WHERE IS FANCY BRED?1 [152]

W.S. McCulloch

Abstract

Since 1952 Dr. Warren S. McCulloch has been a staff member engaged in research in the Research Laboratory of Electronics at the Massachusetts Institute of Technology.

After receiving his degree in medicine from Columbia University in 1927, Dr. McCulloch interned at Bellevue Hospital, the 2nd Division on the Neurological Service. He was also a resident there. The next two years were spent in research on experimental epilepsy at the Neurosurgical Laboratory and Department of Neurology at Columbia. This was followed by a year of work in the field of head injuries at Bellevue Hospital in the Laboratory of Experimental Neurology.

Dr. McCulloch taught physiological psychology early in his career and spent a year at New York University doing graduate work in mathematical physics.

The years 1932 to 1934 were spent in study on the admission service of the Rockland State Hospital at Orangeburg, New York. Following this he became an Honorary Research Fellow and later a Sterling Fellow at Yale University, Laboratory of Neurophysiology studying activity of the central nervous system. He attained the rank of Assistant Professor at Yale.

In 1941 Dr. McCulloch became the Director of the Laboratory for Basic Research in the Department of Psychiatry at the University of Illinois, College of Medicine, where in addition to his own investigations, he was responsible for a large program in numerous areas. Dr. McCulloch is also well known as one of the founders of the group who have developed Cybernetics. He was Chairman of the Macy Conference on Cybernetics during its life from 1946 to 1951 and has published some of the most fundamental principles in this field. At the University of Illinois he became Professor of Psychiatry and Clinical Professor of Physiology.

Introduction

When a physician laughs loud and long at the antiquated folly and vested interests of his fellowsbe they psychiatrists or only neuropharmacologistsit is well for him to find his precedent in words attributed to the father of medicine. The citizens of Abdera wrote to Hippocrates crying for help, because their great atomic scientist had gone mad (with inappropriate effect). Hippocrates was long delayed. When he arrived with his bottle of hellebore, the weeping citizens led him to Democritus, where he sat unshod, dissecting animals and making notes in the book on his knees. Hippocrates asked why he was doing it, and he answered that he was looking for the causes of madness in the parts of beasts, and he demanded what had detained Hippocrates. He answered, Family matters, engagements, money, and other business. Democritus roared with laughterthat men called great so waste their lives, marrying only to fall out of love, seeking wealth without measure, making wars to no purpose, and in peace overthrowing one tyrant to set up another. Hippocrates listened to his railing and, turning to the people, told them to cease their lamentations, for Democritus was not only sane but the wisest man in Abdera.

With this sanction, I continue to cut up animals, write in my book, and laugh at folly. But, Gentlemen, the title of my paper is not facetious. At the behest of the Mathematical Sciences Division of the Office of Naval Research (ONR), I spent two months abroad, questing:

Where is fancy bred? Or in the heart or in the head? How begot, how nourished?

For several years, I had been working on three biological questions raised by von Neumann, who wanted a probabilistic logic to account for the logical stability, the reliability, and the flexibility of brains with components as unreliable as real neurons. You will find the first answers in The Stability of Biological Systems in the Brookhaven Symposium in Biology, No. 10, 1957. It shows how to build circuits of formal neurons, with two inputs each, that are logically stable under a common shift of threshold, are more reliable than neurons, and are more flexible than those made with flip-flops.

This is of importance in neurophysiology, because the threshold of real neurons at the trigger point fluctuates by 10 percent of its resting voltage. But, except for John Abbott, it was engineers, not biologists, who jumped at the new logic. That work was supported by the national Science Foundation, and its theoretical continuation by the National Institute of Health. Hence, I became a consultant to Heinz Von Foerster, Professor of Electrical Engineering at the University of Illinois, in his ONR project on artificial intelligence, which is an offshoot of cybernetics concerned with self-organizing automata that learn.

Like us, they are parallel machines and hence can employ probabilistic logic, one in which not merely the value of the argument but the function itself is only probable. It was Heinz von Foerster who sent me overseas, asking Where is fancy bred?

