The simpler forms of learning both enable dogs to piece the world together, and allow us to train them to behave the way we want them to. But dogs can also think for themselves: they do not just have feelings about the world, but also, in their own way, have knowledge about their physical environment and the other animals around them (including humans, of course). The formal study of canine intelligence dates back to the earliest years of the twentieth century, and the puzzle-boxes of Edward Thorndike. Thorndike’s approach to studying how animals learn was different from Pavlov’s; he was more interested in how they solve problems. Many of his experiments involved placing animals, usually domestic dogs or young cats, into puzzle-boxes of his own devising. These could be opened from the inside when the animal performed some kind of action. For example, in the box illustrated, the door was attached to a weight that would pull it out of the way once the dog had unlatched it. The latch could be one of the wooden toggles on either side of the door, which the dog could paw upright by pushing its foot through the gap between the slats. Or it could be a bolt at the top of the door, connected by a rope to a loop hanging from the roof that the dog could pull on with its teeth. Or it could be a treadle in the middle of the floor, which the dog had to press with its paw in order to release the door (all three are shown in the drawing).
Thorndike’s puzzle-box
Thorndike was interested in finding out how the dogs solved the problem of getting out of the box, and also whether they subsequently remembered how they had done it. At the time, many people believed that animals like cats and dogs were capable of considerable insight – that they could, as it were, sit down and think things out. Thorndike, however, believed that a much simpler explanation would suffice. He considered the possibility that simple associative learning, coupled with the dog’s natural inquisitiveness, might actually explain such apparently intelligent behaviour. He found that his dogs would initially scrabble around the puzzle-box until they blundered on the mechanism that would let them out. He then fed them, as a reward, and put them straight back into the box to see if they could now escape any faster. If the dogs had insight into what they had done, they would have immediately returned to the lever or pedal that had previously let them out. But in fact the dogs rarely did this. However, after repeated sessions in the box, they did take less and less time to escape, and eventually, after several repetitions, did start going immediately to the releasing mechanism.
Largely on the basis of these experiments, Thorndike came up with the concept of trial-and-error learning. Animals faced with a problem to solve will try out a variety of tactics that they would normally use in such a situation (in this case, being trapped). When one of these tactics happens to work, it produces rewards (in this case, being let out and then fed). The next time they are in a similar situation they will be more likely to perform the action that got them out last time, or to focus on the area where they had been when the door had opened on previous occasions. Both are explainable by simple operant conditioning: either repeat whatever action preceded the previous escape, or go to the place where you were immediately before the door opened (or both). This behaviour requires no insight – no problem-solving skills – on the part of the dog. Based on Thorndike’s experiments, and others like them, most scientists now believe that dogs have rather limited powers of reasoning, certainly inferior to those of chimpanzees (and even a few birds).
The point here is not that dogs are stupid, but rather that their brand of intelligence differs from the primate model. Though Thorndike’s dogs showed little evidence of problem-solving skills, they excelled when it came to recalling the correct escape method. In fact, they retained this memory for several months, even without any further exposure to the puzzle-box. (Dogs’ retention of skills they have acquired is generally very good, but they find it much harder to remember for more than a few seconds things they have merely observed.) Memories of events, as opposed to memories of their own actions, may not be of great value to canids; indeed, they may be confusing. Canids, especially foxes, are known to be capable of remembering where they buried food, for days or even weeks afterwards. Since the ability to retrieve food in this way is clearly adaptive, it is believed that they evolved the ability to recall where they buried the food. Conversely, canids are also often faced with the problem of what to do when the prey they are chasing disappears. This is not something that it is useful for them to remember for very long, because prey is unlikely to remain in one place for more than a short time without leaving some other clue. If the prey is not where it vanished from, and its scent is fading fast, then it has probably long gone; better to move on to another hunting site than to hang around hoping that it is stupid enough to return to the very place it had just been chased away from. Thus evolution seems not to have selected for retention of such information for more than a few seconds.
Dogs’ short-term memory has been investigated experimentally. Using a method called visual displacement, scientists have tested how long dogs can remember where something has disappeared from.3 Initially, each dog is taught that its favourite toy is hidden behind one of four identical boxes: it first watches the experimenter hiding the toy, and is then allowed to retrieve it. Once it reliably goes to just the box it has seen the toy disappear behind, and is no longer searching the boxes at random, the dog can be tested to see how long it remembers where the toy has been hidden. In this second phase of the experiment, a screen is placed between the dog and the boxes immediately after the toy has been hidden, so that the dog has to remember which box its toy is behind. Then when the screen is removed, it has to recall that memory in order to locate the correct box. If it cannot, it will go back to searching the boxes at random. If the screen is kept in place for only a few seconds, most dogs will go straight to the correct box, showing that they are indeed capable of both memorizing and recalling the location, provided the interruption is only brief. However, just a thirty-second delay is enough to induce mistakes. (Although dogs make even more mistakes after a minute’s delay, they do not get significantly worse if the screen is left in place for four minutes, at which point many are still performing better than chance.) Subsequent experiments have shown that dogs are better at remembering where things have disappeared from in relation to their own positions (‘to my left’) than in relation to landmarks (‘under the box that has boxes on either side’). Overall, many dogs’ short-term memories of single events appear to be rather fallible, perhaps because they are much more interested in working out what people want them to do here and now than in recalling precisely what happened a few minutes ago. That is not to say, however, that they pay no attention to the more fixed features of their surroundings, otherwise they (or their evolutionary forebears) would quickly get lost.
