Social Cognition: Social Intelligence

Question 1

Introduction

Answer 1

social intelligence hypothesis, the proposal that living
in primate like social groups requires exceptional cognitive abilities and is therefore associated with high intelligence, although it is still being debated. With only a few exceptions, we still do not know very much about the nature of social knowledge, how it compares across species, and how it is acquired. Is social knowledge qualitatively different from nonsocial knowledge in any way, or is navigating a large social network simply a matter of acquiring an unusually large amount of information?

In any case, other individuals are not only social stimuli but have physical features and may administer physical rewards and punishments so their companions can learn about them through domain general mechanisms. But if there are specifically social forms of cognition, highly social species might be expected to have them to an exceptional degree. To date there are few well controlled tests of this prediction. A promising way forward is with species other than primates, not only mammals such as hyenas and cetaceans (whales and dolphins) but birds and fish.

For human adults, knowing about other individuals means not only being able to predict their behavior but understanding their states of mind, that is, having theory of mind. Accordingly, the study of animal social cognition prominently includes tests of the mentalistic underpinnings of social interactions. Do animals, for instance, know that other individuals have beliefs, desires, and intentions? Section 12.3 lays out a framework for approaching it which is then applied to research on animal theory of mind in Section 12.4. Section 12.5 looks at the evolution of cooperative behavior, asking whether any examples of animal cooperation require specialized cognitive abilities or
emotional dispositions. We start, however, with a closer look at the nature of social knowledge.

Question 2

12.1 The social intelligence hypothesis

Answer 2

As set forth by Jolly (1966) and more influentially by Humphrey (1976), the social intelligence hypothesis (also called the social theory of intellect or the Machiavellian intelligence hypothesis, Byrne and Whiten 1988) proposes that social conditions in
primate social groups drove the apparently high general intelligence that monkeys and apes seem to reveal in traditional tests of concept formation, learning set, discrimination reversal, and the like as well as their exceptionally large brains in relation to body size.

Characteristic of its time, the original theory assumed that intelligence is general rather than modular. From a contemporary perspective, it might instead be suggested that cognitive adaptations for social life are, or have evolved to be, accessible to problems with nonsocial content (Rozin 1976). In any case, the original version of the social theory of intellect implies that complex social organization and general problem solving ability go together, whereas a modular
view (e.g., Gigerenzer 1997) implies that they may be independent.

In principle, comparative and phylogenetic data can distinguish among these possibilities. For instance, lemurs have a complex social organization but perform more poorly on tests of physical intelligence than Old World monkeys (Jolly, 1966). Because lemurs are
prosimians, closer to ancestral primates than monkeys, this finding is consistent with Jolly’s suggestion that social intelligence preceded the evolution of equivalent physical intelligence. But in Chapter 10 we saw evidence consistent with some independence between social and physical intelligence in that some corvids whose excellent spatial memory is consistent with their reliance on stored food are outperformed in
socially relevant tasks by less spatially adept species (Balda and Kamil 2006).

An often-proposed alternative to the social theory of intellect might be called the foraging theory of intellect. As an example of how foraging niche might select for high intelligence, consider that tropical forests are a complex mosaic of hundreds of tree species, each with its own schedule of fruit and flower production. Because fruits are typically available for a shorter time than leaves, fruit-eating species may be faced with a harder environmental tracking problem than leaf eaters. In addition, a primate troop of a given size needs a larger home range if they eat fruit than leaves since at any given time there may be less food available in it.

The foraging theory of intellect therefore predicts that fruit eaters should show evidence of greater generalized learning ability than leaf eaters. Comparison of howler monkeys (leaf eaters) and
spider monkeys (fruit eaters) yields evidence consistent with this hypothesis (Milton 1988), though of course data from two species is hardly conclusive. Comparative data on brain size provide more broadly based evidence that fruit versus leaf eating may be correlated with cognitive differences. In bats, rodents, and primates, fruit-eating species have heavier brains relative to their body weights than their leaf eating relatives (Barton, Purvis, and Harvey 1995). 

But it takes a longer gut to digest leaves than fruits, so a difference in brain:body ratios could arise because leaf eaters have relatively big bodies rather than relatively small brains. And type of food could have an indirect effect on brain size in that the young of species that eat things requiring more learning to find or process will remain dependent on their natal social
group for longer and thereby have more complex social relationships and/or opportunities for social learning.

The Social Brain Hypothesis:

As the preceding brief review suggests, the social theory of intellect has become closely bound up with discussion of correlates for the relatively large brains (more specifically, neocortical areas) of primates (Figure 12.1; overall brain:body weight ratios for some primates can be compared to those for other mammals in Figure 2.11). But although overall brain size may be convenient for comparing species, specific cognitive demands might be most strongly reflected in specific areas of the brain (see Chapter 2; Striedter 2005). We need to know more about whether brain areas involved in learning about the physical versus social environments are the same or not before neuroanatomical comparisons can be strong evidence for or against the social theory of intellect (Healy and Rowe 2007).

And in any case, sociality and tracking food availability are only two among a ‘‘bewildering’’ (Healy and Rowe
2007) array of factors that have been proposed as selecting for unusually large brains. For example, on one hypothesis (Barton 2000), much of the enlargement of primate brains is accounted for by visual areas. Unlike other mammals, primates have trichromatic color vision (Box 3.1). Red-green discrimination in particular aids in detecting
both ripe fruits and tender young leaves, so although it evidently now also functions in social behavior, as witnessed by the colorful faces and bottoms of many monkeys, color vision may have evolved in the context of foraging.

In addition, overall primate brain size can be related to innovation rate, an aspect of physical intelligence
(Box 2.2; Lefebvre, Reader, and Sol 2004). One problem is that except in a few well-studied cases we have only a sketchy idea of what social complexity consists of (Kummer et al. 1997). It is not simply a correlate
of group size because not all large groups are socially complex. Animals within a flock of birds or a herd of wildebeest may not distinguish among numerous unique individuals and multiple social roles in the way shortly to be described for some primates. However, few if any birds and few nonprimate mammals have been the subjects of the same kind of long-term field studies as some primates.

Even within primates, which index is the best proxy for overall social complexity to be correlated with brain measures is disputed. For instance, Figure 12.1 shows a nice relationship between mean overall group size in a species and its ratio of neocortex volume to
remaining brain volume, but good correlations have also been shown using other measures such as size of grooming cliques and frequency of deceptive behavior
(Dunbar and Shultz 2007). Even worse, this lack of clarity means that attempts to document the social brain hypothesis in other taxa can verge on circularity.

Thus the observation (Emery et al. 2007) that larger brains in birds are associated with pair-bonded mating systems, together with the assumption that the social brain hypothesis applies to birds, encourages speculation that long-term pair bonds entail special cognitive demands (speculation all too easy to generate from an anthropomorphic perspective). In summary, then, although the social theory of intellect and the associated social brain hypothesis have attracted a lot of attention, thinking in this area is still in flux. The special features of primate brains most likely result from more than one kind of selection pressure (Striedter 2005; Holekamp 2006; Healy and
Rowe 2007).

Progress will likely come through better information about the roles of different brain areas in social behavior and new statistical techniques allowing
multiple factors to be considered simultaneously (Dunbar and Shultz 2007) along with more thorough comparisons of social behavior and cognition within groups of related species, both primates and nonprimates.

Why be social anyway?:

The species-typical size of animal groups reflects tradeoffs among a multitude of factors related to the nature and distribution of both a species’ food and its predators. For example, hamadryas baboons (Papio hamadryas) live in the North African semidesert. Food is sparse, and during the day the baboons forage in small bands. However, the safest way for a baboon to sleep is perched on the side of a cliff, and because suitable cliffs are few and far between, as many as 200 baboons gather at night on sleeping cliffs (Kummer 1995). Many birds also congregate to nest or sleep
and disperse to forage.

In contrast, species dependent on a temporary resource that occurs in large patches, like the grasses of the African plains, forage in large migrating herds or flocks. Still other species are solitary and territorial except for breeding. Other animal social systems were sketched in Chapter 2. The same costs and benefits determine optimal group size in most species
(Section 2.1; Silk 2007a). On the benefit side, an individual foraging in a group can take advantage of others’ vigilance and thereby devote more time to feeding (Section 3.7.3). When a predator does attack, the group may be able to confuse it or drive it off.

In any case, the effect of the predator on any one individual will be diluted by the presence of others. Individuals foraging together may also help each other find food. They may be attracted to others of their species that are feeding, they may follow each
other, as ants follow each other along chemical trails, and they may learn from one another in ways to be described in Chapter 13. But group living may also increase the risk of predation and decrease access to food. For example, a group is more conspicuous
to predators than a solitary individual, and animals foraging together may interfere or compete with one another not only for food but also for mates and other resources.

Among some group-living primates, the time-consuming and potentially damaging effects of continual squabbling are minimized by observing a strict social hierarchy that defines priority of access to food and other resources. In addition, comparatively friendly relations are maintained among subgroups comprised of kin or others with alliances of some sort. These relationships are often expressed by animals grooming each other or spending time close together; their extent can be an important predictor of individual fitness (Silk 2007b).

Among primates and some other animals, all these social arrangements take place in the context of a comparatively long lifespan and prolonged dependence of the young on their mothers or other
adults. Usually members of only one sex, most often males, disperse during adolescence while members of the other stay in their natal area for life. Extended families therefore may contain grandmothers, aunts, cousins, and so on of all ages, the oldest of whom have very extensive knowledge of the local social and physical environment. The slow development that underlies this kind of social group also facilitates growth of a large brain (van Schaik and Deaner 2003; Striedter 2005). Perhaps it is not surprising, then, if large brains and complex societies go together.

Question 3

12.2 The nature of social knowledge

Answer 3

So far we have taken for granted that living in a social group presents distinctive cognitive problems. But what do animals know about their social companions? Is any
aspect of social cognition a distinctive form of cognition, as opposed to a distinctive, social, use of some more general cognitive ability (Gigerenzer 1997)? Tomasello and Call (1997) concluded from their comprehensive review of primate cognition that primates differ from other animals in being able to learn about third-party relationships, a kind of relationship they claimed as uniquely social.

As an example, consider dominance relationships. An animal’s knowledge about the dominance hierarchy of which it is a part might be entirely in terms of its own, that is, first-party, relationships, perhaps acquired through associating rewards and punishments with particular behaviors directed at particular individuals. For instance, a mid-ranking animal might learn, ‘‘If I try to displace Joe from food he moves away and I get the food; if I try to displace Pete, he threatens me.’’ This knowledge implies a third-party relationship, namely Pete is dominant to Joe, which could also be acquired just by watching Pete and Joe interact. As we will see, monkeys are sensitive to many kinds of third-party relationships, but we now know that some nonprimate mammals and some birds and fish are, too.

12.2.1 Social knowledge in primates:

Much of what we know about primates’ social knowledge comes from long-term field studies, sometimes combined with clever field experiments (see Cheney and Seyfarth 1990; Kummer 1995; Cheney and Seyfarth 2007). A sample from such research illustrates the richness of social relationships to which some primates are sensitive.

Relatedness

We saw in Chapter 6 how Dasser (1988a) used operant category learning to test whether Java monkeys had a concept of the relationship mother-offspring. Vervet
monkeys tested in the field also show evidence of associating particular infants with their mothers (Cheney and Seyfarth, 1990). When vervets hear the cries of a familiar but temporarily unseen infant broadcast from a concealed speaker, they are more likely to look toward that infant’s mother than toward some other monkey. Vervets also show evidence that they are sensitive to more remote kinship relationships among troop members.

For instance, a monkey that has recently been the subject of aggression is more likely to behave aggressively toward a relative of its attacker than
toward an unrelated monkey. Such redirected aggression is seen in many nonprimates as well as primates (Engh et al. 2005). However, while all these observations reveal social knowledge, the processes by which it is acquired need not be specifically social.
Mothers and infants are normally seen together and thereby may become associated in the minds of their companions.

Relatives may look alike (Vokey. et al. 2004) or become associated through proximity, promoting generalization from one to another. Redirected aggression suggests that the negative effects of fighting with an individual generalize less widely to similar individuals than does the tendency to fight that with animal in the first place.

