sciences. let us consider two closely related notions : that of the association of ideas and that of the condi tioned reflex. the phy siologists. to resolve an ambi guity which is simply due to its neglect.hathitrust.hardly worthy to be called a force. the geneticists. the notion of mental conlent domin ated that of mental process. we may employ it to throw light on human logic. in a world in which the noun was hypostasised and the verb carried little or no weight. in psychology.org/access_use#pd-google the entire bundle. Tn all this there is a dynamics implied. Has the machine a more eminently human char acteristic as well — the ability to learn ? To see that it may well have even this property. Google-digitized / http://www. . In the British empirical school of philosophy. We thus see that the logic of the machine resembles human logic. the step from these static ideas to the more dyna mic point of view of the present day. contiguity. and to the later authors as ideas and impressions. so that the presence of any one of them would produce Generated on 2011-09-28 15:01 GMT / Public Domain. is perfectly clear. from Locke to Hume. with a point of view quite opposed to that of the evolutionists. the state of mind of the biologists could hardly have been different. The typical biologist of the eight eenth century was Linnaeus. according to the prin ciples of similarity. The simple ideas or impressions were supposed to exist in a purely passive mind.

150 CYBERNETICS

Pavlov worked much more with animals than with men, and

he reported visible actions rather than introspective states of

mind. He found in dogs that the presence of food causes the

increased secretion of saliva and of gastric juice. If then a cer

tain visual object is shown to dogs in the presence of food, and

only in the presence of food, the sight of this object in the

absence of food will acquire the property of being by itself able

to stimulate the flow of saliva or of gastric juice. Theunion by

contiguity which Locke had observed introspectively in the

case of ideas now becomes a similar union of patterns of beha

vior.

There is one important difference, however, between the point

of view of Pavlov and that of Locke, and it is precisely due to

this fact that Locke considers ideas and Pavlov patterns of

action. The responses observed by Pavlov tend to carry a pro

cess to a successful conclusion, or to avoid a catastrophe.

Salivation is important for deglutition and for digestion, while

the avoidance of what we should consider a painful stimulus

tends to protect the animal from bodily injury. Thus there

enters into the conditioned reflex something that we may call

affective tone. We need not associate this with our own sensa

tions of pleasure and pain, nor need we in the abstract associate

it with the advantage of the animal. The essential thing is this :

Generated on 2011-09-28 15:01 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

that affective tone is arranged on some sort of scale from nega

tive « pain » to positive « pleasure » ; that for a considerable

time, or permanently, an increase in affective tone favors all

processes in the nervous system that are under way at the time,

and gives them a secondary power to increase affective tone;

and that a decrease in affective tone tends to inhibit all processes

under way at the time, and gives them a secondary ability to

decrease affective tone.

Biologically speaking, of course, a greater affective tone must

occur predominantly in situations favorable for the perpetuation

of the race, if not the individual, and a smaller affective tone

in situations which are unfavorable for this perpetuation, if not

disastrous. Any race not conforming to this requirement will

go the way of Lewis Carroll's Bread-and-Butter Fly, and always

die. Nevertheless, even a doomed race may show a mechanism

valid so long as the race lasts. In other words, even the most

COMPUTING MACHINES AND THE NERVOUS SYSTEM

151

suicidal apportioning of affective tone will produce a definite

pattern of conduct.

Note that the mechanism of affective tone is itself a feed-back

mechanism. It may even be given a diagram such as the fol

lowing.

Process 1 wTAffective tone

mechanism

Affective

tone totalizer

FIG. 7.

Here the totalizer for affective tone combines the affective

tones given by the separate affective-tone mechanisms over

a short interval in the past, according to some rule which we

need not specify now. The leads back to the individual affec

tive tone mechanisms serve to modify the intrinsic affective tone

of each process in the direction of the output of the totalizer,

and this modification stands until it is modified by later messages

from the totalizer. The leads back from the totalizer to the

process mechanisms serve to lower thresholds if the total affec

tive tone is increasing, and to raise them if the total affective

tone is decreasing. They likewise have a long-time effect, which

endures until it is modified by another impulse from the totali

Generated on 2011-09-28 15:01 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

zer. This lasting effect, however, is confined to those processes

actually in being at the time the return message arrives, and a

similar limitation also applies to the effects on the individual

affective-tone mechanisms.

I wish to emphasize that I do not say that the process of the

conditioned reflex operates according to the mechanism I have

given ; I merely say that it could so operate. If, however, we

assume this or any simular mechanism, there are a good many

152 CYBERNETICS

things we can say concerning it. One is that this mechanism is

capable of learning. It has already been recognized that the

conditioned reflex is a learning mechanism, and this idea has

been used in the behaviorist studies of the learning of rats in a

maze. All that is needed is that the inducements or punishments

used have respectively a positive and a negative affective tone.

This is certainly the case, and the experimenter learns the na

ture of this affective tone by experience, not simply by a priori

considerations.

Another point of considerable interest is that such a mecha

nism involves a certain set of messages which go out gene

rally into the nervous system, to all elements which are in a

state lo receive them. These are the return messages from the

affective-tone totalizer, and to a certain extent, the messages

from the affective-tone mechanisms to the totalizers. Indeed,

the totalizer need not be a separate element, but may merely

represent some natural combinatory effect of messages arriving

from the individual affective-tone mechanisms. Now, such mes

sages « to whom it may concern » may well be sent out most

efficiently, with a smallest cost in apparatus, by channels other

than nervous. In a similar manner, the ordinary communica

tion system of a mine may consist of a telephone central with

the attached wiring and pieces of apparatus. When we want to

Generated on 2011-09-28 15:01 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

empty a mine in a hurry, we do not trust to this, but break a

tube of a mercaptan in the air intake. Chemical messengers

like this, or like the hormones, are the simplest and most effec

tive for a message not addressed to a specific recipient. For

the moment let me break into what I know to be pure faney.

The high emotional and consequently affective content of hor

monal activity is most suggestive. This does not mean that a

purely nervous mechanism is not capable of affective tone and

of learning, but it does mean that in the study of this aspect of

our mental activity, we cannot afford to be blind to the possi

bilities of hormonal transmission. It may be excessively fanci

ful to attach this notion to the fact that in the theories of Freud,

the memory — the storage function of the nervous system —

and the activities of sex are both involved. Sex, on the one hand,

and all affective content, on the other, contain a very strong

hormonal element. This suggestion of the importance of sex

This is largely true of equations of the parabolic type as well. We have already seen thai it is the run rather than the entire existence of the mechanical structure of the computing machine which corresponds to the life of the individual. it is not manifestly absurd in prin ciple. When it comes to equations of the elliptic type. it is highly probable that information is stored largely as changes in the permeability of the synapses . J. It also contains the content of its storage mechanisms. One of the chief of these is in the solution of partial differential equations.hathitrust. the typical problem is that of solving the equation when the initial data are given. as the data involve the accurate description of functions of two or more variables. for example. and thus to alter the numerical value of the summation of impulses which will make the tube or tubes fire. While at present there is no adequate evidence to prove its validity. Even linear partial differential equations require the recording of an enormous mass of data to set them up. Lettvin and Mr. A more detailed account of learning apparatus in computing Generated on 2011-09-28 15:01 GMT / Public Domain. It is perhaps better to devote the rest of this chapter to the more developed. "We' have also seen that in the nervous computing machine. There is nothing in the nature of the computing machine which forbids it to show conditioned reflexes. like the wave equation. and the uses to which it may be put. and this can be done in a progressive manner from the initial data to the results at any given later time.org/access_use#pd-google and control machines. may well be left to the engineer rather than to a preliminary book like this one. Let us remember that a computing machine in action is more than the concatenation of relays and storage mechanisms which the designer has built into it. With equations of the hyperbolic type. It is perfectly possible.COMPUTING MACHINES AND THE NERVOUS SYSTEM 153 and hormones has been made to me by Dr. to cause any message going into storage to change in a permanent or semi permanent way the grid bias of one or of a number of vacuum tubes. Oliver Selfridge. and this content is never completely cleared in the course of a single run. and it is perfectly possible to construct artificial machines where information is stored in that way. normal uses of modern computing machines. the natural methods of solution involve an itera- . Google-digitized / http://www. where the natural data are boundary values rather than initial values.

154 CYBERNETICS

tive process of successive approximation. This process is

repeated a very large number of times, so that very fast methods

such as those of the modern computing machine are almost indis

pensable.

In non-linear partial differential equations, we miss what

we have in the case of the linear equations —a reasonably ade

quate purely mathematical theory. Here computational me

thods are not only important for the handling of particular

numerical cases, but, as von Neumann has pointed out, we need

them in order to form that acquaintance with a large number

of particular cases without which we can scarcely formulate a

general theory. To some extent this has been done with the aid

of very expensive experimental apparatus, such as wind tun

nels. It is in this way that we have become acquainted with

the more complicated properties of shock waves, slip surfaces,

turbulence, and the like, for which we are scarcely in a position

to give an adequate mathematical theory. How many undis

covered phenomena of similar nature there may be, we do not

know. The analogy machines are so much less accurate, and

in many cases so much slower, than the digital machines, that

the latter give us much more promise for the future.

It is already becoming clear in the use of these new machines

that they demand purely mathematical techniques of their own,

Generated on 2011-09-28 15:02 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

quite different from those in use in manual compulation or in

the use of machines of smaller capacity. For example, even the

use of machines for computing determinants of moderately high

order, or for the simultaneous solution of twenty or thirty si

multaneous linear equations,shows difficulties which do not arise

when we study analogous problems of small order. Unless

care is exercised in setting a problem up, these may completely

deprive the solution of any significant figures whatever. It is

a commonplace to say that fine, effective tools like the ultra-

rapid computing machine are out of place in the hands of those

not possessing a sufficient degree of technical skill to take full

advantage of them. The ultra-rapid computing machine will

certainly not decrease the need for mathematicians with a high

level of understanding and technical training.

