Chapter 4 Chapter 3: Stereotypes and Learning: Two Modes of Communication Behavior
As before, some types of machines and some life forms, especially the more advanced life forms, can change their behavior patterns on the basis of past experience, so as to achieve the specific purpose of anti-entropy.In these higher forms of communicating organisms, the environment as an individual's past experience is capable of changing the individual's behavioral patterns in a sense to be more effective in dealing with his future environment.In other words, the organism is not in predetermined harmony with the universe like Leibniz's clock machine monad, but seeks a new balance with the universe and its future contingencies according to the actual situation.Its present is different from its past, and its future is different from its present.In living beings, as in the universe itself, exact repetition is absolutely impossible.As far as the analogy between living organisms and machines is concerned, Dr Ahibe's work may be our greatest contribution to the problem to date.Learning, like simpler forms of feedback, is a process in which seeing the past is different from seeing the past than seeing the future from the past.The whole concept of an apparently purposive organism (whether it be mechanical, biological, or social) is an arrow in the flow of time with a certain direction, not an arrow at both ends. A line segment that can go in either direction.A creature that can learn is not the two-headed snake in ancient mythology, with heads at both ends, and it doesn't matter where it goes.A learning creature moves from a known past to an unknown future that is not interchangeable with the past.
Let me give another example to clarify the role of feedback in learning.When the huge control rooms on the locks of the Panama Canal were at work, they were communication centers in both directions.The control station not only sends signals to control the operation of the traction locomotive, to control the opening and closing of the gates, and to control the opening and closing of the water gates, but the room is also full of instruments, which in addition to indicating that the locomotive, gates and water gates have received their signals commands, but also to show that they have actually effectively carried out those commands.Had this not been the case, the lockkeeper would have been quick to think that perhaps a tractor had stalled, or a warship of tonne had crashed into the gate, or one of many other similar mishaps had occurred.
The control principle can be applied not only to the locks of the Panama Canal, but also to states, armies and individuals.
During the American Revolutionary War, due to the carelessness of the British, the orders already given failed to bring together a British army from Canada under British command and another British army from New York at Saratoga. This led to the disastrous defeat of the British Burgoyne forces, which could have been avoided by a well-thought-out two-way communication program.It follows that administrative officials, whether government, university, or corporate, should participate in a two-way communication flow, not just a one-way communication from top to bottom.Otherwise, superior officials will find that their policy is based on the total misunderstanding of facts by their subordinates.Again, there is no more difficult task for an orator than addressing an expressionless audience.The purpose of applause in the theater is, by its very nature, to induce some two-way communication in the minds of the actors.
The question of social feedback has enormous implications for sociology and anthropology.The modes of communication in human society are extremely diverse.Some societies, such as those of the Eskimos, seem to have no leadership, and the relations of subordination among members are very insignificant, so that the social group is based only on the climate and the supply of food. A common desire to survive under very specific conditions.There are societies divided into many classes, such as are found only in India, in which the means of communication between two persons are strictly limited by their family and social status.Some societies are governed by an absolute monarch, and every relationship between two subjects is subordinate to the relationship between the monarch and his subjects.Then there was the feudal society of hierarchies of lords and serfs with very specific social communication techniques.
Most of us Americans like to live in fairly relaxed social groups where the barriers to communication between individuals and classes are not too great.I do not mean to say that the United States has reached this ideal in terms of communications.In a situation where white supremacy is still the credo of much of the country, this ideal will always fall short.
But such a limited and indeterminate form of democracy may even be considered too anarchic for many who regard efficiency as their highest ideal.These efficiency-worshippers like to let each person move on the social track assigned to him from childhood, performing social functions that bind him like a slave is bound to the earth.
It is shameful that these tendencies, this denial of the opportunities of an unknown future, exist in the American social landscape.It is for this reason that many, though wholeheartedly attached to this well-ordered State, which forever assigns men's social functions, would be discomfited if they were compelled to admit it openly.They can only show their apparent preference by their actions.But these actions are clear and prominent enough.A businessman isolates himself from his employees by surrounding himself with a group of submissives; or the head of a large research institute assigns each subordinate a research topic, but does not give him independence. the right to think, lest he go beyond the scope of the topic and glimpse the full gist of the research work.All this shows that the democracy they respect is not really the order they would like to live in.The well-ordered state they aspire to, in which the social functions of individuals are preassigned, is reminiscent of Leibniz's automaton, which does not provide irreversible activity towards a contingent future, but rather It is the real condition of human life.
