The Thinking of Men and Machines

A physicist with an M.A. from Harvard University. JOHN H. TROLLis a consultant for companies that manufacture electronic computers and has worked for many years on thinking machines of various kinds. He has written for military publications and technical books dealing with radar target intelligence and German electronic progress during the last war. In the article which follows he discusses the furtive rivalry between man and machine.



THE uneasy, half-embarrassed rivalry between man and machine has reached a peak with the thinking machine. We have become used to machines that are more powerful, more durable, more accurate, and faster than we are, but machines that, challenge our intelligence are hard to take. At this point the competition becomes uncomfortable.

Machines and tools have always been created in the image of man. The hammer grew from the balled fist, the rake from the hand with fingers outstretched for scratching, the shovel from the hand hollowed to scoop. As machines became more than simple tools, outstripping their creators in performance, demanding and obtaining increasing amounts of power, and acquiring superhuman speeds and accuracies, their outward resemblance to the natural model disappeared; only the names of the machine’s parts show vestiges of their human origin. The highly complex machinery of the modern industrial age has arms that swing, fingers that, fold, legs that support, teeth that grind, and male and female parts that mate. Machines feed on material, run when things go well, and spit and cough when they don’t.

But the newest machines possess human traits that had always been considered far beyond mechanization. Here we find not only electric eyes that see and sensing devices that feel, but also memories that recall and logic sections that classify, arrange, and select. These machines can make choices, comparisons, and decisions, learn from past experience, and reach logical conclusions on the basis of premises. It may no longer be denied: these machines can really think.

This realization has renewed the furtive rivalry between man and machine. The battle is being fought underground because even to concede the existence of such a contest would be undignified. Like a small child jealous of the attention paid a puppy, men do not often admit openly that this inhuman contrivance of nuts and bolts and evilly gleaming electron tubes is a threat. But as the child will get even with the puppy by tweaking its tail when no one is looking, so man, consciously or unconsciously, likes to throw monkey wrenches into machines and see them get their comeuppance.

Newspaper editors a few years back felt, that there would be interest in a story about a Japanese arithmetician who, with an abacus — a simple device made of a few counting beads — won a race against a mechanical calculating machine. The story was prominently featured in the world press. If the mechanical calculator had won, there would have been no story.

No one likes to depend on a rival. Consequently there is a general desire to distrust and by-pass machines. Pilots during the Second World War preferred to fly by the scat of their pants — a device so notoriously insensitive that it won’t tell the pilot when he flies upside down — rather than by their highly precise and reliable instruments. Many posters and disciplinary actions were necessary to make pilots use their instruments.

When Univac, one of the computers used on election night, made an amazingly accurate prediction of the outcome on the basis of very early returns, it was disbelieved by the experts who designed and constructed it. Even when by all rational standards it becomes evident that the machine knows better, man is reluctant to let it have the last word.

An even more telling sign of this half-secret battle of man and the technical monster of his creation is the character of the Utopias of our time. Where Thomas More of the sixteenth century and Edward Bellamy of the nineteenth found ideal, beautifully harmonious societies in their imaginary travels, with satisfactory solutions to the pressing problems of their days, George Orwell and Aldous Huxley in our age see only a night marishly heightened outgrowth of the modern world. In their Utopias, standardization, an integral part of the machine culture, extends to the hygienically controlled production of humans; machines take all the major roles in human enjoyment, dominating even sex and simple sports; machines write all novels and plays and newspapers and create all art and entertainment; machines watch and spy day and night, destroying all vestiges of human individuality. Is the arrival of the thinking machine the first, sign that these nightmares are about to become a reality? Is man hopelessly outmatched in this bout with the machine?

Take for instance the calculation involved in the design of photographic lenses. Before the arrival of computers, one could design lenses by painstaking pencil and paper work. By this method an experienced lens designer took about six years to design one of the complicated lenses. The desk calculator cut this time to about fifteen weeks, and now a giant computer like the Bureau of Standards SEEAC does the job in a single hour.