In the mechanical or electrical construction of computing

machines, there are a few maxims which deserve consideration.

COMPUTING MACHINES AND THE NERVOUS SYSTEM 155

One is that mechanisms which are relatively frequently used,

such as multiplying or adding mechanisms, should be in the

form of relatively standardized assemblages adapted for one

particular use, and no other ; while those of more occasional

use should be assembled for the moment of use out of elements

also available for other purposes. Closely related to this con

sideration is the one that in these more general mechanisms, the

component parts should be available in accordance with their

general properties, and should not be alloted permanently to

a specific association with other pieces of apparatus. There

should be some part of the apparatus, like an automatic telephone-

switching exchange, which will search for free components and

connectors of the various sorts, and allot them as they are needed.

This will eliminate much of the very large expense which is due

to having a great number of unused elements, which cannot be

used unless their entire large assembly is used. We shall find

this principle is very important when we come to consider

traffic problems and overloading in the nervous system.

As a final remark, let me point out that a large computing

machine, whether in the form of mechanical or electric apparatus

or in the form of the brain itself, uses up a considerable amount

of power, all of which is wasted and dissipated in heat. The

blood leaving the brain is a fraction of a degree warmer than

Generated on 2011-09-28 15:02 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

that'entering it. No other computing machine approaches the

economy of energy of the brain. In a large apparatus like the

Eniac or Edvac, the filaments of the tubes consume a quantity

of energy which may well be measured in kilowatts, and unless

adequate ventilating and cooling apparatus is provided, the

system will suffer from what is the mechanical equivalent of

pyrexia, until the constants of the machine are radically

changed by the heat, and its performance breaks down. Never

theless, the energy spent per individual operation is almost va-

nishingly small, and does not even begin to form an adequate

measure of the performance of the apparatus. The mechanical

brain does not secrete thought « as the liver does bile », as the

earlier materialists claimed, nor"does it put it out in the form

of energy, as the muscle puts out its activity. Information is

information, not matter or energy. No materialism which does

not admit this can survive at the present day.

CHAPTER VI

GESTALT AND UNIVERSALS

Among other things which we have discussed in the last chapter

is the possibility of assigning a neural mechanism to Locke's

theory of the association of ideas. According lo Locke, this

occurs according to three principles : the principle of contiguity,

the principle of similarity, and the principle of cause and effect.

The third of these is reduced by Locke, and even more defini

tively by Hume, to nothing more than constant concomitance,

and so is subsumed under the first, that of contiguity. The

second, that of similarity, deserves a more detailed discussion.

How do we recognize the identity of the features of a man,

whether we see him in profile, in three-quarters face, or in full

face ? How do we recognize a circle as a circle, whether it is large

or small, nearorfar; whether in fact itisin a plane perpendicular

to a line from the eye meeting it in the middle, and is seen as a

circle, or has some other orientation, and is seen as an ellipse ?

How do we see faces and animals and maps in clouds, or in the

blots of a Rorschach test ? All these examples refer to the eye,

but similar problems extend to the other senses, and some of

them have to do with inter-sensory relations How do we put

into words the call of a bird or the stridulations of an insect ?

How do we identify the roundness of a coin by touch ?

For the present let us confine ourselves to the sense of

Generated on 2011-09-28 15:02 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

vision. One important factor in the comparison of form of diffe

rent objects is certainly the interaction of the eye and the mus

cles, whether they are the muscles within the eyeball, the mus

cles moving the eyeball, the muscles moving the head, or the

muscles moving the body as a whole. Indeed, some form of

this visual-muscular feed-back system is important as low in the

animal kingdom as the flatworms. There the negative photo-

or by which we bring an object outside the visual field picked up by some other sense into that . as when the pupil opens in the dark and closes in the light. seems to be controlled by the balance of the impulses from the two eyespots. It would be difficult or impossible for us to compress this mechanism into the dimensions that a flatworm can carry .GESTALT AND UNIVERSALS 157 tropism. It is interesting to note that a combination of a pair of photocells with appropriate amplifiers.hathitrust. Google-digitized / http://www. there is a reflex feed-back to bring it into the fovea. Some of these are of purely homeostatic nature. and to focus the lens so that its outlines are as sharp as possible. These actions are supplemen ted by motions of the head and body. turning the body away from the light.org/access_use#pd-google into the eye between narrower bounds than would otherwise be possible. When the peripheral vision has picked up some object conspicuous by brilliancy or light-contrast or color or above all by motion. This feed-back is accompanied by a complicated system of interlinked subordinate feed-backs. and further amplifiers con trolling the input into the two motors of a twinscrew mecha nism. thus tending to confine the flow of light Generated on 2011-09-28 15:02 GMT / Public Domain. a Wheatstone bridge for balancing their outputs. let us come at once to the eye-muscle feed-backs in man. the tendency to avoid the light. that living mechanisms tend to have a much smaller space scale than the mechanisms best suited to the techniques of human artificers. This balance is fed back to the muscles of the trunk. while its perception of motion is better on the periphery. although on the other hand the use of electrical tech niques gives the artificial mechanism an enormous advantage in speed over the living organism. Others concern the fact that the human eye has economically confined its best form and color vision to a rela tively small fovea. Without going through till the intermediate stages. but here we merely have another exem plification of the fact that must by now be familiar to the reader. would give us a very adequate negatively phototropic control for a little boat. by which we bring the object into the center of vision if this cannot be done readily by a motion of the eyes alone. and in combination with the general impulse to move forward. carries the animal into the darkest region accessible. which tend to con verge the two eyes so that the object attracting attention is in the same part of the visual field of each.

the face of a man. One of the most remarkable phenomena of vision is our abi lity to recognize an outline drawing. the constant maintenance of a stimulus reduces its ability to receive and to transmit that stimulus. This is quite Generated on 2011-09-28 15:03 GMT / Public Domain. Clearly an outline drawing of. In the case of objects with which we are more familiar in one angular orientation than another — writing. yet it may be a most recognizable portrait of its subject. There is considerable evidence that for a considerable number of stages. These later processes occur in the eye and in the visual cortex.org/access_use#pd-google understandable. say. the correspondence on the periphery is such that one optic nerve fibre corresponds to ten or more end organs. Like all senses.hathitrust. has very little resemblance to the face itself in color. so that the visual image which we form of it varies within as small a range as possible. human faces. This does not exhaust the processes which are involved in perceiving the form and meaning of the object. The beginning of these processes is in the eye itself. and brings this information one step nearer to the form in which it is used and is preserved in the memory. but it certainly facilitates all later processes tending to this end. Google-digitized / http://www.158 CYBERNETICS field. and the like — there is also a mechanism by which we tend to pull them into the proper orientation. It will be noted that while in the fovea there is almost a one-one correspon dence between the rods and cones and the fibres of the optic nerve. All these processes can be summed up in one sentence:we tend to bring any object that attracts our attention into a standard position and orientation. or in the massing of light and shade . landscapes. This is most markedly so for the receptors which record the interior of a large block of images with constant color and illumination. the retina is subject to acco- modation : that is. somewhere in the visual process. for even the . The most plau sible explanation of this is that. each step in this pro cess diminishes the number of neuron channels involved in the transmission of visual information. in view of the fact that the chief function of the peripheral fibres is not so much vision itself as a pick-up for the centering and focussing-directing mechanism of the eye. outlines are emphasized and some other aspects of an image are minimized in importance. The first step in this concentration of visual information occurs in the transition between the retina and the optic nerve.

In photogra phy. Here these fluctuations produce an alternation between one sti mulus and another. as we see in the pheno menon of after-images. and that every visual image in fact has something of the nature of a line drawing. We have thus designated several actual or possible stages of the diagrammatization of our visual impressions.GESTALT AND UNIVEHSALS 159 slight fluctuations of focus and point of fixation which are inevi table in vision do not change the character of the image received. and such phenomena. We have now to compare them with one another. and this alternation. We have given a rough sketch which indicates how the Lockean principle of contiguity in association may be mechanized. The different aspects of the same object are often to be seen in those processes which bring it to the focus of attention. Probably not all of this action is peripheral. are certainly not beyond what the nervous system can do. such as « circle » or « square ». They are allied to the phenomena of the telegraph-type repeater.org/access_use#pd-google image. they use an im pression which has not been blurred beyond a certain point to trigger a new impression of a standard sharpness. Let us notice that the principle of contiguity also covers much of the other Lockean principle of similarity. which are of non-linearity. We center our images around the focus of attention. This may be done in several ways. and reduce them more or less to outlines. This is true whether the contrast between the two adjacent regions is one of light-intensity or of color. Like this. It is quite different on the boundary of two contrasting regions. or at any rate with a standard impression stored in memory. but even tends to enhance its sensitivity.hathitrust. We thus find that the eye receives its most intense impression at boundaries. not only does not tend to exhaust the visual mechanism by accomodation. At any rate. and are probably correlated with a part of the reduction of the number of transmission fibres found at various stages of the visual cortex. Google-digitized / http://www. which we have already mentioned. As a comment on these facts. they decrease the total unusable information carried by an Generated on 2011-09-28 15:03 GMT / Public Domain. and of other motions which lead us to see it. it is known that certain treatments of a plate increase its contrasts. let us note that 3/4 of the fibers in the optic nerve respond only to the flashing « on » of illumination. now at one distance and now at .