Male-female relationships:

Knowing who belongs with whom or what kind of behavior to expect from A as opposed to B may be explicable as the products of learning mechanisms that are not specific to social stimuli. However, an ability to categorize interactions among specific individuals in terms of kinds of social relationships such as mother-offspring, ally, and so on, that is, to use social concepts, would enable ready generalization to completely new individuals if group membership changes (Seyfarth and Cheney 1994). A pioneer in designing experiments to tap such knowledge was the Swiss ethologist Hans Kummer, working with hamadryas baboons in Ethiopia.

In this species, males control ‘‘harems’’ of females. When large numbers of males with their females and offspring gather to sleep and rest, the large powerful males seldom fight over access to females. One male’s respect for another’s possession of a female arises
from observing the two interacting, as Bachmann and Kummer (1980) showed in the experiment depicted in Figure 12.2. The subjects were pairs of males from the same troop and females unfamiliar to them. What would the males do when placed together with a female if (a) one had previously seen the other interacting in a friendly manner with the female or (b) they had both seen the female before but neither had
interacted with her?

In the first case, as little as 15 minutes observing the pair inhibited any attempt by the second male to interact with the ‘‘married’’ female and her partner. When introduced into the enclosure with them, he sat in the corner with his back turned and groomed himself or looked at the sky or into the bushes. In the control condition, however, both males tried to interact with the female and occasionally fought over her. These results suggest that the observing male processed what
he saw as a particular kind of third-party relationship, one that dictated his staying out of the way of the second male. Quite possibly past experience had taught him to refrain from approaching a female in the presence of a possessing male, but this too requires generalizing over a class of interactions among different individuals.

Further evidence that male baboons rapidly encode information about mating associations comes from more recent work by Crockford and her colleagues (2007) with chacma baboons (P. cyanocephalus). In this species, males form temporary consortships with sexually receptive females, during which they stay close to a female and repel mating attempts by rival males. Consortships end abruptly after a few hours or days while the female is still receptive, opening mating opportunities for others. Males not currently in a consortship were played a recorded copulation call of a
female in their group preceded by a grunt from her current or recent consort while both of the consorting animals were well out of sight and earshot.

If the consortship was still ongoing and the calls came from the same location or if it had ended and the
calls came from different locations, subjects hardly glanced toward the hidden speaker(s). However, if the consortship had been ongoing and the calls came from different locations—as if the consortship had ended—the subject males looked a long time and some even headed toward the speaker from which they had heard the female’s call.

Social causation:

One might say that the males in the study just described behaved as if performing a kind of causal reasoning: ‘‘Those two are have separated because the consortship is over.’’ Another example of such social causal reasoning, this time in chacma baboon females, was provided by Cheney, Seyfarth, and Silk (1995). When a dominant female approaches a subordinate who is holding an infant in an attempt to touch or
hold the baby herself, the dominant will often emit a grunt vocalization, and the subordinate may emit a fear bark. If instead a subordinate approaches a dominant,
this sequence of vocalizations is never heard.

Dominants do not give fear barks to more subordinate individuals. Cheney et al.’s experiment tested whether baboons understand the causal relationship between status of the approaching, grunting, female and fear barking by the female being approached by comparing their reactions to causally consistent and inconsistent sequences of grunts and fear barks (Figure 12.3). In inconsistent sequences, a grunt by a subordinate individual, say F, was followed by the fear bark of a female dominant to her, say C. A consistent sequence matched to this example would also contain F’s grunt followed by C’s fear bark, but in this case a grunt by an individual dominant to C, say A, preceded C’s fear
bark.

This sequence was causally consistent because C’s fear bark could be caused by the approach of A. The consistent sequence contained more vocalizations, so it might be expected to be more salient, and attract more looking, than an inconsistent sequence, the opposite pattern to that expected if the animals are reacting primarily to the information in the sequence. Each subject heard a consistent sequence on one occasion and an inconsistent sequence on another. On average, subjects looked toward the speaker longer when an
inconsistent sequence was played. Because the stimuli were matched for features like the specific vocalizations they contained, these results seem to show that the baboons do recognize a kind of social causation in the sequences of grunts and fear barks.

To do so, they need to recognize other individuals’ calls and dominance ranks and to know in some sense that fear barks result only from the approach of a dominant
toward a less dominant animal. Another possibility, however, is that the specific consistent sequence was simply more familiar. It would be difficult for a study of this kind to escape from such an objection, since it is only through watching and listening to the interactions of other individuals that a subject monkey learns their ranks in the first place.

Alliances and rules of thumb:

Monkeys’ knowledge about kinship and dominance finds practical expression during agonistic interactions. When two individuals are threatening or fighting each other, bystanders may join relatives or others in alliances. Joining may result from recruitment, a behavior in which an animal in an agonistic interaction looks back and forth between its opponent and a bystander, the potential recruit. Since rank is power, it
makes sense to recruit allies higher ranking than one’s opponent. Similarly, a good way to be on the winning side in a fight to join the higher-ranking animal. Both of
these choices imply knowledge of third-party relationships, and indeed two species of macaques reveal such knowledge in recruitment and alliances (Silk 1999; Schino, Tiddi, and Di Sorrentino 2006).

Behavior consistent with knowledge of third-party relationships can arise for other reasons (Range and Noe¨ 2005). For example, animals might use rules of thumb like ‘‘always recruit the dominant animal’’ or ‘‘always join the winning side’’ (most likely the more dominant). Relatives may be especially likely to be dragged into each other’s disputes simply because they spend a lot of time close together. Unlike in studies with
playbacks of social interactions, information about recruitment and alliances is typically extracted from observations of free behavior. Here the only way to control for potential confounds is to start with so much data that subsets of it can be analyzed meaningfully.

A good example is the study by Silk (1999) showing that the rank of allies recruited by male bonnet macaques varies with the rank of the opponent (Figure 12.4). We see an example from hyenas in a moment. Still, inevitably a goodly proportion of interactions will be as consistent with rules of thumb as with knowledge of third-party relations (Range and Noe¨ 2005).

Multiple relationships and hierarchies:

Taken together, the foregoing information implies that some primates classify their social companions in multiple ways simultaneously, particularly in terms of kinship and dominance, but also in terms of shorter-term relationships like consortship. Evidence for hierarchical classification by family dominance rank and rank within family (Bergman et al. 2003; Schino, Tiddi, and Di Sorrentino 2006) was described in Section 6.5.5 as implying a kind of categorization not yet studied with arbitrary stimuli in the laboratory. As discussed there, learning who belongs to which family has some similarities to learning equivalence classes, but unlike with equivalence classes, the individuals within each class are still differentiated. Seyfarth and Cheney
(2003a) have suggested that what is going on is better described as hierarchical chunking of information.

12.2.2 Social knowledge in nonprimates:

Elephants, whales, dolphins, and hyenas form long-lasting groups with some of the characteristics of primate societies (see de Waal and Tyack 2003; Bshary and Grutter 2006; Connor 2007). And although it remains to be seen whether any birds or fish
form social groups involving such a multiplicity of relationships as those described for primates, some birds and even fish are sensitive to third-party relationships. Not only are studies of social knowledge in such animals fascinating in their own right, but also
they provide information that can potentially distinguish among three possible interpretations of the social theory of intellect (Cheney and Seyfarth 2005a).

(1) Primates exceed all other species in social intelligence. (2) Primate-like social cognition is seen in
any species with similar numbers and complexity of social relationships. (3) Social cognition is similar across species, perhaps because it is qualitatively not different
from physical cognition. Accordingly, studies of social cognition in species other than primates are on the increase. This section samples a few that demonstrate knowledge of third-party relationships.

Alliances in hyenas:

Spotted hyenas (Crocuta crocuta) are carnivores that live in large groups, or clans, similar to primate groups in consisting of individuals from overlapping generations in a network of kin and dominance relationships (Holekamp, Sakai, and Lundrigan 2007). Because carnivores and primates diverged millions of years ago, similarities between them in cognition and brain organization would likely reflect convergent
evolution. Like primates, hyenas join conspecifics engaged in agonistic interactions. The knowledge of third-party relationships used in doing so has been analyzed by Engh and colleagues (2005) from extensive records of free behavior just as Silk (1999)
has done for monkeys. 

In the frequent cases where the aggressor was the more dominant of the two original interactants, support by the joining hyena could reflect a rule of thumb (‘‘join the aggressor’’) rather than knowledge of relative ranks. But in a critical minority of cases a subordinate animal attacked a dominant, and here, too, joiners most often supported the dominant. Hyenas also show redirected aggression after conflicts. As in primates it is more
often directed toward relatives of the former opponent than toward other lower-ranking animals, suggesting that hyenas know about kinship as well as dominance
relationships among third parties. These observations may have been biased by relatives of the former opponent being especially likely to be nearby, but they are corroborated by the results of a playback experiment.

‘‘Whoops’’ of hyena cubs not only are recognized by the mothers of the whooping cubs, but they also elicit more looking by the cubs’ relatives than by other nearby hyenas (Holekamp et al. 1999). Dominance rank of the cub’s mother also influences looking. However, unlike with vervet monkeys, hyenas hearing a cub whooping do not look toward its mother. This could mean they do not recognize the relationship mother-cub, or they may have the requisite knowledge and not express it in looking.

Social transitive inference and eavesdropping and in birds and fish:

Some of the best evidence from any species for knowledge of third-party relationships comes from the study of transitive inference in pinyon jays (Paz-y-Mino et al. 2004) discussed in Section 10.3.3. Although the information gained from watching a familiar jay interact with a dominant stranger did not influence more than the first few seconds of the observer’s own interaction with the stranger, this is one of the few studies with nonprimate species that rises to the level of Bachmann and Kummer’s (1980) experiment with baboons as a well-controlled demonstration that animals can acquire information about third-party relationships by watching. The related study by Grosenick, Clement, and Fernald (2007) provides similar evidence for fish.

As discussed in Section 10.3.3, such data suggest that relative information is being encoded as such, perhaps along an analog scale. But the fact that birds and fish learn about the social relationships between pairs of conspecifics was already well established by studies of ‘‘eavesdropping’’ in vocal communication and territorial behavior (see P. McGregor 2005). The dictionary definition of eavesdropping mentions listening in on a secret conversation, but in animal behavior the term refers to extracting any information from the interactions of others, be it auditory, visual, olfactory, or in some other modality (Peake 2005).

Importantly—as with deception, cooperation, and other terms prominent later in this chapter—this is a functional definition. Deciding whether eavesdropping has occurred does not depend on knowing whether the eavesdropper intended or tried to acquire information, let alone whether it was doing so by stealth, nor does it imply anything about whether the animals eavesdropped upon wanted or intended to provide information. The information acquired by eavesdropping can include absolute features of individuals such as being a good mate or the holder of a certain territory, but some fish and birds also appear to learn about third-party sexual or dominance relationships by eavesdropping.

One example will illustrate the kinds of controls necessary in such studies. Male Siamese fighting fish (Betta splendens) are known for showing vigorous aggressive displays as soon as they catch sight of another male. In a study by Oliveira, McGregor, and Latruffe (1998), five males lived in a large tank subdivided as shown in Figure 12.5, with the subject male in the central compartment. After a preliminary
exposure to each of his four neighbors, the subject watched (eavesdropped on) a fight between two of them through a one-way glass (i.e., the combatants could not see the eavesdropper).

At the same time but unseen by the subject, his two other neighbors had a fight. Following the fights, both of which resulted in a winner and a loser, the subject encountered each of his four neighbors, one at a time in an order balanced across fishes, in a small transparent compartment in his own tank (Figure 12.5). He spent more time displaying toward the winner than the loser of the fight he had witnessed but showed no discrimination between the unseen combatants. The latter data importantly show that discrimination was not based on intrinsic features of the stimulus fish or after effects of their having recently won or lost a fight.

Nevertheless, to discriminate between the winner and loser of a fight an observer need not have encoded the relationship between the combatants as such. He may
instead encode the level of aggression and/or submission shown by each individual. A follow-up study by Peake, Matos, and McGregor (2006) supports this account. In their study, male fighting fish again saw two fish engaged in aggressive behavior, but instead of displaying at one another, each of them was displaying at a mirror between their tanks. When one of the ‘‘combatants’’ was made to appear less aggressive than
the other by placing the mirror farther from his tank, the observer treated that fish like a loser.

