Abstract

A variety of recent experiments suggest that apes know what other individuals do and do not see. The results of each experiment may be explained by postulating some behavioural rule that individuals have learned that does not involve an understanding of seeing. But the postulated rule must be different in each case, and most of these do not explain more than one experiment. This patchiness of coverage gives this kind of explanation a very ad hoc feeling, especially since there is rarely any concrete evidence that animals actually have had the requisite experiences to learn the behavioural rule—there is just a theoretical possibility. It is thus more plausible to hypothesize that apes really do know what others do and do not see in many circumstances. Moreover, and more generally, there is no reason to assume—other than some kind of blind allegiance to behaviourism—that just because an animal has learned something, no cognitive processes are involved.

17.1 Introduction

Observations of naturally occurring behaviour are always open to multiple interpretations, especially when the issue is the underlying cognitive processes involved. Some theorists prefer richer interpretations (so-called boosters) while others prefer leaner interpretations (so-called scoffers). The ultimate boosters are those who think there are no significant differences between non-human animal and human cognition, and the ultimate scoffers are radical Behaviourists who do not believe it is useful to talk of cognitive processes at all. Most of the contributors to this volume fall somewhere in between these two extremes. With particular reference to primate social cognition (‘theory of mind’), there are boosters who think that non-human primates understand much about the psychological states of (p. 372 ) other organisms, and scoffers who think they understand little or nothing. In effect, the scoffers think that non-human animals are behaviourists who see and respond to the behaviour of others but do not conceive of any cognitive processes underlying this behaviour.

Not so many years ago, the data on non-human primate social cognition was so thin that it was possible for sensible people to espouse fairly extreme booster or scoffer positions. But in the last few years a number of new studies have produced data that constrain, in significant ways, the range of interpretive options. In this chapter, we look at these new data and compare richer and leaner interpretations of them. In particular, we focus on data concerning what chimpanzees know about what others can and cannot see—do they have such knowledge, or are they only capable of determining where others are looking, and what they are looking at? There are a number of ways in which the difference between these two interpretations—understanding seeing versus responding to looking behaviour—could be characterized, but for the moment let us proceed as if it were clear enough. We will explicate the intended distinction further by reviewing data from four different experimental paradigms: gaze following, competing for food, begging and gesturing, and self-knowledge. For each paradigm we start with a basic finding and then contrast the booster and scoffer alternatives with the data available in the literature.

17.2 Gaze following

Tomasello et al. (1998) conducted a very simple experiment. A human experimenter (E) waited until a pair of chimpanzees was spatially arranged so that one was facing toward him in the observation tower (the looker) and another was oriented to that looker (the subject). E then held up a piece of desirable food, inducing the looker to look up at him, and observed how the subject responded to the looker's looking behaviour. In control trials E displayed the food in an identical manner but when the subject was alone. The finding was that subjects reliably followed the gaze direction of the looker to the food a greater number of times in experimental than in control trials (Fig. 17.1). Other studies have found that chimpanzees also follow the gaze direction of human beings, and they can even do this on the basis of eye direction alone and independently of the experimenter's head direction (e.g. Povinelli and Eddy 1996a).

There are of course many ways to interpret straightforward gaze-following behaviour, but other studies have helped to reveal how chimpanzees understand the behaviour they are observing. First, chimpanzees follow the gaze direction of human beings to a specific location even if they have to look past and ignore other novel objects along the way in order to fixate the target location. This would seem to indicate that they are not just turning in the same general direction as the looker and then searching randomly for something interesting; they are seeking the target of the looker's perceptual activity (Tomasello et al. 1999). Second, if adult chimpanzees track the gaze of another (p. 373 )

individual to a location and find nothing interesting there, they quite often look back to the individual's face and track her gaze direction a second time (Call et al. 1998; Bräuer et al. submitted). This ‘checking back’—which only adult chimpanzees do—is a key criterion used to assess human infants' understanding of the visual experience of others, since it would seem to indicate that the subject expects to find the looker's perceptual orientation directed at a target (Bates 1979). Interestingly, if a looker looks repeatedly to a location with no salient target adult (but again not juvenile) chimpanzees stop responding to that individual's looking activities (Tomasello et al. 2001) indicating acquired expectations about when it is likely that following the gaze direction of another is likely to lead to an interesting target.