the rotations in two or three dimensions about a point . too specific.hathitrust.org/access_use#pd-google of scale in all directions are preserved. in which we consi der only those transformations which leave the region at infi nity untouched . and doubtless of much importance in the comparison of our more complicated experiences. the homogeneous dilations about a given point. and the equality Generated on 2011-09-28 15:03 GMT / Public Domain. now from one angle and now from a distinct one. in which one point. Under the circumstances. which is by no means confined to a space of . Among these groups the ones we have just mentioned are continuous: that is. it is worlhcon sidering how such a sub-assembly might possibly work. so every region in a group-space. not confined in its application to any parti cular sense. Google-digitized / http://www. just as a region in the ordinary two-dimensional plane is covered by the process of scanning known to the televi sion engineer. but a permanent sub-assembly like the adding and multiplying assemblies of a computing machine. and contain sub-sets of {rans formations which constitute regions in such a space. This is a general principle. and how we should go about designing it. the directions of the axes. It is nevertheless proba bly not the only process which leads to the formation of our specifically visual general ideas. Now. « complex ideas ». the operations belonging to them are determined by the va lues of a number of continuously varying parameters in an ap propriate space. to lead us to suppose that it operates by what is after all a highly generalized mechanism. In such a process. The possible perspective transformations of an object form what is known as a group in the sense in which we have already defined one in Chapter II. It leaves us the impression that we are here dealing with a special mecha nism which is not merely a temporary assemblage of general- purpose elements. They thus form multi-dimensional configura tions in n-space.160 CYBERNETICS another. the set of all translations. can be represented by a process of group scanning. and so on. by which a nearly uniformly distributed set of sample positions in that region is taken to represent the whole. including the whole of such a space. the transformations pre serving length . The structure of our visual cortex is too highly organized. This group defines several sub groups of transformations : the affine group. or as Locke would call them. with interchangeable parts.

and the two regions are said to be alike.LT AND UNIVERSALS 161 three dimensions. or sets of parameters. in some appropriately defined sense. this is recorded. This certainly happens in reality. If this region is not the entire group. Nevertheless. at least in so far as the immediate impression — one not involving any of the higher processes — is concerned. it means that the results of transforming a given figure by these transformations will come as near as we wish to any given transformation of the figure by a transformation operator lying in the region desired. If at any stage of the scanning of the group of transformations the image of the region to be compared under some one of the transformations scanned coincides more perfectly with the fixed pattern than a given tolerance allows. If our scanning is fine enough. and the region transformed has the maximum dimensionality of the regions transformed by the group considered. it may well be that re gion A seems like region B. and that region B seems like region C. Let us then start with a fixed comparison region and a region to be compared with it. It will thus contain posi tions as near to any we wish as may be desired. a net of positions in the space is traversed in a one-dimensional sequence. are actually used to generate the appropriate transformations. while region A does not seem like region C. but shade into one another. at each stage of its in version. and serves as a method to identify the shape of a figure independently of its size or its orientation or of whatever transformations may be included in the group-region to be scanned. There are other more sophisticated means of using group .GESTA. If this happens at no stage of the scanning process.org/access_use#pd-google to be unlike. A figure may not show any particular re semblance to the same figure inverted. there may be a considerable range of neighboring posi tions which appear similar. they are said Generated on 2011-09-28 15:03 GMT / Public Domain. This process is perfectly adapted to mechaniza tion. Google-digitized / http://www.hathitrust. The universal « ideas » thus formed are not perfectly distinct. and this net of positions is so distributed that it conies near to every position in the region. If these « po sitions ». this means that the transformations actually tra versed will give a resulting region overlapping any transform of the original region by an amount which is as large a fraction of its area as we wish.

hathitrust. and even to a single type-face. we shall obtain a quantity which we may write in some such form as. and let us use TS to express the transform of the set S by the transformation T of the group. to work by the use of photo-electric cells. a probability density which depends on Ihe transformation group itself. the centering of the lines. The most dramatic of the attempts to accomplish this has been the design of reading devices for the blind.01) will be identical for all sets S interchangeable with one another under the transformations of the group : that is. if we have any quantity depending on a set S of elements transformed by the group. let us designate the quantity de pending on S by Q j S j.162 CYBERNETICS scanning to abstract from the transformations of a group. It is possible to scan the group in such a way that the density of scanning of any region of a considerable class — that is the amount of time which the variable scanning element passes within the region in any com plete scanning of the group—is closely proportional to its group measure. and if this set of elements is transformed by all the transformations of the group. or to a small number of type-faces. and does not change when all the transformations of the group are altered by being preceded or followed by any specific transformation of the group. We shall also suppose that the alignment of the page. In recent years. there has been a good deal of attention to the problem of the prosthesis of one lost sense by another. (6. The groups which we here consider have a « group measure ». Then Q j TS | will be the value of the quantity replacing Q 1 S j when S is replaced by TS. Quantity Generated on 2011-09-28 15:03 GMT / Public Domain. We shall suppose these efforts con fined to printed matter.org/access_use#pd-google (6. So much for group measure.01) /*Q ! TS \ d'T.01) is over less than the whole group. where the integration is over the group measure. It is possible to obtain an approximate comparability of form where the integration in (6. If we average or integrate this with respect to the group measure for the group of transformations T. Google-digitized / http://www. the traverse . In the case of such a uniform scanning. for all sets S which have in some sense the same form or Gestalt. if the integrand Q j TS | is small over the region omitted.

GESTALT AND UNIVERSALS 163

from line to line taken care of either manually, or as they may

well be, automatically. These processes correspond, as we may

see, to the part of our visual Gestalt determination which depends

on muscular feed-backs and the use of our normal centering,

orienting, focussing, and converging apparatus. There now en

sues the problem of determining the shapes of the individual

letters as the scanning apparatus passes over them in sequence.

It has been suggested that this be done by the use of several pho

to-electric cells placed in a vertical sequence, each attached to a

sound-making apparatus of a different pitch. This can be done

with the black of the letters registering either as silence or as

sound. Let us assume the latter case, and let us assume Ihree

photocell receptors above one another. Let them record as

the three notes of a chord, let us say, with the highest note on

top and the lowest note below. Then the letter capital F, let

us say, will record :

Duration of upper note ;

Duration of middle note ;

— Duration of lower nole.

The letter capital Z will record :

the letter capital 0 :

and so on. With the ordinary help given by our ability to

interpret, it should not be too difficult to read such an auditory

Generated on 2011-09-28 15:04 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

code : not more difficult than to read Braille, for instance.

However, all this depends on one thing : the proper relation

of the photocells to the vertical height of the letters. Even

with standardized type faces, there still are great variations

in the size of the type. Thus it is desirable for us to be able to

pull the vertical scale of the scanning up or down, in order to

reduce the impression of a given letter to a standard. We must

164

CYBERNETICS

at least have at our disposal, manually or automatically, some

of the transformations of the vertical dilation group.

There are several ways we might do this. We might allow

for a mechanical vertical adjustment of our photocells. On

the other hand, we might use a rather large vertical array of

photocells, and change the pitch-assignment with the size of

type, leaving those above and below the type silent. This may

be done, for example, with the aid of a schema of two sets of

connectors, the inputs coming up from the photocells, and lea

ding to a series of switches of wider and wider divergence,

and the outputs a series of vertical lines, as in the following:

Layer of oscillators D

Layer of photo.cells

Fio. 8.

Here the single lines represent the leads from the photocells,

the double lines the leads to the oscillators, the circles on the

dotted lines the points of connections between incoming and

outgoing leads, and ihe dotted lines themselves the leads where

by one or another of a bank of oscillators is put into action.

This was the device, to which we have referred in the introduc

tion, designed by McCulloch for the purpose of adjusting to the

height of the type-face. In the first design the selection be

Generated on 2011-09-28 15:04 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

tween dotted line and dotted line was manual.

This was the figure which, when shown to Dr. von Bonin,

suggested the fourth layer of the visual cortex. It was the

connecting circles which suggested the neuron cell-bodies ot

this layer, arranged in sub-layers of uniformly changing hori

zontal density, and size changing in the opposite direction to

the density. The horizontal leads are probably fired in some

cyclical order. The whole apparatus seems quite suited to the

process of group scanning. There must of course be some pro

cess of recombination in time of the upper outputs.

This then was the device suggested by McCulloch as that

GESTALT AND UNIVERSALS 165

actually used in the brain in the detection of visual Gestatt.

It represents a type of device usable for any sort of group scan

ning. Something similar occurs in other senses as well. In

the ear, the transposition of music from one fundamental pitch

to another is nothing but a translation of the logarithm of the

frequency, and may consequently be performed by a group-

scanning apparatus.

A group-scanning assembly thus has well-defined appropriate

anatomical structure. The necessary switching maybe performed

by independent horizontal leads which furnish enough stimu

lation to shift the thresholds in each level to just the proper

amount to make them fire when the lead comes on. While we

do not know all the details of the performance of the machinery,

it is not at all difficult to conjecture a possible machinery con

forming to the anatomy. In short, the group-scanning assembly

is well adapted to form the sort of permanent sub-assembly of

the brain corresponding to the adders or multipliers of the

numerical computing machine.

Lastly, the scanning apparatus should have a certain intrinsic

period of operation which should be identifiable in Ihe perfor

mance of the brain. The order of magnitude of this period

should show in the minimum time required for making direct

comparison of the shapes of objects different in size. This can

Generated on 2011-09-28 15:04 GMT / Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

only be done when the comparison is between two objects not

too different in size ; otherwise, it is a long-time process, sugges

tive of the action of a non-specific assembly. When direct com

parison seems to be possible, it appears to take a time of the

order of magnitude of a tenth of a second. This also seems to

accord with the order of magnitude of the time needed by exci

tation to stimulate all the layers of transverse connectors in .

cyclical sequence.