In some studies of social eavesdropping in birds, by contrast, researchers have manipulated what a witness is exposed to while keeping constant the total amount of signaling from each interactant. For example, in some species a relationship between two unseen neighbors is expressed in the degree to which a dominant bird’s songs overlap those of a submissive neighbor. Studies in which only song overlap is varied have revealed sensitivity to a relationship as such (review in Peake 2005). Similarly, the pinyon jays in Paz-y-Mino and colleagues’ (2004) study on social transitive inference saw each bird they were to encounter later both win and lose fights.

Many of the studies of eavesdropping by birds and fish were done with species such as territorial songbirds that typically interact socially with only their mate and a few close neighbors, perhaps only during the breeding season (see Cheney and Seyfarth 2005a). They therefore suggest that knowledge of third-party relationships is not confined to species living in large stable social groups. But the social lives of birds and
fish that have larger and perhaps more complex social networks are also beginning to be examined through a primate-centric lens. For example, young rooks in captive flocks form affiliative relationships expressed through behaviors such as mutual preening and food sharing.

Rooks also show redirected aggression, and preliminary evidence suggests that it is directed preferentially against the affiliates of an opponent (Emery, Seed, von Bayern, and Clayton. 2007). We look at other aspects of corvid social cognition later in this chapter, but among birds sophisticated social cognition may not be limited to corvids. Graylag geese (Anser anser) form long-term family relationships, and families form flocks with clear dominance relationships among families (Scheiber et al. 2005). Among other evidence of such relationships, family members support each other in aggressive interactions. And in fish, one of the most interesting examples of complex social networks is interspecific, in the relationships among cleaner fish and their clients (Section 12.5; Bshary and d’Souza 2005).

Conclusions:
In conclusion, the research on social knowledge in nonprimates sampled here seems consistent with the conclusion that primate-like social cognition is not unique to primates. However, primates may still excel in the multiplicity of qualitatively different relationships to which they are sensitive. For example, a female chacma
baboon can be at the same time a mother to a particular youngster, a member of a matriline, a member of a within- and a between-family dominance hierarchy, and in a friendship or consortship with a particular male. Some of her relationships are life-long, others are temporary, some are transitive, others intransitive (Cheney and Seyfarth 2005a, 2007).

Yet she and the many other baboons in her troop seem to know all these things about one other. The ability of primates to acquire and deploy such knowledge may or may not be the same ability reflected in their performance in tests of abstract concept learning and transitive inference in the laboratory. In any
case, it remains to be seen whether long-term, in-depth studies of any nonprimates comparable to those done on a few primate species will reveal comparably sophisticated social knowledge.

12.2.3 Comparing social and nonsocial intelligence:

Having sampled the extensive information about what goes on in primate social groups and similar groups of other animals, we can ask whether social knowledge
differs in any way from knowledge about the physical world. This question is distinct from that addressed in the next sections of the chapter, namely does social behavior involve social causal understanding such as theory of mind? Here we ask simply, do social situations have a distinctive abstract structure and/or do they engage distinctive learning mechanisms by virtue of their social content?

To see that neither of these questions need have an affirmative answer, think back to the discussion of associative learning and performance rules in Chapter 4 and consider fear conditioning and conditioned taste aversion in rats. Like other examples of associative learning, both are engaged by predictive temporal relationships between events, but the nature of those events determines both the relevant temporal
parameters and the behavioral outcomes of experience. Thus, in fear conditioning a close temporal relationship between an exteroceptive signal and shock engages freezing, escaping, and the like, whereas in conditioned taste aversion, experiencing a flavor minutes to hours before gastric distress engages rejection and other disgust responses.

One might similarly try to define a social behavior system or module engaged by predictive relationships among particular social stimuli and ask how the conditions for learning compare to those in conditioning. So far only a handful of provocative examples suggests what such an analysis might yield.
Observations of vervet monkeys in the field suggest that they know much less about the physical than the social world (Cheney and Seyfarth 1990). For instance,
vervets may have watched as a snake slithering by left a trail on sandy ground, but show no apprehension on encountering a trail in the absence of a snake.

Here associative learning expected on the assumption that a fresh trail means a snake is nearby does not seem to occur. Vervets appear to be good social psychologists but poor naturalists (Cheney and Seyfarth 1990). Cheney and Seyfarth (2007) suggest that in addition to possibly learning more quickly about social than nonsocial events, some primates may be predisposed to attend to and learn about dominance and kinship. They support this suggestion with charming accounts of goat-herding baboons that learned spontaneously which kids belonged to each mother goat.

But another legendary baboon learned to help a disabled railway signalman by operating switches on the tracks, seemingly more physical than social learning. And as we see in Chapter 14, various animals learn the meaning of the alarm calls of other species, in
that they apparently associate calls with the presence of particular predators and behave appropriately.
Returning to social and nonsocial tasks with similar logical structures, recall the comparative studies of transitive inference in scrub jays and pinyon jays from
Chapter 10. The highly social pinyon jays learned faster and performed in a more monkey like way on operant transitive inference with colors.

This is consistent with the notion that the operant task taps a social cognitive ability, much as operant spatial
memory tasks tap the same ability shown in retrieving hoarded seeds (Chapter 8). However, pinyon jays acquire genuine social transitive inference far more quickly than even the first few items in the physical task, suggesting that the tasks tap different abilities. Clearly, however, this comparison is confounded by all sorts of differences between the social and physical tasks. For instance, one involved observing conspecifics; the other, operant conditioning. Even if it were possible to make the conditions for acquisition more similar, equating the salience of the stimuli involved could be an insurmountable challenge.

Other jays interacting may grab another jay’s attention
more than any physical objects. This consideration creates serious obstacles to deciding whether transitive inference, at least in nonhuman animals, is a specifically
social cognitive process engaged only weakly by nonsocial stimuli or whether it is a domain-general ability used with any sufficiently salient stimuli.
What it means to compare social and nonsocial reasoning about identical materials is illustrated by studies of the Wason selection task with human subjects (see Cosmides and Tooby 1992).

As originally studied by the psychologist Peter Wason,
this task requires people to look for violations of a logical rule of the form ‘‘If p then q.’’ A subject is given the four cards shown in Figure 12.6A and asked which ones need to be turned over to detect violations of the rule, ‘‘If a card has p on one side it has q on the other.’’ Most people turn over the card with p on the front to see if it has q on the back. Very few turn over only the one necessary additional card, the one with not
q. Familiar, less abstract, content doesn’t always improve performance, but in the example shown in Figure 12.6B, as many as 75% of subjects can detect whether people are drinking illegally by turning over the correct cards.

According to Cosmides (1989) and others (see J. Evans 2002), the reasoning ability needed here evolved to detect cheaters on social contracts in early hominid society, and the drinking age problem taps into it. Reciprocal altruism depends on participants obeying general rules of the form, ‘‘If you take a benefit, you pay a cost.’’ For instance, ‘‘If I share my meat with you, you help me gather wood.’’ Cheaters take the benefit without paying the cost, that is, they satisfy the logical condition, ‘‘p and not q.’’ The view that reasoning in the Wason task reflects a cognitive adaptation for social exchange has been supported by the results of experiments in which people are asked to reason about identical statements in a social context versus another sort of context.

To control for familiarity of content, Cosmides told Harvard students elaborate stories about fictitious tribes and their customs. Similarly, Gigerenzer and
Hug (1992) told subjects stories about people visiting a mountain cabin. When solving the Wason task in these contexts, subjects were much more often correct
when ‘‘If p then q’’—for instance ‘‘People who stay overnight in the cabin must bring firewood’’ (Figure 12.6C)—was framed as a social contract than when it was framed as a description of social customs.

The approach to human reasoning exemplified by studies of the Wason selection task has been applied to reasoning in other areas, to see whether other examples of apparent irrationality are actually ‘‘ecologically rational,’’ that is they make sense when seen as evolved to solve ecologically relevant problems (Todd and Gigerenzer 2007). However, not surprisingly for a task that was already much studied before any evolutionary theorizing about it, the evolutionary psychologists’ view that performance on the Wason task reflects adaptations for social transactions is not universally accepted (cf. J. Evans 2002). Nonetheless, like social versus nonsocial transitive inference in the scrub jay, it shows how one might test the idea that similarly structured problems with different content tap qualitatively different cognitive processes.

Question 4

12.3 Intentionality and social understanding

Answer 4

12.3.1 Levels of intentionality

‘‘I’ll pick up the children from school today,’’ says Max as he leaves for work. We’d normally say that Max’s statement conveys an intention. We can predict that he will drive from his office to the school by a certain route at a certain time and that he will change his behavior if the circumstances change. If he is working away from the office, he’ll travel by a different route and start out at a different time; if the road is blocked, he’ll make a detour; if the car breaks down, he’ll walk or take a taxi. That is to say, his behavior will be flexible, directed by the goal of being at the school on time.

A philosopher might say that Max exhibits intentionality, but she would not mean that Max has intentions in the everyday, folk-psychological, sense. Intention in the
philosophical sense is the property of aboutness (Dennett 1987; Allen 1995; Dennett 1996). Intentionality, being about things, is perhaps the defining property of mental states. Beliefs and desires, plans, understandings, and wishes, as well as intentions, are
examples of intentional states. A belief, for instance, has to be a belief about something. A distinguishing feature of intentional statements is that they do not obey the usual logical rules of substitutability.

For instance, Max is Susie’s father, and Max is a
man born in 1950. It follows logically that Susie’s father is a man born in 1950. However, Susie can believe that Max is her father without necessarily believing that a
man born in 1950 is her father. If we ask whether nonhumans have intentions, beliefs, desires, or the like we are asking whether they are intentional systems. Asking this question means formulating clear criteria for what an animal with a certain sort of intentional state does, just as in the tests of belief, desire, and planning discussed in Chapter 11.

Philosophers distinguish a hierarchy of orders of intentionality (Dennett 1983). In terms of this hierarchy, an animal that does not have beliefs, desires, and the like, that is, one that does not in fact have intentional states, is exhibiting zero-order intentionality. Systems of responses to stimuli have zero-order intentionality.
A creature that has beliefs, desires, and the like about the real or imagined physical world or the behavior of others is a first-order intentional system. When its mental states concern the mental states of others, we have graduated to second-order intentionality.

Thus if Max plans to arrive at the school on time, he has first-order intentionality. If he believes that the children know he is coming for them today, he is exhibiting second-order intentionality. If he wants them to believe that he expects them to be waiting for him, then he is exhibiting third-order intentionality. Level can be piled on level endlessly in this way, but in dealing with animal behavior, it is enough (usually more than enough) to wonder whether one individual is capable of
having beliefs or desires regarding another’s beliefs and desires (i.e., second-order intentionality), regarding only others’ behavior or physical states of the world (first-order intentionality), or neither of those (zero-order intentionality).

In predicting what Max will do when circumstances or his own beliefs and goals change, we are taking the intentional stance (Dennett 1983, 1987). That is, we are
using the assumption that he is an intentional being to predict and explain his behavior. Most of the time the intentional stance accounts very well for the behavior
of other adult human beings. It often provides useful rough and ready predictions of other species’ behavior too, but experiments are needed to test them. In Chapter 11 we saw evidence for a first-order intentional account of rats’ bar pressing: a hungry rat presses a bar because it both knows pressing leads to food and wants the food.

In the arena of social cognition, folk psychology often suggests that animals have second-order intentionality, that is, knowledge or belief about what other individuals know, believe, desire, or intend, but, as illustrated in Box 12.1, a careful theological analysis may provide a full account of the behavior involved without invoking any form of social cognition as such. An example closer to those analyzed later in this chapter is the situation depicted in Figure 12.7: a subordinate male baboon moves behind a rock to solicit sexual contact with a female while out of a dominant male’s sight, as if aware that the dominant will not know what he’s doing there.

The numerous cases of such deceptive behavior in
primates described by field workers have been taken as evidence of ‘‘Machiavellian intelligence’’ (Whiten and Byrne 1988). But how can we tell whether the subordinate male’s behavior is based on a belief about what the dominant sees or knows? Many animals are sensitive to the direction of other animals’ gaze (Box 12.2), as if possessing a low-level perceptual module that detects what other animals are looking at (not
the same as the mentalistic ‘‘what they are seeing’’). 

Our subordinate baboon may well be going behind the rock because he has learned that he escapes punishment for approaching certain females if he is out of the dominant’s line of sight; that is, his behavior can be explained as a response to observable cues such as where the dominant is facing. First-order intentionality is likely involved: the subordinate wants to groom the female undisturbed. But because anything the subordinate might do in response to the dominant’s seeing or knowing is inevitably a response to his looking or other behavior, second-order intentionality—the subordinate knows what the dominant sees or wants the dominant to believe he is just sitting doing nothing—is difficult or impossible to prove.