Although most pet dogs do not have to memorize the features of the environment where they live (because most of the time we humans decide where they can or cannot go), they nevertheless retain their wild ancestors’ ability to find their way around, and can use it if they need to. Indeed, dogs have very good memories for places, as might be expected from the descendants of animals that roamed widely in search of food. They have a variety of cognitive methods at their disposal for this purpose, such as the ability to memorize landmarks, but also more complex skills such as constructing mental maps of how those landmarks are distributed. Landmark-learning does not require a complex brain – it is how many insects find their way around – but it is useful nonetheless and, indeed, an essential part of more complex skills. Just as humans do, dogs effortlessly and continuously memorize the features of their surroundings; unlike us, however, they rely heavily on what things smell like. We might recall turning left around a dark green shrub; a dog would remember that shrub as smelling of orange with grassy top-notes. Yet despite these differences, dogs continually store (and then, presumably, gradually forget) information about features of the environment that they have recently encountered.
The dog’s propensity for memorizing landmarks can
actually impede training. Younger dogs are so good at learning locations that they often spontaneously memorize their surroundings as part of the set of cues that tells them to do something. For example, puppies taught the verbal command ‘sit’ in a training class may appear to have forgotten it as soon as they get home – because, in addition to the command, they have spontaneously memorized as the relevant cue some feature of the room where the class was held and, in different surroundings, do not recognize the command. Many dog trainers therefore repeat a training exercise in a variety of places, in order to break such associations and isolate the intended cue, in this case, the verbal command alone.
Dogs, as the descendants of hunters that roamed far and wide in search of prey, ought to have more refined navigational abilities than simple landmark-learning, and, indeed, it has been shown that they simultaneously construct maps inside their heads. The standard way of investigating mental maps is to see whether animals can work out short-cuts for themselves. In one experiment, animal psychologists examined the mapping abilities of half-grown German shepherd dogs by showing them two caches of food hidden in the undergrowth in a large overgrown field (they chose young dogs because older ones might already have learned something similar to the task they were about to test them with, and they wanted to investigate the dog’s natural abilities).4 Starting at point C on the diagram shown here, one of the experimenters walked the dog to the first cache of food, at A, walked it back to C, and then walked it out and back to the second cache at B. Each dog, still on-leash, was then allowed to take the experimenters wherever it wanted to go. If the dogs had been using landmarks to find the food, they should have retraced the paths they had already been taken on. But in fact they invariably took a short-cut. Often this was a direct path from A to B, but not always; sometimes a dog went to B first and then used the short-cut in reverse to get back to A. This suggests that some dogs, perhaps especially young ones, may spontaneously look for new solutions to a problem once they are comfortable that their first way of solving it works – and can then go back to that solution once they are sure that the new method is not an improvement on the original one.
Bird’s-eye view of an experiment demonstrating dogs’ abilities to take short-cuts. (Left) The dog was first led from C to A, where it was allowed to find hidden food, then back to C, then to and from B, where it also found food. Lines of sight between A, B and C were blocked by vegetation. (Right) A typical track taken by the dog after release from C.
The fact that some dogs seem to be able to use mental maps more flexibly than others suggests that this ability approaches one limit of the dog’s cognitive abilities; specifically, older dogs, or dogs that have been under stress for a long time, seem to lose some of their ability to orientate themselves. In one experiment, my colleague Elly Hiby investigated spatial ability by allowing gundogs to search a square grid of sixteen buckets, placed four feet apart. A few of the buckets always contained food, which the dogs were allowed to eat once they had found it, but most buckets only smelled of food (in this way, she prevented the dogs from simply finding the buckets with food by their smell).5 Once they had had one opportunity to search the buckets, it was possible for the dogs to make two different kinds of mistakes: on subsequent searches, they might either visit buckets where there had never been food, or revisit buckets that they should have known they had already emptied. In the second phase of the experiment, she made the task harder, by releasing the dogs from the corner of the grid opposite to the one they were used to starting from; to succeed, they would then have to effectively turn their mental maps upside down in order to know which buckets contained food. The younger dogs learned the task quickly, and made few mistakes, even when released from the ‘wrong’ corner. By contrast, older dogs, and those whose hormones suggested they had been stressed for a long time, made the most mistakes; they were especially confused when the release point was changed, suggesting that some part of their spatial memory had become impaired.