While this cyclical process then might be a locally determined

one, there is evidence that there is a wide-spread synchronism

in different parts of the cortex, suggesting that it is driven from

some clocking center. In fact, it has the order of frequency

appropriate for the alpha rhythm of the brain, as shown in elec

troencephalograms. We may suspect that this alpha rhythm is

associated with form perception, and that it partakes of the

nature of a sweep rhythm, like the rhythm shown in Ihe scanning

as distinguished perhaps from one congenital!}' blind. It disappears in deep sleep. it will probably not differ very much from a comparison J. but are available to store impressions gathered from other senses than the one to which they normally belong. and the sweep rhythm is acting as something like a carrier for other rhythms and activities. precisely as we might expect. A blinded man. In view of the some what similar organization of the different parts of the sensory cortex. We have just seen that the problem of sensory prosthesis — the problem of replacing the information which is normally conveyed by a lost sense by one which is still available — is important. Personal communication of Dr. He may feel his way around a room. are not locks with a single key. nor mally approached through one sense. It is most marked when the eyes are closed in waking. Walter. and not necessarily insoluble. Thus a part of his normal visual mechanism is accessible to him. hut is even able to store tactile and auditory impressions in a visual form. of Bristol. but with an artificial visual cortex.hathitrust.org/access_use#pd-google has also lost the use of that part of his visual cortex which may be regarded as a fixed assembly for organizing the impressions of sight. Google-digitized / http://www. and seems to be obscured and overlaid with other rhylhms. or when we are staring into space at nothing in particular. when it shows an almost perfect periodicity. not only retains visual memories earlier in date than his accident. not only with artificial visual receptors. This is a comparison of amounts of information. On the olher hand. which will translate the light impressions on his new receptors into a form so related to the normal output of his visual cortex that objects which ordinarily look alike will now sound alike.166 CYBERNETICS process of a television apparatus. as in the condition of abstrac tion of a Yogi 1. Thus the criterion of the possibility of such a replacement of sight by hearing is at least in part a comparison between the number of recognizably different visual patterns and recogniza bly different auditory patterns at the cortical level. he has lost more than his eyes : he Generated on 2011-09-28 15:04 GMT / Public Domain. What makes it more -• hopeful is the fact that the memory and association areas. England. . and yet have an image of how it ought to look. when we are actually looking at something. It is necessary to equip him.

On the other hand. which is substantially perfect. and thus a 10/100 vision means an amount of flow of information about one per cent of normal.hathitrust. If all the auditory cortex were used for vision. .org/access_use#pd-google of sensory prosthesis is an extremely hopeful field of work. we might expect to get a quantity of reception of information about one per cent of that coming in through the eye. In the other direction the picture is even more favorable. our usual scale for the estimation of vision is in terms of the relative distance at which a certain degree of resolution of pattern is obtained. The eye can detect all of the nuances of the ear with the use of only one per cent of its facilities. nor do people with this amount of vision necessarily consider themselves as blind. Thus the problem Generated on 2011-09-28 15:04 GMT / Public Domain. This is about 100 : 1 as between sight and sound.GESTALT AND UNIVERSALS 167 between the areas of the two parts of the cortex. This is very poor vision : it is. definitely not blindness. and still leave a vision of about 95/100. howe ver. Google-digitized / http://www.

I am not a psychopathologist nor a psychia trist. due to the malfunction of individual components.hathitrust. I therefore wish to disclaim in advance any assertion that any particular entity in psychopathology. It is quite possible for a chain of computational operations to involve 10* separate steps. is far from having reached that state of perfection where an a priori theory can command any confidence. it is true. On the one hand. our knowledge of the normal performance of the brain and the nervous system. Google-digitized / http://www. Nevertheless. Similar questions referring to the computing machine are of great practical impor Generated on 2011-09-28 15:05 GMT / Public Domain. as for example any of the morbid conditions described by Kraepelin and his disciples.CHAPTER VII CYBERNETICS AND PSYCHO PATHOLOGY It is necessary that I commence this chapter with a disavowal. and a fortiori.org/access_use#pd-google tance. is due to a specific type of defect in the organization of the brain as a computing machine. Under these circumstances. and even to psychiatrics. the realization that the brain and the computing machine have much in common may suggest new and valid approaches to psychopathology. and lack any experience in a field where the guidance of experience is the only trustworthy one. even though.our knowledge of Iheir abnormal performance. Those who may draw such specific conclusions from the considerations of this book do so at their own risk. the reliability of modern electronic apparatus has far exceeded the most sanguine expectations. the chance that at least one operation will go amiss is very far from negligible. On the other hand. gross miscarriages of activity. These begin with perhaps the simplest question of all : how the brain avoids gross blunders. . for here a chain of operations each covering a fraction of a millisecond may last a matter of hours or days.

perhaps a tube which has burnt out and needs replacement. or the whole effective order of speed of the machine will conform to that of the slower process of checking. and is guided by them in his search for the malfunctioning part. In this case. which finds the first available element of a given sort. the machine stops. the removal and replacement of defective elements need not be the source of any appreciable delay. all data are transferred to permanent storage. and in fact the one gene rally used in practice.hathitrust. Google-digitized / http://www. is to refer every operation simultaneously to two or three separate mechanisms. but there is a signal indicating where and how the minority report differs from the majority report. To do this with a high-speed machine.OGY 169 In ordinary computational practice by hand or by desk machines. If this occurs at the first moment of discrepancy. In this case the colla tion mechanism accepts Ihe majority report. Furlhermore. there will practically always be an Generated on 2011-09-28 15:05 GMT / Public Domain. but at each stage there is a searching process. and when an error is found. quite similar to that used in automatic telephone exchanges. and if there is a discrepancy. it is the custom to check every step of the computa tion. In the case of the use of two such mechanisms. its complica tion and bulk will be increased to an intolerable point.CYBERNETICS AND PSYCHOPATHOI.org/access_use#pd-google agreement between two of the three mechanisms. if the machine is made to keep all intermediate records of its performance. and the machine need not stop. their answers are automatically collated against each other . to localize it by a backward process starting from the first point where the error is noted. no particular element is assigned to a particular stage in the sequence of operations. and this agreement will give the required result. If three separate mecha nisms are used for each stage. and a signal is sent to the operator that something is wrong. In a well-designed machine. the check must proceed with the speed of the original machine. and switches it into the sequence of operations. The operator then compares the results. A much better method of checking. and single misfunctions are as rare as they are in fact. by a factor which is likely to be enormously greater than two or three. the indication of the position of the error may be very precise. It is conceivable and not implausible that at least two of the elements of this process are also represented in the nervous .

and the like.170 CYBERNETICS system. tumors. Google-digitized / http://www. and this distinction seems to contravene the dogma of modern materialism. and probably holds most clearly for the relatively unspecialized cortical areas which serve the purpose of association and of what we call the higher mental functions. Still. it may leave that point and proceed to the next by one or more alternative members of what is known as an « iuternuncial pool ». So far we have been considering errors in performance which are normal. such as injuries. are the sequellae of general bodily disease. where this interchangeability is much limited or abolished. the brain probably works on a variant of the famous principle expounded by Lewis Carroll in The Hunting of the Snark : « What I tell you three times is true ». and that certain mental diseases. nor of a manic-depressive patient. We can hardly expect that any important message is entrusted for transmission to a single neuron. and show a pathological condition of the brain tissue . It is true that specific brain lesions.org/access_use#pd-google Psychopathology has been rather a disappointment to the instinctive materialism of the doctors.and only pathological in an extended sense. Generated on 2011-09-28 15:05 GMT / Public Domain. indeed. and these are likely to be such highly specialized parts of the cortex as those which serve as the inward exten sions of the organs of special sense. These disorders we call functional. who have taken the point of view that every disorder must be accompanied by material lesions of some specific tissue involved. the principle holds. It is also improbable that the various channels available for the transfer of information generally go from one end of their course to the other without anastomosing. There may be parts of the nervous system. may be accompanied by psychic symptoms. nor of a paranoiac. Let us now turn to those which are much more clearly pathological. Like the computing machine. nor that any important operation is entrusted to a single neuronal mechanism. that every disorder in function has some physiolo gical or anatomical basis in the tissues concerned. but there is no way of identifying the brain of a schizophrenic of one of the strict Kraepelin types. This distinction between functional and organic disorders receives a great deal of light from the consideration of the com- .hathitrust. clots. such as paresis. It is much more probable that when a message comes in to a certain level of the nervous system.

or they comprehend more and more neurons in their system. This is what we should expect to be the case in the malignant worry which ac companies anxiety neuroses. This information is stored in some physical form — in the form of memory — but part of it is in the form of circulating memories. Either. In a system containing a large number of neurons. at least — but the combination of this structure with the instructions given it at the beginning of a chain of operations and with all the additional information stored and gained from oulside in the course of this chain. and of the long-time permeability of synapses. they run their course. which must follow such primary injuries. and the significance of this chain as to the ideational content which it records.hathitrust. as in the case of memories belonging to the specious present. As we have already seen. circular processes can hardly be stable for long periods of time. and even if we knew this. as by the secondary disturbances of traffic. and the alteration of synaptic thresholds. which are stored in a way at which we can only guess. dissipate themselves. In such a case. it is possible that the patient simply does not have the room. it is not the empty physical structure of the computing machine that corresponds to the brain — to the adult brain. there maybe less going on in the brain to load . and part in the form of long-time memories. with a physical basis which vanishes when the machine is shut down. or the brain dies. not so much by the destruction of tissue which they involve. and the re-routing of messages. and die out. the sufficient number of neurons.org/access_use#pd-google effects. There is therefore nothing surprising in considering the func tional mental disorders as fundamentally diseases of memory. There is no way yet known for us to recognize in the cadaver what the threshold of a given synapse has been in life. to carry out his normal processes of thought. the overload of what remains of the nervous system. Even the grosser disorders such as paresismay produce a large part of their Generated on 2011-09-28 15:05 GMT / Public Domain. but probably also in a form with a physical basis which vanishes at death. Google-digitized / http://www. there is no way we can trace out the chain of neurons and synapses communicating with this.CYBERNKTICS AND PSYCHOPATHOLOGY 171 puting machine. until they occupy an inordinate part of the neuron pool. of the circulating information kept by the brain in the active state. Under such conditions.