The Military Air Transport Service generously flew me over a day late, to the wrong country, but I humped to London and hopped to Paris in time to discuss the first paper at the Semaine Neurophysiologique de la Salpetriere. It was by Professor Moles, on the mathematics and physics of sensory integration. But it, though able, and my discussion, though much to the point, went unapplauded by the senior physicians. I was in the chair when Dr. Sem Jacobsen spoke well, but gingerly, about implanted electrodes; and not even in Paris did he dare to describe those stimulations that evoked sexual delight. A glance at the audience revealed the hostility of those with vested interests in clinical electro-encephalography and psychiatry, which subsequently found vent in the press, and lost him financial support for implanted electrodes. It was quite different from the British dualistic dislike of physical tickling of the mind, where they think fancy bred. That folly, played up in local councils by the psychiatric kings of hospitals, has made trouble for Dr. Sherwood, who has done the best of this work in England. But France is a Catholic country. The Popes blessing on this as an ethical procedure, not at odds with religion, facilitated the work in New Orleans and at Rochester. Percival Bailey, who was one of the first to implant electrodes, keeps his copy of the Popes letter in his desk. In Boston, Jim White and Bill Sweet have just been blessed, not merely for implanted electrodes, but for destruction of the sensory thalamus, by Sir Wilder Penfield, who considered it an improvement over leucotomy as a last-ditch stand against intractable pain. So much for implanted electrodes. They are here to stay. Through them we will record activities in structures heretofore inaccessible, locating the womb of Fancy. Through them we will stimulate, begetting Fancy. With them we will destroy whatever generates or mediates the diseases of Fancy. To do less would be unethical. Confess youd rather wear them in your head for years than let the cigar-clippers nip your frontal poles.

But let us get back to Salpetriere, where there were many good anatomic and physiologic and psychologic papers. But the work of Antoine Remond was the crowning success of the meeting. Long years ago he had come to my laboratory in Chicago with the intention of plotting two dimensions of the Laplacian over the whole head simultaneously at any instant after stimulation, or at any one place at any succession of instants at any specified delay. Over the years, ably helped by Offner, he has built up the apparatus for this, combined with accumulators that let him use hundreds of repeated stimulations and so raise the signals way above the noise. The resulting maps of the first special derivatives on the surface of the head are impressive. To go from these to the second derivative, which locates the nervous activity as well as possible, is still done by a laborious longhand computation. In spite of his ingenious plotting devices, one set of records may need days of computation. Ours, on the spinal cord, required months for a single experiment. Today it can be done electronically, so that the device plots these second spatial derivatives directly. It may take three years to build the gadgets.

Remonds work was so impressive that the National Institutes of Health and the European Office, Air Research and Development Command, are now backing it financially, but I understand that Bugnard and Alajouanine are of the opinion that it will raise the envy of those who have done nothing new.

The crowning failure was my own. John Lilly failed to show, and I, as his fellow-American, was commanded to speak on a mathematics suited to neurology. I did, and better than ever before or since, but I began to recognize the shut faces of pupils of my good friend Professor Fessard. To them cybernetics is an alien conceit, and McCulloch is one of those that begat it. Nevertheless, a lively discussion began, only to be halted, long before my time was up, by the gavel of the antique president, who thanked me for my obscurity. The next time they hear it they will say, It is not news, and the third time, It is obvious. To Ashbys friend, the psychiatrist Schutzenberger, who heard it only once, it is already obvious, and, when I said farewell to him in Paris, he had put it to work in genetics. Through him I met, at Cambridge, the son of the geneticist L. Penrose, who makes the simplest machines that, merely shaken, make more like themselves. The son, R. Penrose, the algebraist, is the specialist in impossible objects who inspired Gregory to design the optical and mechanical gadget for making three-dimensional enlarged real images of thick Golgi preparationsthe cleverest trick I saw in England. They can be photographed in color with no difficulty.