But perhaps the most telling situation occurs when a human being looks behind a barrier. Following a suggestive finding of Povinelli and Eddy (1996a), Tomasello et al. (1999) had a human E look around various types of barriers (or look straight ahead in a control condition). In this case, a simple gaze following response (turning head to look in the direction E is looking) would not be enough—that would simply lead to the subject fixating the barrier itself. To track E's gaze to its target, subjects needed to move a few meters so as to attain the appropriate viewing angle enabling them to look behind the barrier. And this is just what they did (much more often than in the control condition) for all four of the barrier types investigated.

The booster interpretation of these findings on gaze following is simply that chimpanzees follow the gaze direction of others—checking back when they find nothing there (and eventually stopping if they repeatedly find nothing there)—because they want to see what the other is seeing. When there is a distractor they ignore it; when there is a barrier, they move themselves in order to see what the other is seeing. The scoffer interpretation avoids positing that the subject knows that the other is seeing something and instead relies on some combination of biological predispositions and individual learning. The first proposal is simply that chimpanzees are biologically (p. 374 ) predisposed to orient in the direction in which another individual orients, and to account for the barrier findings this predisposition must be geometrically quite specific. The second proposal is that individuals learn, through personal experience, that when they look in the direction toward which another individual is oriented, they often find something interesting or important (i.e. that is rewarding in some way). It is not clear in these lean interpretations exactly how to account for the facts that chimpanzees check back with the looker when they find nothing and that they eventually stop looking if they repeatedly find nothing, but presumably some account in terms of learning could be constructed.

17.3 Competing for food

In a recent series of studies, Hare et al. (2000) placed a subordinate and a dominant chimpanzee into rooms on opposite sides of a third room. Each had a guillotine door leading into this middle room which, when opened at the bottom, allowed them to observe two pieces of food at various locations—and to see the other individual looking under her door. After the food had been placed, the doors for both individuals were opened and they were allowed to enter the third room. The basic problem for the subordinate in this situation is that the dominant will take all of the food she can see. However, in some cases things were arranged so that the subordinate could see one piece of food that the dominant could not see, for example by placing it on the subordinate's side of a small barrier (with the other piece of food in the open) (Fig. 17.2). The question was thus whether the subordinate knew that the dominant could not see a particular piece of food, and so it was safe to go for it. The basic finding was that the subordinates did indeed go for the food that only they could see much more often than they went for the food that both they and the dominant could see.

One scoffer possibility is that subordinates in these studies may have been monitoring the behaviour of dominants, rather than dominants' perceptual access to the food, and reacting to that. But this possibility was ruled out in some of the studies in this series by giving subordinates a small headstart and forcing them to make their choice between the two pieces of food before the dominant was released into the middle cage.

(p. 375 ) Moreover, in one additional control condition the dominant's door was lowered before the two competitors were let into the room (and again the subordinate got a small headstart), so that the subordinate could not see the dominant at all at the moment of choice—and so could not react to her behaviour. Subordinates still targeted the piece of food the dominant could not see. Another possibility is that subordinates viewed the barrier as a physical impediment to the dominant's access to the food, in that it might slow her down by requiring her to reach around the barrier for the object of her desire. So in another control condition food was placed on the subordinate's side of a transparent barrier. In this case, subordinates chose equally between the two pieces of food, seeming to know that the transparent barrier was not blocking the dominant's visual access to the food—and seeming not to care about any possible blocking of physical access.

