Theory of mind refers to the ability of one person to infer the contents of another person’s mind by observing their behavior. The concept of theory of mind was advanced by Premack and Woodruff as early as 1978 to explain the behavior of chimpanzees. In a landmark article, Povinelli, Nelson, and Boysen reported the classic guesser-knower experiment with chimpanzees. As a chimpanzee subject watched, one experimenter (the “knower”) baited an opaque cup behind a screen, thus preventing the chimp from seeing which cup was baited. Another experimenter (the “guesser”) entered the room after the baiting procedure. The screen was removed, and each experimenter pointed at a different cup, with the knower pointing at the cup that actually contained food. Although chimps performed little better than chance on initial tests, they came to choose the cup pointed at by the knower significantly above chance after several hundred trials. Subsequent studies with apes and monkeys yielded mixed results.
An important criticism of the guesser-knower paradigm is that animals may use external cues instead of mental inferences to choose the object pointed at by the knower. Importantly, the experimenters who play the roles of knower and guesser randomly alternate among trials, making it impossible for an animal to be correct by always choosing the container pointed at by a single individual. Nevertheless, other behavioral cues might tip off the observant subject. Thus, a subject might learn to follow the point of only the person who stays in the room and to disregard the person who leaves the room. As a control for this problem, Povinelli and colleagues had the guesser stay in the room and put a bag over his head while the knower baited a cup. Here again, however, it might be argued that chimps quickly learned to disregard the point of the person who performed this bizarre behavior.
Another concern with tests of nonhuman primates is that they are not naturally attentive to human pointing or human gaze. However, research carried out with domestic dogs over the last 20 years suggests that they may be ideal subjects for the guesser-knower experiment. Numerous studies have shown that dogs are immediately attentive to human pointing and gaze and use such direction to seek out hidden rewards. This attention to human cuing has developed over thousands of years of domestication of dogs by humans, and by human training during the lifetime of a dog. Thus, four different sets of experiments using the guesser-knower procedure have been performed with dogs in independent laboratories, and the results are in surprising agreement.
Cooper and colleagues, in 2003, tested 15 dogs on the guesser-knower paradigm over six trials. On the first trial, 14 out of 15 dogs chose the target container pointed at by the knower, but performance strangely dropped to chance on the subsequent trials. However, overall choice of the knower’s point was significantly better than chance. Maginnity and Grace, in 2014, performed experiments using several versions of the guesser-knower task with dogs. In an initial experiment, the guesser left the room while the knower baited one of four containers. Thirteen of 16 dogs followed the knower’s point on the first trial. In a second experiment, a third experimenter baited one of the four containers while the knower watched with her hands over her ears and the guesser covered her eyes. In a third experiment, the knower watched the baiting but the guesser looked at the ceiling. Both experiments yielded significant preference for the knower from the initial trials onward.
In an article in the special issue of Learning & Behavior on canine cognition that is at the heart of this digital event, Johnston, Huang, and Santos compared dogs’ behavior with that of human children on variations of the guesser-knower task. The authors asked if dogs would behave like children by deferring to human cueing even when that cueing is incorrect. Their dog study was based on an experiment performed with 3- and 4-year-old children by Palmquist, Burns, and Jaswal. Children saw that the knower behind a screen placed a sticker in one of four circular wells while the guesser had her back turned. Children were then tested with the screen removed in one of two ways. In a traditional test, the knower and the guesser each pointed at different wells covered by cups, but in an alternate test, children chose between cups with pictures of the faces of the knower and guesser attached to them. Interestingly, children chose the cup with the knower’s picture on it significantly above chance. When knower and guesser both pointed at different cups, however, children chose indifferently. Apparently children’s choices were strongly controlled by human pointing, although their choice of the cup with the knower’s face showed that they knew the knower knew where the sticker was and the guesser did not.
Johnston and colleagues performed a similar experiment with two groups of dogs. For one group, the guesser and knower pointed at different containers, but for the other group the guesser and knower placed differently colored blocks of wood on the containers as markers. In contrast to children, dogs showed a preference for the knower in the pointing condition but not in the marker condition. Thus, dogs did not show indifference to pointing by the knower and guesser, as was the case with children. Although dogs are strongly controlled by human pointing cues, they are not similarly affected by the knower and guesser. Perhaps most importantly, the experiments by Johnston and colleagues provide yet another replication of knower preference by dogs.
Still a fourth set of experiments was carried out by Catala, Mang, Wallis, and Huber in 2017. In the traditional guesser-absent procedure, dogs showed a highly significant preference for the container pointed at by the knower on the first block of four trials. Most importantly, Catala and colleagues ran an additional condition that involved a clever control for differential behavioral cues between the guesser and knower. Both the guesser and knower turned their heads in an identical fashion during the baiting procedure. However, the positions of the two observers made it possible for the knower to see which container was baited while the guesser could not. Thus, their behaviors were identical, but the views their gazes afforded were different. Dogs again showed a significant preference for the container pointed at by the knower.
Sets of experiments carried out in four different laboratories all show the same results. Dogs followed the point of the knower significantly more that the point of the guesser under a variety of conditions. Importantly, this difference consistently appeared on the initial trials of testing. Further, various controls ruled out the possibility that dogs were using auditory or odor cues to find the baited container.
What do we make of these findings?
Should we attribute human-like theory of mind to dogs? Quite rightly so, the researchers who carried out these studies are hesitant to reach such a conclusion. As Catala and colleagues suggest, dogs may have learned through experience that they are more likely to be reinforced by humans who see where food is than by humans who are not present or who have their eyes covered. They may also be highly sensitive to the direction of human gaze, so that they more readily trust the person who looks in the direction of food reward than the person who looks away. Perhaps we should think of dogs showing a “theory of attention” instead of a theory of mind.