Experience-Projection Methods in Theory-of-Mind Research: Their Limits and Strengths

The Logical Problem

Whether or not animals have theory of mind is an important issue in comparative psychology and philosophy. It has historically been assumed that theory of mind is a defining feature of human cognition (Davidson, 2001) that evolved because of its advantages for predicting and manipulating other individuals’ behavior (Povinelli & Vonk, 2003). For more than 40 years, researchers have vigorously tested the truth of this assumption through studies with animals. Some studies have yielded positive findings (Call & Tomasello, 2008; Whiten, 2013), challenging theory of mind’s uniquely human pedigree and evolutionary origins. However, there is a long-standing controversy over whether these studies show that animals impute mental states or are engaged in some form of behavior reading (Heyes, 2015; Lurz, 2011; Penn & Povinelli, 2007). The controversy stems from the fact that theory of mind is not telepathy but a form of indirect knowledge gained from reading observable cues. If an animal, for example, thinks a predator sees it and intends to eat it, the animal attributes these mental states to the predator not by directly observing them but by inferring them from observable cues (e.g., the predator is staring and drooling).

This creates a methodological problem: If the cues an animal uses to infer mental states are cues it knows to be associated with the behavior it predicts, there is no scientifically compelling reason to conclude the animal is using theory of mind rather than just behavior reading. There is no reason to conclude, for example, that prey animals that selectively anticipate being chased by predators that are looking at them do so because they understand that gawking predators see them rather than that gawking predators typically chase them (Carter, Lyons, Cole, & Goldsmith, 2008). Daniel Povinelli and colleagues (e.g., Povinelli & Vonk, 2003) argued that in such cases there is no “unique causal work” (p. 159) for the theory-of-mind hypothesis to do beyond what can be explained by behavior reading. Thus, the challenge is to design experiments in which animals can predict others’ behavior by using only theory of mind and not a plausible behavior-reading strategy. This is known as “the logical problem” (Lurz, 2011, p. 306; Povinelli & Vonk, 2003, p. 160), and there have been numerous attempts to circumvent it. Recently, for example, Krupenye, Kano, Hirata, Call, and Tomasello (2016) ran a change-of-location false-belief task with chimpanzees that tracked their anticipatory-looking behavior. Subjects did appear to anticipate, with their gaze direction, where an agent with a false belief would search, but some researchers have still argued that the change-of-location task fails to solve the logical problem (Fabricius & Khalil, 2003; Heyes, 2014; Perner & Ruffman, 2005).

The Experience-Projection Method

In 1998, Cecilia Heyes proposed a procedure for chimpanzees that aimed to solve the logical problem (Fig. 1). The basic idea behind Heyes’s experience-projection procedure was (a) to give chimpanzees first-hand experience with a novel apparatus that affected their own states of seeing and (b) to test whether they would project similar mental states onto others encountering the same apparatus. Chimpanzees would first be trained to expect food from an experimenter who saw where the food was hidden but not from one who did not see where it was hidden. In the subsequent self-experience phase, chimpanzees would wear two different pairs of goggles: a blue pair with a one-way mirror lens that allowed the wearer to see objects in the environment (like mirrored sunglasses) and a red pair with a similar mirror lens but an opaque finish on both sides (to prevent the wearer from seeing). After experiencing both goggles, chimpanzees would complete a transfer test in which the experimenters returned, one wearing blue goggles and one wearing red goggles, and directed their faces toward the food-hiding process. The question was whether chimpanzees would use their own prior experience with the goggles to infer that only the blue-goggled experimenter knew where the food was hidden. If they did so, Heyes argued, they should prefer to beg from the blue-goggled experimenter. A behavior-reading strategy would not predict this, since the chimpanzees had no prior experience of being fed by blue-goggled (vs. red-goggled) experimenters. ²

OPEN IN VIEWER

Fig. 1.