Another scoffer interpretation of this line of research was suggested by Povinelli and Giambrone (2001). They suggested that subordinates were attracted to the food behind the barrier because in general they prefer to forage in the vicinity of barriers rather than out in the open where dominants might easily see them (the peripheral feeding hypothesis). But this hypothesis is ruled out by a second line of similar experiments. Hare et al. (2001) used the same basic procedure but with only one piece of food, which was hidden behind one of two barriers; they varied whether or not the dominant witnessed the hiding process. Thus, in some trials the dominant did not witness the food being hidden because her door was down, whereas in other trials she witnessed the hiding process under her partially open door (subordinates always saw the entire baiting procedure and could monitor the visual access of the dominant competitor as well). Subordinates in this case preferentially went for the food that dominants had not seen hidden. This indicates that they know not only what others can and cannot see at the moment, but also what others have just seen in the immediate past. Moreover, since the food was always hidden behind one of the two barriers and the only difference between conditions was whether the dominant had or had not witnessed the baiting the peripheral feeding hypothesis does not apply. It is also interesting that if the dominant individual was switched with another dominant individual (who had seen nothing) just before the moment of choice, subordinates now felt free to go for the food no matter what had transpired earlier. Chimpanzees seemingly know which particular individuals have and have not seen important events. This is important because it rules out the possibility that subordinates have used the mere presence/absence of a dominant during the baiting.

One final scoffer interpretation of these food competition experiments is the so-called evil eye hypothesis. Perhaps subordinates believe that any piece of food observed by a dominant is ‘contaminated’—it is forbidden once the dominant has put the evil eye on it—and so the only safe food is food that she cannot see and indeed has never seen. In one final study in Hare et al. (2001), both the dominant and the subordinate watched the food being hidden behind one of the two barriers; the dominant's evil eye was thus placed on it, and so on this interpretation it should be avoided at all costs. But then in one (p. 376 ) experimental condition only the subordinate watched it being moved to a new location (dominant's door down), whereas in another condition they both watched it being moved. Subordinates went for the food when they alone had watched the moving process, but not when both competitors had watched the moving process. It is thus clear that they did not believe in any dominant evil eye, since they went for the food whose movement to a new location the dominant had not witnessed (even though she had seen it earlier).

These food competition studies are especially important in determining what chimpanzees know about seeing because in these studies subordinates did not just follow the gaze direction of another to a location, rather they demonstrated by their behaviour that they knew something about the actual content of what the dominant could see and had seen in the immediate past. This is clearly indicated by the fact that subordinates made inferences about the dominant's impending action; if the dominant could see the food or had seen it just before, subordinates could infer that she would go for it. They would not make this inference if what the dominant could see or had seen was a rock. This means that they knew not just where the other was looking, but they knew the content of what the others saw—the food—because only with this knowledge could they predict her action. Of course they may have had learning experiences in the past in which they noted a dominant looking at food and then going for it, but conditions involving such things as the transparent barrier and the food whose location was moved after the dominant had seen it hidden are not likely to be common experiences in chimpanzees' lives, and so not things about which they could have learnt.

17.4 Begging and gesturing

Povinelli and Eddy (1996b) trained young chimpanzees to approach a Plexiglas barrier and extend their hand toward one of two human experimenters (each with a hole in the Plexiglas in front of them) to request the food that was on a table between them. The attentional state of the two humans was varied, and it was found that, for example, chimpanzees consistently gestured toward the human who was facing toward them, in preference to one facing away from them (thus confirming the naturalistic findings of Tomasello et al. 1994, 1997). However, they did not gesture differentially for a human who wore a blindfold over his eyes (as opposed to one who wore a blindfold over his mouth), or for one who wore a bucket over his head (as opposed to one who held a bucket on his shoulder), or for one who held his hands over his eyes (as opposed to one who held his hands over his ears), or for one who had his eyes closed (as opposed to one who had his eyes open), or for one who was looking away (as opposed to looking at the subject), or for one whose back was turned but who looked over his shoulder to the subject (as opposed to one whose back was turned and he was looking away) (Fig. 17.3). To explain these results, Povinelli and Eddy proposed the valence hypothesis: chimpanzees have a general notion of perceptual access (communicative access) based (p. 377 )

mainly on body orientation, with the face and eyes playing basically no role. Thus in this scoffer hypothesis they know little, if anything, about seeing as a psychological process.

