In Defense of Bird Brains

BY RICHARD J. HERRNSTEIN

SCIENTIFIC psychology promises to predict and control behavior, but compared to the commonsense psychology we all know, the formal science may well seem superfluous. Are not the farmer, the hunter, and the organ-grinder skilled manipulators of behavior when they teach a mule to drag a plow, a dog to track a deer, or a monkey to pass the hat? And does not each one of us, in our human contacts, constantly predict and control the behavior of others with greater success than failure? We are painfully aware of our failures and turn then to those who claim expertise, but we are curiously unaware of the broader implications of our successes.

The implications are especially clear where people have harnessed animal behavior for specific purposes. The farmer knows how to use the mule’s powerful muscles; the hunter, the dog’s sense of smell; the organ-grinder, the monkey’s ability to delight and entertain. Successes won so long ago are taken for granted, but they prove that behavior, at this level at least, can be precisely controlled. The farmer, hunter, and organ-grinder are behavioral engineers, putting to practical use the scientific laws of behavior implicit, although unstated, in their successes. In psychology, as in some of the older sciences, the specific skills of the engineer have antedated the generalized laws of the scientist. But in the course of time, the behavioral scientist will overtake the behavioral engineer, giving him the wherewithal to undertake new tasks and to do the old ones better.

The conscious application of psychology to the training of animals is a recent phenomenon, occasioned by the Second World War, when the need for expendable and well-trained specialists reached an all-time high; and relative expendability, as medical science has known for hundreds of years, is one of the useful qualities of animals. Of all the practical uses that a science of behavior may ultimately serve, the engineering of animal behavior may well be the least important, but it is the one that is likely to be fully exploited first.

During the Second World War, the New York Times reported that the Russians used trained dogs to destroy German tanks. A dog carrying a charge of dynamite on its back and strongly disposed by previous training to seek out and approach tanks made a lethal antitank weapon that homed in on its prey relentlessly. The dogs were put to work too late in the war for them to be a significant factor. There was also a report of a Swedish scheme in which a sea lion would cut a mine cable with an automatic clipper in order to get a fish out of a sealed tank carried on its back. The sea lions were trained to return to the home base to get the tank replenished and the clipper refueled.

So scanty are the accounts that I have found of these bits of behavioral engineering that I would surely discount them as psychological apocrypha were it not for the even more fantastic, but fully described, venture into animal warfare undertaken by B. F. Skinner, then at the University of Minnesota. In recent years at Harvard, Skinner has promoted something of a revolution in educational technique with his teaching machine, a venture into behavioral engineering at the human level. In 1939, however, he was trying to perfect a method for training birds to steer winged bombs and missiles.

The story has been well documented, by Skinner in a technical journal, and also in the pages of the Atlantic in 1947 by a physicist named Louis Ridenour, who undoubtedly did not know that he was revealing then classified information. The heart of Skinner’s scheme was to put into the nose of a missile a pigeon trained to peck directly at the image of a target on a screen in front of it. As long as the pigeon pecked accurately on target, the missile would be steered toward its destination by a pneumatic or electronic system that translated the location of the pigeon’s pecks into guidance instructions to the wings and rudders of the missile. The scene that the pigeon viewed was whatever the missile was headed toward, and the target could be virtually anything —a battleship, a particular intersection of streets in a city, a spot along a beach. Skinner assumed that in actual practice the military would find it useful to train different groups of pigeons to specialize in particular classes of targets.

Skinner’s scheme worked. That is to say, he showed that pigeons can be taught to peck at targets with sufficient accuracy and persistence to guide a simulated missile to a simulated destination. He used a variety of targets, specified by the military authorities, and found them ail within a pigeon’s conceptual grasp. The military did not, however, put the pigeons to work, perhaps because the era of pinpoint bombing seemed about to end with the atom bomb.

V\^UAT was the end for the military, however, proved to be the beginning for a number of psychologists. The practical demands of Skinner’s plan, particularly the necessity for behavior that was reliable, accurate, persistent, and rapid, raised issues of fundamental importance. In a limited sort of way, the behavior of the pigeon had to be as predictable and controllable as the aerodynamics that governed the flight of the missile or the chemistry that was to be exploited in the detonation of the charge. This was something new in psychology: to have to produce behavior the way physicists produce flight and chemists produce explosives. Such practical demands exercised a healthy discipline on Skinner and his associates, and later, on many others who learned of their work. Psychology had taken another step in its evolution into a natural science.

Marian and Keller Breland, who were among Skinner’s associates, decided to pursue the practical rather than the purely scientific possibilities of the work that had been done. The war was over. The inherent comedy of a trained animal was now a better field for exploitation than its expendability. The Brelands went into show business, with trained animals as their specialty. At first they promoted the products of General Mills, Incorporated, at county fairs, with chickens that would play a fivenote tune on a toy piano, or pigs that would turn on a radio, eat at a table, and operate a vacuum cleaner. Fhese days, they sell trained-animal acts in which the audience rather than the sponsor pays the bills. An impressively varied menagerie has been put to work, including —besides chickens and pigs — lambs, turkeys, rats, ducks, porpoises, crows, rabbits, and of course, pigeons.

