Abstract
Experiments involving dog-human social cognition have shown dogs’ close attention to human behavioral cues. Dogs remained near a fallen owner, avoided a deceptive human, and preferred a human that provided valid information about the location of a reward over an uninformative human. On the other hand, dogs showed no evidence of going for help in an emergency, having theory of mind or metacognition, or performing successive numerical discrimination. When tested for spatial memory and simultaneous numerical discrimination, dogs showed evidence of these abilities but also showed lower performance levels than found in other species.
The domestic dog has had an unusual history in psychological research. Two historical giants—Darwin and Pavlov—took great interest in the domestic dog. Darwin’s observations, along with those of his one-time research assistant Romanes, were almost exclusively anecdotal. Pavlov’s discovery of classical conditioning, on the other hand, was a fortunate observation made while doing unrelated work on the digestive system. Although dogs were periodically used as subjects in psychological research throughout the 20th century (Feuerbacher & Wynne, 2011), they were largely considered an “artificial” species because of their domestication (Miklósi, 2014) and were less often used than rats or pigeons. The past two decades, however, have seen the rise of the domestic dog as a new darling in animal cognition research.
Why has intense interest recently been focused on dog cognition? Much of this interest seems to center around the extended-domestication hypothesis. Dogs descended from gray wolves (Pollinger et al., 2010) at least 10,000 years ago. During much of this period, they were in the company of humans, who selectively bred them for work, protection, and companionship. Given this unique relationship with humans, dogs acquired many human-like traits, including communicative, social, cooperative, and attachment behaviors. Much of the recent research with dogs has explored the possibility that dogs might also have some of the cognitive abilities found in humans. Thus, it is not surprising that the bulk of dog research has predominantly concerned topics in social cognition, often dogs’ understanding of human behavior.
Do Dogs Help in an Emergency?
Indeed, our first venture into the world of dog cognition fits into this category. In an early anecdote, Romanes (1882) reported the case of a hunter who fell through the ice on a frozen river. His dog
made many fruitless efforts to save his master, and then ran to a neighboring village, where he saw a man, and with the most significant gestures pulled him by the coat and prevailed on him to follow. The man arrived on the spot in time to save the gentleman’s life. (p. 447)
It is not uncommon to hear even today anecdotes of canine heroes—dogs who have alerted their owners of impending danger or who have sought help for an owner in distress. Are these instances of coincidence or expanded accounts of dog behavior with a simpler explanation, or do dogs actually understand the nature of an emergency and take appropriate action? We carried out two controlled experiments to examine dogs’ response to a human emergency.
An important aspect of our experiments (Macpherson & Roberts, 2006) was that we borrowed the bystander-apathy paradigm (Darley & Latané, 1968) from human social psychology research. Bystander-apathy experiments have shown that although people recognize an emergency situation when another person is in danger, they are less likely to help when there are many bystanders available to help than when there are few (Fischer et al., 2011). This is explained as a diffusion of responsibility: The more bystanders that are available, the less responsibility any one individual feels to act. We used conditions in which one or two bystanders were available; the bystander(s) did not respond to the emergency, and the question was whether the dog would go to a bystander for help.
In the first of our dog emergency scenarios, an owner walked into a field with her dog, as if on a normal walk. Once she reached the center of the field (indicated by a target painted on the ground), the owner clutched her chest and feigned symptoms of a heart attack. The owner then collapsed to the ground and remained motionless for 6 minutes. One or two bystanders seated nearby could be gone to for help (see Fig. 1). At the other end of the field, a cameraman was hidden behind a tree, recording the dog’s behavior.
A dog owner lies motionless after feigning a heart attack in Experiment 1 of Macpherson and Roberts (2006). One or two bystanders remained nearby throughout the trial.
In reviewing video footage from the Macpherson and Roberts (2006) experiment, we measured how long dogs engaged in certain behaviors (e.g., time spent in proximity to their owner, time spent in proximity to a bystander) as well as the frequency of certain behaviors (e.g., how often they touched the owner or bystander, how often they barked). Contrary to popular belief, we found no evidence that the dogs were attempting to seek help for their owner: They spent most of their time in proximity to the owner, rather than the bystander. Only in one instance did a dog touch the bystander—this was a toy poodle that jumped in the bystander’s lap and appeared to be seeking comfort or attention for itself. There were no significant differences between dogs in the one-bystander versus two-bystander conditions.