Between visits to laboratories and lectures to theoretical physicists, engineers, psychiatrists, and physiologists, I found time to work with Sherwood on the third component of the Laplacian of the cerebral cortex. He was able, as I was not, to construct an electrode with seven equally spaced 10 micron tips in a straight line less than one millimeter long, which we thrust into the cortex of a cat. By taking the difference between a given tip and the average of the ones below and above it, we obtained a first crude picture of the second spatial derivative of the voltage, but his amplifiers were not then suitable for a good approximation. I hope to finish this work with him this summer. This is the complement of Antoine Remonds work and, if successful, it may prove whether the slow rhythms of the cortex have their sources and sinks there or are piped up from below.

I did attend the two-day meeting at Queens Square, where the diffidence of the senior staff left the cleverest lectures undiscussed. Bill Mayer was just back from working with Rexed on localized lesions, placed anywhere in the head, made by spinning it (unopened) around that point in the deuteron beam of a cyclotron.

But my principal business in England was the study of artificial intelligence. In industry and banking this grows out of Operational Research, which began during the war as an application of von Neumanns theory of games, first to tactics, later to strategy. Its minimax procedures raise statistical problems that are by no means mathematically trivial and are so cumbersome as to require large-scale digital computers. I had not been in London three days before I was forced to run to the nearest Lyons for lunch, dreading its weak tea and tasteless fish and chips. But the place was large, clean, well-lighted. The cafeteria had p large selection of good foods. Even the hardware was at the right end of the assembly line. And the prices had not risen. Every item of that business, from the location of the restaurants, purchasing, hiring, or the shape of the teaspoon, is now the decision of a computer, called Leo. Management has learned not to interfere with Leo, for every attempt to do so has lost profit and displeased customers. Now Lyons is not yet up to those tricks which Albert Sperry built into American automation; first, because man has to feed it data and construct its programs, and second, because it has not, in Leo, the self-organizing properties of Ashblys ultrastability. But Leos creator is well aware of his problem.

Though in some points I think Donald MacKay was ahead of him, Gordon Pask is generally regarded as the genius of self-organizing systems of the variety called cybernetical; that is, those which coupled to a given environment, or a teaching machine, are able to rearrange their insides so that regenerative couplings are altered into inverse feedback, and so stabilize themselves. Though he is by training a biologist with a couple of years of medicine behind him, he works in an electronic engineering firm, evolving his strange creatures, threads of iron-salt crystals in shallow dishes coupled to his teaching machines. I spent many hours watching them grow from a genetic constitution, determined by half a dozen leads, into well-learned behavior, and what surprised me most was that, like man and beast, they learn faster and remember better if they are rewarded often but not always. While their decisions are basically digital, their memoriesas MacKay would have themare analogical. When a mode of behavior is inhibited one sees the disjointed trace waiting to be reconstituted as soon as it is disinhibited. Yet, with disuse, all traces fade as the hours go by.

Let me try to make clear the theoretical importance of Pasks gadget. What we need for psychology and psychiatry is an explanation of how brains work. Craik showed that we have an explanation to the extent to which we can embody our theory of behavior in a working model, and Quine proved that if all we want is a logical theory, then the only hardware we need are the natural numbers. Turing showed that a computing machine having a finite number of parts and states can compute any computable number. Pitts and McCulloch (1943) proved the theoretical equivalence of all Turing machines, whether they be made of neurons or any other hardware. From this it follows, as von Neumann said, that we can build a machine that will do with information anything that brains do with informationsolve problems, suffer emotions, hallucinate on sensory deprivation, what you willprovided we can state what we think it does in a finite and unambiguous manner. Now Ashby, in Design for a Brain, showed that any gang of miscellaneous, loosely coupled systems are capable of learning to match a variable environment and so stabilize themselves. What changes will be stored, and in which systems, are determined only by the sequence of particular events in the experience of that gang; yet repetition leads to storage in more central systems. Hence he is able to prove that a mechanical chessplayer is capable of learning to play a better game of chess than its inventor. To this I have added that, therefore, it can learn the rules of the game when they are only given ostensively and so become an ethical citizen in the society of game-playing machines. Such a machine could certainly learn checkers more easily. Its maker may not have known what game it was to learn; and it may invent one that is more fun and more difficult than Go itself. Clearly, here is a machine whose structure is unknown to its maker, except by an infinite or ambiguous specification, and so is what it is to do with information. The difficulty is in the number of trials it must make to learn, for it has not evolved to match the world into which it is bom.

