Chapter 11 Chapter Ten: Several Communication Machines and Their Future

I have devoted the last chapter to the industrial and social influence of certain controlled machines which have begun to show important possibilities for the substitution of human labour.But there are still many problems with automata which have nothing to do with our factory system, which are either used to illustrate and reveal the possibilities of communication mechanisms in general, or to be used for semi-medical purposes. Supplement or substitute for lost or aging physiology in unfortunate patients.The first type of machine we will discuss below is designed for theoretical purposes, and it was developed by me some years ago with my colleague A.Rosenblueth, J.An example of earlier work co-authored on a paper by two Ph.D.s in Bigelow.In this work, we conjecture that voluntary mechanisms are feedback in nature and, accordingly, we look for dysfunctional features of feedback mechanisms in human voluntary activities that manifest themselves under overload.

The simplest type of obstacle is an oscillation in the pursuit of an object, which occurs only when the process is actively stimulated.This phenomenon is quite similar to the so-called intention tremor phenomenon in people. For example, when a patient uses his hand to fetch a glass of water, his hand will shake more and more violently, so that he cannot hold the glass. There is another type of tremor in man, which is in some respects the opposite of the purposive tremor.It's called Parkinsonianism, and it's a paralytic tremor common to all of us in the elderly.People with this condition even experience tremors at rest, and in fact, if the condition is not too severe, tremors only occur at rest.When the patient tries to accomplish a definite purpose, the tremor subsides to such an extent that an early Parkinson's patient can even become an accomplished ophthalmologist.

The three of us associated this Parkinsonian tremor with some form of feedback that was slightly different from that of accomplishing a purpose.In order to successfully achieve a goal, although many of his joints are not directly related to the purposeful movement, they must also maintain a moderate rigidity or tension, so that the final purposeful contraction of the muscles can just keep up.In order to do this, there must be a second level feedback mechanism, which does not appear to be in the cerebellum, because the cerebellum is the central control station of that mechanism, and if it is damaged, it will cause tremors of purpose .This second type of feedback is called postural feedback.

Mathematics can show that in both cases of tremor, the feedback is overdone.Now, when we look at the kind of feedback that is important in Parkinson's syndrome, we figure out that the voluntary feedback that regulates the primary movement and the postural feedback go in opposite directions so far as the movement of that part of the organ is regulated by the postural feedback .Thus, the presence of purpose prevents the tendency of the postural feedback to be overemphasized, and keeps it from oscillating altogether.We know these things well in theory, but until recently we have not resolved to create a model of their behavior.More and more, however, we want to build a machine that can be shown to work according to our theory.

It is for this reason that Professor J. B. Wiesner of the Electronics Laboratory of the Massachusetts Institute of Technology discussed with me the possibility of making directional machines or machines with a single fixed purpose, requiring The various parts of this machine can be adjusted enough to account for the fundamental phenomena of voluntary feedbacks, the fundamental phenomena of postural feedbacks of which we have just been speaking, and how they behave when they are broken.At our suggestion, H.Sir Henry Singleton solved the problem of making such a machine, and concluded with brilliant success.This machine has two main modes of activity, one is positive photosensitive, that is, light-oriented; the other is negative photosensitive, that is, light-proof.We call these two functions of the machine "moth" and "bug".The machine consists of a small three-wheeled cart with a propeller on the rear axle.The front wheel is a small caster, operated by a lever.The car has a pair of forward-facing photocells, one of which checks the left half for light and the other checks the right half.These two photocells are the two handrails of a small bridge.The output of the bridge is non-inverting, it goes to an amplifier which can be adjusted.The output of the amplifier goes to a servo motor to adjust the position of a junction connected to a potentiometer.The other joint is also adjusted by a servomotor that also moves the lever that operates the front wheels.The output of the potentiometer represents the difference between the positions of the two synchronous motors, and this output is passed via a second adjustable amplifier to a second servo motor, which in turn adjusts the lever that operates the front wheels.

Whether this instrument goes towards the brighter half or away from it depends on the direction of the output of the bridge.But in either case, it automatically balances itself.So it's a feedback that depends on the light source, that is, the light travels to the photocell, and from the photocell to the rudder control system, through which the instrument finally adjusts its direction of motion and changes the angle of incidence of the light. This feedback tends to achieve the purpose of phototaxis or avoidance.It is a simulation of voluntary feedback, because we believe that human voluntary activities are essentially a choice among various tropisms.When this feedback is overloaded by increasing magnification, the little cart, the "moth" or "bug," will seek or enter light in an oscillating manner according to the direction of its tropism, and it oscillates more and more .This is very similar to the purpose tremor phenomenon caused by cerebellar injury.