In an ant society, each member performs a specific function.There may also be a full-time soldier class.Certain highly specialized individuals perform the functions of emperor or empress.If this society were to be adopted as the model of human society, we would live in a fascist state, in which each individual was destined by birth to have a specific occupation: rulers always rulers, soldiers always soldiers, peasants They are nothing more than farmers, but workers are destined to be workers.
One of the themes of this chapter has been to point out that the Fascist desire to use the ant society as a model for a state is the result of a gross misunderstanding of the nature of both ants and humans.I would like to point out that the physiological development of an insect itself determines that it is an essentially stupid, unlearned individual, doomed to change to any great extent.I would also like to show how these physiological conditions make insects a cheap, mass-producible thing with no more individual value than a paper pie plate that is thrown away after a single use.On the other hand, I would like to show that man is biologically equipped for a great deal of study and research, which takes up almost half his life, while ants lack it.Variety and possibility are inherent properties of the human senses, and indeed they are the key to the grand leap in understanding man, since both variety and possibility are peculiar to the very fabric of man.
Even if we can ignore the superiority of human beings far over ants, and use human beings as materials to organize an ant-like fascist country, I am convinced that this kind of practice is the devaluation of human nature, and economically speaking, it is also unreasonable for human beings. A waste of great value.I think I believe that human society is far more useful than ant society; and if man were judged and confined to perpetually repeating the same function, I fear he would not even be a good ant, let alone a good man.Those who want to organize us in terms of constant individual functions and constant individual limitations are condemning the human race to move forward with far less than half its power.They have eliminated almost all human possibility, and by limiting the ways in which we can adapt to future contingencies, they have destroyed our chances of surviving on this earth for a considerable period of time.
Let us now return to the limitations of the ant's individual structure that make ant societies such special things.These limitations have deep roots in the anatomy and physiology of individual ants.Both insects and man are air-breathing biological forms, representatives of the long transition from the convenient living conditions of aquatic animals to the extremely demanding products of terrestrial confinement.Wherever this transition from the aquatic to the terrestrial world takes place, it entails a fundamental modification of the respiratory system, the circulatory system, the mechanical framework of the organism, the organs of perception, etc.
The mechanical enhancement of the bodies of terrestrial animals proceeded along several unrelated paths.In most molluscs, as in certain other groups of organisms (which, although not related to molluscs, have essentially mollusk-like forms), they secrete from some part of their outer skin a A lifeless, calcium-containing tissue called a carapace.This thing is constantly added from the earliest stages of the animal to the end of its life.Those groups of organisms which develop in a spiral form need only be explained by this process of addition.
If the carapace is protective to the animal, and the animal grows to considerable size in its later stages, the carapace must be a very considerable burden, suitable only for the slow-moving, quiet-living terrestrial creatures of the snail-like kind. animal.In other shelled animals, the lighter the shell, the less burdensome it is, but at the same time the less powerful the defense.The construction of the shell, with its heavy mechanical loads, is a modest achievement among terrestrial animals.
Man himself represents another direction of development, which can be seen in all vertebrates; and in invertebrates, at least as highly developed as the horseshoe crab and the octopus, this direction is also marked.In all this biological form, parts of the connective tissue inside have coalesced, and instead of being fibrous, they have become a dense, hard jelly.These parts of the body are called cartilage and are designed to attach the powerful muscles needed for the animal's active life.In higher vertebrates, this primitive cartilaginous skeleton acts as a temporary scaffolding, which is then replaced by a harder material called bone, which is more suitable for attaching strong muscles.These bones made of bone or cartilage are not composed of a large number of living tissues in any strict sense, but within this whole piece of intercellular tissue, they are full of living structures such as cells, cell membranes and nutrient blood vessels.
Vertebrates not only develop endoskeletons but also other traits to suit their active lives.Their respiratory system, whether in the form of gills or lungs, is well adapted to an active exchange of oxygen between the external medium and the blood, much more efficiently than the blood of ordinary invertebrates, because the vertebrae The blood of animals contains respiratory pigments concentrated in blood cells for transporting oxygen.This blood circulates in a closed vascular system under the more efficient pumping of the heart, rather than in an open system of irregular sinuses.
Insects and crustaceans and all arthropods are built in entirely different ways of growing.The body of arthropods is surrounded by a layer of chitin, which is secreted by epidermal cells.Chitin is a dense substance very close to celluloid.In the joint parts of the animal body, the chitin layer is very thin and relatively soft, but in the rest of the body, it is a hard exoskeleton, which we can see in prawns and cockroaches.An endoskeleton, such as that of a human, is able to grow along with the growth of all tissues.Exoskeletons can't do that unless they grow by addition like snails.The exoskeleton is dead tissue with no intrinsic ability to grow.Its function is to provide strong protection to the body, and it is also used for muscle attachment, but it is equal to a tights.
