Fido Might Not Know What You Do and Do Not See

You've probably had a conversation that goes something like this:
Person A: "My dog is sooooo amazing!"
You: "I mean, dogs are awesome and all, but what's so amazing about this particular dog?"
Person A: "He just understands me. It's like he knows what I'm thinking and what I need."
You: "Do you think he's just maybe responding contingently do your overt displays of emotion?"
Person A: "Listen, man, I'm telling you: my dog can read my mind!"

No matter on which side of this sort of argument you tend to fall, the question of whether or not domestic dogs can read human minds is an interesting and important one. More specifically, do dogs have a theory of mind? Can they take the perspective of a human?

Dogs show a wide range of "human-like" social behaviors. They respond appropriately to human body language and to verbal commands, and are capable of facilitating joint attention with humans. It some cases, their social skills, at least in terms of communication with humans, surpasses those even of chimpanzees.

Also, impressively, domestic dogs tend to perform particularly well in theory of mind experiments. The problem is that the experimental participants in these studies could potentially be responding to other sorts of contextual social or environmental cues when succeeding at theory of mind tasks, rather than relying on theory of mind skills per se. Ever the empiricists, some also argue that associative learning, rather then an innate theory of mind module, could account for dogs' apparent theory of mind skills. If either of these sorts of arguments were indeed the case, then theory of mind would not strictly be necessary to explain these sorts of complex behaviors.

The best sorts of experimental designs to use in non-human animal theory of mind tasks derive from what I call the "Hare Task." The Hare Task was originally designed for chimpanzees, and is in essence a food competition task. Two chimpanzees, one dominant and one subordinate, are placed at opposite ends of the testing area. Between them are two barriers, one clear (like a window), and one opaque. Just behind each of the barriers, on the side of the subordinate chimp, is a piece of food. In typical interactions, the dominant chimp always has first access to the food. However, the dominant individual can only see the food behind the clear barrier, while the subordinate individual can see both pieces of food.

The set-up looks something like this (source):

The idea is that if the subordinate chimpanzee has a theory of mind - if he (1) knows what the dominant individual does and does not see, and (2) assumes that the behavior of the dominant individual will be consistent with that knowledge - then the subordinate chimp should immediately retrieve the food behind the opaque barrier, and leave the food behind the clear barrier to the dominant chimp.

This sort of task can be easily modified to address slightly different questions for different animals, and recently in the journal Learning and Behavior, psychological scientist Monique Udell and colleagues from the University of Florida, Gainesville, have modified the Hare task to ask if domestic dogs can take the visual perspective of humans. Given that domestic dogs tend to do well on theory of mind tasks, Udell wanted to know if their success emerges due to a theory of mind, or whether their success can be explained by learning processes (such as classical or operant conditioning).

More specifically, some have argued that human-like social skills in dogs are the result of selection through the process of domestication, even if as a correlated by-product of the selection for or against something else. If these social skills are the result of domestication, then they are not simply the result of experience or learning. If this is true, Udell reasoned, then "all healthy populations of domestic dogs should be expected to outperform non-domesticated canids on human attentional-state tasks. Furthermore, this should hold true independent of the age, condition, home environment, or experience level of the dog under test."

In order to investigate this question, Udell tested the ability of three different groups of canids, each of which varied on evolutionary origins or developmental experiments: pet dogs, shelter dogs, and human-reared wolves. All pets dogs were brought into the lab by their owners. The researchers were careful to exclude any shelter dogs that were owner-surrenders, and only tested strays who were comfortable around humans, but clearly had significantly fewer social experiences. The human-reared wolves were from Wolf Park, in Battle Ground, Indiana.

If dogs' success at theory of mind tasks is the result of domestication, then both groups of dogs should outperform wolves, regardless of experience. However, if dogs' success is the result of experience or learning, then the pet dogs should outperform the shelter dogs, and the human-reared wolves might outperform the shelter dogs as well.

In each of four conditions in this experiment, the dog or wolf was allowed to approach one of two researchers holding a piece of food in her hand. One of the researchers was designated the "seer," while the second one always had her vision obstructed ("blind"). The "blind" researcher could be blind in one of four ways: back turned to the subject, holding a book in front of her face, holding a camera in front of her eyes, or holding a bucket over her head.

In each trial, the wolf or dog was kept 6 meters away from either of the two researchers by a third research assistant, with his or her back to the researchers. Then, in unison, both researchers would call the name of the dog or wolf (or, for the shelter dog, they'd say something like "here, puppy!"). The question was which researcher would the dog or wolf approach? Once the wolf or dog stayed within 1 meter of the chosen researcher for three seconds, the trial was concluded. In addition, the "seer" had to visually track the wolf or dog throughout the trial so that eye-contact could be initiated. If the subject approached the "seer," he or she was rewarded with the food; if the subject approached the "blind," even though she always held a piece of food in her hand, the subject was never rewarded.nd on every trial. Each dog participated in only two of the four possible conditions, though the wolves participated in all four experimental conditions (though they were distributed over the course of a year, to minimize the possibility of within-experiment learning). Altogether, each individual completed a total of ten trials.