The positioning mechanism of the rudder contains a second level of feedback, which can be regarded as attitude feedback.This feedback is routed from the potentiometer to the second motor and back to the potentiometer, and its zero point is adjusted by the output of the first stage feedback.If this feedback is overloaded, the rudder starts to exhibit type 2 chatter.The second type of tremor occurs when there is no light, that is, when we give the machine no purpose.Theoretically, this is due to the fact that as far as the second mechanism is concerned, its feedback is opposed to the activity of the first, so that the latter tends to weaken the former.This phenomenon is what we describe as Parkinson's syndrome in humans.

I recently received a letter from Dr. G. Walter, Budden Institute of Neurology, Bristol, England, expressing his interest in "moths" and "bugs" and telling me that he A similar machine has been constructed, which differs from mine in that it has a definite yet changeable purpose. In his own words, "we have taken into account all the properties except the negative feedback which gives the machine an inquiring and moral attitude towards the universe as much as it has a tropism ".I have already discussed the possibility of such a change in behavioral patterns in the chapter of this book on learning, a discussion which is directly relevant to Walter's machine, although I do not yet know what instrument he actually used to achieve this. type of behavior.

At first glance, both moths and Dr. Walter's further developed tropism machines appear to be subjects of technical skill, or at best mechanistic interpretations of philosophical themes.But even so, they all serve some definite purpose.The U.S. military medical corps has compared photographs of "moths" with photographs of actual cases of nervous tremors, and they are so certain that these photographs are useful for training military doctors in neurology. We have also studied a second class of machines, which have a medical value of much more immediate and obvious importance.These machines can be used to compensate for the various defects of crippled limbs or insufficiencies, and can also provide those organs that are otherwise healthy with new and possibly dangerous capabilities.We can generalize the use of machines and make artificial limbs that perform well; we can design assistive devices for the blind that convert visual patterns into auditory signals to read ordinary books page by page; and other similar aids can be designed. Make the blind aware of danger and make it easier for them to walk.In particular, we can use machines to help the totally deaf.Aids of this kind are perhaps the easiest to manufacture; partly because telephone technology is well-researched and the most familiar of communications technologies; In most cases, it is the loss of the ability to freely participate in conversations with people?

It is also partly due to the fact that the useful information conveyed by speech can be compressed within such small boundaries that it does not exceed the conveyance capacity of the organ of touch. Some time ago, Professor Wiesner told me that he was interested in the possibility of making an instrument to aid the total deafness, and that he would like to hear my opinion on the subject.I stated my opinion, and it turned out that we were mostly on the same page.We all know the work that Bell Telephone Laboratories had done on visible speech, and how that work related to earlier work on automatic speech synthesizers (Vocoders).We already know that the work of automatic speech synthesizers has given us a superior measure of the amount of information necessary to transmit speech intelligibility than any previous method.We find, however, two disadvantages of visible speech, namely, that it does not seem to be easily portable, and that it places too great demands on sight, which is more important to the deaf than to us.