In arthropods, internal growth can be transformed into external growth by shedding the old tights and growing a new one underneath the old one, which is initially soft, bendable and able to take Slightly newer and roomier in style, but it quickly becomes worn and hardened.In other words, their growth stages are marked by certain moulting cycles, which are more frequent in crustaceans and much less so in insects.The larval stage may consist of only several molt stages.The pupal stage is its transitional form, at this time, the wings, which were originally inactive in the larval stage, develop internally towards a functional state.This developmental process is not complete until near the end of the pupal stage, and this moulting completes the adult.Once you are an adult, you will never peel your skin again.This is the sexual stage of insects. In most cases, although it still has the ability to eat food at this time, the mouth and digestive tract of some insects stop developing, so this insect is called an imago.Adults can only mate, lay eggs, and then die.
In the process of shedding old clothes and growing new ones, the nervous system is involved.Although we have a certain amount of evidence that some kind of memory is maintained at the transition from larva to adult, this memory cannot be very extensive.
The physiological conditions of memory, and thus of learning, appear to be a certain continuum of organization, that is, the transformation of changes produced by sensory impressions from the outside into relatively permanent changes in structure or function.The metamorphosis of insects is too complete to preserve any long-term record of these changes.It is indeed difficult to imagine that after such a severe process of internal transformation, a memory of any degree of precision can be maintained.
Another limitation of the insect has to do with its method of respiration and circulation.The heart of an insect is a very weak tubular structure, which communicates not with blood vessels of definite shape, but with cavities or sinuses of indeterminate shape, which carry the blood to various tissues.This blood is devoid of red blood cells, but contains hemoglobin in solution.This method of oxygen delivery certainly seems to be much inferior to the method of oxygen delivery through blood cells.
Furthermore, methods of oxygenating insect tissues utilize blood locally at best.In the body of this animal there is a branched tracheal system, which draws air directly from the outside into the aerobic tissues.These tracheae are protected from damage by spiral fibers of chitin, so that they open passively, and we find nowhere evidence that insects have an active and efficient air-pumping system.Insects breathe only by diffusion.
Remarkably, it is the same tracheal system that brings in fresh air by diffusion and carries used, dirty air containing carbon dioxide out of the body.In the mechanism of diffusion, the diffusion time does not vary with the length of the trachea, but with the square of the tube length.In general, therefore, the efficiency of the system decreases very rapidly as the size of the animal increases, and for animals of considerable size the efficiency of the system decreases below the point of survival.Therefore, from the structure of the insect, it is not only impossible to have the best memory, but also impossible to grow larger.
To illustrate the significance of the aforementioned constraints on the size of insects, let's compare two artificial structures—a hut and a skyscraper.The ventilation condition of the hut can be properly ensured by the air circulation near the window frame, without considering the problem of duct ventilation.On the other hand, in a skyscraper divided into many units, switching off the forced ventilation system can make the workplace air unbearably foul in a few minutes.For such structures, ventilation methods such as diffusion or even convection are not sufficient.
The maximum overall size of insects is smaller than that of vertebrates, but those basic elements of which insects are composed are not always smaller than those of man or even of whales.The nervous system of an insect is also small in size according to its body size; however, it contains not much smaller neurons than that of a human brain, although they are much smaller in number and its structure is far less complex.In matters of intelligence we should expect that it is not only the relative size of the nervous system but, to a large extent, its absolute size that matters.In the small and tiny structure of an insect, there is certainly no room for a very complex nervous system, no place for a huge amount of memory to be stored.
From the point of view of the impossibility of storing large amounts of memory, insects have no opportunity of learning much, as can also be seen from the fact that, during the course of growth, by such catastrophes as physiological deformations, An insect such as an ant spends its infancy in a form of life unrelated to its adulthood.Moreover, the fact that the behavior of an insect in adulthood must be complete in its nature from the outset makes it clear that the instructions received by the insect's nervous system must be essentially the result of the way it is structured rather than any personal experience product of.Thus, the insect is very much like a computer that states all its instructions on a "paper tape" in advance, with little feedback mechanism to help it take action in an uncertain future.Ant behavior is primarily a matter of instinct rather than intelligence.The physical straitjacket in which an insect grows up directly determines the psychological straitjacket that regulates its behavioral patterns.
Readers may want to say here: "Okay! We already know that ants are not very intelligent as individuals, so why bother to explain a lot of reasons why they can't be intelligent?" The answer lies in that cybernetics adopts such a Viewpoint: The structure of a machine or organism is the index by which its performance can be seen.Just as the mechanical stereotypes of an insect limit its intelligence, the mechanical variability of man offers almost unlimited prospects for its development.This fact is closely related to the point of this book.Theoretically, if we can build a machine whose mechanical structure is a copy of the physical structure of a human being, then we can have a machine whose intelligence is a copy of the human intelligence.