The similarity between getting a pigeon to steer a missile and getting a pig to use a vacuum cleaner is much greater than might be imagined. Both processes involve a single set of scientific principles, of which the central one is that an animal tends to repeat any act that has been rewarded in its past. The importance of reward has long been used by those who train animals for any purpose, but psychological research has sharpened our understanding of it. We now know that even so small a delay as one second between the act and its reward interferes with training. We know that the animal must be in a state of considerable need for the reward, ordinarily a small quantity of food or water, for a contented animal is unfeelingly indifferent to the machinations of even the most skillful trainer. To get a bit of food or water to an animal executing a complex series of movements may often pose a serious logistic problem, especially if the food or water is to be delivered at just the right moment. Psychologists have, however, found that the problem is readily solved if, instead of the reward itself, the animal is given a token of the reward — a flash of light or a burst of sound that is consistently paired with the reward and thus signals to the animal that its reward is now available. Training becomes vastly more efficient with these tokens, for they can be delivered remotely.

A little more of the lore of the scientific laboratory is involved in the practical training of animals. Psychologists have discovered that behavior, once learned, can be sustained even when the reward is given only occasionally when the behavior occurs. For example, a pigeon taught to peck at a switch for food will continue to peck even if it is successful only a small fraction of the time. Its successes may come at prearranged intervals — every five minutes, for instance — or whenever it fulfills a prearranged requirement of work — for example, every fiftieth time it pecks. If the pigeon gets its wages in accordance with the passage of time, it is likely to work slowly but be tolerant of very low wages. If, on the other hand, it is paid for the work it accomplishes, it will tend to work rapidly as long as the rate of pay is sufficiently high, but cease working entirely if the rate is too low. If pay comes at regular intervals, the pigeon’s work tends to be cyclic, slackening off after each payment and hastening as the next one approaches. If pay comes irregularly, the work is steady. This sensitivity to the schedule of reward is not peculiar to pigeons, for it is shared with animals above and below on the phyletie scale. How animals adjust so subtly to conditions of payment is well beyond the scope of this article.

From a practical viewpoint, schedules of reward are useful in retarding satiation, and also because a given pattern of intermittency in reward produces a related and dependable pattern of repetition of the behavior. The trainer can get his animal to make a movement quickly or slowly, at a steady rate or a fluctuating one, simply by picking the appropriate schedule of reward. Ordinarily, a learned movement stops when it is no longer rewarded, but certain schedules of reward produce such persistence that the movement will occur literally thousands of times in the absence of any reward at all. Men thousand unrewarded pecks by a pigeon have become an experimental commonplace, and in one study a pigeon pecked unrewarded more than 200,000 times before it stopped. Chimpanzees have recently been shown to be equally persistent under comparable conditions.

Finally, the scientilic laboratory has found that a learned movement can be brought under the control of anything the animal can perceive. If, for example, the animal is rewarded when a bell is sounded, but not in silence, it learns to execute the movement only when the bell is sounded. In this way, the animal’s behavior can be triggered remotely, automatically, and reliably by anything the trainer chooses, just as long as the animal’s sensory machinery is capable of responding to it. By taking advantage of this feature of behavior, psychologists are able to answer long-unanswered questions about what an animal perceives.

An example of such a question with a venerable tradition of unresolved debate is whether any bird can smell. J. J. Audubon, the great nineteenthcentury ornithologist, in opposition to folklore and legend, concluded that birds of prey find their food exclusively by vision and are insensitive to the accompanying odors. He showed that buzzards and vultures would tear to shreds a dried, odorless deerskin laid out in a held but would ignore a reeking carcass nearby if it were hidden from sight. The experiment was repeated by many others, including Charles Darwin, always with the same results. Yet there was persistent anecdotal evidence that birds could smell. It was, for example, widely known that pheasant and partridge virtually never approached hidden hunters except from the windward side, as if they were being warned by the hunter’s odor.

After more than a century of debate, the question seems finally to have been answered a few years ago by W. J. Michelsen, who was at the time an undergraduate at Harvard. He succeeded in training pigeons to peck at one switch or another depending upon whether or not a stream of air contained an odorant, showing that there is at least one kind of bird that can respond to an odor. There is a rapidly growing number of studies like this, in which long-debated scientific questions are being answered by the exploitation of a characteristic of the learning process, but exploitation in a practical sense has hardly begun.

EVEN disregarding special sensory powers such as the olfaction of dogs or the vision of birds, trained animals could play a useful role in industry. Wherever human being are paid a wage for the use of their sense organs rather than for their intelligence or judgment, it is likely that they could be replaced by animals, economically and easily. In fact, there have recently been a couple of demonstrations of animals effectively replacingpeople. In one, pigeons were taught to look at small electrical devices called diodes and report whether the paint cover on them was satisfactory and not chipped, scratched, nonuniform, or absent.