A rough estimate shows that the principles employed by the instrument of visible speech may be transferred to the sense of touch, and this, we decided, should be the basis of our apparatus. As soon as we started working on this work, we discovered that the researchers at Bell Telephone Laboratories had also considered the possibility of receiving sound by touch, and had included it in their patent application.They kindly told us that they hadn't done any experiments with it, and let us do our research at will.Therefore, we entrusted the design and manufacture of this instrument to L.Mr. Leon Lavine, an engineer in the electronics laboratory.We foresee that much work will be necessary to bring our apparatus to practical use, among which the question of training in the use of this apparatus will occupy the foreground, and on this we have obtained our School of Psychology, Ley A. Barvras. (Alexander Bavelas), who came up with many ideas. The problem of explaining speech by means of other senses than the auditory organs, such as the tactile organs, can be explained from the point of view of language as follows.As mentioned earlier, we can broadly distinguish three language stages and two mediation processes between the outside world and the subject receiving information.The first stage is the stage of acoustic symbols, which is physically the stage of air vibration; the second stage is also called the phonetic stage, which is the various phenomena that occur in the relevant parts of the inner ear and the nervous system; the third stage is also called semantics stage, when acoustic symbols transform into experiences of meaning. For the deaf, the first and third stages still exist, but the second stage is missing.However, it is quite conceivable that the second stage could bypass the organ of hearing and replace it with another organ, for example, the organ of touch.At this time, the transition from the first stage to the new second stage is not done by our natural physical-neural apparatus, but by an artificial, that is, a system made by man.The transition from the new second stage to the third stage, which we cannot directly examine, represents new systems for forming habits and responses, such as those we develop when we learn to drive a car.The status of the instrument we have designed is such that the transition between the first stage and the new second stage is sufficiently controlled, although there are still several technical difficulties to be overcome.We are studying the learning process, that is, studying the transition from the second stage to the third stage; in our opinion, these studies have great promise of success.The best we have been able to demonstrate so far is that when using a learning vocabulary consisting of 12 words, only 6 mistakes were made during 80 random repetitions. In our work we always keep certain facts in mind.The first of these is, as already stated, the fact that hearing is not merely a communicative organ, but a communicative organ whose chief use is to undertake communication with other people. It is again an organ which acts on our part for certain communications, speech.Other uses of the sense of hearing are also important, such as receiving sounds in nature and enjoying music, but they are not so important that if a person does not use his sense of hearing for any other purpose than to participate in daily communication between people with speech use, then we have to treat him as socially deaf.In other words, the sense of hearing is of such a nature that if we were deprived of all uses of our hearing, except as a means of communication with others, we would still be reduced to inconvenient . In order to compensate for the sensory deficit, we must regard the entire speech process as a constituent unit.The importance of this is immediately observable when we consider the speech of the deaf.For most deaf-mute people, the training of lip-reading is neither impossible nor extremely difficult, so deaf-mute people can be extremely proficient in receiving speech signals from others.On the other hand, with rare exceptions, the best and most up-to-date training methods have resulted in the fact that although the vast majority of deaf people can learn to use their lips to make sounds, what they utter is a strangely raspy sound. Tone, which is a highly ineffective form of sending a message. The difficulty lies in the fact that, for the deaf-mute, the act of talking has been split into two completely separate parts. We can very easily simulate this situation for the normal man, if we give him a telephone communication system to talk to other people, so that the telephone does not transmit his speech to his own ears.It is very easy to create such a transmission system whose microphones do not work, in fact the telephone companies have studied them, and these systems have been abandoned simply because they caused a great sense of confusion, especially not The confusion of knowing exactly how much of your own voice is fed into the line.People using these systems