In terms of behavioral stereotypes, there is no mammal in general, and humans in particular, that differ most from ants.We often see that man is in a neoteinic form, that is to say, if we compare man with his close relatives, the great apes, we will find that adults are different in hair, head shape, body shape, and body proportions. In terms of body, bone structure and muscles, etc., they are more similar to the newly born apes than to the adult apes.Among animals, man is Peter Pan who never grows up.
This anatomical immaturity corresponds to a long childhood.Physiologically, man does not reach puberty until he has lived a fifth of his normal lifespan.Let's compare this with the corresponding ratio for rats.A mouse can only live for three years, but after three months, it begins to reproduce.This is the ratio of twelve to one.Among most mammals, this ratio is nearly normal in mice compared with humans.
Puberty in most mammals either marks the end of the protective period or marks the onset of puberty well after the end of the protective period.In our society, a person is not considered an adult until the age of twenty-one, and the education time required for modern higher vocational education lasts until about thirty years old, which is actually past the period of the strongest physical strength.Therefore, a person can spend up to forty percent of his normal lifespan in being a student, and the reason is again related to his physiological structure.It is quite natural for human society to be based on learning, just as ant society is based on inherited patterns.
Like other organisms, man lives in a universe of chance, but his superiority over other creatures lies in his physical and therefore intellectual equipment that enables him to adapt to major changes in his environment.The human race is powerful only because it exploits the innate capacity for adaptive learning, a possibility afforded by its physiology.
We have pointed out that an effective action must obtain information through some kind of feedback process, so as to know whether its purpose has been achieved.The simplest feedback is that which deals with the overall situation of performance success or failure, such as whether we actually grabbed what we wanted to grab, or whether a vanguard arrived at the specified place at the specified time.However, there are many other forms of feedback of a more complex nature.
It is often necessary to know whether a general strategy of conduct, such as a strategy, has proven successful.When an animal is taught to navigate a maze to find food or avoid an electric shock, it must be able to fully document the success or failure of its master plan for navigating the maze, and it must also be able to modify this plan in order to navigate the maze efficiently.This form of learning is certainly a kind of feedback, but it is a higher level of feedback, that is, it is strategic feedback rather than simple action feedback.Just B．In terms of what Russell called "logical types", this feedback is different from those more basic feedbacks.
This pattern of behavior can also be found in machines.Recent innovations in telephone wiring technology provide an interesting mechanical analogy to human adaptability.Throughout the telephone industry, automatic switches quickly outperformed hand-operated switches, and people still seem to think that automatic switching devices are now a near-perfect thing.However, with a little thought, one can understand that the current wiring process is very wasteful of equipment.
There are only so many people I really want to reach on the phone, and most of the people I spoke to today are the same people I spoke to yesterday, day after day, week after week.I used telephone equipment to establish communication with this group of people.Now, with current wiring technology so pervasive, the process of connecting to someone who calls us four or five times a day is indistinguishable from the process of connecting to someone who perhaps never spoke to us before.From the point of view that the telephone line load should be equal, we can see that the telephone equipment used is either too little for frequent paging users or too much for infrequent paging users. This situation reminds me of Holmes's "Single The poem "The Carriage".You will probably remember that this ancient carriage, after a hundred years of use, showed itself to be so well designed that neither the wheels, the roof, the bars, nor the seats showed any incongruity. Economical, where it wears out more than other parts.In fact, the "one-horse carriage" is a representative of cutting-edge technology, and it is not just a humorous fantasy.If the hoops outlasted the spokes or the fender bars, some economic value would be wasted in these durable parts.These economic values can either be saved without compromising the durability of the entire vehicle, or they can be evenly distributed over the entire vehicle to make it more durable.Indeed, any structure that is not of the nature of a "one-horse carriage" is wastefully designed.That is to say, as far as the most economical service is concerned, the process of contacting me with A (who I call three times a day) and the process of contacting me with B (who is in my phone book is A household that is not noticed) is equal, not ideal.If I allocate a little more means for me to contact A directly, then even if I spend double the waiting time to connect with B, it can be fully compensated.If this is the case, then we can design an instrument to record my past conversations without spending much money, and redistribute a service level to me according to the amount of lines I used in the past, then it will serve me better. Or spend less, or both.Philips Electric Light Company of the Netherlands has successfully done this.Its service has been improved with feedback from what Russell calls "advanced logic types."This kind of equipment has more changes, greater adaptability, and works more effectively than common equipment, because common equipment has an entropy trend, and the ones with high probability overwhelm the ones with low probability.