Diagram depicting Heyes’s (1998) original description of the experience-projection method. In the training phase, chimpanzees observe that one experimenter sees food and the other does not; the chimpanzees learn to beg from the experimenter who sees the food. In the self-experience phase, chimpanzees gain experience wearing transparent (blue) and opaque (red) goggles. In the transfer phase, experimenters return wearing the goggles, and the chimpanzees decide from which experimenter to beg. Proponents of the traditional experience-projection method argue that subjects learn in the self-experience phase that the red goggles prevent them from seeing but the blue goggles allow them to see (visual awareness), and they subsequently use this information in the transfer phase to decide to beg from the experimenter they believe sees the food. Critics (Lurz, 2011; Perner, 2012) argue that what subjects may actually learn in the self-experience phase is that the red goggles prevent and the blue goggles allow an unobstructed line of gaze (geometrical relation), and that subjects subsequently use this information in the transfer test to decide to beg from the experimenter they believe has an unobstructed line of gaze to the food. As a result, critics maintain that the traditional experience-projection method cannot determine whether subjects use a theory-of-mind strategy or a behavior-reading strategy to discriminate between the experimenters in the transfer phase.

Use of the Experience-Projection Method

The experience-projection method conceived by Heyes has been used to investigate seeing attribution in humans and animals, using transparent and opaque barriers of various types. Meltzoff and Brooks (2008) and Senju, Southgate, Snape, Leonard, and Csibra (2011) used transparent and opaque blindfolds to assess infants’ understanding of seeing in adults. In both studies, infants who had self-experience with the transparent blindfold were more likely to respond to an adult wearing a blindfold as if the adult could see than were infants who had worn the opaque blindfold (but see Teufel, Clayton, & Russell, 2013).

Teufel et al. (2009) and Furlanetto, Samson, Becchio, and Apperly (2016) first used the experience-projection method on adults to investigate whether they automatically encode what other people see. Subjects made judgments about various stimuli (e.g., the number of dots on a screen) while observing another person wearing transparent or opaque goggles. If the adults automatically encoded the other person’s mental state of seeing, this circumstance should influence their judgments (e.g., they should be slower to judge how many dots they themselves see if that information conflicts with the number the other person sees). Adults’ judgments were, indeed, affected when the other individual was wearing transparent (but not opaque) goggles, suggesting that they automatically represent what others see, not simply their head or body direction.

Vonk and Povinelli (2011) published the first experience-projection study of animals. Two young chimpanzees were given different buckets to wear over their heads: One bucket had a transparent visor, and the other had an opaque visor. When the chimpanzees later encountered experimenters who were wearing the different visor-buckets, the chimpanzees (perhaps because of their immaturity) did not treat them differently. In 2015, Karg, Schmelz, Call, and Tomasello used the experience-projection method with a larger sample of adult chimpanzees. Subjects first saw two screens that were identical except that one was transparent when viewed directly, whereas the other was opaque from all angles. Subjects then saw an experimenter facing either a transparent or an opaque screen. Although subjects were no more likely to follow the experimenter’s gaze in the transparent condition, they were more likely to steal food from her in the opaque condition. The researchers concluded that the chimpanzees used their experience with the screens to infer that the experimenter facing the transparent screen, but not the opaque screen, could see.

Bugnyar, Reber, and Buckner (2016) ran an experience-projection test with ravens. In the self-experience phase, ravens entered a competitor’s empty enclosure, where they could peer through a peephole into their own enclosure. They were then returned to their own enclosure and allowed to cache food while hearing sounds made by their competitor in its adjacent enclosure. The ravens used the same strategy they had employed when a competitor watched them through an open window: They cached food quickly and did not return to the cache site. The researchers concluded that the ravens generalized from their own experience with the peephole to infer that the competitor could observe them caching food through the hole.