are always yelling and raising their voices to the top, lest the other end of the line can't hear them. We now return to ordinary speech.We know that the speaking process and the listening process in normal people are by no means separate; the learning of speech itself depends on the fact that everyone can hear themselves speaking.It is not enough for the best speech to come from a man who only hears what he says in bits and pieces, filling in the gaps from memory.Speech is of good quality only when it is in a state of continuous self-monitoring and self-criticism.Any assistive device for the totally deaf must take advantage of this fact, and while it is true that such a device may have recourse to other senses, such as the organ of touch, rather than to the organ of hearing which is already crippled, it must be made in a manner similar to the present one. Portable, durable electric hearing instruments are similar. A further theory for compensating for auditory deficits has to do with the amount of information effectively available for hearing.The most rough estimate of the maximum value of this quantity can be transmitted in the range of 10,000 hertzes and an amplitude of about 80 decibels.Although this load of communication marks the maximum that the ear can achieve, it is too large to express the effective information given by actual speech.First of all, speech coming in and out of the telephone does not have a sound above 3000 Hertz, and its amplitude range certainly does not exceed 5 to 10 decibels; Exaggerates the range of sounds used by the ear and brain to reconstruct intelligible speech. We said that the most accomplished work ever done on the problem of estimating information volumes was Bell Telephone Laboratories' work on automatic speech synthesizers.This work can be used to illustrate: if the human language is properly divided so that it does not exceed five frequency bands, and if these frequency bands are detected so that only their appearance is perceived, then these appearances or appearances are used for Modulate completely arbitrary sounds in their frequency range; and if these sounds are finally added together, then one can still recognize the original speech as a speech, or even recognize it as someone's speech.However, the amount of useful or useless information that may be conveyed has now been reduced to less than one-tenth or one-hundredth of the original information that might be presented. We distinguish useful information from useless information in speech, that is, the maximum encoding capacity of the ear to receive speech from the maximum encoding capacity through the cascade network of successive stages formed by ear and brain.The former ability is only related to the transmission of speech through the air and through an intermediary instrument such as the telephone, which only imitates the ear, not any organ in the human brain for understanding speech.The latter concerns the conveying capacity of the entire complex system of air-telephone-ear-brain.Of course, there can be subtle differences in pitch that don't pass through the entire narrow-band transmission system we use when we speak, and estimating the amount of information loss caused by these nuances is a difficult thing to do, but this loss The amount does not appear to be large.This is the idea on which automatic speech synthesizers are based.The valuation of information by engineering in the past is flawed. They ignore the last link in the chain from the air to the brain. When using the other senses of the deaf, we must realize this; except vision, all other senses are lower than hearing, that is, they transmit less information per unit time than hearing.The only way to make the lower sense organs such as the touch organ work with maximum efficiency is not to send all the information we hear through hearing to it, but only the processed part, that is, the part suitable for understanding speech. Send it aurally.In other words, we replace the part of the cortex that normally performs the function of receiving sound with an information filter before the information passes through the tactile receptors.This is how we transfer part of the cortex to an artificial extrinsic cortex.In the instrument we are working on, the detailed method we use to achieve this is to separate several frequency bands of speech like an automatic speech synthesizer, and then use these bands to modulate the frequencies that are easily detected by the skin. Following the perceptual