To reiterate: Feedback is a way of controlling a system by inserting its past performance into it.
If these results are used only as data to qualify and tune the system, that is simple feedback for the control engineer.
But when the information describing the performance is sent back and can be used to change the general method of operation and the mode of performance, then we have a process that may well be called learning.
Another example of the learning process is found in the design of predictive machines.In the early days of World War II, anti-aircraft fire was so ineffective that it was necessary to invent an instrument capable of tracking the position of an aircraft, calculating its distance, and determining the distance a shell experienced in the air before hitting it time, and the position the aircraft will reach at the end of that time.If the plane is capable of evasive maneuvers at will, no skill we have can grasp the motion of the plane between the time the anti-aircraft guns start firing and the time when the shell reaches the target, which we do not know.
However, in many situations, the driver is not or unable to take voluntary evasive maneuvers.He is limited by the fact that if he makes a quick turn, centrifugal force will knock him out; habit, even in its evasion maneuvers.These regularities are not absolute, but the statistical advantages shown by repeated practice.It may be different for different pilots, but it must be different for different aircraft.Let's remember: When tracking something as fast as an airplane, the calculator doesn't have time to pull out the instruments and figure out where the airplane is going.All calculations had to be programmed into the control system of the anti-aircraft gun itself.This calculation must include data from our past statistical experience with certain types of aircraft under different flight conditions.All anti-aircraft guns in use today are equipped with a calibration instrument, which either uses such fixed data, or selectively uses these limited fixed data.Their correct selection can be changed at will according to the needs of the gunner, just by pressing a button.
However, there is another level of control problem which can also be solved mechanically.Obtaining statistical data from the actual observation of aircraft flight, and then transforming these statistical data into rules for controlling anti-aircraft guns, this problem itself is a specific mathematical problem.Tracking an aircraft according to a given rule is relatively slow compared to tracking an aircraft through actual observations because it requires a large number of observations of the aircraft's past flights.But even so, it is not impossible to mechanize this long activity as well as the short one.Therefore, it is possible to build an anti-aircraft weapon that itself can make statistical observations of the movement of flying targets, then process these materials, program them into a control system, and finally use this system as a method of rapid adjustment , aligning the position of the weapon with the observed position of the aircraft and its motion.
As far as I know, this has not been done yet, but it has been included in our research, and it can hopefully be applied to other forecasting problems.The formation of a master plan for anti-aircraft weapons to be aimed and fired in response to specific movements of the aircraft is, in its essence, an act of learning.This is a change in the program band in the calculation mechanism of anti-aircraft weapons, which is not very different from the process of interpreting numerical data.In fact, it is a very general feedback that can change the entire method of instrument operation.
The high-level learning process we are discussing here is still limited by the mechanical conditions of the system in which it is located, and it is obviously not comparable to the normal learning process of humans.From the reading process, however, we can deduce quite different ways of mechanizing complex types of learning.The guiding ideology of these methods is given by Locke's association theory and Pavlov's conditioned reflex theory respectively.Before I discuss them, however, I shall answer some criticisms of the insights I present below with some general explanations.
Let me say a little more about the grounds on which the theory of learning might be based.The vast majority of the work of neurophysiologists is concerned with the conduction of impulses in nerve fibers or neurons, and this process is given by the phenomenon of "all or nothing".That is to say, if a stimulus reaches a certain point or end along a nerve fiber, it will propagate down a certain nerve fiber there, as long as it does not disappear within a relatively short distance, then the stimulus The effect produced at the more distant points of the nerve fiber is essentially independent of the co-initial strength.
These nerve impulses travel along one fiber to another through the connection point between them, the so-called synapse, where one input fiber can be seen to connect with multiple output fibers, and one output fiber A fiber can also be connected to multiple input fibers.At these synapses, the impulse provided by a single input nerve fiber is often insufficient to produce an effective output impulse.In general, if too few individual impulses reach a given output fiber from an input synapse, the output fiber does not respond.I have said so little here that it does not necessarily mean that all input fibers function the same, nor does it even necessarily mean that there are a number of synapses engaged in input activities as connection points, so that whether the output fiber responds or not The problem can be solved once and for all.I also don't want to ignore the fact that some input fibers not only do not produce stimuli in the output fibers to which they connect, but have a tendency to prevent these output fibers from receiving new stimuli.