I have shown that with idealized neurons which alter the logical function step by step for every three per cent shift of threshold, and each receives signals from but five sources, the input-output functions number about 1010, which is enough for chess. Given a net of five input neurons and one output neuron, the number of proper circuits it might try is about 10200. The age of the universe is some 1032 or 1033 microseconds; so, at one try per microsecond, it would take 10168 times the age of the universe. Clearly even organic evolution could not have found its solutions by such a brute-force method of trial and error. Other constraints must have worked to engender viable beasts. I believe they are sought in the nature of her building blocks, subatomic particles, atoms, and molecules, proceeding discretely through well-regulated autocatalytic reactions to produce cells and cell aggregates, or, as in Pasks example, crystals. His gadget does work; it does take habits by a mechanism that Charles Peirce proposed. So, in building models for the soul or Psyche, it behooves us to employ the preordained harmony of natural objects.

Pasks closest friend, Stafford Beer, was, I believe, the first who ever solved a minimax problem using animalculi as components in his machine. Today he is in charge of Operations Research and Cybernetics for the British steel industry in Sheffield. With him I spent a busy day learning how profitable cybernetics is, whether it be a matter of cobbing in a rolling mill, or stockpiling forgings, or controlling energy in blast furnaces. In English medicine cybernetics is still a dirty word, but in their industry it has been washed in the holy water of filthy lucre. You will hear more of Stafford Beer.

I left Sheffield in a midland fog that followed me into a vacant compartment, and sat down with its grey face six inches from my nose. The engine failed, the light went out, and for eight hours we were bumped from behind to Birmingham. Slowly, on the fog before me, I began to see Minskys drawing of a Venn diagram for a neuron with an infinite number of inputs, and above it an infinite input rank of diagrams of the same kind. Only the central ones were clear; the rest faded off. I began trying to put numbers into them to keep track of the sequence of appearance of jots with every step of threshold. At first, the numbers came almost at random, but, finally, I forced them to appear so as to prescribe an infinite net of neurons, each with an infinite number of inputs, which would compute the same input-output function through all infinity of thresholds, though each neuron went wrong with every common shift of threshold. Before I went to sleep in Birmingham, I had written in closed form the rule of formation of one of an infinite number of possible modes. Only then did I realize who She was that, as a fog, joined my wily wanderings; for thus is Fancy bred. And She stayed with me until I saw the sun in Washington.

Birmingham is obviously the poorer for want of Elkes fancy, and it is small comfort that Sir Henry Dale says that things are better than they were when he was forced out of the clinic to pursue research in a drug house. I lectured to Mayer-Gross group on The Stability and Reliability of Biological Computers. In the audience was the neurosurgeon Turner, who makes small lesions in the fibers leaving the hippocampus, toward Amygdala and temporal lobe, to stop meaningless rages or epileptic furors; and it works without untoward consequences. The next day I had a chance, with McLardy, at Northampton, to examine the brain of one of Turners patients and to study McLardys beautiful stains of the strange two-dimensional braiding of the fine axonssheathed in bunches in something not myelinthat pass from the granular layer to the pyramidal cells of the hippocampus, and to discuss with him their possible time-bridging role in this quaint concourse of the fore-brain. The other man at Birmingham whose work I should mention is Eayrs, for he has extended Sperrys work from the transfer of learning from one hemisphere to the other via the corpus callosum, to the learning of a task with one stimulus to one hemisphere and the other to the other and has shown that these tasks cannot be learned after section.