vibrations, these filtered distinct frequency bands are delivered to spatially separated tactile areas.For example, five frequency bands can be sent to the thumb and the remaining four fingers on one hand. The above is the basic idea of ​​the instrument we need, that is, to transfer the sound vibration to the touch through the electrical method to receive intelligible speech.We have long since learned enough that the patterns of large vocabularies are so different from each other that they are so consistent across many speakers that it does not require much speech training to recognize them.From this point of view, the main direction of research should be to comprehensively train the deaf-mute to recognize and reproduce sounds.From a technical point of view, what lies before us is the question of whether the instrument should be portable or not, and how to reduce its energy requirements without compromising its basic performance, and these issues are still at the stage of discussion.I do not want to spread false hopes, especially immature hopes, among those suffering from physical defects and their relatives, but I think it is safe to say that the prospect of making is by no means hopeless. Since the publication of the original edition of this book, other workers in communication theory have produced new specialized instruments by which some of the basic principles of communication theory have been elucidated.In a previous chapter I have dealt with Dr. Ashby's steady-state machine and with Dr. Walter's somewhat similar machine.Let me now return to several early inventions of Dr. Walter's machines, which are not unlike my "moths" or "bugs," but were made for different purposes.In the case of these phototropic machines, each element has light, so it can stimulate other elements.If, therefore, a series of elements acting at the same time appear as groups and reciprocal reactions, animal psychologists would, in most cases, treat such elements as encased not in steel and brass but in flesh and blood. Phenomena explained as social behavior.This is the beginning of a new science of mechanical behavior, although its full development is yet to come. Over the past two years, for various reasons, MIT has made it difficult to manufacture hearing gloves, although the possibility of manufacturing still exists.During this period, theoretical research has led to the improvement of the instrument, which can allow the blind to pass through the intricate streets and buildings, although the details of the instrument have not yet been designed.This study focuses on C. M．The work of Dr. Clifford M. Witcher, himself congenitally blind, but a preeminent authority and expert in optics, electrical engineering, and other fields necessary for this work. One device that seems promising but has not had any real development or final identification so far is the artificial lung.In artificial lungs, the actuation of the breathing motor is determined by electrical or mechanical signals from the patient's weakened but not yet destroyed lung muscles.This situation illustrates that normal feedback in the spinal cord and brainstem of a healthy person can be applied to a stroke patient to help him control his breathing.The so-called iron lung, therefore, may no longer be a prison in which the patient forgets how to breathe, but an exercise tool for maintaining the patient's remaining ability to breathe, and possibly even improving it to the point where he can The degree to which you breathe independently without the help of a machine.There is hope for this to happen. The machines we have discussed so far are those of general public concern, either because they have the directly human character of theoretical science, or because they are certainly aids for the benefit of the handicapped.Now we return to a class of machines with some very ominous possibilities.Oddly enough, such machines include automatic chess machines. Some time ago, I proposed a method for playing chess with modern computers, which was at least passable.In this work the lines of thought I follow have their historical backgrounds which cannot be ignored. A．