Pet dogs and wolves had the best performance, with median performance levels of 9 out of 10 trials correct. Shelter dogs had the worst performance, with median performance levels of 7 out of 10 trials correct. For one condition, all three groups of canids performed significantly above chance: the back-turned condition. The results for the more nuanced conditions were, understandbly, more variable. For the book condition, pet dogs performed above chance, and significantly outperformed the shelter dogs and the wolves. While pet dogs performed above chance in the bucket condition, there was no significant group difference, and all three groups did similarly poor in the camera condition (no group performed, on average, above chance levels).

Udell and colleagues' interpretation of the results is pretty straightforward, and is quite reasonable given the data (emphasis is mine):

The hypothesis that domestication is sufficient for dogs to engage in perspective taking is rendered problematic by the observation that dogs' performance on these tests depends on the type of occluder used and on the life conditions of the canid. Furthermore, the success of wolves on one condition demonstrates that domestication is not essential for above-chance performance under all conditions. The alternative position is that domestic dogs and other socialized canids are not taking the mental perspective of the seeing or blind individual, but instead making a discriminative choice based on past reinforcement histories, in which certain human actions or orientations served as predictors of reinforcement upon approach. This hypothesis is consistent with the patterns of success and failure found in this study.

They reason that all individuals in the study, whether pet dogs, shelter dogs, or human-reared wolves, have experience approach individuals from the front or from the back, and all would have been equally likely to learn that it is statistically more likely to receive food from a human who is facing them than one whose back is to them. They argue that pet dogs would have had experience with humans reading books, and were therefore successful in that condition, while shelter dogs and wolves presumably had less or even no experience with book-reading humans on which to base their behavior. They explain the failure of even the pet dogs to respond appropriately in the camera condition to the possibility that the camera was simply too small of a visual barrier, making it less salient. To their credit, they also note that "on the other hand, humans can and do see through the lens of a camera and may even actively reinforce dogs or wolves for orienting toward them if they are the subject of the photograph. Therefore, the failure to discriminate between experimenters in this condition might even be considered the correct response." When it comes to the bucket condition, it might seem surprising that even the pet dogs were unable to perform accurately. The researchers argue that each of the three groups of canids were unlikely to have had any experience interaction with humans who had buckets over their heads, and, as with the book, were unlikely to have learned the appropriate behavioral response to that stimulus.

The researchers conclude that domestication is not necessary, nor sufficient, to explain the performance of domestic dogs on theory of mind tasks. This is not all that surprising to me, though I would not take the position of the empiricist, and argue that their performance is more likely the result of learning. Rather, I am simply unsurprised that wolves have at least some rudimentary perspective-taking ability. Domestic dogs' theory of mind is probably just a little bit sharper, or a little bit more fine-tuned due to experience, than that of wolves.

Further, Udell and colleagues hypothesize that "dogs' ability to follow human actions stems from a willingness to accept humans as social companions, acquired early in [development], combined with conditioning to follow the limbs and actions of humans to acquire reinforcement." My main problem with this line of reasoning is that this experiment doesn't actually involve the movement of limbs, or any human actions. The only change in the body-state of the human researchers in the experiment is the tracking of the dog or wolf by the "seer," so I'm not clear where any sort of action-related conditioning could enter into the dogs' mental processing in the first place.

So, I suppose I'm just unconvinced with the larger argument put forth in this paper. Rather, I tend to agree with the notion that domestication has allowed for dogs to engage socially with humans in a way that wolves are not incapable of, but for which they are relatively unprepared. The fact that human-reared wolves had at least minimal success in this experiment does not surprise me, especially since they're human-reared. The interesting finding, however, is the relative underperformance of pet dogs in the bucket and camera conditions: this is something, I think, that warrants further investigation. Why does performance in this task vary based on how the eyes of a human are occluded? It stands to reason that it ought to vary based on whether a human's eyes are occluded or not.

It is also possible that some of the failures of the dogs and wolves in this experiment is simply due to the human presence. I'd be interested in seeing what would happen if the dogs and wolves were placed in a more competitive context, similar to the original Hare task. And, while we're at it, how about comparing domesticated and undomesticated Siberian foxes? Now there are two populations of critters with highly controlled experiences, and precisely defined genetic relatedness.

Udell MA, Dorey NR, & Wynne CD (2011). Can your dog read your mind? Understanding the causes of canine perspective taking. Learning & behavior PMID: 21643852

Hare B, Call J, & Tomasello M (2001). Do chimpanzees know what conspecifics know? Animal behaviour, 61 (1), 139-151 PMID: 11170704

For more:
Wolves Are Smart, But Dogs Look Back
How Specific Are The Social Skills of Dogs?
Did Dogs Get Their Social Intelligence By Accident?

Dog photo by the author.