LET US look at this lens design problem a little closer to learn something about the way such a machine operates. A good optical lens like those used in the best cameras differs from simple lenses or from eyeglasses mainly in that it consists of many glasses of various shapes all cemented together. The designers must prescribe the exact shape of each of these glasses making up a lens so that all rays originating from a point, say from a star we want to photograph, will meet in another point behind the lens, forming an image of the star. Actually, these rays cannot be made to meet in a point, which would be ideal, but will all fall within a circle. The smaller the circle, the better defined the image and the better the lens.

The design procedure is part calculation, part trial and error. There are, of course, an infinite number of angles at which the rays may enter the lens. A good many of these must be traced through the lens. That is, we must find the change in angle for each ray as it enters and leaves each glass. As a result, we know the angle of the ray when it leaves the last glass surface and therefore where it will meet the other rays. Though the arithmetical procedure to find these changes of angles for each ray is not complicated, it requires accuracies to about seven decimals, and many rays have to be considered. After all the required rays are traced we find the diameter of the circle within which they meet. If we find it small enough to suit our requirements, the job is done. But if it appears too large, we must change by a slight amount one of the shapes of the glass surfaces. Now we trace all the rays for the new condition and see whether we have improved the design or made it worse. It used to take a man six years to complete such a job.

How much of this work can the computer take over for us? Almost all of it. It requires only an adequate set of instructions. These must contain a formula which shows what the angle of a ray is when leaving a surface if we know the angle of entrance, some properties of the glass, the shape of the surface, and the color of the ray. In addition, the instructions tell the computer how good a lens it must design and what initial shapes to start with.

Next, we tell the machine how to proceed. Our program may read: “Start with a ray 45° off to the center axis. Figure its entrance and exit angles through each of t he eleven surfaces. Note the angle of exit from the last surface; do the same with the ray at 44°, then 43°, and so forth, in intervals of one degree until the rav at 0° has been traced. Compare the resulting circle where the rays meet with the desired one; if it is the same size or smaller, print out the answer; if it. is larger, change the shape of the first surface and repeat the ray tracing. If the new answer is better than the old one but still not right, change the surface again in the same direction. If the new answer is worse, change in the opposite direction. When the best answer is still not right, change the second surface the same way, and so through all other surfaces until the answer is right.”

The actual instructions to the computer appear not in words but in a mathematical shorthand written on magnetic tape or in the form of punched holes in a paper tape very much like the good old player piano roll — quite a remarkable device in days when no one thought of computers. It could memorize long piano pieces, know which notes to play, when and how loud, and yet no one worried about its being a thinking machine.

Now that the machine has received its instructions, it can go to work. Strangely enough, it performs in an eerie silence. There are no motors whirring, no bells clanging, not even a hum as it races through millions of trial-and-error calculations with a speed that is literally close to that of lightning. Only the even red glow of the tubes shows that anything is going on. When the computer is finished, there is the clacking of an electric typewriter printing out the solution.

If anything goes wrong, the machine stops and types out what is the matter. Often it can tell which of its many tubes has failed or what additional information it requires to complete the problem. Most computers are designed so that, they never give wrong answers; if something fails, the machine gives no answer. Once an answer is printed, you can depend upon it. Moreover, computers constantly cheek their work and will repeat any calculation that appears incorrect.

Can we call such a process thinking? We have seen that it involves remembering, sorting, classifying, and choosing alternatives on the basis of logic. When men do this sort of work, it has always been considered thinking. And so in fairness to the machine we must concede that within the usual meaning of the word it can and does think. And since in the course of its work the machine discards solutions in favor of better ones, acting on past experience, it cannot be denied that it also learns. Since it thinks fast, it learns fast — much faster than man. Moreover, it makes no mistakes and while working on a problem never forgets. Does this mean that the machine is more intelligent than man?