Poe had discussed Maelzel's deceitful chess machine and exposed it, pointing out that the machine's ability to play chess was operated by a crippled man with a broken leg inside.But the kind of machine I'm referring to is real, taking advantage of the latest advances in computer development.It is easy to build a machine that can only play chess step by step but poorly; it is hopeless to try to build a machine that can play chess perfectly, because such a machine requires too much. combination of moves.Professor von Neumann of the Institute for Advanced Study in Princeton has discussed this difficulty.However, it is necessary to manufacture a machine that can guarantee that it can have the best move within a limited number of steps after each move, for example, within two steps, so as to ensure that it can move according to a relatively easy estimate. The method puts itself in the best position, which is not easy, but not hopeless. Today's fast computers can be modified to work like chess machines, but if we are determined to have a machine play chess, we can also build a better machine, although it may be expensive.The speed of these modern computers is fast enough that they can estimate the various possibilities of the next two moves within the specified time of each move.The number of chess move combinations generally increases geometrically.Therefore, there is a huge difference between calculating all possibilities within two steps and calculating all possibilities within three steps.To calculate in any reasonable time a game of, say, fifty moves, is beyond the reach of machines.However, for organisms that live long enough, as von Neumann pointed out, it is possible to do it, and both sides play a perfect chess game, the self-evident conclusion, or white always wins , or sunspots often win, or, most likely, often draw. C. Bell Telephone Laboratories.Mr. Shannon once proposed a machine on the same principle as the two-movement machine I had thought of, but he made considerable improvements.First of all, his method of estimating the final chess position after taking two steps includes the estimation of factors such as the control of the game, the mutual protection between chess pieces, and the number of chess pieces, checkmates and checkmates.Then, if, after two moves, the game appears unstable due to the check or an important piece being captured or due to "two-headed generals", the robot player will automatically move one or two more pieces, Until the chess game is stabilized.How much this would prolong the whole game, and how much each move would exceed the prescribed time, I do not know; though I do not believe that we can follow this direction very far without encountering the The time issue is difficult. I am willing to accept Shannon's speculation that such a machine can play chess at the level of a good amateur player, or even at the level of a good player.It played chess stiltedly and tediously, but far more robustly than anyone else.As Shannon pointed out, in the operation of the machine we can add enough chance to prevent the failures that are often encountered in purely systematic methods due to the rigid sequence of moves.This chance or uncertainty can be added to the estimation method of the final chess position after two moves. The machine will also use the standard defense as the attack and the knowledge about the tricks to play the kind of defense as the attack and make the possible tricks like a human being.A relatively perfect machine will record every game played in the past on the paper tape, and will supplement the various chess moves we have determined, and these moves are the result of the machine's study of all past games. Chess game and then found some kind of knack.Simply put, it relies on the learning capabilities of machines.Although we now know that learning machines can be built, the technology for making and using them is still far from perfect.The time is not yet ripe to design a chess-playing machine according to learning principles, though perhaps not for long. A chess machine that can learn can exhibit a wide range of chess-playing abilities, determined by the skill of the players it has played against in the past.Perhaps the best way to build a good chess machine is to have it play against a chess master with a wide variety of chess skills.On the other hand, a well-conceived machine can be more or less damaged through a lack of careful choice of opponents.A horse can be badly ridden, too, if bad riders can spoil it. Among the machines that can learn, we should distinguish which things the machine can learn and which ones cannot.To make a machine, or to make it have a statistical tendency to perform a certain kind of behavior without excluding the possibility of other behaviors, requires that certain characteristics of its behavior be determined