Indeed, as discussed in Section 12.4, Povinelli and his
colleagues (Povinelli and Vonk 2004; Penn and Povinelli 2007b) have argued at length that no existing data can distinguish between inference about unobservable mental states of others and inference based on their behavior, facial expressions, and
so forth. Although, this argument does not deny second-order intentionality to animals so much as assert the extraordinary difficulty of proving it, the same group (see Chapter 15; Penn, Holyoak, and Povinelli 2008) now claim that no animals possess the requisite representational abilities for theory of mind.

In any case, saying that animals respond to each other on the basis of behavioral cues rather than mental
states inferred from those cues need not imply that animals treat animate beings as they treat physical objects. Indeed, as we see next, certain kinds of motion trigger perception of animacy and intentionality in human babies, and other primates may also possess such a social perceptual module.

12.3.2 Perceiving animacy and intentionality

Infants’ and toddlers’ implicit knowledge of physical causality has been tested by showing them a cartoon in which, say, a red ball moves in from the left and collides
with a stationary green ball and comparing their looking times to physically impossible versus possible sequels to this event. A possible sequel might be Red stopping and Green moving away to the right, as if Red transferred its momentum to Green. A physically impossible sequel might be Red starting back toward the left with Green close behind it. Even very young infants display considerable implicit knowledge of
physical causality in such looking time tests (Spelke and Kinzler 2007).

More to the present point, the physically impossible sequence just described would be characterized by adults as a social interaction, Green chases Red (Heider and Simmel 1944). Young children, too, attribute intentional states to very simple inanimate objects
moving in certain ways (Scholl and Tremoulet 2000).
Animate objects—most importantly conspecifics, predators, and animal prey— differ from inanimate ones in that they are self-propelled, they can be influenced from a distance without physical contact, and they have goals and intentions. Even infants have expectations specific to self-propelled objects (Scholl and Tremoulet 2000), and this has been taken as evidence for a low-level social module triggered by perception of certain kinds of motion (Gigerenzer 1997). 

Stimuli for one test of this notion are shown in Figure 12.8. Infants watched a small circle ‘‘jump over a barrier’’ and approach a large one. After habituating to this display, they saw one of two test displays in which the barrier no longer separated the two circles. In the old action condition, the small circle still jumped, whereas in the new action condition, it approached the large circle in a straight line. The infants looked longer at the old action than at the new one, but if they had been habituated to the ball jumping without a barrier, this pattern of data was reversed. In effect, they behaved as if representing the moving ball as a rational being approaching a goal and expecting it to take the shortest path available.

In one of the few attempts to test for the same kind of encoding in another species, Uller (2004) found similar results in four young chimpanzees. Not only does self-propelled motion of a lone object trigger a perception of animacy and goal-directedness, displays with more than one such object trigger perception of social interactions (Heider and Simmel 1944; Scholl and Tremoulet 2000; Barrett et al. 2005). Materials from one study with children are sketched in Figure 12.9 (Dasser, Ulbaek, and Premack 1989).

In an experimental sequence, the big ball and the small one entered the screen together, the smaller one ‘‘fell down the cliff’’ and bounced around frantically, the big one descended and ‘‘helped it up,’’ and they left the screen together. A control sequence consisted of this series of events in reverse order. Children of about three years old looked longer at the experimental
than at the control sequence. Furthermore, when the roles of the balls were reversed, children previously shown the experimental sequence looked longer than those previously shown control sequences. Even preverbal infants seem to discriminate between simple shapes (with eyes) that ‘‘help’’ as in the sequence just described, or ‘‘hinder.’’ They prefer a ‘‘helper’’ (Hamlin, Wynn, and Bloom 2007).

No tests of primates with similar simplified social stimuli appear to have been reported, but some with pigeons have. Reasoning that interactions between predators
and prey should be salient to a vulnerable animal like a pigeon, Goto, Lea, and Dittrich (2002) trained pigeons on a food-rewarded discrimination between displays
with four dots moving around at random and displays in which one dot slowly approached one of three others, a display that to people evokes a predator stalking
prey. Even after more than 2000 trials, the birds averaged less than 65% correct, suggesting that if pigeons do discriminate intentional from random movement, it is not a salient feature of these displays.

When it comes to perception of intentionality, human gestures have a special status for human babies as young as five or six months (Woodward 1998). In the elegant study of looking times depicted in Figure 12.10, babies saw two toys, here a teddy bear and a ball. The babies were habituated to a hand reaching in from the side and grasping a particular toy. On the test trial the positions of the objects were switched, and the hand reached in again. Now it either grasped the same toy as before, which required a new action, or it performed the old action and grasped the other toy.

Babies looked more at the ‘‘new object, old action’’ event, as if they had encoded the action in terms of its goal (grasping a particular toy) and were more surprised to see the goal change than to see a new action performed. This effect did not occur when
the hand was replaced by a sponge on a stick or a mechanical claw, suggesting that it is specific to human actions. In an attempt to see if nonhuman primates behave similarly, a person touched one of two containers or performed an ‘‘unintentional’’ action such as letting their hand flop against it (J. Wood et al. 2007).

The animals (cotton-top tamarins, rhesusm macaques, and chimpanzees) were then tested to see whether they would look for food in the container that had been touched. Aside from the fact that it is not clear why they would be expected to prefer that container anyway, preferences for the one touched ‘‘intentionally’’ could have been based on past experience seeing people use similar actions to put food into containers (indeed the tamarins tested had been trained extensively to use the intentional action as a cue). In any case, the fact that some animals can apparently predict the outcomes of interactions from a partner’s body language and show signs of frustration when the predicted outcome fails to occur does not mean they ‘‘understand intentions’’ in a mentalistic way. The same obviously goes for the infants in similar studies.

In summary, the evidence sketched here shows that animate, potentially socially relevant, objects are discriminated from inanimate ones at a very basic level even by very young infants. The tendency to treat self-propelled objects as goal-directed, a ‘‘teleological stance’’ (Gergely and Csibra 2003), may contribute to the later development of a mentalistic understanding of others’ goals and desires but is distinct from it and could be shared with other species. In any case, knowing an individual’s goals is distinct from understanding their knowledge or beliefs and from understanding that they have a mental representation of the goal (Perner and Ruffman 2005), but here too, direct perception of simple cues has a role. Individuals of many species acquire knowledge visually, by directing their gaze at things. Accordingly, as shown by a large body of comparative research summarized in Box 12.2.

Question 5

12. 4 Theory of mind

Answer 5

12.4.1 What is theory of mind?

Research on theory of mind stems from a single innovative article (Premack and Woodruff 1978) that inspired a veritable industry of research in developmental and comparative psychology (Carruthers and Smith 1996; Povinelli and Eddy 1996a; Heyes
1998; Wellman, Cross, and Watson 2001; Penn and Povinelli 2007b). As introduced by Premack and Woodruff (1978), having a theory of mind means imputing mental states to others. Theory of mind is evident in intentional deception, using others to gain
information by imputing goals, knowledge or belief to them, switching roles, and communicating with intent to inform, among other ways.

In the classification of intentional states, theory of mind implies second-order intentionality. Premack and Woodruff (1978) described a series of tests of whether the chimpanzee Sarah imputed intentions to humans. Sarah, a very special animal with more than 10 years’ experience in laboratory tests of cognition, watched short videos in which an actor was thwarted in accomplishing a goal like reaching a banana outside a cage, plugging in a heater, or washing a floor with a hose. The video was stopped and Sarah was allowed to choose between two photographs, one showing the actor about to reach the goal and one not.

For instance, the actor might be picking up a long stick to reach the banana or a short one, connecting an intact or a broken hose to a tap. More often than not, Sarah chose the picture showing the action and/or object appropriate to the goal, as if she imputed desires and beliefs to the actor. The fact that she did this in a variety of physically different situations is consistent with behavior arising from a theory of mind. But because she had extensive experience watching people do everyday tasks, she may have been simply choosing the picture that completed a familiar sequence.

Premack and Woodruff sketched several other methods for testing whether a creature’s theory of mind extends to imputing knowledge and ignorance to others,
but researchers in child development reported the first relevant data (Wimmer and Perner 1983) using what is still (Newton and de Villiers 2007; Penn and Povinelli
2007b) regarded as the acid test of theory of mind, the false belief test. Importantly, to pass the false belief test the child must understand that others’ beliefs can differ from their own and from the true state of the world.

For example, a young child is introduced to a puppet or a person, say a puppet clown. The child and the clown
watch as the experimenter hides a treat or a toy (Figure 12.11). (‘‘Where is the teddy? … In the green box.’’) Then the clown leaves the scene and the child alone sees
the experimenter move the object. (‘‘Now where is the teddy? … In the purple box.’’) Then the clown returns and the child is asked, ‘‘Where will he look for the teddy?’’
Surprisingly, until they are about four years old children predict the ignorant stooge will look where they themselves know the object is (‘‘He will look in the purple box’’).

They do not seem able to separate their own representation of the situation from another’s, or at least they are unable to inhibit the tendency to report on their own knowledge. This finding appears in a variety of situations, and has stimulated much research and theorizing about the young child’s theory of mind and how it develops (Wellman, Cross, and Watson 2001; Perner and Ruffman 2005).

12.4.2 Do chimpanzees have theory of mind?

Object choice tasks:

One way to test whether, like four-year-olds, nonverbal animals appreciate that seeing leads to knowing is to give them the choice of using information provided by
two informants, one of whom has been observed getting access to that information while the other has not (Premack 1988). Thus the subject chooses between two objects as in the false belief task, but the demands of this object choice task are simpler because there is no need to keep in mind both one’s own and another’s beliefs or inhibit reporting on the true state of the world. Povinelli and his colleagues pioneered use of object choice tasks with chimpanzees in an influential series of
studies. In the initial experiment (Povinelli, Nelson, and Boysen 1990), there were four food containers, each with a handle that the animal could pull to get the food.

As a trial began, the containers were behind a screen, and the animal watched as one experimenter baited a single container in view of a confederate (the Knower), while a second confederate (the Guesser) was out of the room. Then the Guesser returned, and Knower and Guesser each pointed to a container as the chimpanzee was allowed to make a choice. A creature whose theory of mind encompasses the understanding that seeing conveys knowledge would obviously choose the container indicated by the Knower, and in fact, all four chimpanzees tested chose correctly some of the time.

However, because the experiment went on for 300 trials, the animals had plenty of opportunity to learn a conditional discrimination: choose the person who was in the room when the container was baited. Consistent with this interpretation, choices of the Knower increased across trials. The study therefore concluded with a novel
transfer test in which both Knower and Guesser were in the room when the container was baited, but the Guesser had a paper bag over his head. Over all 30 trials of this test, three of the chimpanzees still chose the Knower. However, they chose randomly on the first two such trials (Povinelli 1994), a result more consistent with the conclusion that they quickly learned to choose the person who had not worn a bag than with knowledge attribution.

Importantly, object choice does seem to be a good test of theory of mind development in children: in a similar procedure more four-year-olds than three-year-olds consistently chose the Knower (Povinelli and deBlois 1992). These negative findings were only the beginning of the quest for theory of mindlike abilities in chimpanzees. The next stage was to simplify the Knower/Guesser paradigm into a more direct test of knowledge attribution that did not require remembering where different individuals were looking. In this test, animals need only appreciate that gazing at something (Box 12.2) means seeing it, that is, having some
knowledge about the thing being looked at. This approach was taken by Povinelli and Eddy (1996a, 1996b, 1996c) with young chimpanzees.

The animals were trained to gesture through a hole in a transparent plexiglas wall toward an experimenter
holding food on the other side (Figure 12.12). The wall had two holes, with the experimenter positioned in front of one. The animal received the treat only for reaching through the hole closer to the experimenter. To be sure the animals were attending to what was being offered, occasional probe trials were administered in which one experimenter sat near each hole, one holding a block of wood and the other, a food treat. Both experimenters looked straight ahead, at the plexiglas wall, not attempting to meet the chimpanzee’s gaze.