If a dog truly had insight into an emergency scenario, the dog might assume that it did not need to alert the bystanders, as they could plainly see the endangered owner for themselves. Alternatively, it is possible that the dog simply thought the owner was sleeping or playing a game with it. In light of these limitations, we designed a second experiment (Macpherson & Roberts, 2006, Experiment 2) in which a more explicit emergency was created. This experiment took place indoors, and in two separate rooms. When the dogs entered the building with their owner, they went into the first room, where the owner met and shook hands with a bystander. The dogs then proceeded to the second room with their owner, now aware of the availability of the bystander, who could not see any of the activity in the main room. Once in the main room, dogs in the experimental group watched as a bookcase toppled onto their owner, leaving him or her trapped beneath. For 6 minutes, the owner explicitly appealed to the dog for help. In a control group, the owner simply stood next to the bookcase for 6 minutes looking at books, and no emergency occurred. In both conditions, the dogs’ behaviors were filmed by an overhead camera.
Even with an owner explicitly appealing for help, we found that in no case did a dog ever enter the adjoining room to seek the help of the bystander. One pronounced difference was that control dogs spent much more time roaming or exploring their surroundings, whereas dogs in the experimental group, like those in Experiment 1, tended to remain in close proximity to their owner. Dogs are social animals, so it seems to us reasonable that they would prefer to stay by the side of their master. However, both of our experiments suggested that dogs had no insight into the nature of the emergency and thus did not seek help for their owners. This is not to say that dogs are not capable of “heroic” acts—this is seen every day with police dogs and various types of service dogs. These heroic behaviors, however, appear to be the product of training rather than understanding emergency situations. Interestingly, even young children of kindergarten through second-grade age, with no prior training in responding to an emergency, readily respond to another child in distress (Staub, 1970).
A major limitation of these experiments remains the fact that we could not mimic any type of olfactory cue that might be available to a dog when a human is experiencing a legitimate emergency. For example, a human in imminent danger is likely to produce increased levels of stress-related hormones such as adrenaline. It is therefore possible that dogs may “smell” danger in legitimate emergency situations. At the time of the publication of this study, a limited number of studies were available to support the idea that dogs may detect cancer in humans through olfaction (Pickel, Manucy, Walker, Hall, & Walker, 2004; Willis et al., 2004). Since this time, however, further studies have provided evidence that dogs can detect seizures and hypoglycemia (Wells, 2007) in addition to cancer (Cornu, Cancel-Tassin, Ondet, Girardet, & Cussenot, 2011). Our study was an exploratory study, and while it was not able to address limitations related to olfactory cues of distress, our research in this area may serve as a baseline for future studies investigating the conditions under which dogs may or may not go for help.
Can Dogs Detect Human Deception?
In a series of experiments designed to study dogs’ understanding of human intentions, Petter, Musolino, Roberts, and Cole (2009) asked if dogs could detect human deception. Dogs were allowed to choose between two opaque containers, one containing food and the other empty. The containers were 3 meters apart, and the left-right position of the baited container changed randomly among trials. Two people, previously unknown to the dog, alternately stood behind one container and verbally encouraged the dog to approach that container. One person served as the “cooperator.” This person always stood behind the baited container, and thus approaching this person always led to a food reward. The other person served as a “deceiver.” This person always stood behind the empty container. When the dog approached and examined the empty container, the deceiver went to the other container, took the food, and pretended to eat it. Our interest was in how dogs would respond to deception over 200 trials given over five sessions. We found that dogs continuously approached the container in front of the cooperator. On deceiver trials, however, dogs began to avoid the deceiver. On trials when they did so, they often went to the alternative container containing the food reward.
One explanation of these findings is that dogs have theory of mind. That is, they impute positive and negative intentions to humans cuing them. In a final experiment, Petter et al. (2009) used inanimate cues (black and white boxes) as the “cooperator” and “deceiver.” The results were the same as those from the experiment using humans to cue the dogs: The dogs learned to approach the “cooperator” box more often than the “deceiver” box. The findings indicated that dogs were sensitive to the correlation between cues and their outcomes, but offered no support for the idea that dogs understand human intentionality.