sleep comes the nearest to a non-pathological clearing. for it is possible that what yet remains may be adequate tor our purpose. the permanent memory becomes more and more deeply involved. there is no normal process except death which com pletely clears the brain from all past impressions . nor indeed is a . These contingencies may depend on a highly improbable instantaneous configuration of the system. Pathological processes of a somewhat similar nature are not unknown in the case of mechanical or electrical computing machines. when they occur. Furthermore. sleep does not clear away the deeper memories. so that they are all the more readily involved in the expanding process. in the hope that we may reach the inaccessible par). Now. may never — or very rarely — repeat themselves. subject it to an abnormally large electrical impulse. may build itself up into a process totally destructive to the or dinary mental life. and throw it into a position where the false cycle of its activities will be inter rupted. and the pathological process which occurred at first at the level of the circulating memories may repeat itself in a more intrac table form at the level of the permanent memories. Generated on 2011-09-28 15:05 GMT / Public Domain. Failing this. How do we deal with these accidents in the use of the machine? The first thing which we try is to clear the machine of all infor mation. and after death.172 CYBERNETICS up the neurons not yet affected. it is impossible to set it going again. Thus what started as a relatively trivial and accidental reversal of stability. or if it is electrical.hathitrust. A tooth of a wheel may slip under just such con ditions that no tooth with which it engages can pull it back into its normal relations. Of all normal pro cesses. they temporarily put the machine out of action. if the difficulty is in some point permanently or temporarily inaccessible to the clearing mechanism. we may disconnect an erring part of the apparatus. we shake the machine. in the hope that when it starts again with different data. How often we find that the best way to handle a complicated worry. or a high-speed electrical computing machine may go into a circular process which there seems to be no way to stop. However.org/access_use#pd-google the difficulty may not recur. is to sleep over it! However. and when remedied. If even this fails. or an intellectual muddle. Google-digitized / http://www.

as the mammalian central nervous system seems to possess no powers whatever of regeneration.CYBERNETICS AND PSYCHOPATHOLOGY 173 sufficiently malignant state of worry compatible with an adequate sleep. insulin. This does not prevent its being in many cases the best thing we can do at present. metrazol — are less drastic methods of doing a very similar thing. imperfectly controlled method to interrupt a mental vicious circle. but they do have a decidedly damaging effect on the memory. though it is not impossible that they may have some effect here too. As we have said. They do not destroy brain tissue or at least are not intended Generated on 2011-09-28 15:06 GMT / Public Domain. prefrontal lobotomy does seem to have a genuine effect on malignant worry. The various forms of shock treatment — electric. More generally it appears to limit all aspects of the circulating memory. the ability to keep in mind a situation not actually presented.org/access_use#pd-google to destroy it. The principal type of surgical intervention which has been practiced is known aspre- frontal lobotomy. known in the termi nology of another profession as the conscience. However. and in so far as this memory is chiefly damaged for the recent period of mental disorder. in long-established cases . but it is not always free from deleterious effects on the permanent memory and the personality. leaving behind it permanent damage. As it stands at present. but by damaging or des troying the capacity for maintained worry. mutilation. Google-digitized / http://www. and consists in the removal or isolation of a portion of the prcfrontal lobe of the cortex. We are thus often forced to resort to more violent types of intervention in the memory cycle. shock treatment has something definite to re commend it as against lobotomy . In so far as this concerns the circulating memory. imperfectly under stood. and the abridgementof the powers of the victim. It has recently been having a certain vogue. probably not unconnected with the fact that it makes the custodial care of many patients easier.hathitrust. The more violent of these involve a surgical intervention into the brain. not by bringing the patient nearer to a solution of his problems. and is probably scarcely worth pre serving anyhow. Let me remark in passing that killing them makes their custodial care still easier. it is another violent. Lobotomy and shock treatment are methods which by their very nature are more suited to handle vicious circulating memories and malignant worries than the deeper-seated perma nent memories.

affective though generally invo luntary . the permanent memory is as badly deranged as the circulating memory. why there are circumstances where a joint use of shock treatment and psycho therapy is indicated. and is much richer and more varied than that which is accessible by direct unaided introspection . We do not seem to possess any purely pharmaceutical or surgical weapon for intervening differentially in the permanent memory.174 CYBERNETICS of mental disorder. our treatment is clearly based on the concept that the stored information of the mind lies on many levels of acces sibility. It has been commented on by many writers. Thompson. or because they have been buried by a definite mechanism. . Growth and Form. to make the patient accept them for what they are. modify. and a psychological therapy for the long-time memories which. It perhaps explains. and that the content of these stored experiences as well as their affective tone condition much of our later activity in ways which may well be pathological. beyond which it will not function. if not their content. or whether our psychoterapy is not strictly psychoanalytic at all. combining a physical or pharmacological therapy for the phenomena of reverberation in the nervous system. too. All this is perfectly consistent with the point of view of this book. 2°d Ed. Google-digitized / http://www. Whether psychoanalysis is taken in the orthodox Freudian sense or in the modified senses of Jung and of Adler.org/access_use#pd-google them less harmful. without interference. either because they never were made explicit in our adult language. The technique of the psychoanalyst consists in a series of means to discover and in terpret these hidden memories. We have already mentioned the traffic problem of the nervous system. might re-establish from within the vicious circle broken up by the shock treatment. and by their acceptance. and thus make Generated on 2011-09-28 15:06 GMT / Public Domain. D'Arcy. This is where psy choanalysis and other similar psychotherapeutic measures come in. at least the affective tone they carry. that it is vitally conditioned by affective experiences which we cannot always uncover by such introspection. Thus the insect organization is limited by the length of tubing over which the spiracle method of bringing air by diffusion directly to the brea- 1. (1945).hathitrust. such as d'Arcy Thompson Mhat each form of organization has an upper limit of size.

the important limiting factor is the frac tion of the time during which a subscriber will Gnd it impos sible to put a call through. we must have 98 per cent chance of performance at each stage. a tree is limited by the mechanism for transferring water and minerals from the roots to the leaves. the elevator space needed for the upper stories consumes an excessive part of the cross- section of the lower floors. in order to get a probability of total success equal to p. the probability of success at each stage must be p1/«. a land animal cannot be so big that the legs or other portions in contact with the ground will be crushed by its weight. To obtain 50 per cent of performance. Simi larly. Google-digitized / http://www. To obtain 90 per cent of perfor mance. and this limitation has been very thoroughly studied by telephone engineers. we must have about 95 per cent chance of success per stage. and the products of photosynthesis from the leaves to the roots. A success of Generated on 2011-09-28 15:06 GMT / Public Domain. and probability of failure is inde pendent. we must have 87 per cent of success at each stage.org/access_use#pd-google 75 per cent is already annoying. In a telephone system. and so on. the best possible suspension bridge which can be built out of materials with given elastic properties will collapse under its own weight. 90 per cent of successful calls is probably good enough to permit bus iness to be carried on with reasonable facility. while if half the calls end in failures. and extremely good when this . and beyond a certain greater span. The same sort of thing is observed in engineering constructions. the size of a single telephone central. and equal for each stage. A 99 per cent chance of success will certainly be satisfactory for even the most exacting . Beyond a certain span.CYBERNETICS AND PSYCHOPATHOLOGY 175 thing tissues will function. is limited. Skyscrapers are limited in size by the fact that. Thus to obtain a 75 per cent chance of the comple tion of the call after five stages. If the calls go through n distinct stages of switching. non-expanding plan. but will permit business to be carried on after a fashion . any structure built of a given material or materials will collapse under its own weight. built according to a constant. the more rapidly the service becomes extremely bad when a critical level of failure for the individual call is exceeded.hathitrust. subscribers will begin to ask to have their telephones taken out. when they exceed a certain height. these represent all-over figures. Now. It will be seen that the more stages which are involved.

when he will give way in a serious and catas trophic way. This will first affect the faculties or operations involving the longest chains of neurons. or by Ihe excessive occupation of such channels by undesirable syslems of traffic. and we have a catastrophic traffic jam. Man. very possibly amounting to insanity. is then likely to perform a complicated lype of behavior efficiently very close to the edge of an overload. and we shall have a form of mental breakdown. with behavior that probably depends on the longest chains of effectively operated neuronic chains. Thus a switching service involving many stages and designed for a certain level of failure shows no obvious signs of failure until the traffic comes up to the edge of the critical point. There is another more specific way of considering a very similar matter. Thus the amount of assistance a process receives through a rise in temperature is a rough measure of the length of the neuron chain it involves. like circulating memories which have increased to the extent of becoming pathological worries.hathitrust. any facilitation of a process in a single neuron-synapse system should be cumulative as the neuron is combined in series with other neurons. In all these cases.176 CYBERNETICS critical level of failure is not quite reached. There is appreciable. We thus see that the superiority of the human brain to others in the length of the neuron chains it employs is a reason why mental disorders are certainly most conspicuous and probably most common in man. This is greater for the higher pro cesses. by a physical removal of channels for the carrying of traffic. Let us first consider two brains geometrically similar. with the weights of gray and of white matter related by the same factor of proportionality.org/access_use#pd-google a rise in temperature within nearly physiological limits is known to produce an increase in the ease of performance of most if not of all neuronic processes. with the best developed nervous system of all the animals. This overload may take place in several ways : either by an excess in the amount of traffic to be carried. Now. when it goes com pletely to pieces. Google-digitized / http://www. evidence that these are precisely the processes which are recognized to be the highest in our ordinary scale of valuation. a point will come — quite suddenly — when the normal traffic will not have space enough allotted to it. roughly in the order of our usual estimate of their degree of « highness ». but . The evidence is this: Generated on 2011-09-28 15:06 GMT / Public Domain.