Critique of the Experience-Projection Method

Although the experience-projection method has been embraced by many researchers, others have questioned its ability to deliver on the promise of distinguishing theory of mind from behavior reading (Andrews, 2005; Hurley & Nudds, 2006; Perner, 2012; for replies, see Heyes, 2015; Penn & Povinelli, 2007). One behavior-reading alternative it fails to rule out is the line-of-gaze hypothesis (Lurz, 2011). Line of gaze is a straight line between an agent’s eyes and nonoccluded objects in front of its eyes. The computer screen in front of your eyes is in your line of gaze; the coffee mug behind the screen is not, because the screen occludes it. The mental state of seeing, on the other hand, is a state of visual awareness distinct from line of gaze. That they are distinct is evident in the fact that many things exist in your line of gaze that you cannot see (e.g., small dust particles), and you can sometimes see things that are not literally in your line of gaze (e.g., your own face while looking into a mirror). Many species represent other individuals’ line of gaze (Bugnyar, Stöwe, & Heinrich, 2004; Okamoto-Barth, Call, & Tomasello, 2007), but this is not the same as representing others’ mental state of seeing.

The problem with the traditional experience-projection method is that transparent and opaque barriers do not dissociate seeing and line of gaze. A glass window allows you to see objects outside, but it also allows an unobstructed line of gaze to those objects. Similarly, a closed door prevents you from seeing objects on the other side, but it also obstructs your line of gaze to those objects. Consequently, what subjects in experience-projection tests may implicitly learn in the self-experience phase is not which barriers allow them to see, but which barriers allow them an unobstructed line of gaze. Armed with information about the differential effects the barriers have on line of gaze, subjects can then infer that the same holds true for other individuals. Thus, the traditional experience-projection method fails to experimentally discriminate between the use of a theory-of-mind strategy and a plausible behavior-reading strategy.

New Experience-Projection Methods and Their Application

Second-generation experience-projection methods have been carefully designed to overcome the problem with the traditional method: They use barriers designed to differentially affect individuals’ experience of seeing but not their line of gaze (Fig. 2). In some second-generation studies, both barriers are transparent. Krachun and Lurz (2016), for example, had 4- to 5-year-olds look at objects through magnifying and minimizing lenses. The children then predicted the behavior of a naive agent who was looking at the objects through the lenses but believed them to be regular glass. Most children expected the agent to behave as if he were seeing a large object behind the magnifying lens and a small object behind the minimizing lens, even though the children knew both objects were the same size. Moreover, because the transparent lenses allowed the agent an equally unobstructed line of gaze to both objects, children’s performance on the task could not be explained by a line-of-gaze hypothesis.

OPEN IN VIEWER

Fig. 2.

Schematic representation of the similarities and differences between traditional and second-generation experience-projection (EP) methods. In the training phase of both EP methods, subjects learn to discriminate between individuals who can see and those who cannot see a relevant object or event. In the self-experience phase of the traditional EP method, subjects are exposed to transparent and opaque barriers (e.g., goggles, blindfolds, screens). In the transfer phase of the traditional EP method, subjects watch individuals who are either behind a transparent barrier (allowing subjects to see) or behind an opaque barrier (preventing subjects from seeing). In second-generation EP methods, subjects in the self-experience phase are exposed to barriers (e.g., lenses, telescopes, mirrors) that manipulate their visual experiences without correspondingly manipulating their line of gaze. In some studies, both of the barriers are transparent (Conway, Lee, Ojaghi, Catmur, & Bird, 2017; Krachun & Lurz, 2016), and in others they are both opaque (Lurz, Krachun, Mahovetz, Wilson, & Hopkins, 2018). In the transfer phase of second-generation EP methods, subjects encounter individuals looking at the same types of barriers (opaque or transparent), and these individuals can (depending on the barrier) either see or not see the subject or relevant object in the study.