Metacognition in Dogs?
Closely related to theory of mind in human cognition is metacognition, or awareness of the contents of one’s own memory. In particular, a metacognitive individual for whom needed information is unavailable will seek out that information from another source. In an initial study, we asked if dogs would seek out information needed to obtain a food reward when that information was not immediately available to them (McMahon, Macpherson, & Roberts, 2010). Dogs were trained to choose among four different boxes, each with a food tray and potential food reward underneath. The boxes were all black, with the exception of one box that had a white side. The box with the white side always had a food reward underneath it, whereas the black boxes did not (see Fig. 2). Initially, the dogs were trained with the boxes facing them front-on, such that the white-sided box was fully visible. The dogs quickly learned to choose the box with the white side. The boxes were then rotated 45° (such that the white side was still easily visible) and 90° (such that the white side was still partially visible). The dogs’ percentage of correct choices progressively declined across these conditions. Finally, the boxes were rotated 135°, making it impossible for the dogs to see the white side from their starting position. If the dogs were aware that they could seek information, then, given their extensive training, they should have realized that they could walk around to the back of the boxes to determine which box had the white side. The dogs failed to do this and did not choose the correct box above chance level. Even when the dogs were given guidance training, in which they were walked behind the boxes to show them that it was simply the orientation that had changed, the dogs still failed to seek information about the white side on test trials.
Setup for boxes used in Experiment 1 of McMahon, Macpherson, and Roberts’s (2010) information-seeking study. Food was located under the box with one white side, which gradually rotated out of view.
Dogs have been selectively bred and trained to pay attention almost exclusively to humans. In a follow-up study, McMahon et al. (2010) attempted to determine whether dogs might seek information in a human-oriented context. In this task, dogs had to choose one of three identical boxes to knock over, only one of which had a food reward hidden beneath it. Prior to selecting a box, the dog first had to choose between two humans—one was an informant who would point to the location of food, and the other was a non-informant who would turn her back to the dog, providing no information. Both humans were females of similar height and dress, and both had no previous association with the dog (see Fig. 3). Dogs rapidly developed a significant preference for the informant. Thus, this study suggested that dogs may seek needed information when they are unaware of the location of a food reward, but only when the source of that information is a human informant.
Schema showing the setup for Experiment 2 of McMahon, Macpherson, and Roberts’s (2010) information-seeking study. Dog had to choose between an informant (who would point to the location of food) and a non-informant (who would not point to the location of food). The gray rectangle represents a large cooler that was used to block the dogs’ view of the boxes while they chose between the humans.
Spatial Memory in Dogs
The interest in social cognition with respect to the domestic dog is intuitive, given our close relationship with this species. There remain, however, comparatively few studies that have examined fundamental aspects of cognition in dogs. The processing of time, space, and number are crucial elements of cognition that help an animal to survive in its day-to-day life. They are also processes that have been studied for decades in rats, pigeons, monkeys, and other “typical” lab species. To have a full understanding of canine cognition, an understanding of these processes in dogs is required. Given that dogs have been selectively bred to attend to humans, it is entirely possible that we have “bred out” cognitive traits that would have been necessary for their wolf ancestors.
Since its inception, the radial maze has become a staple apparatus in the study of spatial memory in rats (Olton & Samuelson, 1976). In order to study spatial memory in dogs, we therefore created a large-scale, collapsible radial maze measuring 4.56 meters in diameter (see Fig. 4). In an initial experiment (Macpherson & Roberts, 2010), we baited each arm of the maze with a piece of food and recorded how many arm visits it took the dogs to deplete the maze of food. An animal with good spatial memory should deplete the maze quickly, without revisiting arms, whereas an animal with poor spatial memory should make more mistakes (repeat trips to arms already visited). We found that although dogs learned to enter all eight arms with progressively fewer arm visits over 24 trials, their performance was still far inferior to that of rats or pigeons on a similar task (Beatty & Shavalia, 1980; Roberts & Van Veldhuizen, 1985).