strictly and invariably.We call the first type of decisions optional and the second type of decisions restrictive.For example, if we do not add the prescribed rules of chess to the chess machine as a limitation, and if the machine is made to have the ability to learn, then the chess machine will unconsciously become a machine that performs different tasks. machine.Conversely, making a chess machine with rule constraints is still a learning machine in terms of chess playing tactics and strategies. Readers may wonder why we have lost interest in chess machines.Aren't they just harmless little gadgets by which designers want to show off their skills to the world, and hope people will be dumbfounded and reduced to amazement at their achievements? There's at least some smug, self-absorbed element about him.But, as you will immediately see, this emotion is not the only factor in my presentation of the matter here, and besides, this skill is not of primary importance to the reader who is not a professional chess player. Mr. Shannon has given several reasons why his research is more important than just designing a game that can only reduce the interest of gamers.Among these reasons he pointed out that this machine could be the precursor of machines for estimating various military situations and for determining the best course of action in any particular phase of them.No one thought he wasn't serious enough.Von Neumann and Morgenstern's famous book "The Theory of Games" once left a deep impression on the world, and in Washington, this impression was not shallow.When Mr. Shannon speaks of the development of military technology, he is not talking of a wild fantasy, but of an event that is imminent and extremely dangerous. In the December 28, 1948 issue of the famous Paris magazine (Le Monde), a Dominican monk P.Pere Dubarle wrote an insightful review of my book Cybernetics.Below I quote a few passages from him to illustrate some of his thoughts on the dire consequences that chess machines have facilitated and that have been embedded in the arms race. One of the most glamorous vistas thus opens before us, that of the rational management of human affairs, especially those concerning social interests and which appear to have a certain statistical regularity, such as social The phenomenon of public opinion development is statistically regular.Can't one imagine a machine that can collect information of this type or that, e.g. to determine the most probable development of events?Can it not be further conceived that there is such a state apparatus, which is under the control of the polity of the many countries of the earth, or under the control of the government of this planet, which is obviously much simpler. Is it a political ruling system?There is nothing currently preventing us from thinking like this. We can dream of a day when a machine a gouverner will supplement—for good or for evil—the apparent inadequacy of our brains at present to participate in ordinary political institutions. Generally speaking, various practical problems of human beings cannot be determined as clearly as numerical calculation data.We can only decide their probable values.A machine that deals with these processes and problems connected with them must therefore have that probabilistic rather than deterministic mindset, as, for example, a modern computer exhibits.This makes the machine's task more complicated, but not impossible.An example of this is the predictive machine that determines the effectiveness of anti-aircraft guns.Theoretically speaking, time prediction is not impossible, and the determination of the optimal decision is not impossible, at least within a certain range.A gaming machine such as a chess machine is made possible in order to establish such predictions.As for the various personnel processes that are the objects of government management, they can be integrated with the game in the sense that von Neumann has studied with mathematics.While there is already a partial set of rules for this type of game, there are other games involving a large number of players whose data are extremely complex.The state management apparatus can define the state as a player who can obtain information in an optimal way at each specific stage, and the state is the only supreme regulator of all local decisions.These are unique privileges; if these privileges are used scientifically, they will enable the state to defeat in all cases all players in the game of personnel except itself. It is enough to put forward the following two poles Explain the problem: Either destroy the opponent immediately, or cooperate with the opponent in a planned way.This is the inevitable result of the game itself without outside interference.