In any event, even though the problem of impulse conduction along nerve fibers can be described in rather simple terms, for example, in terms of all-or-none phenomena, the problem of the transmission of an impulse through the synaptic layer still depends on complex response Patterns in which certain combinations of input fibers fire for a limited time, enabling further transmission of the message, while other combinations do not.These combinations are not static or even solely dependent on how that synaptic layer has received messages in the past.They are known to vary with temperature, and probably with many other factors.
The above insights about neural theory are consistent with those of machines consisting of a series of switching devices.In such machines, the closing of the subsequent switches depends on such a precisely coordinated action as to close simultaneously the preceding batch of associated switches.
This "all or nothing" machine is called a digital computer.It has great convenience in solving various communication and control problems.In particular, since it's only deciding between "yes" and "no," it accumulates information in such a way that we can distinguish extremely small differences among very large numbers.
Besides these machines that work on the principle of yes and no, there are other computers and control machines that measure, not count.Such machines are called analog computers because their operation is based on the analogy between the quantities to be measured and the numerical quantities representing them.An example of an analog computer is the slide rule, which is quite different from a desktop computer that does number crunching.Anyone who has ever used a slide rule knows that both the printed scale and the accuracy of our eyes place significant limits on the precision we can read on the ruler.These limitations are not as people think, as long as the ruler is made larger, they can be easily resolved.A ten-foot-long slide rule adds only one-tenth the precision of a one-foot-long slide rule, and to achieve this precision we not only have to make each foot of the large slide rule the same precision as the small slide rule , and this foot and that foot must be aligned in the same direction as the expected precision of the small slide rule.除此以外，保持大尺的刚性这个问题要比保持小尺的刚性麻烦得多，这就使得我们依靠增大尺寸来增加精密度的办法受到了限制。换言之，从实用目的看来，用作测量的机器不同于用作计数的机器，因为它的精密度受到了极大的限制。把这一点加到生理学家对全或无活动的偏爱上面，那我们就不难理解为什么对人脑的机械模拟所做过的大部分工作都要在不同的程度上以计数作为基础。