Then the number of cell-bodies in the two cases bears the ratio A3: B3.hathitrust. This is indicated by the remarkably small damage done by cutting some of the long distance cerebral loops of white matter. That is. and a considerable number of association areas. These are precisely the pro cesses which we should normally class as higher. Thus the human brain would seem to be fairly efficient in the matter of the short- distance connectors. Google-digitized / http://www. It almost seems as if these superficial connections were so inadequate that they only furnish a small part of the connections really needed. but it is spread over a far more involved system of gyri and snlci. when it comes to the connectors between different gyri. the processes involving parts of the brain quite remote from one another should sutler first.CYBERNETICS AND PSYCHOPATHOLOGY 177 with different linear dimensions in Ihe ratio A : B. The effect of this is to increase the amount of grey matter at Ihe expense of the amount of white matter. Within a gyrus. There is some evidence that the long-distance paths in the brain have a tendency to run outside of the cerebrum altogether. we shall find that it is much more convoluted. If we compare the human brain with that of a lower mammal. This means that in the case of a traffic Generated on 2011-09-28 15:07 GMT / Public Domain. a number of different motor processes. Let the volume of the cell-bodies in the gray matter and the cross-sec tions of the fibres in the white matter be of the same size in both brains. as the opposing folds of a gyrus are nearer together than they would be on a smooth-surfaced brain of the same size. processes involving several centers. the density of activity in the fibres is A : B times as great in the case of the large brain as in that of the small brain.org/access_use#pd-google jam. and to traverse the lower centers. and we obtain another confirmation of our expectation. the distance they have to run is increased if anything by the convolution of the brain. should be among the least stable in cases of insanity. This means that for the same density of activity in the cells. and the number of long-distance connectors the ratio A2:B2. . this decrease of the white matter is largely a decrease in length rather than in number of fibres. which seems to be verified by experience. On the other hand. but quite defective in the matter of long distance trunk lines. The relative thickness of the gray matter is much the same. that the higher processes deteriorate first in insanity.

the choice between the right and the left side in muscular skill does actually seem to be less than in innn even in the lower primates. for example — are represented. a hemiplegia. » There certainly were extensive lesions of the parietal and temporal regions. the effect of an extensive injury in the secondary hemisphere is far less serious than the effect of a similar injury in the dominant hemisphere. so that the patient ap pears far more nearly normal than he would be. an extensive injury to the dominant hemisphere may compel the normally secondary hemisphere to take its place.178 CYBERNETICS With reference to this. though it is less conspicuous than in man.org/access_use#pd-google with a moderate degree of one-sided paralysis. and one of these. is generally associated with a left-brainedness. However. the cerebral functions are not distributed evenly over the two hemispheres. « he had only half a brain. and that in the first six months of life. the dominant hemisphere. This is quite in accordance with the . It is said that the situation is considerably better in early infancy. in the adult. Nevertheless. Handedness seems to occur in the lower mammals. so extensive that it has been said that after his injury. Nevertheless. has the lion's share of the higher functions.hathitrust. his brain was examined. When he died. each i« its appropriate hemisphere. had the injury occurred at a later stage. Google-digitized / http://www. Pasteur suffered a cerebral hemorrhage on his right side which left him Generated on 2011-09-28 15:07 GMT / Public Domain. the phenomena of handedness and of hemispheric dominance are interesting. after this injury. The right-handedness of the normal man. as is well known. For example. and he was found to be suffering from a right-sided injury. he did some of his best work. most of the « higher » areas are confined to the dominant hemisphere. though this is not true for all bilateral functions. That is. At a relatively early age in his career. It is true that many essentially bilateral functions — those involving the fields of vision. and the left-hand- edness of a minority of humans with a right-brainedness. and would certainly reduce the patient into an animal condition of mental and nervous crippled ness. A similar injury of the left side in a right- handed adult would almost certainly have been fatal. probably in part because of the lower degree of organi zation and skill demanded by the tasks which they perform.

speech. and stuttering is the most natural thing in the world. the natural handed ness and cerebral dominance are established for life. Now. In the past. It used to be thought that left-handedness was a serious social disad vantage. and sports equipment primarily made for the right-handed. they must do this again and again. the neuron chains involved in processes of the sort must cross over from hemisphere to hemisphere and back . and even succeeded. Since however these motions are carried out in the closest possible association with reading. It is possible that. scanty. but which are cer tainly long. and the inter- hemispheric traffic must go by roundabout routes through the brain-stem . the processes associated with speech and writing are very likely to be involved in a traffic jam. in changing the external handedness of their children by education. It was then found that in very many cases these hemispheric changelings suffered from stuttering and other defects of speech. How ever. With most tools. to the extent of seriously wounding their prospects in life and their hopes for a normal career. As a conse quence. From a combination of motives. and in a process of any complication. and other acti vities which are inseparably connected with the dominant hemis phere. the direct connectors between the hemispheres — the cere bral commissures. are so few in number that they are of very little use. reading.CYBERNETICS AND PSYCHOPATHOLOGY 179 general great flexibility shown by the nervous system in the early weeks of life. • in a brain as large as that of man. and the great rigidity which it rapidly devel ops later. moreover.org/access_use#pd-google hemisphere which deals with skilled motions. such as wri ting. it was viewed with some of the superstitious disapproval that has attached to so many minor variations from the human norm. long before the child is of school age. short of such serious injuries. which we know very imperfectly. and subject to interruption. .hathitrust. With the education of the secondary hand there has been a partial education of that part of the secondary Generated on 2011-09-28 15:07 GMT / Public Domain. Google-digitized / http://www. school desks. handedness is reasonably flexible in the very young child. though of course they could not change its physiological basis in hemispheric dominance. many people have attemp ted. it certainly is to some extent. We now see at least one possible explanation for the phe nomenon. such as birthmarks or red hair. and writing.

At any rate. Generated on 2011-09-28 15:07 GMT / Public Domain. In man. the gain achieved by the increase. the destruction of the dominant hemisphere seems to produce relatively less damage than in man. and the destruction of the secondary hemisphere probably more damage. is partly nullified by the fact that less of the organ can be used effectively at one time. Google-digitized / http://www. in which highly specialized organs reach a level of declining efficiency. It is interesting to reflect that we may be facing one of those limitations of nature. In a cat. The human brain may be as far along on its road to this destructive specia lization as the great nose horns of the last of the titanotheres. in the size and complication of the brain.180 CYBERNETICS That is.hathitrust. and ultimately lead to the extinction of the species. the apportionment of function in the two hemispheres is more nearly equal. the human brain is probably too large already to use in an efficient manner all the facilities which seem to be anato mically present.org/access_use#pd-google f .

and the support of the colony as a whole. such as the blood cor puscles. The Levia than of Hobbes. the Holy Roman empire and its similarly constituted feudal contemporaries. It is in fact scarcely more than a philosophical anticipation of the cell theory. is neither unfamiliar nor new. the hives . is an illustration of the same idea one stage lower in scale. Strictly speaking. the United States of America and the many United States to the south of it. the support.org/access_use#pd-google the several individuals are modified in different ways to serve the nutrition. Google-digitized / http://www. are all examples of hierarchies of organizations on the political sphere. are made up of units. the locomotion.hathitrust. the reproduction. The loose federations of ancient Greece. which have many if not all the attributes of independent living organism. AND SOCIETY The concept of an organization. where Generated on 2011-09-28 15:07 GMT / Public Domain. the Man-Slate made up of lesser men.CHAPTER VIII i INFORMATION. the Swiss Companions of the Oath. according to which most of the animals and plants of moderate size. while Leibniz's is treatment of the living organism as being really a plenum. the excretion. the Union of Socialist Soviet Republics. cells. wherein other living organisms. which is complex structure of differentiated coelenterate polyps. the beaver colonies. have their life. It is very different with man and the olher social animals — with the herds of baboons or cattle. The multicellular organisms may themselves be the building-bricks of organisms of a higher stage. such as the Portuguese Man-of-War. the United Netherlands. is but another step in the same direction. the elements of which are themselves small organizations. LANGUAGE. and all of those of large dimensions. such a physically conjoint colony as that poses no question of organization which is philosophically deep er than those which arise at a lower level of individuality.