From the Midlands, I returned by car in fog to London. Graham Russell was driving, and I learned from him much of the art of enhancing the contingent probability of successful behaviors, given the goal. He used to work with Littley on learning machines and knows the detailed ways of constructing electronic devices that form an internal representation of the world in terms of the contingent probabilities of signals over many channels, an art in which they were some years ahead of the American work on perception.

Now he and McLardy and I had spent much time on the third problem of learning machinescall it insight if you will. We began by guessing at hippocampal function, and were of one mind as to so-called insight being a long shot that pays off. It is made by us in terms of what is sufficiently familiar to appear simple to us. We evolved, in this, our familiar world, and our long shots do pay off surprisingly well. Even the hypothesis of the scientist are too often right for chance in an ensemble of infinite possibilities.

As David Hume says, only in logic and arithmetic and scarcely in geometryfor we must import its metriccan we hope to argue indefinitely without confusion. Our guesses must be formed in terms of concepts that are not noisy. The value of π can be calculated to an infinite number of places.

On the way to London with my head hanging out the left window, calling the distance to the curb and identifying forms in the fog, I could spot familiar obstructions more easily than peculiarly English ones. To be at all clever the machine must invent its own noiseless concepts and general rules of conduct in closed formnot merely muddle through. Somehow, somewhere, Fancy must be bred!

When I reached the National Physical Laboratory for Uttleys symposium on The Mechanization of Thought Processes, I expected to hear something clever on this score, but I cannot see any advance since our article (Pitts and mine) on How we Know Universals, in 1947.

Generally the contributions, which were precirculated, might have been clever two years ago. Alex Andrew and Uttley have pushed ahead in perception. Pasks work on cybernetics was, of course, significant. Ashby was incisive as to what remains of statistical mechanics when the constraint of energy is removed, as it is in all computing machines, notably in brains. He proved that these systems must show accommodation.

Instead of rehashing my paper, called Agathe Tyche, I presented in closed form the prescription for one mode of making nets that have the same input-output function under common shift of threshold over the whole range, regardless of the number of inputs per neuron, as I had learned it of the fog. With the permission of the chairman, J. Z. Young, I asked, first, how many had read my Agathe Tyche”—a dozen hands came upand second, how many had had trouble understanding it. Almost every hand rose, which produced sheepish laughter. But, naturally, the discussion was as poor as it was generally.

The exception was that by Stafford Beer. As at the international meeting on cybernetics at Namur, so in Teddington, he was able to spot and annihilate the irrelevant, the false, and the trivial. His discussion of the Perceptron will be well worth reading even if its creator withdraws or rewrites his paper.

The outstanding American contribution was by Oliver Selfridge. It was rightly entitled Pandemonium, for his devices consist of little demons, each of which spots something in the input and shouts as loud as he is sure of what he spots. Behind this rank of demons is a second rank, who listen to them, looking for longer syntheses, and again shout as loud as they are sure. And so it goes, each superior demon listening to all below him, until Satan himself, listening to the whole pandemonium, decides where the noise is loudest and acts accordingly. What is crucial in this parallel procedure is that decision is delayed until after all possible computations have been made and weighted for their likelihood, thus, preserving flexibility and seriatim. The existing Pandemonium at Lincoln Laboratory reads man-sent Morse code with better than 95 per cent correctness for letters and 99 per cent for words; and it can be made to do better for sentences. Any demon who does not contribute can be spotted and replaced till one is found that helps. So Pandemonium evolves to match his task. But his present demons have digital tricks.

In Agathe Tyche, I demonstrated that for parallel computers things resembling neurons are better suited, as Von Neumann forecast in his lecture to the American Psychiatric Association in 1955. The Bell Laboratories and other companies are now building these formal neurons. The best of them is that of the psychiatrist, Jerome Y. Lettvin, of my group in the Research Laboratory of Electronics, whose circuit is published in the Quarterly Progress Report of the Research Laboratory of Electronics, M.I.T., January 15, 1959. It was designed for Selfridges Pandemonium. Pasks and Selfridges gadgets can muddle their way rapidly to solutions because their tasks do determine their structure. They evolve naturally, each in its own world.