Conway, Lee, Ojaghi, Catmur, and Bird (2017) ran a second-generation experience-projection procedure with transparent lenses inserted into distinctly colored telescopes. One telescope allowed users to see a target object in their line of gaze; the other was designed to prevent users from seeing the object. The aim was to test whether adults in a dot-perspective-taking task automatically encode what another subject is seeing or simply what the subject has a line of gaze to. Because the telescopes manipulated which objects subjects could see but not which objects were in their line of gaze, the procedure—the researchers claimed—would provide “unambiguous evidence” (p. 454) for theory of mind, which earlier dot-perspective-taking tasks could not. However, subjects’ responses were the same regardless of which telescope the other individual was using. The researchers concluded that these results undermined earlier findings (e.g., Furlanetto et al., 2016) supporting the automatic encoding of seeing.

In Lurz, Krachun, Mahovetz, Wilson, and Hopkins’s (2018) second-generation experience-projection test for chimpanzees, both barriers were opaque. One barrier was a mirror, which allowed the experimenter looking at its reflective side to see the chimpanzee behind him (mirror-forward condition). The other barrier was the nonreflective side of the mirror, which prevented the experimenter looking at it from seeing the chimpanzee behind him (mirror-away condition; Fig. 3). ³ After looking at themselves in the mirror, chimpanzees were given a chance to beg from an experimenter whose line of gaze was directed either at the mirror or at the nonreflective barrier. Chimpanzees produced significantly more visual gestures (e.g., pointing, lip protrusion) toward the experimenter in the mirror-forward condition, suggesting that they understood that the experimenter could see them (i.e., they ascribed a true mental state to the experimenter). Because the experimenter’s line of gaze in both conditions was directed away from the chimpanzee, the results are not explainable by a line-of-gaze hypothesis. A control condition ruled out the possibility that subjects were just responding to the mirror image of the experimenter looking at them without understanding that the experimenter could see them.

OPEN IN VIEWER

Fig. 3.

Illustrations of the (a) mirror-forward condition and (b) mirror-away condition from the study by Lurz, Krachun, Mahovetz, Wilson, and Hopkins (2018). In both conditions, the experimenter’s face was directed away from the chimpanzee, and his line of gaze was instead to the barrier. However, in the mirror-forward but not in the mirror-away condition, the experimenter could see the chimpanzee. This setup thus dissociated line of gaze from seeing. Figure reproduced from Lurz, Krachun, Mahovetz, Wilson, and Hopkins (2018).

Concluding Remarks

The experience-projection methodology has given researchers a way around the logical problem of designing experiments in which other individuals’ behavior can be predicted by using only theory of mind and not a plausible behavior-reading strategy. This is no small achievement, as disentangling mental-state inference from behavior reading has been an enormous challenge for decades. We suggest that experience-projection methods be used to investigate, in humans and other animals, additional types of mental-state inference beyond seeing. For example, researchers have been investigating what animals and young children understand about the mental state of hearing in other individuals (Moll, Carpenter, & Tomasello, 2014; Santos, Nissen, & Ferrugia, 2006). Unfortunately, these experiments also fall prey to the same logical problem. They make it difficult to determine whether participants understand that other agents heard a noise (theory of mind) or simply that they were in a particular spatial proximity to it (behavior reading). Experience-projection methods have the potential to discriminate between these competing hypotheses, as they have done in studies on seeing.

Recommended Reading

Conway, J. R., Lee, D., Ojaghi, M., Catmur, C., & Bird, G. (2017). (See References). An innovative second-generation experience-projection study whose findings challenge the idea that adults automatically encode what others see.

Heyes, C. (1998). (See References). The locus classicus of the traditional experience-projection method, providing the rationale for the method and its use to test for theory of mind in chimpanzees.

Karg, K., Schmelz, M., Call, J., & Tomasello, M. (2015). (See References). The first study using the traditional experience-projection method with chimpanzees that yielded positive results.

Lurz, R., Krachun, C., Mahovetz, L., Wilson, M., & Hopkins, W. (2018). (See References). A second-generation experience-projection study showing that chimpanzees gesture at individuals who can see them, not simply at those who have a line of gaze to them.

Povinelli, D. J., & Vonk, J. (2003). (See References). A clear description of the logical problem in animal theory-ofmind research.