To state it generally, today’s thinking machines are in their element and truly superior to men when they draw conclusions about particular cases to which a general rule applies. There are computers in development that can make quick and accurate strategic decisions in air battles, taking into consideration the positions of the friendly and enemy aircraft — provided they are given a basic tactical rule they can follow. And by the same token, there is no reason why tomorrow’s computer could not predict the sales volume for an article corrected for season, weather, the general state of prosperity, Mr. Dior’s dictates, and the prevailing feminine mood, as long as it has past sales trends that it can use as a rule.

But the unquestioned obedience to the initial rules which makes for the machine’s superhuman precision also sets a limit to its general intelligence. For the results of its thinking can only he as good as the rules that it has been taught to follow. If the rules showed themselves to be totally wrong for the situation, the machine would cling to them stubbornly, threatening, like the broom of the sorcerer’s apprentice, destruction for its master and itself.


THERE is another kind of thinking — the thinking that sees relations between individual events and forms rules on this basis, and, having formed them, discards or modifies what no longer fits. Men do this kind of thinking so effortlessly that we often do not even consider it thought. If we see a circle, for instance, we immediately recognize it as such regardless of its material or its size. We need not examine each point on the circle separately and compare it with a formula. Moreover, we can tell things that are approximately circular without much strain. Machines cannot sense shapes that are not given point by point or as a mathematical formula.

It is this form of thinking that we use when we recognize someone on the street. We do not, computer fashion, check a lot of details: “5 feet 7 inches tall, size 32 blouse, brown eyes, blond hair, arm length 33 inches, finger lengths 3 inches, 4 inches,” and so forth; we can say immediately, “Hello, Mary.” It matters little whether Mary has lost or gained weight, has grown taller or dyed her hair.

In fact, we need no precise quantitative information about her at all. On a purely statistical basis, the amount of information required to distinguish her definitely among the 75 million females living in this country would be formidable. Yet we need to know astonishingly little to be quite certain that this is Mary. We may recognize her on a cold winter day though she is covered with bulky clothes from head to foot and nothing shows but the tip of a red nose — or we might recognize her from the rear without even this meager clue. People can recognize one anolhor at unexpected meetings after twenty years, when they have last seen one another in grammar school and when they have grown, acquired beards or figures, changed their voices and their clothing — when, in fact, not a particle of their bodies is the same.

Despite the nearly miraculous feat involved in recognition, it requires no outstanding mental ability. Children and even pets are quite good at it. Yet such an activity exceeds the capabilities of the most complex thinking machines. It depends entirely on forming a general picture, an idea — something more than a simple checking off, or adding, or averaging of all the individual parts.

How we form such ideas or generalizations has always been considered one of the most puzzling aspects of the human mind. The ancient Greeks and particularly Plato saw it closely related to the recognition process. He believed that true reality in the form of ideas was stored in a place visited by man’s soul before birth, and that the earthly realization of part icular objects was a recall of memories acquired during this prenatal experience. Ideas can not only serve in helping us to recognize what we have seen but can be applied to predict the unknown on the basis of similarity. A cab driver in New York told me that he was able to cut his working day to a respectable eight hours while most of his colleagues had to work ten or twelve. Yet he made just as much money and had as many fares as they did. His secret: ho learned to recognize the peculiar characteristics of people making up their minds to take a cab. He could spot such people in a crowd or walking out of a building. Before he let me off, he pointed to a man who was just walking along and said, “He wants a cab.” He pulled up next to him and the man got in as I got out.

Most good salesmen know who can be called by his first name and slapped on the back after a few minutes’ acquaintance and who must always be addressed as ‘Mr.”and treated with formality. Confidence men are very adept at determining what kind of man makes a good “mark,” and they don’t have at their disposal a set of standardized psychological tests. Their occupation is safe from the intrusion of the thinking machine.

All of us form definite first impressions and adjust our behavior accordingly. We feel whether the new acquaintance is friendly, whether he is a threat or harmless, whether he is bright or dull, and how we may best be able to get along with him. We recognize and adjust to behavior just as we recognize a person, not by the busy examination of many detailed facts but by organizing these facts into a new entity.