Kaminski et al. (2004) used a related but different methodology to test this hypothesis in a modified way (see also Call and Tomasello 1994). The difference was that chimpanzees did not have to choose between human communicators, but were always faced with only one communicator. Moreover, unlike Povinelli and Eddy (1996a, b) there was no training required at the beginning of the test. We simply took advantage of the begging behaviour that apes display in these situations. In different experimental conditions, experienced by the subject at different times, the communicator was oriented in different ways. For example in one condition the human communicator faced the subject, and in another she turned her body away and looked over her shoulder back at the subject. In this new one-communicator design, chimpanzees clearly demonstrated that they were sensitive to whether the communicator was looking at them (perhaps because the situation involving only one potential recipient of a communicative act is more natural). Specifically, when the communicator was bodily facing them, chimpanzees gestured differently depending on whether her face was also oriented toward them. This sensitivity to the face contradicts the valence hypothesis. Kaminski et al. (2004) argued, therefore, that body orientation and face orientation indicate two different things for an ape when it begs food from a human being. Whereas the human being's body orientation indicates his disposition to give the subject food (i.e. when it is oriented so as to transfer food effectively), face (p. 378 ) orientation encodes information about whether the human being is able to see the subject's begging gesture.

Similarly, Hare et al. (in press) found that when chimpanzees attempt to steal food from a human being whose body is facing one piece of food but whose face is directed at a second piece of food (he is looking over his shoulder), they take the food in front of the body and avoid the one in front of the face. The finding of sensitivity to the face orientation in this case also contradicts the valence hypothesis, and strongly suggests that chimpanzees know which piece of food the human being can see. In a related study, Hare et al. (in press) used the same competitive set up but put one piece of food behind a transparent barrier and one behind an opaque barrier while the human competitor stared straight ahead. Chimpanzees preferentially selected the piece that was behind the opaque barrier, the one that the competitor could not see, thus providing further confirmatory evidence for the findings of the food competition studies of Hare et al. (2000,2001). Moreover, and importantly, chimpanzees in this competition-with-a-human-being paradigm sometimes attempted to hide the beginning of their approach to the food by using circuitous routes that hid them from the human being's view (behind the barrier) from as early in their approach as possible—which would seem to indicate a fairly sophisticated understanding of the viewing angle of the human experimenter.

The scoffer hypothesis for these competition-with-a-human-being results is that chimpanzees do not compute what the human competitor can and cannot see, but rather they simply avoid approaching food if they themselves can see a competitor's face. In a third study, Hare et al. (in press) found something different. In this study, subjects had to choose to approach food either behind: (1) an opaque barrier that obstructed the human competitor's view of the subject and the food; or (2) a split barrier (each part half the size of the barrier on the other side, so that overall area covered was the same) that occluded the subject's view of everything at eye level (including the human competitor's face) and also at lower body level—but with a gap allowing the competitor to see the subject's body approaching. This condition thus compared the hypothesis that subjects were simply going to the side where they themselves could not see the human competitor—in which case they would show no preference—with the hypothesis that they were going to the side where the competitor could not see them—in which case they should approach behind the opaque barrier. The finding was that chimpanzees preferred approaching the food that was behind the opaque barrier, indicating their knowledge that the competitor could see them approaching behind the split barrier. It is possible that in this condition chimpanzees could see the competitor's torso or feet (through the split) and so they simply tried to avoid any sight of their competitor at all. But note that this explanation does not apply to the Hare et al. (2000, 2001) studies in which chimpanzees' competed with one another while looking directly at one another the whole time; seeing the competitor in these studies did not prevent their approaching the food when they thought the other could not see it. Moreover, seeing the dominant looking during the baiting period did (p. 379 ) not prevent subordinates from approaching and taking the food when the original dominant was switched by a naïve dominant.