Once the animals discriminated very reliably between the two experimenters on probe trials, Povinelli and Eddy started a series of tests in which both experimenters held out food but one could clearly see
the chimpanzee as before, whereas the other could not. For instance, the ‘‘non-attending’’ experimenter might be wearing a blindfold, have her hands or a cardboard screen over her eyes, or her back to the chimpanzee. Controls for having something unusual on the face included the ‘‘attending’’ experimenter having a
blindfold or hands over her mouth or a cardboard screen beside her face. Such tests were intermixed with normal trials and occasional probes with a block of wood versus food.

The chimpanzee was always rewarded for begging from the experimenter who could see. The surprising result of these experiments was that in nearly every type of test
the chimpanzees gestured as often to the experimenter who could not see them as to the experimenter who could. This was despite the fact that the animals performed well in the continuing regular trials and probes with the block of wood. The exception to random behavior was that when one experimenter had his back turned, the chimpanzees tended to choose the experimenter who was facing them. Even this did not indicate an understanding of seeing as attention because when both experimenters turned their backs while one looked over her shoulder toward the subject, choice reverted to random.

Over trials of all kinds, however, the proportion of choices of the ‘‘attending’’ experimenter gradually crept above 50%, indicating that the chimpanzees were learning, perhaps to choose the person whose eyes were visible. Povinelli and Eddy concluded that although their subjects were very good at detecting where someone was looking (Box 12.2), chimpanzees do not understand seeing as attention and/or knowledge but can learn to direct behavior selectively to people who are looking at them.

Even on the view that experiments like Povinelli and Eddy’s cannot in principle establish more than chimpanzees’ sensitivity to behavioral cues, their animals’ failure to respond more readily, if not spontaneously, to such cues is surprising in the light of the sophisticated social cognition generally attributed to chimpanzees (e.g., by Byrne and Whiten 1988; Whiten and Byrne 1988). Predictably, then, many reasons were found to challenge their conclusion (e.g., Gomez 1996; P. Smith 1996; Tomasello, Call, and Hare 2003). Two
questionable aspects of their procedure turn out to be key.

First, the animals in Povinelli and Eddy’s experiments were confronted with potentially helpful individuals whereas in nature chimpanzees may more often compete than cooperate over food (but see Penn and Povinelli 2007b). Moreover, because the helpful individuals were humans rather than other chimpanzees, the experiments tested the (captive)
chimpanzees’ theory of the human mind, or at least their ability to take behavioral cues from humans. Hare and colleagues tackled both of these issues in a new series
of experiments.

Chimpanzees compete for food:

In the setup developed by Hare and colleagues (2000) two chimpanzees, one dominant to the other, are in cages on either side of a central area with one or more barriers or containers where food can be placed. In the study depicted in Figure 12.13, two pieces of food are placed in the central cage while the doors to the side cages are closed. One is visible to both animals whereas the other is visible only to the subordinate because the barrier shields it from the dominant’s view. When the doors are open, will the subordinate prefer to head toward the piece of food the dominant
cannot look at? This question was answered in the affirmative for several variants of the situation in Figure 12.13 (Hare et al. 2000).

The same was not true of capuchin monkeys in a similar test (Hare et al. 2003; see also Burkart and Heschl 2007). With the chimpanzees, various controls ruled out possibilities such as the subordinate preferring to eat near a barrier. However, these effects are not evident in all measures of the subjects’ behavior and depend on details of the setup that influence the intensity of competition. For example, if the pieces of food are both closer to the dominant or are so close together that one animal can easily reach both, subordinates show no
preference (Karin-D’Arcy and Povinelli 2002; Bra¨ uer, Call, and Tomasello 2007). Not only may chimpanzees respond as if knowing what another can see in a
competitive situation, they may remember what others have looked at, leading them to behave as if knowing what others know (Hare, Call, and Tomasello 2001).

Food was hidden in a setup like that depicted in Figure 12.13 while the subordinate animal watched. The critical variable was whether the dominant was also watching, that is, whether it could know where the food was and thus be a strong competitor when both animals were released. Subordinates were more likely to obtain the food when the dominant had not seen it hidden. Moreover, subordinates discriminated between a
dominant present during the baiting and one that had not seen it, getting the hidden food more often in the latter case as if knowing what a particular animal knew.

They did not, however, discriminate in a more demanding situation that required remembering which of two pieces of food the competitor had seen being hidden. Chimpanzees’ sensitivity to others’ behavior in these tests may indeed be favored by the situation being one of competition rather than by the fact that it involves conspecifics instead of humans. This conclusion is supported by the finding (Hare and Tomasello 2004) that subject chimpanzees were somewhat more successful in choosing the correct container in a simple object choice task when a human or a chimpanzee ‘‘informant’’ behaved in a competitive rather than in a cooperative manner toward them.

Because some of the reaching and pointing cues used by the humans were similar in the two contexts, these findings suggest that competition enhances either the salience of such cues or chimpanzees’ motivation to attend to or use them. Notice that this says nothing about theory of mind but a great deal about
predispositions to respond to and/or remember certain kinds of behavioral cues. The importance of such predispositions is underlined by the success of dogs in
similar tasks.

Dogs take cues from humans:

Pet dogs are very good at locating hidden food when a person points to it or gazes toward it (reviews in Hare and Tomasello 2005; Miklosi 2007; Udell and Wynne 2008; Reid 2009). Indeed, they perform substantially better with a variety of human communicative cues than do apes tested in a comparable way. For example, in one study (Bra¨uer et al. 2006) dogs chose the container a person pointed to in 90% of trials
whereas chimpanzees and bonobos chose it only 60% of the time.

Several explanations have been proposed for dogs’ skill in such tasks: canids versus primates, domesticated
versus nondomesticated species, more versus less experience with human cues (Rei 2009). One prominent early study (Hare et al. 2002) indicated that domestication was key. Puppies from a kennel immediately responded to gazing, pointing and the like, but captive wolves with extensive exposure to humans did not. However, when the testing conditions are more rigorously equated across groups, wolves with exposure to people can outperform pet dogs, and stray dogs from shelters perform very poorly (Udell, Dorey, and Wynne 2008).

These findings implicate pet dogs’ extensive experience with people, especially with their hands as sources of food (Wynne, Udell, and Lord 2008; Reid 2009). But this does not mean the genetic changes accompanying domestication are unimportant. Domestication of dogs may have involved increasing their usefulness to people by selecting for responsiveness to cues from humans. Alternatively, highest reproductive success may have gone to the animals least fearful and aggressive toward
humans, with responsiveness to human behavioral cues as a byproduct. Evidence for this hypothesis comes from the finding that foxes selected for 45 generations for low fear and aggression toward people, but not control foxes, behave like dogs in an object choice task with human cues (Hare and Tomasello 2005).

Most likely both domestication and experience are important: selection for attentiveness to human actions may have led to a propensity for rapid early learning about human cues (Reid 2009). Clearly more extensive studies are needed of how dogs’ and other canids’ responsiveness to humans develops from a very early age. In the meanwhile, the discussion has strayed
away from theory of mind into analysis of how animals respond to cues from people. One way back toward theory of mind is to look at natural situations in which animals behave as if knowing what their conspecifics see or know.

12.4.3 Food storing birds remember who was watching

Some socially living food-storing birds are able to remember where they saw their companions caching food (Section 7.4.2). To protect its caches from thieves, a store in such a group should attend to whether others can see it while it is caching and have strategies for reducing the chances that observers later pilfer its caches. A number of food storing species use such strategies (Dally, Clayton, and Emery 2006; Pravosudov
2008). The cognitive processes involved have begun to be analyzed in two of them, ravens and Western scrub jays. The examples summarized here (for others see
Clayton, Dally, and Emery 2007) show that food-storing corvids are very good at detecting and remembering what others have watched.

As a result they can equal chimpanzees in behavior consistent with theory of mind. Figure 12.14 depicts the setup for an experiment in which a captive raven cached meat in a large aviary while two flockmates, both subordinate to it, were in separate cages at the side (Bugnyar and Heinrich 2005). One could see the cacher and thus could potentially pilfer the caches, whereas the other’s cage was enclosed by curtains.
Five minutes after the caching trial the cacher was returned to the aviary either alone or with the observer or nonobserver. As would be predicted if the cacher remembered which bird observed the caching and treated it as a potential pilferer, subjects retrieved more of their caches in an observer’s presence than when alone or with a nonobserver.

This effect was evident primarily when the second bird was close to caches. That is, subjects were quick to retrieve caches that a knowledgeable competitor was approaching but, if anything, in the presence of an ignorant competitor they selectively retrieved caches at a distance from the second bird. Observers tested alone
did in fact know where the caches were and nonobservers did not, as evidenced by differences in their latencies to pilfer the caches. Latencies of observers and nonobservers did not differ significantly in trials when the cacher was present, mainly
because the observer was slower to approach the caches in those conditions.

Thus cachers may not have been able to detect observers on the basis of their approach behavior but rather needed to remember that they had been watching. Nonetheless, the raven subjects themselves might have been detecting behavioral cues from potential pilferers that were not obvious to people. The same reservation applies to the second experiment in this study, in which the focus was on whether potential pilferers remembered which other ravens had observed caches being made and thus might compete to pilfer them.

In any test of any animal’s ability to discriminate between a ‘‘knower’’ and a ‘‘guesser,’’ if the roles of knower and guesser are played by animals that do
genuinely differ in their knowledge, it is difficult to rule out the possibility that subjects are responding to some subtle differences in those animals’ behavior in
the test. This problem was tackled by Dally, Emery, and Clayton (2006) in a study with scrub jays that, like Bugnyar and Heinrich’s (2005) study with ravens, tested birds’ memory for who had witnessed a particular episode of food caching. Here, the birds cached in distinctively marked trays in their home cage.

Observers in an adjoining cage never actually competed to retrieve caches but simply watched (or not) at retrieval. The primary measure of whether they were treated as potential competitors was the number of items subjects moved to new locations during retrieval. This recaching behavior is thought to reduce
pilfering by observers by overloading or interfering with memory (DallyClayton & Emery, 2006). An earlier study (Emery and Clayton 2001) indicated that scrub jays need experience pilfering the caches of others to show
this response to potential pilferers of their own caches, as if projecting their own experience onto others.

However, because only small amounts of recaching were shown in this study, and the birds all had extensive past experience in other studies, this provocative finding needs replicating with naive birds pretested for
recaching and then assigned randomly to pilfering or control experience. To return to the study of Dally, Emery, and Clayton (2006), here jays recovering
items in private recached more of them when caching had been observed by a dominant bird than by its partner, by a subordinate, or no bird at all. Because these differences occurred in constant conditions, with the subject alone, they are consistent with memory for the type of bird present during caching.

To test whether that memory was for a specific individual, in a further experiment jays cached in two
different trays, each watched by a different observer, one of which was present while the subject could retrieve from both trays (Figure 12.15). Subjects recached significantly more items from the tray that bird had observed than from the tray observed by a different bird. Of course an observer may have looked in some special way at the tray that it knew had been cached in or otherwise behaved differently toward it than toward the second tray.

However, when subjects were watched at recovery by a bird that had seen a different subject cache in one of the same trays, recaching was at a low level and did not differ between the trays. These findings indicate that in addition to remembering what, where, and how long ago they cached (Chapter 7), scrub jays remember who was present when they cached in particular places and behave as if aware of other individuals’ knowledge.

Of course the cacher’s behavior could instead have reflected different behaviors by the control and the actual observers, but this difference in itself would mean that scrub jays (in this case observers) know who
cached where. In any case, it appears that scrub jays, like ravens and probably other corvids (see Clayton, Dally, and Emery 2007), have detailed social knowledge that they deploy in defending their caches from potential pilferers.

The procedure for the experiment with scrub jays by Dally, Clayton, and Emery (2006). Circles around caching trays indicate Plexiglas covers, forcing the subject to cache in just one tray. Square surrounding a tray in the recovery phase indicates the tray in which caching has been observed by the observer present for that test. After Clayton, Dally, and Emery (2007) with permission.

12.4.4 Behavioral abstractions or theory of mind?

The fundamental question about animal theory of mind is whether animals reason about others’ mental states or respond to their behavior alone. As Premack and
Woodruff (1978) put it, ‘‘Is the chimpanzee a behaviorist or a mentalist?’’ But because inferences about mental states are based on current or remembered behavior it is impossible to be only a mentalist. As we have seen, no existing data demand explanation in terms of theory of mind, but neither do they conclusively rule it out (see also Cheney and Seyfarth 2007; Penn and Povinelli 2007b; Emery and Clayton 2009). Is it even possible in principle to decide whether or not any nonverbal creature has a theory of mind? We look here at three proposed answers to this question, starting with a fresh look at what having a theory of mind entails.