The most important thing I learned from that meeting cannot appear in its transcript. It was never said in so many words. For the first time I heard Russian scientistsconcerned with programming computers, with machine translation of natural languages, with computer design and with artificial intelligencetalking science with their Western fellows. As if Komogorof were not Russian, and there had been no Sputnik, many Americans have asked me whether the Russians have enough computer know-how to launch a war with intercontinental ballistic missiles. If they are behind us today in the computer field, it can only be in hardware; and you can take it for granted that they will soon be abreast or ahead of us. But what I saw in Russian faces was that their scientists, like ours, know they are confronted by the problem of the Rabbi of Chelm with his Golem, and they are as unhappy as we are.

Distressed by the persecution of his people, the Rabbi had gathered as much clay as he could mold into a robot to defend them. On its forehead he wrote the secret name of God. So it came alive with the desire to do what the Rabbi told it to do. The difficulty was to tell it in such a way that the order could not be misunderstood. There are endless stories of the ensuing mishaps, worse than the sorcerers apprentice, and Talmudic arguments as to the legal form of the orders; but, finally, the Rabbi of Chelm had to put an end to his Golem.

We do not even know the secret name; we can only install computers to launch intercontinental ballistic missiles. We have the energy and the hardware. Defense is obviously impossible. Only Massive Retaliation remains to deter Aggression. To be effective, Retaliation must take off before Aggression arrives. The only split-minute solution is pushbutton warfare. It can be realized by both countries all too soon, and it can only result in mutual destruction. Men go trigger-happy; Failsafe devices go off by accident: Sabotage occurs: A third party springs the traps. We cannot yet build an ethical Psyche into the Golem. I am sure that if the politicians would let us, weour scientists and theirs would follow the example of the Rabbi of Chelm.

But we, I mean all scientists in the field of communication, have a necessity begot of our own invention. We saw it first at Pearl Harbor, but it was already foreseen in 1587. Call it The Brass Head.

That year was gloomy. Da Vincis engines of death and destruction ended the hope of defenses against aggression. Autocratic Spain had butchered the bulk of European Jews and Protestants; true, she had been halted in the democratic Netherlands, but by English aid. The augursthe astrologerswere full of dire predictions. Spain was gathering the greatest fleet ever assembled to attack that fortress of freedomEngland. The greatest information theorist of those days they called him a magicianFriar Bacon of Oxford, and his fellow Friar Bungay, made a brass head to tell them how to build a brass wall around the South of England; but it would not speak. Perhaps they had used a wrong model for neurons. They finally persuaded the devil to make it talk, and he agreed, provided they would be there to listen. After a vigil of 26 days they had to sleep, and left a technician in charge. From his pranks with the heir apparent one cannot tell whether he was just stupid or perverted by the soldier or politician, but he had positive instructions to wake his master when the head spoke. It said: Time will be, and he answered: For this I should wake my master. Then it said: Time is, and he: That I know well and if you have nothing more to tell, let them sleep. Finally it said: Time has been and exploded. Little as we like it, the Friar Bacon and Friar Bungay of science know well they can never again relinquish informational computers to technicians or clinicians. The vigil must be endured even if it entails Q clearance.

But one problem never appeared in Teddington in any form or with any name by which I could recognize it, and no cybemetical muddling through can solve it. In the dread year 1588, the Invincible Armada arrived in the narrow seas with the pageantry of death and bottled up the English fleet with its Mortal Moon, a crescent of shipsthe formation inherited from antiquity but, information-wise, no further evolved than the sea-cucumber that engulfs its prey. The English ships of the line emerged for the first time in single file and cut the crescent to bits, leaving the Spanish broadsides to sink their own galleons. Thereafter, every ship became a quasi-automaton with its own multiple closed-loop servosystems of information, like the segments of the caudata, with a chain of command from the front, whose distance receptors are first to sight the enemy and pass the word aft. This has been the formation until the double defeat of World War Icalled the battle of Jutland. Thereafter, every fleet has grown a reticular formation (CAMIC was its old name). Every ship of any size or consequence receives information from the others and sweeps the sky for hundreds of miles and water for tens of miles with its own sense organs. In war games and in action, the actual control passes from minute to minute from ship to ship, according to which knot of communication has then the crucial information to commit the fleet to action. This is neither the decentralized command proposed for armies, nor a fixed structure of command of any rigid sort. It is a redundancy of potential command wherein knowledge constitutes authority.