但是，假若我们过分坚持人脑是一部值得推崇的数字计算机，那我们就要受到某种非常公正的批评了。批评可以部分来自生理学家，部分来自心理学家，后者是跟那些不喜欢用机器作对比的心理学家们多少持着相反意见的。我讲过，数字计算机中有程序带，它决定所要完成的操作程序，而程序带在过去经验基础上的变化就和学习的过程相当。
在人脑中，最最显见的类似于程序带的地方就是突触阈的确定性，即激发一个与之相连的输出神经元的那些输入神经元要在彼此之间作出精确组合的确定性。我们已经知道，这些阈值随着温度而变化；我们没有理由认为，它们不随血液的化学成分以及许多自身根本没有全或无性质的现象而变化。因此，在考虑学习问题时，采用神经系统的全或无理论，我们就必需特别当心，如果我们对这个概念还没有做过理论上的批判，而又没有特定的实验证据来支持这个假设的话。
常常有人说，任何一个适用于机器的学习理论都是不存在的。也有人说，就我们目前的认识水平而言，我所能提出的任何一种学习理论都不免为时过早，它和神经系统的实际情况大概不对头。我希望从这两种批判意见的夹缝中穿过去。一方面，我希望提出一种制造学习机器的方法，要求这个方法不仅能够使我造出一些特定的学习机器，而且能够给我提供一种制造多种多样学习机器的一般工程技术。只有在我达到这种一般性的程度时，我才能够免除下述的批评：我所主张的类似于学习的那种机械过程事实上是某种与学习的本质完全无关的东西。
另一方面，我希望使用与描述神经系统以及人和动物的行为的实际过程不太不同的语言来描述这种机器。我充分了解到，我在表述人的实际机制时不可能期望在每个细节上都是正确的，我甚至可能在原则上发生错误。但虽然如此，只要我提出一种能够用那些属于人心和人脑方面的概念对之进行文字描述，那我就是给出一个免于受到批评的起点，也是给出一个用以和其他理论所能得到的成果进行比较的准绳。
在十七世纪末叶，洛克认为人心的内容就是他称之为观念的那种东西构成的。对他说来，人心完全是被动的，是一块干干净净的黑板，是tabula rasa（一张白纸），个人经验就是他在这张白纸上面所写下的印象。如果这些印象经常地出现，或是同时地出现，或是在某一序列中出现，或是在我们往往归之于因果联系的那些情况中出现，那么，按照洛克的意见，这些印象或观念便具有某种能动的趋势把各个组成部分粘合在一起而形成复合观念。观念粘合的机制就在于观念自身之中，但是，洛克在其所有的著作中，有一个令人感到奇怪的反对描述这种机制的意图。他的理论与现实的关系只能是火车的照片与行进中的实际火车的关系。它是一张任何部分都是静止不动的图表。如果我们考虑到洛克学说产生的时代，这一点就不值得惊奇了。动力学的观点，动态地描述事物的观点，首先是在天文学中而非首先在工程学或心理学中获得其重要性的，这项工作要归功于牛顿，但牛顿不是洛克的先驱者，而是他的同时代人。
在许多世纪中，科学在亚里士多德冲动的驱使之下，着重于分类工作而把现代的研究冲动即研究现象发生作用的方式扔在一边。的确，当植物和动物还有待于调查研究的时候，要是不经过一个不断搜集材料以描述自然史的过程，我们就很难理解生物学如何能够进入真正的动力学时代。伟大的植物学家林耐（Linnaeus）就是一个例子。对于林耐讲来，种和类都是固定不变的亚里士多德式的形式，而不是进化过程的路标；但是，我们只有根据林耐的全面描述，才有可能找到令人信服的进化实例。早期的自然史家都是知识领域中的实干的拓荒者，他们围攻和占领新领域的欲望太强烈了，以致对于他们所观察到的新形式不能十分细致地作出解释。拓荒者之后来了讲究操作的农场主，自然主义者之后来了现代的科学家。
在上做纪最后四分之一和本世纪最初四分之一的年代里，另一位伟大学者，巴甫洛夫，以其独特方法从本质上研究了以前洛克研究过的同一个领域。但是，他的条件反射的研究是实验地进行的，而不是象洛克那样理论地进行的。除此以外，他认为条件反射是在下等动物中出现的东西，而不是在人体中出现的东西。下等动物不会讲人的语言，但是，它们能讲行为语。在它们的比较显眼的行为中，就其动机而言，大多数都是情绪方面的行为，而它们的情绪又大部和食物相关。巴甫洛夫正是从食物和唾液的生理征候而开始其研究的。我们可以很容易地把一根小管插入狗的唾腺中，并观测唾液在食物刺激出现时的分泌情况。
通常，许多东西都和食物没有什么联系，例如，视的对象、听的声音等等，它们对唾液不会产生任何的影响。但是，巴甫洛夫观察到，如果狗在进食时系统地出现某种对象或某种声音，那么，这个对象或声音单独出现时也足以激起唾液。这就是说，唾液的反射受到过去联想的制约。
这里，在动物反射的水平上，存在着某种类似于洛克的观念联想的东西，即反射应答所产生的联想，其情绪内容显然是很强烈的。我们现在考察一下那些性质相当复杂的为产生巴甫洛夫型的条件反射所必需的前提。首先，它们一般是动物生活中居于重要地位的东西，在上述情况下，就是食物，虽则在反射的最后形式中食物因素可以全部消除掉。我们还可只用畜牧场周围的电网为例，说明原始刺激在巴甫洛夫的条件反射中的重要性。
在畜牧场上，要建造一个足够牢固的线网来圈住牲口，不是一桩容易做到的事情。
因此，比较经济的办法就是用一两根较细的、通有高压电流的导线来代替这种笨重类型的线网，一当动物身体触及电网从而使电流短路时，动物就受到了一个十分可观的电击。