Theory of mind as an intervening variable:

We—and perhaps monkeys too—infer that Monkey B wants bananas not only because he looks avidly at Monkey A’s banana, but also because he eats bananas whenever possible, he climbs tall trees to get bananas, and so on. The situation parallels that facing motivation theorists deciding, for example, when to describe a rat as thirsty as opposed to merely drinking in response to external stimuli (Whiten 1996).

In the traditional language of experimental psychology, a theory of mind or a motivational state is an intervening variable (Sober 1998). The animal as psychologist
and the human as animal psychologist have the same problem (Whiten 1994). It becomes defensible to infer such a variable if behavior can be described more
economically and predicted more effectively by doing so than by not doing so (Figure 12.16). Thus we cannot tell if Monkey A is imputing desire and belief to B
if all we observe is A concealing bananas from B. An animal that has a theory of mind should act appropriately in a variety of situations, including physically novel ones where simple stimulus generalization from past learning will not work.

Such a device would be expected to evolve if it supports fitness-increasing generalization from one
social situation to another (Seyfarth and Cheney 1994). For instance, seeing B climb tall trees for bananas and snatch bananas from others gives A no grounds to fear B as a banana thief if she cannot generalize from those
physical situations to one in which B is watching when she is eating a banana. Folk psychology assumes that theory of mind mediates this generalization, but the representation in Figure 12.17 is equally consistent with what Povinelli and colleagues (e.g., Povinelli and Vonk 2004) call behavioral abstraction.

To illustrate this distinction with the false belief test
(Figure 12.11), a child who correctly identifies the box where the ignorant stooge will look might explain his answer by saying, ‘‘Because that’s where he saw it last’’
(behavioral abstraction) or ‘‘Because he thinks it’s there’’ (theory of mind). That is to say, the cognitive structure that connects perception and/or memory of others’
behaviors to one’s own responses in Figure 12.17 need not be an explicit representation of others’ states of mind (Sober 1998; Penn and Povinelli 2007b).

The information in this section so far indicates that through some combination of predispositions for reading and remembering species-specific behaviors
and learning like that responsible for mediated generalization (Chapter 6) animals come to categorize behavioral cues together as relevant to given behavior
systems or functionally related responses. Just as with physical events, memory for social events engages adaptively relevant behaviors. For example, corvids
and chimpanzees treat an individual that gazes at or is remembered as gazing at a desirable piece of food as a competitor or perhaps the owner of the food (Burkart and Heschl 2007).

This means that encountering that individual when
the food is available to be retrieved engages a species-specific suite of defensive, functionally deceptive, or avoidant behaviors that vary flexibly depending on the
spatial setup and the social status of the competitor. Similarly, when the male baboons in Bachmann and Kummer’s experiment (Figure 12.2) had seen a
particular male and female interacting in a friendly manner, they avoided contact with the couple rather than attempting to gain access to the female.

Inference from self to other:

If behavioral abstractions can do the same job as theory of mind in all the tests described so far, maybe a new approach to isolating theory of mind is called for. One
such proposed approach rests on a particular interpretation of human theory of mind, namely that it is based on inference—not necessarily explicit—from self to other. Such inference is of the form, ‘‘When I look toward something with my eyes open, I see it; when I grab something, I want it; therefore, when others like me do the same things, they must have the same mental states.’’ In a proposed test of chimpanzees
based on this notion (Heyes 1998; Penn and Povinelli 2007b), subjects would first be exposed to two distinctively colored visors attached to helmets of some sort.

One, for example the red one, would be transparent and the other, for example the blue one, would be opaque. By putting them over its eyes the animal would learn that it can see through the red one but not the blue one. The test phase would resemble Povinelli and
Eddy’s tests sketched in Figure 12.12 except that people from whom the chimpanzee can beg would be wearing the visors. Choice of the person with the transparent visor would be evidence that the animal imputed its own experience of seeing to another individual.
Informal reports indicate that this has been tried with chimpanzees and they chose randomly (see Penn and Povinelli 2007b), whereas 12- to 18-month-old
toddlers behave as if projecting their experience with an opaque or see-through blindfold onto an adult wearing it (Meltzoff 2007).

However, not all agree that this would be a powerful test of theory of mind for chimpanzees. The animal
need not use the visors’ effects on seeing but on its ability to do things while wearing them in order to choose the person who could respond to its request
(Andrews 2005; Penn and Povinelli 2007b). Moreover, notwithstanding that ‘‘self-recognition’’ in front of mirrors has been taken as evidence for theory of
mind in chimpanzees (Box 12.3), whether the ability to generalize from self to other is predictive of behaviors consistent with theory of mind seems to be an
empirical question. Behavioral abstractions If theory of mind (or behavioral abstraction) allows its possessor economically to encode information and generalize about others’ behavior, then any attempt to assess theory of mind must use more than one behavioral test.

Moreover, the results of a set of such tests should be statistically nonindependent (Sober 1998). That is,
passing one should predict passing others judged to be of similar difficulty. Heyes (1993b) has called this method triangulation because it is designed to point to the
same conclusion from different metaphorical angles. Just as in any test of a concept (Chapter 6), in triangulation an animal acquires information in one set of conditions and is tested in conditions that are conceptually but not perceptually similar. Penn and Povinelli (2007b) have proposed a new series of false belief tests for chimpanzees based on this logic, using food competition.

The proposed setup would be like that in Figure 12.13, but more locations for hiding the food would be available to permit discriminating each of two predicted choices from random behavior. Two locations per trial would be baited, each with a different amount
of food. Subordinate subjects are trained to go for the smaller amount in direct competition with a knowledgeable dominant. Then they would experience an elaborate series of tests in which the competitor sees or does not see the food placed and the food is or is not moved or the two food amounts swapped when the competitor is or is not watching.

Penn and Povinelli (2007b) suggest that combined
results of their proposed tests could distinguish among various possible behavioral rules. Most importantly (see Heyes 2008), some of them would directly contrast
predictions from theory of mind with predictions from specific plausible behavioral already has based on tactile and proprioceptive feedback. This representation allows detection of a mismatch, as when dye is applied in the mark test, but then the animal must also be
motivated to explore the altered parts of its body. It is not always clear whether such motivation is comparable across species in comparative studies in this area (de Veer and van den Bos 1999; Bard et al, 2006).

Nearly 40 years after Gallup’s (1970) seminal article, most issues surrounding apes’ ‘‘self-recognition’’ in mirrors are unresolved. There are still occasional reports of mirror-directed body inspection in other species (e.g., Plotnik, de Waal, and Reiss 2006; Prior, Schwarz, and Gu¨ ntu¨ rku¨ n 2008). Just as with apes, not all subjects of a given species ‘‘pass.’’ With primates, new insights have been contributed by looking at specific elements of mirror-directed behavior in novel ways. Apparently for the first time with any species de Waal and colleagues (2005) directly compared the
responses of mirror-naı¨ve animals (capuchin monkeys) to a same-sex stranger in a neighboring cage, a familiar same-sex conspecific, and a mirror.

The monkeys immediately showed more positive,
friendly, responses and fewer threatening or anxious ones to the mirror than to the stranger. Thus although capuchins do not show mirror self-recognition, for reasons not yet understood the monkey in the mirror is not entirely a stranger either. And in a test of whether learning to use the mirror as a tool would enhance its use in self-grooming, Heschl and Burkhart (2006) trained marmosets to use a mirror to locate things (out-of-sight pieces of food) in the real world. Their skill did not transfer to marks on their heads. Indeed, when the mark was a dab of chocolate cream, rather than touching their own face, most of the animals tried to lick it off the mirror image.

By now, some readers may be inclined to dismiss the vexing question of what animals do in front of mirrors as overblown and misguided. Although animals in the wild may occasionally see their reflections in pools of water, how they behave toward them may not have much adaptive significance. Some intriguing connections have been made between mirror guided body inspection
and other aspects of cognition, but it is still not clear that behavior in front of mirrors reflects any fundamental cognitive processes.

As Penn and Povinelli acknowledge, however, their proposed protocol is complex and might be difficult to implement for that reason. Whether it ever will be remains to be seen.

Looking in the wrong places?:

In the hypothetical tests just discussed, as in those that have actually been done, theory of mind is conceived as a device for predicting the behavior of others. The
philosopher Kristin Andrews (2005) has suggested that such tests are ‘‘looking in all the wrong places’’ because humans use theory of mind not to predict others’ behavior but to explain it verbally. Indeed, people often do not explicitly reason about others’ mental states before acting but react unthinkingly to behavioral cues. This claim is in line with evidence that human theory of mind is closely tied to language, both during
development (Perner and Ruffman 2005) and in adults as well.

For example, when people watch a cartoon and do an interfering verbal or nonverbal task at the same
time, their ability to answer a question about false beliefs of a character in the cartoon is selectively impaired by the verbal task (Newton and de Villiers 2007). This finding is consistent with evidence from tests of implicit memory (Chapter 7) that false beliefs
are not tracked automatically, unlike true beliefs/the true state of the world (Apperly et al. 2006). Andrews (2005) suggests that whether chimpanzees use theory of mind as explanation could be tested nonverbally with a variant of the colored visor experiment.

After experience with the distinctive opaque and transparent visors, the chimpanzee would interact with other individuals wearing the visors, but now their color coding is reversed. If, for example, someone with the ‘‘transparent’’ visor behaves as if unable to see, the chimpanzee should find this surprising and perhaps seek an explanation. One acknowledged drawback of this proposal is that it is difficult to specify exactly
what animals should do in such a situation and what would count as seeking an explanation. For example, increasing attention or looking are ways of getting information about something unexpected, but they would probably not count as explicit information-seeking.

12.4.5 Conclusions: Misled by folk psychology or denying continuity?

In the absence of evidence that chimpanzees pass a false belief test (see Kaminski, Call, and Tomasello 2008) or any other data agreed to discriminate conclusively between use of behavioral abstractions and reasoning about theory of mind, controversy in this area will surely continue. Are proponents of animal theory of mind being led too far beyond the data by their own folk psychology or are those who
conclude that chimpanzees do not reason about mental states denying evolutionary continuity?

Morgan’s Canon (Chapter 1) is being severely tested here: maybe it’s simplest to attribute humanlike theory of mind at least to chimpanzees and other great apes, though it’s not clear where this line of reasoning leaves ravens and scrub jays. One reply to this proposal is that there may be a genuine discontinuity here because
human language underlies reasoning about others’ states of mind in the first place.

But evolutionary continuity is not all or none, especially not when it comes to such a multifaceted ability as theory of mind. Maybe human theory of mind is modular and only some aspects of it are shared by some other species (Lyons and Santos 2006). Indeed, very early on, Premack himself (see Emery and Clayton 2009) suggested three classes or subdivisions of theory of mind: understanding others’ attention and perception, their desires and intentions, and their knowledge and beliefs.

Much of the research reviewed here seems to have treated animal theory of mind as an all-ornothing issue, but it is becoming clear that nonhuman species may share some but not all of these separate human competences. For example, the analysis of gaze-following summarized in Box 12.2 shows that many animals follow gaze in subtle ways but without necessarily understanding that the gazer is acquiring knowledge.

A prominent suggestion of this kind is that chimpanzees and perhaps some other species understand others’ intentions, or ‘‘understand others as intentional beings’’ (Tomasello, Call, and Hare 2003; Tomasello. et al. 2005; Cheney and Seyfarth 2007; Call and Tomasello 2008). However, one can ‘‘understand intentions’’ in the sense of predicting what others are about to do or try to do from behavioral cues such as what they are looking at or moving toward without understanding their underlying mental representations or goals. This latter kind of understanding could be useful in cooperating with others, but the next section—on cooperation—provides no more evidence for it than does this one.

Question 6

12.5 Cooperation

Answer 6

12.5.1 The evolution of altruism

In behavioral ecology, altruism refers to behavior that increases the reproductive success of others at a cost to oneself. When selection can operate between groups, as is increasingly acknowledged (see Box 1.2), cooperative behaviors such as those to be discussed in Section 12.5.3 are expected, but on the view that the main force in evolution is individual selection, altruism is evolutionary puzzle: to be selected, behaviors need to increase the representation of the performer’s own genes, not somebody else’s. However, with individual selection alone altruism can still evolve under at least three conditions: kin selection, mutualism, and reciprocal altruism (Trivers 1971).