As we would expect from their evolutionary past as wide-ranging hunters, dogs appear to memorize their surroundings continuously and effortlessly, and also to cross-reference different memories to construct mental ‘maps’ that enable them to navigate efficiently. However, they are less skilled than we are at re-orienting themselves when viewing familiar landmarks from an unexpected direction; the ‘maps’ themselves are probably accurate enough, it is the ability to think about the maps that they appear to lack.
When they are finding their way around, dogs probably use their acute sense of smell in preference to relying on what things look like, as we do. Their memories, too, are probably based as much on odour as on visual appearance, or even more so: dogs can remember a particular odour – say, that of a previous owner or a dog they have lived with before – for many years, possibly as many as ten.
Even scientists sometimes overlook the dog’s preference for focusing on smell, and consequently think that they have demonstrated complex abilities in dogs that are in fact more probably merely evidence of how well dogs detect odours. In one experiment, dogs were outfitted in blindfolds and ear defenders and taken on a short walk. They were nonetheless able to retrace their footsteps. The scientists took this to mean that the dogs had memorized each turn, left or right, and how many steps they had taken in between, much as you or I might under these circumstances. But crucially the scientists who performed this research did not allow for the dog’s acute sense of smell. In short, they failed to account for the possibility that the dogs could have retraced their steps either by following the odour of their own footprints, or by using olfactory ‘landmarks’ that they had memorized while their vision and hearing were blocked.
Another experiment that illustrates the dog’s ability to pick out subtle differences in smell that mean little or nothing to us – though again not designed for this purpose – involved a border collie called Rico, living in an apartment in Leipzig, who had been trained to retrieve his toys based on their ‘names’. As far as the dog was concerned, these were probably not literally the names of the toys but just commands, one for each toy (thus in his mind ‘sock’ was not an abstract label for a sock, but actually meant ‘Fetch your sock’). The experimenters laid out some of Rico’s own toys, adding one that they had brought with them, different from any of those in the apartment. When Rico’s owner then called out a word that Rico had never heard before, the dog often retrieved the novel object. Although the experimenters claimed that this was evidence for linguistic skills in this dog,6 and therefore by extrapolation in dogs in general, a simpler explanation is that Rico retrieved the novel objects because he found them fascinating, since they smelled different to everything else in the apartment (having never been handled by his owner). In other words, he was able to categorize toys as ‘mine’ and ‘not mine’, an interesting cognitive ability in itself, but apart from that all his behaviour was explainable by simple associative learning.7
In sum, dogs find their way around by a combination of abilities that overlap with, but are distinct from, our own. They have a good memory for locations, and a capacity for integrating these memories into ‘maps’ that they carry in their heads, so their intuitive skill at finding their way around is probably rather similar to ours. As in our species, old age and chronic stress impair these functions, eventually to the point that the dog may appear confused when it loses its usual terms of reference. However, the features on dogs’ cognitive ‘maps’ are at least as likely to be olfactory as visual, whereas the representations of the environment that we carry in our heads are almost entirely visual.
Dogs’ ability to construct mental maps suggests that they understand how things are connected together in the physical world, in the same way that we do, but, when tested experimentally, this hypothesis has been found not to be true. Their intuitive understanding of the ways in which objects connect together, their ‘folk physics’, is quite different from ours; they remember connections between actions and consequences, but without necessari
ly understanding how those consequences come about. One of the standard ways that psychologists use to test the ability to comprehend physical connections is the means–end test. For dogs, this involves animals retrieving inaccessible food by pulling on a string. In the simplest form of the test, one piece of food is attached by a single string to a wooden block. The food is made detectable but inaccessible (for example, under a mesh cover) whereas the block is left accessible. Most dogs can learn by trial and error that pulling on the block results in the food at the other end of the string coming out from under the cover, and can then repeat this whenever they are asked. A casual observer would conclude that the dog had understood the reason for what it had achieved; specifically, that the food was connected to the block by the string. However, if the task is made a little less straightforward, dogs are soon flummoxed. If there are two strings that cross one another, and only one has food on the end, then the dogs should, if they understood the physics, choose the one connected to the food. But they do not. Some pull on the block nearest to the food. Others just give up and try to dig the food out from underneath the wire mesh (some find even one string a problem, if it goes under the mesh at an angle).8 The implication is that when dogs do learn to get the food, they do so through straightforward operant learning, and not through understanding that the food and the string are physically connected together. What they learn seems to be simply this: pulling on a wooden block near the smell of food produces food. Other ‘intelligent’ animals, such as monkeys, parrots and crows, perform much better at this task, but does this mean that dogs are stupid? More likely, the experiment in question is just not a suitable test of their intelligence. The canids’ hunting lifestyle does not require a detailed understanding of precisely how things work, unlike the more opportunistic foraging strategies of monkeys, parrots and crows.
In Defence of Dogs Page 23