From the Golgi stains of Scheibel and the records of Amassian it is clear that the reticular formation is such a structure, neither an undifferentiated mess, nor a rigid chain of command. Every large cell certainly receives signals from almost every source, coded in pulse-interval modulation, to convey whence the signal came and what happened there. Clearly the reticular formation decides what he ought to do, what he should heed, how vigilant we ought to be, and whether we have time for that idle fancy that inspires our future action. Short of the battle fleet, we have yet to build into any computer a replica of that little structure in Whose will, to misquote Dante, lies our peace.

My last date in England was a requested speech at Mill Hill on the action of strychnine, in which I showed the inapplicability of pharmacological terms such as excitant, stimulant, depressant, inhibitor, and disinhibitor, to that rise in voltage of axonal terminations that prevented gating of signals by signals before they reached the recipient neuron, and ended with a dirge for pharmacological centresExitus Acta non Probat.

Thence I flew to Amsterdam, met by Van der Tweel, who made the first servosystem analysis of any reflex on the human pupil. I lectured there to medical physicists and, at Groningen, to neurologists; and I came to the end of the trail of nuclear magnetic resonance that I had followed from Saclay through England and Holland by finding the young collaborator I require to study the condition of water in neuronswhere gasses do not dissolve, ions will not play ball with electrodes, and dyes act as if in ice.

I spent the next afternoon, at Dean Farnsworth's suggestion, with Prof. Amolf at the Institute d'Optique in Paris, on the monocular diplopia of presbyopia. It is no fancy, merely the physics of an old lens indulging in plastic flow beyond the elastic limit as appears, at once, in his slit and edge photography of the optic system.

That evening I was in the operating room with Tony Remond for his stereotatic operation on a unilateral Parkinsonian, which was televised live. A ventriculogram is X-rayed on two cassettes on the cage, which is then removed from the head. The X-ray tubes are replaced with light sources, and the telescopic electrode so placed that the shadow of its tip falls in the proper spot in both photographs. The depths are measured, and the guide tube clamped. The cage is replaced on the patient's head. Through the guide tube, with a sonic frequency, the electrode carrier and obturator are driven fast through the skull. The obturator is withdrawn and the electrode inserted. Its position near the third ventricle is checked by stimulation short of the final position. High in this region the patients usually say: Ughstop it; low: Mndo it again. When the tip should be in the inner edge of the globus pallidus, a series of long pulses is given. If it blocks tremor and melts rigidity, the lesion is made electrically. So it went with this patient, as it had with a hundred before him.

The next day, there appeared a diatribe in the press. It ended with this paragraph:

Le grand triomphateur de la soirée a été le professeur Mac Culloch de Chicago, un des fondateurs de la cybemetique, qui lui est persuadé qu'il en est a peu près ainsi. Son succés a été d'autant plus complet qu'avec son air goguenard, fait de malice et de geni, et sa belle barbe blanche, il ressemble comme deux gouttes d'eau au Père Noel. Mais ne vous y trompez pas, c'est le diable.

The psychiatric aroma of such personalities is all too familiar to be missed. It led me promptly to the author. . . Before I could get out of town, Radio Paris and Figaro waylaid me at LeGrillon, and Remond's sisters had great difficulty in persuading them that, considering its source, I took it as a complement and desired neither time nor space for public retaliation. Thanks to Dr. Fegersten, who battled my recalcitrant taxi-driver, we reached Orly in time for me to get through red tape and sit down facing aft next to the window of the plane.