Altruistic behavior directed toward relatives evolves through kin selection as long as those helped bear a large enough proportion of that individual’s genes (see Section 5.4). Mutualism refers to cases in which unrelated individuals all concurrently achieve a net benefit from the interaction, as in ‘‘you scratch my back
while I’m scratching yours.’’ Thus its evolution is similarly unmysterious. When benefits are delayed relative to costs, as in ‘‘If you scratch my back now, I’ll scratch yours later’’ or ‘‘If you scratch my back now, I’ll support you in a fight later,’’ we have reciprocal altruism.

Until relatively recently, kin selection, mutualism, and reciprocal altruism were thought to exhaust the conditions for the evolution of altruism (see West, Griffin, and Gardner 2007). Notice that these are functional terms. Their significance in evolutionary theory must not be confused with their cognitive or emotional implications: altruists need not experience empathy for those helped, nor as we saw in Chapter 5 do those helping kin need to understand relatedness.

But each kind of altruism does have specific cognitive (Stevens, Cushman, and Hauser 2005) and emotional (Silk 2007c; de Waal 2008) implications. Kin selection implies discriminative behavior toward kin, which can arise through a variety of recognition mechanisms discussed in Chapter 5. It is easy to imagine that mutualism might be maintained by immediate
reinforcement of participants’ acts, and this supposition is sometimes correct. Reciprocal altruism has traditionally (e.g., Trivers 1971) been thought to be the most cognitively demanding because reciprocal altruists must recognize each other and remember each other’s altruistic acts as if keeping sophisticated mental balance sheets.

In addition, interactions of the form ‘‘If you scratch my back now I’ll support you in a fight later’’ seem to imply behavior maintained by delayed reinforcement, which as we know from Chapter 11 is not very effective. As these considerations predict, there seem to be few good examples of reciprocal altruism (Stevens, Cushman, and Hauser 2005; Silk 2007d). Cooperative behavior can also be maintained by current punishment (harassment) of noncooperators or the threat of future
punishment (sanctioning) for example, ‘‘If you don’t scratch my back now, I’ll take your food later’’ (Clutton-Brock and Parker 1995; Stevens, Cushman, and Hauser
2005).

Sanctioning requires the same kind of memory for past interactions as reciprocal altruism and is accordingly rare (Stevens, Cushman, and Hauser 2005). In this section we look first at a few examples of naturally occurring cooperative behavior for which there is at least a hint about underlying mechanisms (for more
extensive discussion see Silk 2007d). We then look at new models of the evolution of human sociality showing how cooperative behavior can evolve under conditions not encompassed by classic models of altruism and at some experiments designed to test
whether the mechanisms implied by these models are shared with any other primates.

12.5.2 Altruism in the wild

Mutualism:

The example of mutualism probably the best-analyzed mechanistically is that among fish and their cleaners (Trivers 1971; Bshary 2006). Cleaner fish species subsist on the ectoparasites they eat from the surfaces of other fish. Cleaners closely approach their ‘‘clients,’’ which may be larger predatory fish and even swim in and out of their mouths, but they are seldom eaten. Cleaners may have specialized coloration and behaviors that signal their approach; clients likewise have special behaviors of resting in a trancelike state while being cleaned and then signaling to the cleaner when they
are about to depart. The client benefits by getting rid of parasites, and the cleaner gets a meal.

Cleaners have fixed stations on the reef, which their clients visit regularly. Contact with cleaners is reinforcing for clients, which learn not the identity of their cleaners as such but the locations where they are found. In the laboratory, fish will learn to enter an area where they are contacted by a cleaner model (Losey 1979). In the Australian reef fish studied by Bshary and his colleagues, cleaner wrasse, Labriodes dimidaitus, sometimes cheat by eating the client’s mucus, a food they prefer to parasites. Clients respond by attacking the cleaner and/or swimming away.

Bshary and Grutter (2005) simulated this interaction in the laboratory by letting Plexiglas plates coated with food play the role of clients. When both shrimp and fish flakes were offered, cleaners ate the shrimp first, but if an attempt to take shrimp from a plate caused it to ‘‘chase’’ them or ‘‘dart away,’’ they learned to take
flakes first. On the reef cleaners interact with clients up to 2000 times a day, so there would be plenty of opportunity for the learning demonstrated in this study to shape their behavior, even toward individual clients.

Because clients evidently sense when a cleaner is eating mucus and find it aversive, they may also learn which cleaners are reliable. In effect, this would be an example of direct reciprocity, that is, one individual reciprocates a known other individual for past benefits or costs. Here it can result from associative learning through positive reinforcement and punishment
(being attacked or having mucus eaten) or negative reinforcement (the client withdrawing). Current excitatory strength is in effect a tally of net value of past
interactions with a particular client or cleaner. Any learning specializations here lie in the special events that reinforce or punish cleaners and clients.

Clients also learn about cleaners by eavesdropping on their behavior toward other clients (Bshary and d’Souza 2005). On the reef, clients more readily approach a
cleaner if it has just been seen cleaning another client without conflict than causing it to dart away. In the laboratory setup shown in Figure 12.18, clients spend more than half their time near the end of a tank where a cleaner is eating from a Plexiglas model fish when the other end of the tank has a cleaner just swimming around, that is, with unknown behavior toward clients.

In effect, such preference is an example of indirect reciprocity, in that cleaners who do not cheat are reciprocated for their cooperative behavior by eavesdroppers that in turn become their clients. Models like those sketched in Section 12.5.3 show that indirect reciprocity can evolve in species with social networks in which individuals acquire an image score or reputation. Those seen to be good cooperators are reciprocated by being cooperated with. So far in the cleaner-client system such image scoring has been shown only to affect immediate choice between waiting a turn at a given cleaning station or going elsewhere. As one might therefore expect, cleaners are more cooperative when they have an audience (Bshary and d’Souza 2005), another likely effect of reinforcement contingencies that has been reproduced in the laboratory (Bshary and Grutter 2006).

Reciprocal altruism in vampire bats?

Unlike mutualism, in reciprocal altruism acts are exchanged over delays. A muchdiscussed candidate involves vampire bats (Desmodus rotundus). Vampires fly out each evening from communal roosts to seek a meal of blood. A substantial blood meal allows a bat to survive another 50–60 hours before feeding again, but not every bat succeeds in getting a meal every night. Unfed bats may starve within 24 hours, but a starving bat can be rescued if a recently fed bat regurgitates blood for it. To test whether feedings were reciprocal, groups of unrelated bats were kept in the laboratory, and each night one was kept without food while the others fed (Wilkinson 1984).

When the hungry bat was returned to the group, in 12 of 13 cases it was fed by another bat from the group it had come from in the wild. Moreover, the recipients of
regurgitations tended to reciprocate the donation on a later night. These observations suggest that vampire bats recognize unrelated individuals and retain some memory of past interactions with them. However, the relatedness of the bats involved was not always known; because some may have been closely related the possibility remains open that the behavior is kin selected (Stevens, Cushman, and Hauser 2005).

Cooperative sentinels?

Some group-living species appear to resolve the conflict between feeding and vigilance as humans might, by posting sentinels who watch for predators from an exposed location while others forage. At first glance, reciprocal altruism might be in operation here to ensure an equitable sharing of dangerous guard duties. However, it turns out that at least for meerkats not only is a basic assumption of this theory wrong—guards are less vulnerable to predation than their busily foraging companions, not more—but there is no regular rotation of sentinels as would be expected if opportunity to feed is regularly being repaid by time on guard.

Instead, meerkats are more likely to guard when they are near satiation (Clutton-Brock et al. 1999). Individuals experimentally given extra food guarded more often and for longer, and individuals that were unusually hungry because they had been babysitting at the
burrow temporarily guarded less. Here, then, a simple kind of social organization arises from largely individual processes. A meerkat’s top motivational priority is
feeding; once fed, it will guard if no one else is on guard at the time.

Reciprocity and alliances in wild primates

As we saw in Section 12.2, members of a primate troop have friendly relationships expressed in mutual grooming and support in agonistic interactions. Such observations suggest the participants are reciprocal altruists who are exchanging grooming for agonistic support (see Silk 2007d). Indeed this does seem to be the case in a group of captive Japanese macaques studied by Schino, Sorrentino, and Tiddi (2007). They
analyzed a large number of grooming and agonistic interactions, statistically removing effects of kinship and proximity, and found that not only did monkeys groom
most those that had groomed them most, they also groomed most those individuals that had supported them most.

Support was similarly predicted by past receipt of
both support and grooming. The relevant correlations were apparent over the long term but not when only events in the past 30 minutes were analyzed. The fact that, for example, grooming monkey A today is repaid by A’s support tomorrow or the next day seems to imply either learning over long delays or a detailed memory of specific interactions. The improbability of either has been claimed to be a cognitive constraint on reciprocal altruism (Stevens, Cushman, and Hauser 2005).

However, as with the cleaner fish, the net effect of past interactions with a particular individual can just as
well be encoded as a single current value or attitude, similar to associative strength in models of learning (Chapter 4), resulting in what de Waal (2000) calls attitudinal reciprocity. As nicely put by Schino et al. (2007, 186) ‘‘it is necessary to assume only that the exchange of services triggers partner-specific emotional variations and that monkeys make their behavioral decisions on the basis of the emotional state associated to each potential partner.’’

12.5.3 Cooperation and other-regarding behavior

The evolution of human cooperation

Cooperation is a hallmark of human society. Not merely do people behave considerately toward complete strangers, they sometimes make substantial financial and even physical sacrifices for them. Such behavior seems impossible to explain with classic models of the evolution of altruism in which cooperation between unrelated others can arise only when the same individuals interact repeatedly. However, new models
of the evolution of human sociality and the results of experiments to test them suggest that a sense of fairness and other foundations of morality have deep evolutionary roots.

Such models consider processes at the level of groups of individuals, but without relying on the discredited notion of group selection in which individuals
act ‘‘for the good of the group’’ (Wilson and Wilson 2007, 2008). Genes promoting prosocial tendencies in individuals can arise when groups compete in ways argued to be characteristic of the early stages of human evolution. In particular, such conditions may have promoted the evolution of strong reciprocity, a tendency to cooperate with anonymous unrelated others and to punish those who do not do the same even when doing so is costly to the punisher (Nowak 2006; review in Gintis et al. 2007).

A good deal of the evidence that people actually have such tendencies comes from simple stripped down social situations, economic games, a key one of which is the ultimatum game. The rules are as follows. One individual, the proposer, is given a sum of money, say $10, to divide between himself and an anonymous stranger, the responder. He can offer the responder any amount from $1 to $10, and if the responder accepts, they both keep whatever is proposed; if the responder rejects, they both get nothing. In either case, they do not interact again. Notice that because the players are anonymous and interact only once, neither one’s behavior should be influenced by expectations of the other’s approval, reciprocation, or retaliation.