热爱美好世界的人们确实是有某些东西让他们到梦乡中去寻找的。 不管这一切怎样，值得庆幸的也许是：国家管理机器不会在不久的未来出现。因为除有种种非常严肃的问题仍需搜集大量信息并从速处理外，预测的稳定性问题仍然处在我们的控制能力所能认真梦想的范围之外。这是因为人事过程可以比拟为规则不完全确定的博奕，尤其可以比拟为规则自身为时间函数的博奕。规则的这种变化，既取决于博奕自身所发生的种种情况的有效细节，又取决于博奕者们每一瞬间面对所得结果的心理反应所构成的系统。 还有比这些情况甚至变化得更加迅速的情况。在1948年的选举中，盖洛普民意测验所发生的情况看来就是一个极好的例子。这一切，不仅使得种种预测因素受到影响的复杂性增大，它也许还使得人事状况的机械操作根本破产。就我们所能作出的判断而言，这里只有两个条件可以保证人事问题取得数学意义上的稳定性。这两个条件是：一方面，广大的博奕者是十分愚蠢无知的，他们受到一位精明的博奕者的愚弄，而他甚至还可以计划出麻痹群众意识的方法来；或者，另一方面，有足够的善意允许某人为了稳定全局起见而把自己的决定提供给一位或为数无多的几位在全局中具有任意特权的博奕者作为参考。这是一门艰苦的课程，其中都是冷冰冰的数学，但它可以对我们这个世纪的冒险事业——彷徨于人情世事变幻莫测和可怕的大海兽的到来之间——指点迷津。和这种情况比较起来，霍布士《利维坦》只不过是一个有趣的笑话而已。今天，我们去创建一个庞大的“世界国家”是冒着风险的，在这样的国家中，能使群众统计地得到幸福的唯一可能条件恐怕就是存心蓄意作出粗暴不公之举了：对于每个头脑清楚的人讲来，这是一个比地狱还要坏的世界。对于目前正在创建控制论的人们而言，给他们的技术干部增加上述的思想也许不无好处，这些技术干部现在已经从所有各门科学的地平线上出现了，其中有些是严肃的人类学家，也许还有一位对世界问题表现出某种好奇心的哲学家。 P．杜巴勒的国家管理机器并不因为它有自动控制人类的任何危险而令人感到恐怖。 这种机器过于粗糙，过于不完善了，它不足以表现人类合目的的独立行为的千分之一。 不过，它的真正危险却是完全另一回事，那就是，这类机器虽然自身不会兴风作浪，但可以被某人或某一伙人所利用，以之来增强他们对其余人类的控制；或者是，某些政治领导人不是企图借助机器自身来控制人民，而是企图通过政治技术来控制人民，这种政治技术对人的可能性显得如此之狭隘，如此之漠不关心，就好象它们事实上是用机器制订出来的一样。机器的最大弱点——正是这个弱点使我们远不至于被它统治住的——就是它还计算不出表征人事变化幅度甚大的几率性。用机器来统治人类就预先假定了社会已经处在熵增加的最后阶段，其中几率性可以略而不计，各个个体之间的统计偏差等于零。幸而，我们现在还没有达到这样一种状态。 即便现在还没有P．杜巴勒的国家管理机器，但就本世纪五十年代的种种发展所已经表明的情况看来，我们还是发展出了新的战争概念，新的经济竞争概念以及以冯?诺意曼的博奕论（它自身就是一种通讯理论）为依据的宣传概念。我在前面的一章中已经讲过，这种博奕论有助于语言理论的研究，但是，现在有些政府机构却热衷于把它应用在军事和半军事的攻守目的上面了。 博奕论依其本质而言乃是以博奕者之间的协议或结合为基础的，每个博奕者都力图制订一种策略来达到自己的目的，都假定自己的敌手和自己一样地为了争取胜利而各自使用最优的策略。这种大规模的博奕已经机械地实现了，而且大量制造出来了。纵使这种理论所依据的哲学也许不为我们的对手共产主义者所接受，然而，有种种明显的迹象表明：在俄国也象在我们这里一样，对于它的可能性已经作了研究，俄国人不满足于接受我们所提出的理论，已经在它的若干重要方面作了可能的修正。具体说，我们在博奕论上所完成的大部分工作（虽然不是全部工作）都是以下述假定为依据的：敌我双方都有无限的才干，我们博奕所受到的限制唯一地决定于分配到我们手上的牌或者棋盘上的明显局势。有相当数量（事实方面而不是文字方面）的证据表明：俄国人给世界赌局的这个态度补充了一个看法，即考虑到了博奕者的心理限制，特别是考虑到了他们作为赌局自身的组成部分的疲劳性。因此，现在世界矛盾的双方本质上都在使用着某种国家管理机器，虽然它从任一方面说来都不是一部独立的制订策略的机器，但它却是一种机械技术，这种机械技术是适应于那群醉心于制订策略的、象机器般的人们的紧急需要的。 P．杜巴勒吁请科学家注意世界上的军事和政治方面的日益增长的机械化，其情况就跟一部巨大的按照控制论原理进行工作的超人般的机器一样。为了避免这种机械化所带来的多方面的（外在的和内在的）危险，他之强调需要人类学家和哲学家是十分正确的。 换句话说，作为科学家，我们一定要知道人的本性是什么，一定要知道安排给人的种种目的是什么，甚至当我们一定得去使用象军人或政治家之类的知识时，我们也得做到这一点；我们一定得知道为什么我们要去控制人。 当我说到机器对社会的危险并非来自机器自身，而是来自使用机器的人时，我的确得强调一下S.巴特勒的预见。在《爱理翁》中，他认为，机器只有被人用来作为自己的附属器官时才能征服人类，否则，它就无所作为。但虽然如此，我们还是不宜把巴特勒的这个预见看得过分认真，因为事实上在他的那个时代，他和他周围的任何人都无法理解自动机行为的真正性质，而他所讲的话，与其说是科学方面的评论，勿宁说是言词方面的尖锐夸张。 自从我们不幸发现了原子弹以来，我们的报纸一直在大事渲染美国人“懂得如何做”。 但是，还有一种比“懂得如何做”更加重要的品质，而这，我们就无从责备美国有任何不当之处了。这个品质就是“懂得做什么”，我们不仅据此来决定如何达到我们的目的，而且据此决定我们的目的是什么。我可以举出一个例子来说明二者之间的区别。若干年前，有位知名的美国工程师买了一架高价的钢琴。一两个星期以后，事情明白了，该物之被购买并非因为他对钢琴演奏的音乐特别减到兴趣，而是因为他对钢琴的机械结构有着不可抗拒的好奇心。对于这位先生讲来，钢琴这种乐器并非产生音乐的工具，而是给某位发明家提供机会来表明他在乐器生产中如何巧妙地克服若干困难的工具。这种态度对于中学二年级学生讲来是值得尊敬的，但对于国家的整个文化前途赖以决定的人物之一讲来，这种态度如何值得尊敬，我留给读者去考虑。 在我们童年时代读过的神话故事中，我们学到了一些比较单纯、比较浅显的生活真理，例如，当我们发现瓶中装有妖魔时，最好的办法是把瓶子扔下；如果渔夫在自己妻子的唆使之下向上天祈求恩赐的次数太多时，那他就要回到原先由之出发的状态的；如果让你满足三个愿望，那你就要对你所希望得到的东西十分当心。这些单纯浅显的真理是从儿童语言表达出来的人生悲剧感，它是希腊人和许多现代欧洲人都具有的观点，但它不知何故却是这个富饶国家所缺少的东西。 希腊人是以极端矛盾的情绪来对待大的发现这桩事情的。一方面，他们和我们一样，认为火是给予全人类的巨大恩益。另一方面，把火从天上取到人间乃是对奥林普斯诸神的反抗，而这就不能不因冒犯诸神的特权而受到他们的谴罚。于是，我们看到了取火者普罗米修斯的伟大形象——他是科学家的原型，一位英雄，然而却是应该受罚的英雄——被锁在高加索山上，让兀鹰来啄食他的肝肠。我们都读过伊斯奇拉斯（Aescnylus）的音韵铿锵的悲剧诗章，诗中讲到，这位被囚禁的神在祈求着阳光普照之下的全世界为他作证，证明他在诸神手中遭受到何等的苦难。 悲剧感意味着世界不是一个快乐的、为了保护我们而创造出来的小窝巢，而是一个具有巨大敌意的环境，在这样的环境里，我们只有反抗诸神才能取得伟大的成就，而这种反抗又必然地给它自己带来了谴罚。这是一个危险的世界，在这个世界里，除了谦卑顺从、知足常乐可以得到某种消极的安全外，再也没有任何安全了。我们的世界是这样一个世界，其中理所当然的谴罚不仅要落到有意犯罪者的头上，而且要落到其唯一罪过就是对诸神和周围环境措然无知者的头上。 一个人如果怀着这种悲剧感去对待另一种力之本源的显现，不是火，例如，去对待原子分裂，那他就会怀着畏惧颤栗的心情。他不会冒险进入天使都害怕涉足的地方去的，除非他准备接受堕落天使的折磨。他也不会心安理得地把选择善恶的责任托付给按照自己形象而制造出来的机器，自以为以后不用承担从事该项选择的全部责任。 我讲过，现代人，特别是现代美国人，尽管他可以有很多“懂得如何做”的知识，但他的“懂得做什么”的知识却是极少的。他乐意接受高度敏捷的机器决策，而不想较多地追问一下它们背后的动机和原理为何。他这样做，迟早是要把他自己置身于w． W．贾可布斯（Jacobs）的《猴掌》（The Monkey's Paw ）一书中那位父亲的地位上的，这位为父者企望得到一百金镑，结果只是在他家门口碰到他儿子工作的那家公司的代理人，给他一百金镑作为他儿子在厂里因公死去的抚卹金。或者，他还可以象中阿拉伯渔翁在那只装有愤怒妖魔的瓶子上揭开所罗门的封印时所做的那样地做去。 让我们记住：猴掌型的和瓶装妖魔型的博奕机都是存在的。任何一部为了制订决策的目的而制造出来的机器要是不具有学习能力的话，那它就会是一部思想完全僵化的机器。如果我们让这样的机器来决定我们的行动，那我们就该倒霉了，除非，我们预先研究过它的活动规律，充分了解到它的所作所为都是按照我们所能接受的原则来贯彻的；另一方面，瓶装妖魔型的机器虽然能够学习，能够在学习的基础上作出决策，但它无论如何也不会遵照我们的意图去作出我们应该作出的或是我们可以接受的决策的。不了解这一点而把自己责任推卸给机器的人，不论该机器能够学习与否，都意味着他把自己的责任交给天风，任其吹逝，然后发现，它骑在旋风的背上又回到了自己的身边。 我讲的是机器，但不限于那些具有铜脑铁骨的机器。当个体人被用作基本成员来编织成一个社会时，如果他们不能恰如其分地作为负着责任的人，而只是作为齿轮、杠杆和连杆的话，那即使他们的原料是血是肉，实际上和金属并无什么区别。作为机器的一个元件来利用的东西，事实上就是机器的一个元件。不论我们把我们的决策委托给金属组成的机器抑是血肉组成的机器（机关、大型实验室、军队和股份公司），除非我们问题提得正确，我们决不会得到正确的答案的。肌肤骨骼组成的猴掌就跟钢铁铸成的东西一样地没有生命，瓶装妖魔作为描述整个团体的综合形象时，就跟惊心动魄的邪法一样地可怕。 时已近矣，善恶抉择之机已经迫在眉睫了。