17.5 Self-knowledge

There is one last paradigm indicating something about chimpanzee's understanding of seeing—as opposed to just looking behaviour. Call and Carpenter (2001) presented chimpanzees and orang-utans with two hollow tubes perpendicularly oriented toward the subjects and then they placed a reward inside one of the tubes. To get the reward, subjects had to touch the baited tube (at which point the experimenter gave them the contents, if any). There were two baiting conditions. In the visible condition, the experimenter placed the food inside the tube in full view of the subject. In the hidden condition, the experimenter baited one of the tubes but prevented the subjects from witnessing the baiting process—so that subjects knew there was food in one of them but they did not know which one. Subjects looked into the tubes before choosing more often when they had not witnessed the baiting (as was necessary for them to be successful) than when they had witnessed the baiting (Fig. 17.4).

From a scoffer point of view, one could argue that somewhere in their past or in experiments subjects learned a conditional discrimination of the kind: if the location

(p. 380 ) of the food is unknown, then bend down and look inside the tubes until you find the reward. However, in another condition it was found that when the tubes are blocked from the front (and the subjects sees this barrier), apes do not attempt to look (Call, unpublished data). One could still argue that although the sight of the tubes paired with not witnessing the baiting makes subjects bend down, this response is cancelled if something is blocking the front of the tubes; from a behavioural point of view, the blockage is a cue that modulates the conditional discrimination. This is possible, but of course the alternative to this complex conditional discrimination (a skill not easily mastered by chimpanzees), based on hypothesized but not observed learning experiences is that subjects know what they have and have not seen and they look to seek information when they need it. They know something about their own seeing.¹

17.6 So what do chimpanzees really understand about seeing?

The current review has shown that chimpanzees follow the gaze of conspecifics and humans, follow it past distractors and behind barriers, ‘check back’ with humans when gaze following does not yield interesting sights, stop following the gaze of a looker who keeps looking at nothing, use gestures appropriately depending on the visual access of their recipient, know when they themselves have seen something, and select different pieces of food depending on whether their competitor has visual access to them— understanding transparent barriers and split barriers in the process. Taken together, these findings make a strong case for the booster hypothesis that chimpanzees have some understanding of what other individuals can and cannot see. The scoffer position, on the other hand, requires about a dozen different hypotheses to account for all of these phenomena—no one of which accounts for more than a few phenomena. Table 17.1 summarizes the main experimental findings, the scoffer explanations needed to account for them, and the one booster explanation that, if correct, explains them all. Note that the booster explanation is analogous to the concept of intervening variable used by Whiten (1994, 1996) to discuss the issue of‘mind reading’ in animals. Whiten argued that one viable alternative to postulating a one-to-one correspondence between sets of stimuli and responses, is to invoke a common intervening state (produced various different stimuli) that regulates the various responses.

In the debate between boosters and scoffers, parsimony is often raised as an issue. But parsimony is a notoriously protean concept. Inspecting Table 17.1, it would seem that the boosters clearly have parsimony on their side. The number of different explanations required to explain the evidence available is sensibly smaller (12 versus 1). Of course, looking at it another way one could say that the scoffers need only invoke behaviouristic principles of learning, whereas the boosters need to posit an understanding of various psychological constructs—perhaps in addition to principles (p. 381 )

of learning. But what seems to be really going on when a scoffer claims parsimony here is that they are making an appeal to something akin to Morgan's Canon: learning is a simpler mechanism—that is, a lower level mechanism shared with many animal species—whereas understanding psychological states is a more complex or higher level mechanism. However, the distinction between lower and higher mechanisms is plagued with interpretative problems. For one thing, there is no objective basis for classifying learning as a lower or simpler mechanism and insight or representational as higher or complex. In general, we are not strong proponents either of parsimony (unless one clearly defines the criteria for parsimony) or of Morgan's Canon—certainly not as substitutes for grappling with data when there is plenty of it.