这种电网要能在开头一两次承受住了牲畜的压力，但继此以后，电网的作用，不在于它能够承受机械压力，而在于牲畜已经养成力图避免与电网接触的条件反射。在这里，反射的原始扳机乃是痛苦，而避免痛苦对于任何动物的生命延续讲来则是一枚极为重要的事情。形成该反射的扳机是牲口对于电网的视觉。除饥饿和痛苦外，还存在着其他的产生条件反射的板机。对于生物的这些情绪状态，我们可以用拟人的语言来讲述它们，但我们用不着这样一种的拟人主义，即把这些东西说成具有动物经验中所不具有的重要意义。动物的这些经验，无论我们可否称之为情绪的，都能够产生强烈的反射。在形成一般的条件反射时，反射应答使转移到这些扳机状态之一。这个扳机状态经常伴随原始扳机而出现。对于引起给定的应答，刺激物可变，这在神经方面一定是互为相关的：导致应答的突触通路是开着的，不然的话，就应当关着，或者说，不导致应答的突触通路是关着的，不然的话，就应当开着；这样就构戍了控制论所讲的程序带中的变化。
程序带中的这种变化是在旧的、强有力的、引起特定反应的自然刺激和新的、伴随而来的刺激之间经过多次反复的联系而后产生的。看来旧刺激在其活动的同时似乎具有一种能力，即改变其消息通路的渗透性。有趣的是，新的、起作用的刺激除了重复伴随原始刺激这一事实外，几乎没有其他的要求。所以，原始刺激在其出现之时似乎对于所有输送消息的通路都产生了一个长期的效应，至少对其中的大多数通路是这样的。刺激的代替物之具有任意性表明了原始刺激的变形效应极为丰富多样，它不是被限制在少数特定的通路上面的。因此，我们可以认为，原始刺激能够释放出某种一般性的消息，但它仅在原始刺激起作用之时才在那些消息通路中起到作用。这种作用的效应也许不是经久的，但它至少是相当长期地存在着。看来发生这种第二级活动的最合理的场所就在突触之中，因为这个地方最便于改变它们的阈值。
非直达消息这个概念，大家并不生疏。这种消息在找到接收者之前不停地传播出去，然后，它使接收者受到它的刺激。这类消息经常被用作警报。火警就是通知全城居民的信号，特别是通知消防人员的，不论他们呆在何处。在矿井中，当我们发现沼气而要求远处通道上的所有人员离开时，我们便在通风口处把一瓶乙基硫醇打破。我们没有理由认为神经系统中不会有这类消息。如果我去建造一架普通类型的学习机，那我就非常乐意采用这样的方法：把一般地传播“敬告所有与此事有关者”式的语息和特定通路的消息两者结合起来。设计种种电方法来执行此项任务，应该不难。当然，这完全不等于说，动物的学习实际就是采取传播性的和通道性的两种语息相结合的方式。坦白地说，我认为动物的学习完全可能就是这样，但是，我们的证据不足，所见它还只是一种猜想。
至于说到这些“敬告所有与此事有关者”式的消息的性质时，那我还是站在玄想较多的基础上面而假定它们的存在的。它们也许的确是神经性质的，但我宁愿倾向于把它们看作非数字的、类似于产生反射与思想的机制。把突触的活动归因于化学现象，这是自明之理。实际上，在一根神经的活动中，我们不可能把化学势和电势分开来；说某一特定活动是化学的，这几乎是毫无意义的话。但虽然如此，假定突触变化的原因或伴随物中，至少有一个原因或伴随物，不论其来源为何，可以局部地表现为化学变化，这跟流行观点不相抵触。这种变化的出现完全可以局部地取决于神经所传送出来的信号的。
我们至少可以同样地设想：这类变化可以部分地起因于化学变化，而化学变化一般是通过血液而非通过神经来传送的。我们又可以设想：“敬告所有与此事有关者”式的消息是由神经来传送的，这些消息自身局部地表现为化学活动的形式，伴随着突触的变化而出现。作为一个工程师，我认为，比较经济地传送“敬告所有与此事有关者”式的消息的办法似乎是通过血液，而不是通过神经。但是，我没有证据。
让我们记住：“敬告所有与此事有关者”式的消息所起的作用，在一定程度上，是和那种把全部新的统计资料都送进仪器的防空控制装置中的变化相似，而不是和那些只把特定的数字资料直接送进仪器的防空控制装置中的变化相似。在上述两种相似的情况中，都有一种也许是长期积累起来的活动，由于长期持续之故，这种活动将有种种效果产生出来。
条件反射对其刺激作出迅速的应答，并不必然地标志着条件反射的建立过程也是比较迅速的。因此，我认为，下述看法是适宜的：使得这种条件化得以产生的消息乃是通过血液流的缓慢而又普遍的传播作用带来的。
设定饥饿、痛苦或任何其他刺激的固定影响可以通过血液引起条件反射，这就已经把我所需的观点作了相当的限制了。要是我企图去确定这种未知的由血液带来的影响的性质，要是这种影响存在的话，那我的观点就要受到更大的限制了。血液自身带有种种物质，可以直接地或间接地改变神经的活动，这在我看来是一桩非常可能的事情；某些荷尔蒙或内分泌的活动至少暗示了这个事实。但是，这并不等于说，决定学习阈值的那种作用就是特定荷尔蒙的产物。此外，它虽然引导我们在饥饿和电网所引起的痛苦之间找出某种可以称之为情绪的共同物，但如果认为情绪就是决定反射的全部条件，而对反射条件的特殊性不作任何进一步的讨论，那就肯定跑得太远了。
但虽然如此，了解下述一点是有意义的：那种被我们主观地称之为情绪的现象，也许并不单纯是神经活动中的一种没有用处的附带现象，它很可能控制着学习过程中的以及其他类似过程中的某一重要阶段。我并不是说，它一定是这样的，但是，我要说，那些在人与其他生命体的情绪和现代类型的自动机的应答之间截然划上一条不可逾越的鸿沟的心理学家们，在他们作出否定的结论时，应当象我作出肯定的结论时那样地小心谨慎。