Because it should be obvious to both players that the responder will do better by
accepting any proposal than rejecting it, selfish proposers should consistently make
very low offers and responders should accept them. But contrary to these expectations, proposers generally offer more than the minimum, often near 50% of the total
on average, and responders reject very low offers, a costly act that punishes the
proposer. Indeed, people report feeling angry at very low offers. There are differences
among individuals and also across cultures (see Gintis et al. 2007), but the findings
are clearly better described as strong reciprocity than as uniformly self-regarding
behavior (i.e., behavior that maximizes the actor’s own gain in the short run). These
developments have stimulated tests of whether monkeys or apes show such prosocial
or other-regarding tendencies, that is do they seem to have a sense of fairness or
cooperate with unrelated others? Or is strong reciprocity a uniquely human trait?
Inequity aversion: Do monkeys have a sense of fairness?
One of the first and most controversial tests of other-regarding behavior in primates
was a study by Brosnan and de Waal (2003; see also van Wolkenten, Brosnan, and de
Waal 2007) with capuchin monkeys. The capuchins had learned to exchange tokens
(small rocks) for food. The experimenter would give the monkey a token, and then
offer food which the monkey could obtain by handing back the token. In the main
experiment, two capuchins in neighboring cages could watch each other engaged in
this game; importantly, each could see if its neighbor was getting a grape (a preferred food) or a cucumber slice (a nonpreferred food) for its efforts. Isolated capuchins
would play for cucumber, but seeing its neighbor get a grape greatly increased a
capuchin’s tendency to reject cucumber (Equality vs. Inequality tests in Figure 12.19),
a phenomenon Brosnan and deWaal called inequity aversion.
But what is really going on here? To some extent cucumber is simply less attractive
in the presence of grapes irrespective of whether the grapes are being received by
another animal (Wynne 2004b). The data from two control conditions in the original
study provide evidence for such a contrast effect (Figure 12.19). When a grape was
either given to the neighbor for no effort or simply placed in the empty neighboring
cage (Effort and Food controls in Figure 12.19), failures to exchange were elevated
about as much as when the partner ‘‘received unequal pay for equal work.’’ A less
immediately obvious source of contrast is that many of the trials in which monkeys
rejected cucumber occurred after trials in which those same subjects received grapes
(Brosnan and de Waal 2006; Roma et al. 2006). Moreover, although the original
report of this effect (Brosnan and deWaal, 2003) was called ‘‘Monkeys reject unequal
pay,’’ the ‘‘work’’ implied by the title seems trivial. Handing the experimenter a token
hardly seems to require more effort than reaching out for food, and accordingly the
role of ‘‘work’’ in the effect is also debatable (Fontenot et al. 2007; van Wolkenten,
Brosnan, and de Waal 2007). Finally, what is being claimed here? Some discussions of
inequity aversion seem to suggest that monkeys have a humanlike emotional reaction
to unfairness (cf. Brosnan and de Waal 2003; Wynne 2004b). A more measured
interpretation (e.g., Silk 2007c) is that sensitivity to differences between one’s own
rewards and those available to others could be one of the evolutionary building
blocks of human responses to unfairness. Which nonhuman primates show such
sensitivity and under what conditions remains to be better understood.
Are chimpanzees altruistic?
Whatever else is going on in tests of inequity aversion, subjects seem averse only to
getting less than a companion (Henrich 2004). Getting more than a companion, an
equally unfair allotment, does not seem to bother them, whereas tests like the
ultimatum game suggest that people are averse to any form of inequity. Another
line of research supports the same conclusion. Here, food is out of reach outside a
cage, and primate subjects can use a rope or handle to pull it in. Proportion of trials in
which subject monkeys refused to
exchange a token for a piece of
cucumber when a neighbor was
getting cucumber (Equality test) or a
preferred grape (Inequality test).
Control conditions are described in
the text. After Brosnan and de Waal
(2003) with permission.
460 COGNITION, EVOLUTION, AND BEHAVIORthey can move additional food toward a second animal, do they choose this altruistic
act over delivering food to themselves at the same cost? Tests of chimpanzees from
several captive groups answer this question resoundingly in the negative (Silk et al.
2005; Jensen et al. 2006; Silk 2007d; Vonk et al. 2008). In each case, chimpanzees
could choose to operate either of two pairs of trays. Trays in one pair had a piece of
food for the subject and one for a familiar group member in a neighboring cage; the
other pair of trays had food for the subject and an empty tray for the neighbor.
Subjects were indifferent between these options. They similarly chose only on the
basis of personal gain when one choice prevented delivery of food to another (Jensen.
et al. 2006).
One of the most clever and elaborate illustrations of chimpanzees’ pure self-regard
and apparent insensitivity to fairness in such situations is the behavior of pairs of
animals in a simplified ultimatum game (Figure 12.20; Jensen, Call, and Tomasello
2007). Again one animal, here in the role of proposer, chose between two pairs of
trays. Each pair had 10 raisins, but they were allocated differently between the
proposer’s tray and the responder’s. For example, the choice might be between an
8:2 allotment and 5:5. The proposer could pull one tray closer to both animals, and
the responder could then complete the delivery of the chosen raisin allotment to both
or reject it. Proposers preferred options with more for themselves, and as long as they
got at least one raisin, responders hardly every rejected any offer no matter how
inequitable. Control procedures showed that the animals could see what their companion was getting and could discriminate among the amounts of food offered.
Unlike in an earlier study that provided some (albeit weak, Stevens, Cushman, and
Hauser 2005) evidence for altruistic choices in cotton top tamarins after extensive
training (Hauser et al. 2003), here the animals had comparatively few trials with a
given condition and companion. Thus, even though the interaction was not anonymous as it usually is with humans in the ultimatum game, it came close to testing the
animals’ spontaneous preferences, and in any case repeated interactions with known
individuals would have been expected to increase displays of fairness. It can be argued
that other factors besides species difference such as the desirability of the food reward
contributed to the difference between these findings and those that would be expected
for humans, but they are consistent with the conclusion that chimpanzees do not share our sense of fairness but are concerned only with maximizing their own
economic gains (Jensen, Call, and Tomasello 2007).
Do chimpanzees and monkeys cooperate?
Of course maximizing individual gain is not inconsistent with cooperating in mutualistic situations, particularly those that involve large rewards not easily obtainable by
individuals acting alone. Indeed, some chimpanzees are wonderful cooperative hunters.
Boesch and Boesch-Acherman (2000) vividly describe how males in the Tai forest go
after colobus monkeys, some driving a potential victim down from the trees while
others wait on the ground. The animals are clearly cooperating in a complex way, but
because skilled hunting takes up to 20 years to develop, it is difficult to say exactly what
cognitive skills they are using and how they come to use them. This issue is addressed by
experiments in which pairs of apes or monkeys are presented with tasks in which they
must act together on an apparatus like that in Figure 12.21 to obtain reward (review in
Noe¨ 2006; Silk 2007d). Of course it is not at all surprising if one animal can use
another’s behavior or the results of it as a discriminative stimulus in a learning task. The
questions about specifically social cognition here are therefore something like the
following. Do animals ever cooperate spontaneously? If so what animals under what
conditions? Even if animals must learn to cooperate, are there specific cognitive
prerequisites such as a tendency to attend to others’ behavior or to give communicative
signals? Or are the prerequisites primarily emotional or temperamental?
Tests of cooperation with brown and tufted capuchins in different laboratories
have produced mixed results (see Noe¨ 2006). These tolerant monkeys seem to
cooperate in the wild, and they can learn cooperative tasks in the laboratory.
Attending to the partner’s behavior may contribute to solving them, in that blocking
the view of the partner degrades performance (Mendres and de Waal 2000; but see
Visalberghi, Pellegrini Quarantotti, and Tranchida 2000). Chimpanzees can also
learn to pull such an apparatus together, but they seem to do so entirely through
learning the contingencies between their own and the partner’s behavior. Unlike in one case where the partner was a human (Hirata and Fuwa 2007), they do not
attempt to induce cooperation with communicative signals and gestures.
Importantly, even when chimpanzees’ actions result in separate food for each animal,
as would be the case with the apparatus in Figure 12.21, they succeed more often
when working together with a partner that they more readily share food with in
independent tests (Melis, Hare, and Tomasello 2006b). As this finding predicts,
bonobos, which are more socially tolerant, are more successful than chimpanzees
on this task (Hare et al. 2007), again supporting the notion that emotional or
temperamental rather than specifically cognitive attributes underlie species differences in cooperation.
Empathy and the evolution of helping
The findings sketched so far all seem to point to the conclusion that regard for others’
welfare is a uniquely human trait, evolved since humans and apes separated from
their common ancestor, possibly in response to the conditions in early human social
groups (Silk 2007c). Insofar as regard for another’s welfare requires sensitivity to
what they are thinking or feeling, this conclusion is entirely consistent with the lack of
evidence for theory of mind in chimpanzees. But to some (de Waal 2008) this
conclusion flies in the face of countless observations that in naturalistic social groups
chimpanzees seem to empathize with others in distress. For example, an animal that
has witnessed a fight may appear to console the loser by putting an arm around the
loser’s shoulders. Moreover, in experimental settings chimpanzees do respond spontaneously to signals that a conspecific or a person needs help, for example handing
them an object that they are unsuccessfully reaching toward. Young children do the
same thing (Warneken and Tomasello 2006; Warneken et al. 2007).
These situations differ in two important ways from the economic games in which
chimpanzees fail to show regard for others: neither food nor learned responses are
involved, and natural responses to natural signals are. Evolution may have produced
proximate mechanisms for species-specific helpful behaviors in response to specific
sign stimuli from body language or vocalizations (de Waal 2008). Responses like
those described as consolation or helping may even be accompanied by humanlike
affect, but such affect, if present, apparently cannot support functionally similar
behavior such as delivering food to another via arbitrary learned responses. Indeed,
a wide variety of species show emotional contagion, that is, seeing a conspecific in a
certain emotional state arouses the same emotional state in the witness. For example,
in the company of a mouse in pain, other mice exhibit a lower threshold to react to a
painful stimulus, an effect that could be described as empathy (Langford et al. 2006).
As we see in Chapter 13, such reactions can support witnesses’ learning what caused
their companions’ distress.

In conclusion, whether or not animals cooperate or help others depends on not only the species but on what is meant by helping or cooperating. Animals can learn to cooperate in various ways, but it seems that self-interest generally prevails when small amounts of food are involved. However, some group-living animals actively signal the availability of large amounts of shareable food. Examples include ravens (Heinrich 1989) and the food-calling chickens discussed in Chapter 14. Mechanisms underlying apparent cooperation in such species might repay further analysis. In any case, looking for signs of altruism only in economic, food-related, decisions fails to recognize the importance of what de Waal (2008) calls the altruistic impulse, the spontaneous display of species-typical helpful responses.
Rather than resulting from a conscious calculation of benefits to others, helping may be an unconscious reaction to sign stimuli. Such responses are surprisingly important even in human social life. As an example, when a photograph of eyes was placed beside the box where people placed contributions to a coffee pool, payments more than doubled compared to weeks when flowers were present instead (Bateson, Nettle, and Roberts 2006; Milinski and Rockenbach 2007).

Question 7

12.6 Summary

Answer 7

In the course of the chapter we have seen evidence for a number of specifically cognitive mechanisms, some of them very simple. Many species respond to the gaze
of conspecifics and/or predators, perhaps tracking the direction in which they are looking and/or using gaze as a cue to search behind barriers. Some primates and birds also retain this information for use in later competitive encounters. Response to gaze is thus a phylogenetically widespread component of social cognition.
Human babies and most likely some other primates (at least) share a second such low-level component of social behavior, namely a propensity to respond to self-propelled objects as i they are animate and goal-directed. In the next chapter we see evidence for another candidate component of social cognition in the responses of mirror neurons in primate brains.
We have also encountered three candidates for higher level components of social cognition.

Although it turns out, contrary to earlier suggestions (Tomasello and Call 1997), that sensitivity to third-party social relationships can be demonstrated in nonprimates including fish and birds, some primates classify their social companions in multiple ways simultaneously, perhaps using an ability for hierarchical classification that goes beyond anything yet demonstrated in laboratory studies of categorization in any species. This ability may go along with a particularly fine-tuned ability to decode social relationships.

At the same time, however, there is as yet no evidence in any nonhuman species for two other important human social cognitive abilities— theory of mind and a sense of fairness (or a propensity to take other individuals’ welfare into account in economic decision-making). The focus of this chapter has been very much on primates, often on our closest living relatives, the chimpanzees. In the case of social knowledge this reflects the primacy of field research on primates in suggesting that social knowledge and social brains are special.

While this conclusion may still prove to be correct, research has been tempering it with studies of species as diverse as hyenas, geese, and fish. Similarly, in research on the mentalistic aspects of social cognition, studies of chimpanzees together with those on young children originally predominated, but research
has more recently moved on to look at other primate species as well as birds. To some extent the focus here on chimpanzees and other primates reflects a compelling interest in the question of what makes us human.

Indeed, the early years of the twenty-first century have seen a veritable epidemic of attempts to characterize human cognitive uniqueness (e.g., Premack 2007; Penn, Holyoak, and Povinelli 2008). Based on the
evidence in this chapter we might conclude that only humans have theory of mind or understand the unseen causes of social events just as they understand unseen physical causes. The final section of this chapter adds to this tentative catalog of species differences the suggestion that only humans govern economic decisions with a sense of fairness or understand the sufferings of others. Chapters 13 and 14 suggest further
candidates for human uniqueness. But there are many reasons to doubt that the question ‘‘What makes us human?’’ has a simple answer. We look at it again in
Chapter 15.