It is also important to note that when learning explanations are applied to the range of phenomena in Table 17.1, the scoffer must propose many and extremely complicated (p. 382 ) learning scenarios for which there is no direct evidence. For example learning that transparent barriers do not block the visual access of others would require a series of experiences in which each subject approached food behind a barrier, took it while the dominant was looking, and was punished by the dominant when the barrier was transparent but not when the barrier was opaque. Similarly, to learn to stay away from food that has been moved from behind one barrier to another while a dominant watches, but to go for the food if the dominant was not watching, requires another set of possible but extremely unlikely learning experiences for each subject. And it is important that there is no evidence for any of these hypothesized learning experiences. They are just raised by scoffers as theoretical possibilities to account for certain experimental results. Indeed, if the true situation was that chimpanzees individually learn relations between the looking behaviour of others in certain situations and their subsequent behaviour, one would expect to see some individuals learning important things during the experiments themselves. But in none of the studies reported above are there any signs of individuals improving over trials within the experiment.

We believe that imagining possible learning scenarios in a subject's past, or even during a particular experiment, is an important exercise for making sure that experimental designs aimed at cognitive processes measure what they aim to measure and also take account of other potential hypotheses. But when we have many different experiments using many different methods—as we have in the current case—scoffer hypotheses positing all kinds of different learning accounts for different studies begins to become implausible. It is much more plausible to simply credit chimpanzees with understanding seeing.

One final point—learning explanations and cognitive explanations are often presented as mutually exclusive alternatives to explain some range of behavioural phenomena. We believe this is a theoretical mistake. For example it is often supposed that if an organism could have learned something on the basis of personal experience, then it is ipso facto not cognitive. If chimpanzees could have learned about transparent barriers through rewards and punishments, for instance, then we do not need to invoke an understanding of seeing to account for their behaviour with respect to them. But many of the most complex, sophisticated, and abstract human cognitive skills are acquired through learning. For example children take many years to learn a language, algebra, reading, and so on. When we view learning more broadly in interaction with other cognitive skills, then we can see that the real issue is: when an organism learns something, does it do so on the basis of blind associations or on the basis of some causal or intentional understanding of the situation? In the current case, it is very likely that chimpanzees individually learn some things about what others can and cannot see, but this does not mean that they do not understand seeing. This might be tested by conducting experiments to see whether an individual learns something (for example to avoid food behind a transparent barrier) more quickly or more slowly than it would learn some arbitrary contingency (for example to avoid food behind a red barrier). (p. 383 ) We are willing to bet that a naive chimpanzee in the presence of a dominant would learn to avoid food behind a transparent barrier much more quickly (if indeed learning was required at all) than it would learn to avoid food behind a red barrier if it was rigged so that the dominant could actually see the food (say, if there were a red one-way mirror) and so punish the subordinate accordingly when she took it. In fact, there is some evidence suggesting that this preference for causal as opposed to arbitrary connections is also found in physical cognition (see Call 2003, Chapter 10, this volume; Premack and Premack 1994).

Science is open-ended, and the case is certainly not closed on the issue of whether chimpanzees understand seeing. But now that the evidence is overwhelmingly in favour of the booster position, we believe that that they do. Of course negative evidence— for example failure to infer what another could see in some situation—would provide support for the scoffer position. Leaving aside the thorny issue of interpreting negative evidence, currently there are more positive than negative results, only a motley collection of alternative hypotheses—no small set of which explain all of the data—and a generic appeal to possible learning experiences that individuals may have had previous to their participation in an experiment. There are also the negative findings of Povinelli and Edy (1996b), but these have been to some degree superseded by the positive findings of Kaminski et al. (2003). We are therefore boosters on this specific question. At the same time, however, we must reiterate that understanding seeing does not mean that chimpanzees understand other psychological states; this might be the only one. The scientific situation is that we must look at each psychological state—perceiving, intending, desiring, believing, and so on—separately, and with the same kind of variety of experimental methods reported here. Whether chimpanzees understand any particular psychological state is, in each case, an empirical question.