A similar process is involved when a doctor makes a diagnosis. There are really an infinite number of possible diseases that a doctor may be faced with, and if he had to proceed entirely on serial examination of all the symptoms, most of his patients would die—most likely of old age before he was able to make a single diagnosis; yet the good diagnostician often identifies a disease immediately, and at other times requires only relatively few specific tests to come to a conclusion. His mental picture of the disease is a whole, not a collection of many details, and he can therefore recognize it when he sees something that matches this mental pietlire.

A singular human attribute is not only the formation of ideas but the ability to connect such ideas in a useful fashion. The human memory is a filing system that has a far greater capacity than that of the largest thinking machine built. A mechanical brain that had as many tubes or relays as the human brain has nerve cells (some ten billion) would not fit into the Empire State building, and would require the entire output of Niagara Falls to supply the power and the Niagara River to cool it. Moreover, such a compuler could operate but a fraction of a second at a time before several thousand of its tubes would fail and have to be replaced.

One of the largest of today’s computers, the Eniac, has about 10,000 tubes and has therefore about as many brain cells as a flat worm.

The human brain, with one million times as many cells, is unique not only for its ability to store vast amounts of information in a small storage space and for requiring vanishing amounts of operating power, but also for the speed and ease with which any remembered item can be produced. The human filing system is so flexible that it. can be reshuffled instantly from an infinity of new viewpoints. The most elaborate filing systems or library catalogues are arranged by author, subject, and sometimes date of publication, with cross references belween these files. The human file of ideas, however, classifies each idea in an infinite variety of ways; the word “red” can be connected with “green” or “hot ' or “blush” or “Skelton” or “Communist” or “blood or “herring,” to mention only a few. Computers can refer to their memories only in a systematic fashion well planned and explained beforehand but cannot create new cross indexing for themselves.

Yet connection of ideas forms an important aspect of thinking. Without it, Newton could not have associated the apocryphal apple with the motion of the planets because the cross index, “apple falling — see rate — see square law — see planets’ motion,”had not existed. Nor could Norbert Wiener and Shannon have seen that there is a similarity between the way a message loses intelligibility in transmission and an object loses heat to the surrounding area. Nor could physicists have seen that there are similarities in the ways sound, light, and heat behave, so that picturing them as waves would work for all. Nor could Freud have recognized a connection between accidental slips of the tongue and jokes, dreams, and neuroses.

The sort of thinking that can be called truly creative is such forming and organizing of ideas and the connecting of these ideas into new larger entities. And this is precisely what falls beyond the computer’s scope. With its electronics, memories, logic systems, lightning speeds, accuracy, and infallibility, a computer cannot create an idea or ask a question that could form a basis for a new outlook.

Nor does it seem likely that tomorrow’s computers will do this. The machines of the future may overcome some of the other handicaps, such as their enormous size and power requirements. There are signs that they may even beget, their own kind — but never ideas.

True, a computer could be designed which would randomly and madly connect all sorts of facts and then test them for internal consistencies. It would certainly come up with a million theories. But it would have no criterion for selecting the ones that are meaningful.

For what is meaningful is a function of man’s need to survive and to create a world for himself that he can manage physically and mentally.

Thinking machines, more than any other invention in the history of mankind, can aid this creation of a workable and understandable environment by checking man’s ideas for validity and internal consistency, by saving him millions of trials and errors, and by speeding up immeasurably the acquisition of new facts and knowledge. But it always takes a human to come up with the approach, the generalization, the idea which furnishes the basis for the machine’s lightning checking, applying, and finding of new facts. How such basic ideas are conceived we do not know. Yet only they can be called truly creative thought — a process which must forever remain in the province of the human spirit. The bad dreams of our Utopians will not come true; even the most complex, advanced thinking machines will not replace or dominate this spirit.