As we took off for Washington via the Azores, the fog rolled in. This time I recognized her thankfully and let the figure form. It was an X with a circle around its center, the diagram for a neuron with three inputs. As ones and zeros and p's for probability began to fill it in, I realized that with neurons of three or more afferents it was possible to construct nets, not merely more reliable than their components, but infallible, with great random perturbations not only of thresholds but even of synapsis.

When we landed in Iceland, I had it in my note book; and I have been sweating out the general rules ever since.

On the landing here, I made my bow to ONR and began to contact my Human Factors friends in Astronautics. They could use these circuits. I am neither Daddy Christmas nor the devil, only an early space cadet, and my purpose is to sail among the westering stars before I die. So ends this odyssey.

Let me recapitulate its discoveries.

  1. For the good of patients, implanted electrodes are here to stay; and through them, whether psychiatrists like it or not, we will learn where is fancy bred.
  2. For machines that learn we can now build devices that learn to represent the contingent probabilities of associated inputs.
  3. We know ways to maximize probabilities of success of conduct, given the goal.
  4. We have both theory and model for self-organizing cybernetical devices that muddle through to self-preserving structures in a hostile world.
  5. In Agathe Tyche we have parallel circuits that are flexible, logically stable, or even infallible under gross perturbations of threshold and connections.
  6. In Pandemonium we have an evolving system that preserves information until the instant of final decision. Its components will become increasingly like neurons.
  7. We have never yet built, in a single machine, a reticular formation with redundancy of potential command.
  8. We have still no way of engendering fancy, that imaginative leap beyond data into the formation of a noiseless concept, that has a reasonable chance of being right.
  9. It is too late for us to invent an ethical soul for the Golem, or even install common sense. So, for us, we know there is only Armageddon or disarmament.
  10. We will be there when the brass head speaks.

So ends the brief.

In deference to my host, who used always to teach his psychodynamics out of Shakespeare's writings, I would like to quote him. With confidence based on the common sense of the common man, I expect peace. And so, at sixty, I will give youperhaps prematurelythe sonnet Shakespeare wrote at twenty-five.

The Mortall Moone
(Sonnet 107)

Not mine owne feares, nor the prophetick soule,
Of the wide world, dreaming on things to come,
Can yet the lease of my true loue controule,
Supposde as forfeit to a confin'd doome.
The mortall Moone hath her eclipse indur'de,
And the sad Augurs mock their owne presage,
Incertenties now crown them-selues assur'de,
And peace Proclaims Oliues of endless age.
Now with the drops of this most balmie time,
My loue looks fresh, and death to me subscribes,
Since spight of him He liue in this poore rime,
While insults he ore dull and speachlesse tribes.
And thou in this shalt finde the monument,
When tyrants crests and tombs of brasse are spent.

(1589)

Footnote

 

For further research:

Wordcloud: Appear, Began, Better, Brains, Bred, Build, Command, Common, Computers, Cybernetics, Devices, Discussion, Dr., Electrodes, End, England, Evolved, Fancy, Fog, Form, Formation, Generally, Given, Head, Infinite, Information, Inputs, Laboratory, Learn, Logic, Machine, McCulloch, Natural, Neurons, Number, Pandemonium, Possible, Problem, Research, Signals, Spent, Stimulation, Structure, Systems, Threshold, University, Work, World, Years

Keywords: University, Electronics, Work, Fellow, Research, Conference, Member, Hospital

Google Books: http://asclinks.live/lozk

Google Scholar: http://asclinks.live/cxg3

Jstor: http://asclinks.live/dco9


1 Reprinted from Lectures on Experimental Psychiatry, sponsored by the Western Psychiatric Inst. and Clinic, Pittsburg, PA, March 5-7, 1959. Proc. of the Bicentennial Conf. on Exp. Psychiatry. Published by the Univ. of Pittsburgh Press, March 1961.