The company, which undoubtedly prefers anonymity in this instance, had built a machine to assemble the diodes and found that its machine, unlike the people who had previously done the assembly, had no inclination to reject defective parts. Automatic inspection seemed prohibitively expensive, but the company, either through compassion over the tedium of the task or dismay over the high cost of human labor, was reluctant to put people back to work at the job. A psychologist at Columbia University, W. W. Cumming, heard about the problem and offered to solve it with pigeons, which come at about $1.50 per eyeball and are easy to train if one knows how. Before too long, Cumming had pigeons trained to inspect diodes at the rate of about 1000 per hour, to report defective parts with an error rate of 2 percent, which is slightly better than the company had ever gotten from its human inspectors, and to work for four-hour periods showing no signs of fatigue, unlike human inspectors, who get progressively more lax in their standards as a workday passes. But confronted with the weird prospect of animals inspecting their precision products, the company backed down from the whole idea, nervously predicting the reactions of the labor unions, the ASPCA, and the general public. The irresistible tendency to laugh at a trained animal, which the Brelands have so fruitfully put to work, is apparently not what a large electrical company wants to awaken in the public.

The other demonstration had a similar history of success and failure. T. Verhave, a psychologist working for a large pharmaceutical company, got the idea of using pigeons to inspect empty capsules, a task then done by teams of several dozen women. A capsule can be defective by being broken, incomplete, torn, poorly tinted, out of size, or double. Verhave simplified the problem for his pigeons by choosing only the last fault, the double capsule. Since he was merely trying to convince his employers that pigeons can be thus employed, and since the company had recently wasted a fair amount of money on an unsuccessful automatic system to detect double capsules, his strategy was reasonable. With just one week of practice, there were two pigeons whose reports of double capsules were 99.9 percent correct. Again, company management, at the threshold of a miniature industrial revolution, backed down because of various fears, in particular the hazard that would attend incorporating trained pigeons into the company image.

More recently, some associates at the General Atronics Corporation and I have been training pigeons for the military, to be used for purposes that are at present under security restrictions. Our preliminary work, however, may be reported, and it suggests still more ambitious uses of animals than have so far been attempted. We have taught our pigeons to look at projections of 35 millimeter colorslides and to report whether or not a slide contains any human beings. We have found that this job is easily within the capacity of the bird, who gets as good at the game as the experimenters themselves within a month of practice, even with pictures it has never seen before. The pictures come from all parts of the world, in every imaginable sort of setting, with all the races of man, woman, and child represented. Some pictures contain people in the great distance or people severely obscured by intervening objects. More than once we have found that we had misclassified a picture ourselves, either failing to see a hidden person or seeing one where there was none, only to be corrected by our pigeons.

With this work we have shown that a pigeon, and presumably other animals, can be taught to employ a concept that is defined at a higher level of abstraction than simple geometrical form. What human beings call an image of a human being cannot be reduced to a mere set of geometrical rules, for a person is a person whether he is large or small, standing, sitting, or doubled over with his head between his legs, facing us or standing on his head looking away from us, or seen from the waist down or the neck up. The older work on the perception of animals concentrated on sensory capacity, such as whether pigeons can smell. This new work suggests a direction that might be characterized as conceptual capacity. One can think of human concepts as arranged in a hierarchy, going from those which are defined in terms of fixed physical properties, such as “triangle,” to those whose definition we might well despair of ever stating, such as “justice.” Our pigeons have established their claim to a level in the hierarchy somewhat above the base, and have proved themselves able to replace men in tasks we might have considered beyond their grasp. How high in the hierarchy they or other animals can go has yet to be found.

It is not at all certain that the scientific methods of training will have any real importance to the traditional applications of trained animals. In farming, the trained animal is more likely to be replaced by farm machines than by more skillfully trained animals. In hunting, the old methods of training have a ritualistic value that will make them hard to replace with the scientific methods. In entertainment, improved efficiency may be self-defeating, for it will destroy the novelty of remarkably trained animals and thus the fascination they hold for us.

But the future of behavioral engineering is far from bleak. The commercial and military applications are still awaiting exploitation and seem to be held back only by negative attitudes that oppose the dictates of good business and human welfare. Whatever the temporary disadvantages of filling jobs with trained animals, it seems wrong to keep human beings occupied with tasks that animals can do as well if not better. No one laments the men who were put out of work by the mule, and we have come to regard the tractor as a social good. In time, the real competitor of trained animals in industry and war will not be man, but machine. Machines, however, have a long way to go before they can match some of the sensory and conceptual capacities of animals, even disregarding considerations of cost and size. While we wait for machines to catch up with the outcome of eons of biological evolution, psychology offers some immediate and highly practical potentialities.