It certainly can distinguish an ant from its own nest from an ant from a foreign nest. This intercommunication can vary greatly in complexity and content. chitin-bound. All the nervous tissue of the bee hive is the nervous tissue of some single bee : how then does the beehive act in unison. is Generated on 2011-09-28 15:08 GMT / Public Domain. It is hard to see how it can be purely hormonal in quantities as small as those which are readily perceivable . indis pensable. general and undirectional as it is. adapted. destroy the other.182 CYBERNETICS of bees. exterior hormones. Google-digitized / http://www. and to a large extent. once they reach the organ of smell. casto- reum. and the like sexually attractive substances in the mam mals may be regarded as communal. By this I do not mean to assert that the inner action of these substances. Within a few outside reactions of this kind. while the community consists of individuals with shifting relations in space and time. and may cooperate with the one.org/access_use#pd-google rigidly separated from the phase of mature activity. and at that in a very variable. civet. the nests of wasps or ants. the ant seems to have a mind almost as patterned. is not unlike the hormonal influences within the body. we know too little . especially in solitary animals. With the ants. The degree of integration ot the life of the community may very well approach the level shown in the conduct of a single individual. whose learning phase. one of the chemical senses. with permanent topographic relations between the elements and permanent con nections. is hormonal rather than nervous. on the other hand. organized unison? Obviously the secret is in the intercommu nication of its members. It is very improbable that an ant can distinguish one ant from another. and very much besides. and no permanent.hathitrust. as its body. Let it be remarked parenthetically that musk. it pro bably does not cover much more than a few smells. for the bringing the sexes together at the proper time. Indeed smell. The only means of communication we can trace in them are as general and diffuse as the hormonal system of communication within the body. With man. it embraces the whole intricacy of language and literature. It is what we might expect a priori from an animal whose growing phase. yet the individual will probably have a flxed nervous system. unbreak able physical connections. and serve for the conti nuation of the race.

but on the whole nervous constitution of the sender and the receiver of the stimulus as well. Moreover. Whatever means of communication the race may have. who cannot speak my language. the long.intrinsic content may acquire meaning in his mind by what he observes at the time. and to distinguish it from the amount of information available to the individual. I do not care to pronounce an opinion on this matter. active attention is in itself a language as varied in possibilities as the range of impres sions that the two of us are able to encompass. All 1 need to do is to be alert to those moments when he shows the signs of emotion or interest. Certainly no informa tion available to the individual is also available to the race. Thus social animals may have an active. it is possible to define and to measure the amount of information available to the race. The ability thathe has to pick out the moments of my special. and some of the carcino gens. LANGUAGE. Even without any code of sign language common to the two of us. I then cast my eyes around. not because he has communicated them to me by lan Generated on 2011-09-28 15:08 GMT / Public Domain. some of the vitamins. It will not be long before I discover the things which seem important to him. twisted rings of carbon atoms found inmuskone and civetone do not need too much rearrange ment to form the linked ring structure characteristic of the sex hormones. not only on the information conveyed by the stimulus itself.INFORMATION. perhaps paying special attention io the direction of his glance. I leave it as an interesting speculation. and whose language I cannot speak. flexible means of com munication long before the development of language. and fix in my memory what I see or hear. but the value of a simple stimulus such as an odor for convering information depends. but because I myself have observed them.org/access_use#pd-google guage. Google-digitized / http://www. and may aquire mea ning in my mind by what I observe at the time. AND SOCIETY 183 of the action of the hormones to deny the possibility of the hormonal action of vanishingly small quantities of such sub stances. Suppose I find myself in the woods with an intelligent savage. unless it is distin- . The odors perceived by the ant seem to lead to a highly standardized course of conduct . In other words.hathitrust. nor is even that behavior of racial significance. intelligent. a signal without an . unless it modifies the behavior of one individual to another. 1 can learn a great deal from him.

and so on. temporarily assembled.184 CYBERNETICS guishable by olher individuals from other forms of behavior. There is a well-known tendency of libraries to become clogged by their own volume. but they are limited by the impossibility of classifying a book under an unfamiliar heading unless some particular person has already . A group of non-social animals. Vannevar Bush has sugges ted the use of mechanical aids for the searching through vast bodies of material. A group may have more group information or less group information than its members. in all probability.hathitrust. of the sciences to develop such a degree of specialization that the expert is often illiterate out side his own minute specialty. This is really too broad a term for the scope of most communal information. I have spoken of the race. Thus the question as to whether a certain piece of information is racial. depends on whether it results in the individual assuming a form of activity which c«n be recognized as a distinct form of activity by other members of the race. even though its members may possess much information as indi viduals. A measure of the effective size of a group is given by the size which it must have to have achieved a certain stated degree of autonomy. and is acted on by them in a way that Generated on 2011-09-28 15:09 GMT / Public Domain. in the sense that it will in turn affect their activity. Properly spea king. As in the case of the individual. These probably have their uses. contains very little group information. the human orga nism contains vastly more information. On the other hand. Dr. We can thus measure the autonomy of the group. not all the information which is available to the race at one time is accessible without special effort. There is thus no necessary relation in either direction between the amount of racial or tribal or community information and the amount of information available to the individual.org/access_use#pd-google goes further in the group. the community extends only so far as there extends an effectual transmission of information. than does any one of its cells. or of purely private availability. by comparing the number of decisions entering a group from outside wilh the number of decisions made in the group. This is because very little that one member does is noticed by the others. Google-digitized / http://www. It is possible to give a sort of measure to this.

current in many countries. however. plays in accordance with a completely intelligent policy. the evidence. this still requires some individual with an almost Leibnizian catho licity of interest. This is associated with the very comforting view that the individual entrepreneur. which has indeed received a simulacrum in the family game of Monopoly. turncoatism. developed by von Neumann and Morgenstern. Google-digitized / http://www. In many cases. There is a belief. AND SOCIETY 185 recognized the relevance of that heading for that particular book. It is thus the market game as played between perfec tly intelligent. although it often leads to the choice of a definite line of play. or the closely related lives . and has thus earned the great rewards with which so ciety has showered him. is in some manner a public bene factor. and in the overwhelming majo rity of cases. in seeking to forward his own interest. when the number of players is large. but belong to widely separated fields. LANGUAGE. which will in the end assure him of the greatest possible expectation of reward. the individual selfishness of the bargainers. The market is a game. In the case where two subjects have the same techniques and intellectual content. one of the most surprising facts about the body politic is its extreme lack of efficient homeostatic processes. where there are three players. The individual players are compelled by their own cupidity to form coalitions. It is thus strictly subject to the general the ory of games.org/access_use#pd-google This theory is based on the assumption that each player. is against this simple-minded theory. determinate way. Generated on 2011-09-28 15:09 GMT / Public Domain. will result in the end in a stable dynamics of prices. and deception. the theory is complicated. at every stage. in view of the information then available to him. such as it is. but these coalitions do not generally establish themselves in any single. Unfortunately. In connection with the effective amount of communal infor mation. the result is one of extreme indeterminacy and instability.hathitrust. which has been elevated to the rank of an official article of faith in the United States. which is only too true a picture of the higher business life. Even in the case of two players. and with redound to the greatest common good. each seeking to sell as high and buy as low as possible. perfectly ruthless operators. that free competition is itself a homeostatic process : that in a free market.INFORMATION. and usually terminate in a welter of betrayal.

the fool operates in a manner which. We are involved in the business cycles of boom and failure. sociologists. straw votes. Luckily for us. is an abstraction and a perversion of the facts. we have our machinery of radio fan-ratings. pornography. It is rare to find a large number of thoroughly clever and unprincipled persons playing a game together. these merchants of lies. and where the fools are present in sufficient numbers. Where the knaves assemble.hathitrust. after the fashion of von Neumann's game sters. and the great rewards are reserved for the one who watches for an opportune time to break his agreement and betray his companions. induce him to vote for a particular candidate — any candidate — or to join in a political witch hunt. these exploiters of gullibility. The average man is quite reasonably intelligent concerning subjects which come to his direct atten- . This policy of lies — or rather. The psychology of the fool has become a subject well worth the serious attention of the knaves. In the long run. completely ruthless person.186 CYBERNETICS of politics. that policy will. in the wars which everyone loses. or so Generated on 2011-09-28 15:09 GMT / Public Domain. Google-digitized / http://www. even the most brilliant and unprincipled huckster must expect ruin . and pseudo-science will sell an illustrated newspaper. there will always be fools. There is no homeostasis whatever. and intimidation will induce a young scientist to work on guided missiles or the atomic bomb. This is because no man is either all fool or all knave. they offer a more profitable object of exploitation for the knaves. and there are always the statisticians. diplomacy. have not yet arrived at such a pitch of perfection as to have things all their own way. To determine these. which are so real a feature of modern times. A certain precise mixture of religion.org/access_use#pd-google the party hopes. Instead of looking out for his own ultimate interest. bribery. and economists available to sell their services to these undertakings. and war. von Neumann's picture of the player as a completely intelligent. by and large. A certain blend of wheedling. in the successions of dictatorship and revolution. is as predictable as the struggles of a rat in a maze. and other psychological investigations with the com mon man as Iheir object. of statements irrelevant to the truth — will make him buy a particular brand of cigarettes. but let the hucksters become tired of this. opinion samplings. and agree to live in peace with one another. Naturally.

the telephone system. of tolerance for those who have offended once or twice against society. or villages of primitive savages. retention.hathitrust.INFORMATION. whether they are highly literate communities in a civilized country. from private criticism by the laws of libel and the possession of the means of communication. both as it concerns books and as it concerns newspapers. and quite reasonably altruistic in matters of public benefit or private suffering which are brought before his own eyes. the movies. On the other hand. It is only in the large community. and transmission of information. Besides their intrinsic importance as means of communication.org/access_use#pd-google wealth. Google-digitized / http://www. The newspaper is a vehicle for advertisement. in such a community. Of all of these anti-homeostatic factors in society. the schools. In a small country community which has been running long enough to have developed somewhat uniform levels of intelligence and behavior. Strange and even repugnant as the customs of many barbarians may seem to us. the theater. where the Lords of Things as They Are protect themselves from hunger by Generated on 2011-09-28 15:09 GMT / Public Domain. Thus small. LANGUAGE. these people are there. After all. use. There are ways of making him feel the weight of public opinion. they generally have a very definite homeostatic value. the telegraph. these means are the press. the control of the means of communication is the most effective and most important. and the church. AND SOCIETY 187 tion. there is a very respectable standard of care for the unfortunate. One of the lessons of the present book is that any organism is held together in this action by the possession of means for the acquisition. so unavoidable. that ruthlessness can reach its most sublime levels. it does not do for a man to have the habit of over-reaching his neighbors. he will find it so ubiquitous. and the rest of the community must continue to live with Ihem. After a while. of administration of roads and other public facilities. which it is part of the function of anthropolo gists to interpret. so restricting and oppressing. that he will have to leave the community in self defense. closely knit communities have a very considerable measure of homeostasis. and this. and an instrument for the monetary . secondary functions. the radio. In a society too large for the direct contact of its members. each of these serves other. the posts. from public opinion by privacy and anonymity.