Large radial maze from Macpherson and Roberts’s (2010) experiment examining spatial memory in dogs.
In a follow-up study, dogs were tested using a two-phase paradigm (Macpherson & Roberts, 2010). In the study phase, dogs visited four randomly selected baited arms, while the other four arms were blocked. In the test phase, all eight arms were available to the dogs. Half of the dogs were placed in the win/stay condition, meaning that they would find food in the same four arms as they did in the study phase. Dogs in the win/shift condition, however, would find food in the four locations they had not visited in the study phase. Studies with rats and pigeons (Beatty & Shavalia, 1980; Roberts & Van Veldhuizen, 1985) have shown that these animals performed better in the win/shift condition. This observation makes sense in terms of optimal-foraging theory, as it is not efficient to return to locations that are known to have been depleted. This paradigm raises an interesting question with regard to dogs: Given that domestic dogs typically eat out of a food bowl at a single location, might they be more efficient using a win/stay strategy? Contrary to this suggestion, we found that dogs, like other species previously tested, adopted a win/shift strategy on the task. In neither condition, however, did they perform much above chance level; once again, their performance was far inferior to that of rats or pigeons on a similar task (Beatty & Shavalia, 1980; Roberts & Van Veldhuizen, 1985).
A possible explanation for our findings is that dogs operate cognitively much like other species but, through selective breeding, have had some abilities weakened that would aid their survival in the wild, in favor of social cognitive traits that allow them to better interact with humans. The idea that dogs essentially show a downgraded version of the cognition of the physical world possessed by their wolf ancestors in favor of superior social cognition was originally put forward by Frank (1980), who found that Alaskan Malamute pups outperformed wolf pups on several human-directed tasks (Frank & Frank, 1983), whereas wolf pups outperformed dog pups on a detour task (Frank & Frank, 1982). More recent research, however, has shown that wolves can outperform dogs in using human social cues (Udell, Dorey, & Wynne, 2008).
Another possibility is that the radial maze is not an ecologically valid approach to the study of spatial memory in dogs. The radial maze was originally created by Olton and Samuelson (1976) to mimic the natural subterranean environment of wild rats—a central chamber below ground, with multiple tunnels extending from that central hub. Because this maze does not at all resemble a dog’s environment, it simply may not be the best apparatus for testing dogs’ spatial memory. It should be noted, however, that in a subsequently published study (Craig et al., 2012) using a radial maze similar in dimensions and construction to our own, superior spatial-memory performance was found in dogs in several tasks (though the performance of dogs in this task was still not on par with what is typically seen in rats). These differences in findings may be the result of subtle differences in environmental or extramaze cues surrounding the respective mazes. For example, whereas our maze was often moved between daily sessions, the Craig et al. maze remained in one location, which may have provided more consistent cues to be utilized by the subjects. Thus, further research will be needed to completely assay dogs’ spatial-memory abilities, as well as the effectiveness of the radial maze as an apparatus in the study of canine spatial memory.
Numerical Discrimination by Dogs
Numerical discrimination is another area of basic cognition that has been widely studied in nonhuman animals. An animal in the wild, for example, needs to be able to make a relative numerosity discrimination between a small number or large number of predators approaching it. Numerical discrimination has been extensively studied in monkeys and chimps by Beran (2001; Beran & Beran, 2004). Typically, Beran’s experiments involved dropping pieces of a preferred food item into each of two bowls, one piece at a time. After observing this, the subject is then allowed to choose one of the bowls and consume its contents. If the subject is sensitive to number, then it should choose the bowl with the greater number of items. Given the simplicity of Beran’s paradigm for the study of sequential counting, we decided that this would be an ideal way to approach the study of numerical discrimination in dogs. We placed two bowls 3.65 meters apart and dropped pieces of food, one at a time, into each of the bowls (Macpherson & Roberts, 2013). The buckets had false bottoms containing additional food in order to control for olfactory cues. With the exception of the 1-versus-0 ratio (a situation in which the animal must choose between receiving one piece of food or no food at all), dogs failed to discriminate the number of food objects above chance level, even when different variations of the task (in which buckets were placed farther apart or closer together, non-food objects were used, and weighted lids forced the dogs to work harder to get to the food) were attempted. Interestingly, in a similar procedure (Utrata, Virányi, & Range, 2012), wolves were able to discriminate number in a sequential task, although their performance did not improve with a decreasing ratio between the smaller (S) and larger (L) items, as would be predicted by Weber’s law.