and thus natu rally express the opinions of that class . which we have already seen to be one of the chief anti-homeostatic elements in the community. The college president and the Bishop. This is aided by the very elaboration and the consequent expense of the means themselves. as are also the movies and the radio. That system which more than all others should contribute to social homeos- tasis is thrown directly into the hands of those most concerned in the game of power and money. to say nothing of the . the man who pays the piper calls the tune. The book that does not earn money for its publisher probably does not get printed. and the further fact that. its syndicated features. The great news services cost too much to be available to the publisher of moderate means. these secondary aspects of the means of com munication tend to encroach further and further on the primary ones. in which all natural and human resources are regarded as the absolute property of the first business man enterprising enough to exploit them. Google-digitized / http://www. Thus on all sides we have a triple constriction of the means of communication : the elimination of the less profitable means in favor of the more profitable. subject to this disruptive influence.org/access_use#pd-google institutions to run. that the larger communities. its political opinions. contain far less communally available information than the smaller communities. The school and the church are not merely refuges for the scholar and the saint: they are also the home of the Great Educator and the Bishop. as one of the chief avenues to political and personal power. The country paper may continue to use its own reporters to canvass the villages around for gossip. avowedly based on buying and selling. The book publishers concentrate on books that are likely to be acceptable to some book club which buys out the whole of an enormous edition. and can only seek their money where the money is. even if they have no personal ambitions for power. the fact that these means are in the hands of Ihe very limited class of wealthy men. It is no wonder. The radio depends on its advertisers for income.188 CYBERNETICS gain of its proprietor. they attract above all those ambitious for such power. In a society like ours. and certainly does not get reprinted.as stereotyped « boiler plate ». and as everywhere. then. but it buys its national news.hathitrust. have expensive Generated on 2011-09-28 15:09 GMT / Public Domain.

the methods of the natural sciences. they come to believe it possible. In this.INFORMATION. we cannot attribute too much value to this type of wishful thinking. Some of it is due to a sense of the possibilities of a large organization for good. of economics. and the like. in the hope of achieving a like measure of success in the social fields. however. to assume that because the community is larger than the individual. they consider that the main task of the immediate future is to extend to the fields of anthropology. Google-digitized / http://www. Like the wolf-pack. and in whom an optimistic feeling that Ihere must be some way out has led to an overvaluation of the possible homeostatic elements in the community. All the great successes in precise science have been made in fields where there is a certain high degree of isolation of the . Undoubtedly it would be very pleasant for us mice if the predatory cats of this world were to be belled. they show an excessive optimism. is nothing more than an eye for the main chance and a lusting after the fleshpots of Egypt. There is another group of those who see nothing good in the anarchy of modern society. From believing this necessary. Much as we may sympathize with these individuals.hathitrust. hopes which some of my friends have built for the social efficacy of whatever new ways of thinking this book may contain. it is also more intelligent. and I think false. the State is stupider than most of its components. of sociology. Some of this opinion is due to no more than a childish delight in the large and the lavish. because of the considerable. and appreciate the emotional dilemma is which they find themselves. Not a little of it. but — who is going to do it ? Who is to assure us that ruthless power Generated on 2011-09-28 15:10 GMT / Public Domain. and a misunderstanding of the nature of all scientific achievement. LANGUAGE. heads of great laboratories. This runs counter to a tendency much voiced among business executives. although let us hope to a lesser extent. It is the mode of thought of the mice when faced with the problem of belling the cat. They are certain that our control over our material environment has far outgrown our control over our social environment and our understanding thereof.org/access_use#pd-google will not find its way back into the hands of those most avid for it? I mention this matter. I maintain. AND SOCIETY 189 human elements of which all communities are built up. Therefore.

not to speak of his mere glance. and the events that might be of the greatest significance from the point of view of an observer conforming to their scale of existence. It is in the social sciences that the coupling between the observed phenomenon and the observer is hardest to minimize. In modern atomic physics. appear to us —with some exceptions. we do not live on the scale of the particles concerned. we are too small to influence the stars in their courses. There is much in the social habits of a people which is dispersed and distorted by the mere act of making inquiries about it. We have seen in the case of astronomy that this may result from the enormous scale of certain phenomena with respect to man. the social scientist has not the advantage of looking down on his subjects from the cold heights of eternity . On the one hand.190 CYBERNETICS phenomenon from the observer. In another sense from that in which it is usually staled. In short. it is true that anything we do will have an influence on many individual particles which is great/rom the point of view of that particle. and honesty of purpose of my anthropologist friends. As far as these effects are concerned.hathitrust. so that man's mightiest efforts. However. on the other hand. atoms. the observer is able to exert a considerable influence on the phenomena that come to his attention. Many a missionary has fixed his own misunder standings of a primitive language as law eternal in the process of reducing it to writing. and electrons. With all respects to the intelligence. Google-digitized / http://www. In both cases. traduttore traditore. skill. and our statistical theories have an admirably adequate basis.org/access_use#pd-google not be loose enough for us to be able to ignore it altogether. On the other hand. I cannot think that any community which they have investigated will ever be quite the same afterwards. either in space or in time . cannot make the slightest visible impression on the celestial world. and too large to care about anything but the mass effects of molecules. it is true. only as average mass effects in which enormous populations of particles cooperate. we achieve a sufficiently loose coupling with the phenomena we are studying to give a massive total account of this coupling. the intervals of time concerned are large from the point of view of the individual particle and its motion. the science of the unspeakably minute. as in the Wilson cloud- chamber experiments. although the coupling may Generated on 2011-09-28 15:10 GMT / Public Domain.

In other words. but this is not a sociology in which we. pleasures and agonies. LANGUAGE. In short. whether we like it or not. in short. observed like the populations of Drosophila in a bottle. nor can we be sure that a considerable part of what we observe is not an artefact of our own creation.« scientific ». significant information Generated on 2011-09-28 15:10 GMT / Public Domain. can never furnish us with a quantity of verificable. whether our investi gations in the social sciences be statistical or dynamic —and they should participate in the nature of both — they can never be good to more than a very few decimal places. career. are particularly interested. and death of people whose life-scale is much the same as his own. and. . We are too much in tune with the objects of our investigation to be good probes. An investigation of the stock market is likely to upset the stock market.hathitrust.sub specie aeternitatis.org/access_use#pd-google which begins to compare with that which we have learned to expect in the natural sciences. or at least.INFORMATION.education. Your economist is most interested in predicting such business cycles as run their course in less than a generation . Google-digitized / http://www. Few philosophers of politics nowadays care to confine their investigations to the world of Ideas of Plato. We are not much concerned about human rises and falls. neither should we build exaggerated expectations of their possibilities. have repercussions which affect a-man differentially at different stages of his career. It may be that there is a mass sociology of the human animalcule. We cannot afford to neglect them . in the social sciences we have to deal with short statistical runs. to the un. narrative method of the professional historian. There is much which we must leave. Your anthropologist reports the customs associated with the life. who are human animalcules ourselves. AND SOCIETY 191 and ubiquity.

Generated on 2011-09-28 15:10 GMT / Public Domain.hathitrust.org/access_use#pd-google . Google-digitized / http://www.

To each sequence of moves it should assign a certain conventional valuation. to Generated on 2011-09-28 15:10 GMT / Public Domain. At the other end. checking.hathitrust. At the stage at which the machine is to play once and the opponent once.it is unquestion ably possible to construct a machine that will play chess in the sense of following the rules of the game. taking opponents' pieces. The machine must actually play — at a high speed if possible — all its own admissible moves and all the opponent's admissible ripostes for two or three moves ahead.NOTE There is one question which properly belongs to this chapter. but not altogether trivial. should receive valuations not too remote from those which good playerswould assign them. though it in no sense represents a culmination of its argument. the valuation of a play by . This is essentially no more difficult than the construction of a system of interlocking signals for a railway signal tower. Here. and other recognizable situations. Note that we need not raise the question as to whether it is possible to construct a machine which will play an optimum game in the sense of von Neumann. The real problem is intermediate : to construct a machine which shall offer interesting opposition to a player at some one of the many levels at which human chess players find themselves. apparatus for this purpose. and whether this sort of ability represents an essential difference between the potentialities of the machine and the mind. The first of an entire sequence of moves should receive a valuation much as von Neumann's theory would assign it. to be checkmated the lowest . irrespective of the merit of the play. I think it is possible to construct a relatively crude.org/access_use#pd-google checkmate the opponent receives the highest valuation at each stage. Not even the best human brain approximates to this. It is the question whether it is possible to construct a chess- playing machine. while losing pieces. Google-digitized / http://www.

Such a machine would not only play legal chess. but a chess not so manifestly bad as to be ridiculous. This does not mean that it would reach the degree of proficiency of Maelzel's frau dulent machine. if there were a mate possible in two or three moves. Google-digitized / http://www. .194 CYBERNETICS the machine is the minimum valuation of the situation after the opponent has made all possible plays. the machine would make it. but for all that. This it makes as its definitive play. It would pro bably win over a stupid or careless chess player. Then the machine chooses any one of the plays giving the maximum valuation for the stage n plays ahead. it may attain a pretty fair Generated on 2011-09-28 15:11 GMT / Public Domain. the machine would avoid it. At any stage. At the stage where the machine is to play twice and the opponent twice. where n h»s some value on which the designer of the machine has de cided. and so on. it might very well beasgood aplayer as the vast majority of the human race. This process can be exlended to the case when each player makes three plays. the valuation of a play by the machine is the minimum with respect to the opponent's first play of the maximum valuation of the plays by the machine at the stage when there is only one play of the op ponent and one by the machine to follow.hathitrust. and if it were possible to avoid a mate by the enemy in two or three moves. Inot her words. and would almost cerliiinly lose to a careful player of any considerable degree of proficiency.org/access_use#pd-google level of accomplishment.