It has been noted in studies using single-cell recordings (Nieder, Diester, & Tudusciuc, 2006) that although monkeys can discriminate number in sequential-counting tasks, they perform better in simultaneous-counting tasks, in which they can see all of the items to be counted at the same time, presumably because simultaneous counting is less taxing on the working memory system. To study numerical discrimination in dogs using a simultaneous-counting task, we fastened differing numbers of shapes to each of two boxes, using magnet boards (see Fig. 5). If the dog chose the box with the greater number of shapes by pushing it aside, it found a food reward in a container under the box. If it knocked over the box with fewer shapes, no reward was found. Importantly, we varied the size of the shapes and their configuration, so that only number could be used reliably to determine the correct display. In this experiment, dogs were able to discriminate the numbers 1 through 9 above chance level. Importantly, they also demonstrated ratio effects, which are a signature of counting studies in both human and nonhuman animals (see Fig. 6). Thus, as the ratio of the smaller number divided by the larger number (S/L ratio) increased, numerical-discrimination accuracy dropped. That is, dogs could discriminate between 1 versus 2 (S/L ratio = .5) much better than between 8 versus 9 (S/L ratio = .89). The success of dogs in the simultaneous-counting procedure suggests that there may be a limit on working memory that prohibits dogs from keeping track of number in the sequential version of the task. Sensitivity to reinforcement may also be an issue—whereas dogs received reinforcement regardless of which bucket they chose in the sequential task, they received a reward only if they selected correctly in the simultaneous task. Although this made no difference in studies with monkeys, it may have compelled the dogs to work harder in the simultaneous task.
Setup for the simultaneous-counting task in Macpherson and Roberts (2013).
The performance of one dog on the simultaneous-number-discrimination task in Macpherson and Roberts (2013). The graph shows the percentage of correct choices of the larger number as the ratio between the smaller number of objects (S) and the larger number of objects (L) increased. The numbers of objects were 0 versus 3, 1 versus 9, 3 versus 9, 4 versus 8, 2 versus 4, 6 versus 9, 3 versus 4, 6 versus 8, and 8 versus 9.
An important control in our simultaneous-counting procedure was the manipulation of the size of the stimuli used, such that the dogs could not base their decision on the surface area or overall size of objects as opposed to their numerosity. Recently, Petrazzini and Wynne (2016) compared the salience of numerosity of food items to overall amount of food in a numerical task. Dogs were given choices between two quantities of food items in a congruent condition (the total amount of food covaried with the number of food items), an incongruent condition (the total amount of food contrasted with the numerosity of the food items), or a control condition (the total amount in the two sets was equal). Results showed that dogs based their choice on the overall amount of food rather than on the number of food items, suggesting that, in food-choice tasks, amount counts more than number. The Petrazzini and Wynne study suggested that dogs are sensitive to both the quantity and number of items presented to them, reinforcing the need to control for variation in quantity or surface area in numerical-discrimination experiments.
Studies of dog cognition in our lab and others are making strides toward understanding how dogs view the world and process information. Our studies have suggested that dogs are highly attentive to human cues, but they have failed to indicate that dogs have a deeper understanding of human behavior or show human-like reflective consciousness. Thus, we have found no evidence that dogs could understand an emergency situation or that they show theory of mind or metacognition. Surprisingly, we found that although dogs showed spatial-memory effects similar to those found in other species, their spatial memory was generally weak. Similarly, they failed to be able to make numerical discriminations when items were presented sequentially. On the other hand, they clearly showed numerical discrimination when objects were presented simultaneously.
Footnotes
Declaration of Conflicting Interests
The authors declared that they had no conflicts of interest with respect to their authorship or the publication of this article.
Funding
Support for the research reported here was provided by a Discovery Grant to the first author from the Natural Sciences and Engineering Research Council of Canada.