Abstract

Soci al cognition refers to the processes that subserve behavior in response to conspecifics (other individuals of the same species), and, in particular, to those higher cognitive processes subserving the extremely diverse and flexible social behaviors that are seen in primates. Its evolution arose out of a complex and dynamic interplay between two opposing factors: on the one hand, groups can provide better security from predators, better mate choice, and more reliable food; on the other hand, mates and food are available also to competitors from within the group. An evolutionary approach to social cognition therefore predicts mechanisms for cooperativity, altruism, and other aspects of prosocial behavior, as well as mechanisms for coercion, deception and manipulation of conspecifics. The former are exemplified in the smallest groups, in the bond between mother and infant; the latter in the largest groups by the creation of complex dominance hierarchies. It is clear that primates are exceedingly adept at negotiating the social environment. This ability is most striking in the most social primate, Homo sapiens, suggesting the hypothesis that our exceptional cognitive skills may be traced back to evolution in an environment in which there was a premium on social skills. In support of this idea, there is a correlation between mean group size among various primate species and their neocortex volume (specifically, the ratio of neocortex volume to the rest of the brain 1 ). Such a correlation has been found also for several other mammals that all feature a complex social structure (e.g. bats, carnivores and toothed whales) -the larger the social groups, the larger the brains (relative to body size). Although it has been proposed that brain size correlates with a number of other factors, including dietary foraging strategy, tool use and longevity 2,3 , it might be that large brain size is at least a partial consequence of the fact that primates have a complex ecological niche with respect to social structure (including its effect on food and mate availability). This hypothesis, variously dubbed the 'Machiavellian Intelligence Hypothesis' 4 or the 'Social Brain Hypothesis' 1 , depending on what theorists take to be its most salient features, suggests that the complexity of primate social structure, together with certain of its unique features, such as cooperativity and deception, led to an advantage for larger brains. Aside from sheer brain volume, one would of course like to know more about the specific neural systems that subserve various aspects of social cognition. A seminal review 5 argued for the importance of the following set of structures: amygdala, temporal cortex, anterior cingulate cortex, and orbitofrontal cortex 6 . The neurobiological underpinnings of social cognition in humans, the topic of this review, are being investigated using various methods, including lesion studies and functional imaging, and can be situated in the context of what we know about social cognition from anthropological, comparative and developmental studies. An overview of the neurobiology of social cognition in primates Non-human primates Two sets of findings, one at a macroscopic level, the other at a microscopic level, first suggested that the primate brain might contain neural systems specialized for processing socially relevant information. In the 1930s, Kluver and Bucy made large bilateral lesions in monkey brains, encompassing amygdala, temporal neocortex, and surrounding structures 7 . The animals subsequently appeared able to perceive and respond to objects in their environment, but they behaved inappropriately with respect to the emotional significance that objects would normally signal. This included a compulsive examination of objects, especially with the mouth, hypersexual behavior, unusual tameness, and a complete lack of awareness of the emotional significance of stimuli ('psychic blindness'; e.g. handling of snakes). Selective neurotoxic lesions of the monkey amygdala result in more subtle impairments; however they do still appear to impair disproportionately those behaviors normally elicited by social cues [8][9][10] . Although the amygdala is a heterogeneous collection of nuclei that participate in several different functional systems 11 , at least some of its components thus appear to contribute disproportionately to social behavior. The other set of findings that first sparked interest in the neural basis of social cognition pertains to the level of single neurons. Neurophysiological studies in non-human primates have shown that single neurons in the monkey inferotemporal cortex respond relatively selectively to the sight of faces 12 . Moreover, specific neurons modulate their response preferentially with specific information about faces, such as their identity, social status or emotional expression [13][14][15] . There are also neurons whose responses are modulated by viewing complex scenes of social interaction 16,17 , as well as by specific features of faces that can signal social information, such as gaze direction 18 . A neural code in which the responses of individual neurons are tuned relatively selectively to highly specific feature conjunctions may permit a neuronal ensemble to distinguish among complex, similar members of a large class of stimuli, such as the faces of conspecifics. Current information-theoretical approaches are providing more detail on how such socially relevant information might be encoded in a neuronal population 19 . Humans Human social cognition has received extensive attention from cognitive, developmental and social psychologists. Some important current issues that might be informed by findings from cognitive neuroscience concern how social cognitive abilities develop in infants, and to what extent genetic factors might influence such abilities. Clearly, the emotional and social development of humans is extraordinarily complex, involving a multi-factorial interplay between genes, parental behavior, and the influence of culture. There have been two major sets of studies that first argued for neural systems critical to social cognition in humans: social impairments following damage to the frontal lobe, and, more recently, social impairments in subjects with autism. The observation that the frontal lobes can contribute relatively specifically to behavior in the social domain was first made on the basis of a rather horrible accident: the injury of the railroad worker Phineas Gage 20 . Gage received a large bilateral lesion of his frontal lobe, including the ventromedial prefrontal cortex, from an accidental explosion that shot a metal rod through his head (see Review A d o l p h s -H u m a n s o c i a l c o g n i t i o n 470 T r e n d s i n C o g n i t i v e S c i e n c e s -V o l . 3 , N o . 1 2 , D e c e m b e r 1 9 9 9 Fig. 1. Summary of neuroanatomical structures involved in social cognition. These renderings of a normal human brain, reconstructed from serial magnetic resonance (MR) images, show the neuroanatomical structures highlighted in this review. The same brain is shown in different views and with differing amounts of transparency, to permit visualization of interior structures. The images were generated by first tracing the structures on 2-D MR images, and then co-rendering these regions of interest in various colors together with the rest of the brain. Highlighted are ventromedial prefrontal cortex (green), amygdala (red), right somatosensory cortex (blue) and insula (purple), all of which play key roles in various aspects of social cognition discussed here. Additionally, other structures (not colored), such as the cingulate cortices, visual association cortices in temporal lobe, and structures in hypothalamus, thalamus, and brainstem contribute to social function. All of these structures also play varied roles in regulating emotion. A second line of evidence that has been used to argue for the functional modularity of social cognition (see Box 1) comes from a developmental disorder, childhood autism. Interest in the social cognitive abilities of subjects with autism was fueled by the argument that autism features a disproportionate impairment in one specific aspect of social cognition: the ability to attribute mental states, such as beliefs, to others 21,22 . While there is debate on the basic hypothesis, and while the link between autism and brain systems is also not well understood, the data point towards neural components that appear to have a high degree of domain-specific function. This idea is strengthened by comparison with another psychiatric disorder, entirely genetic in etiology: Williams syndrome. Subjects with Williams syndrome exhibit social behavior that comes close to being the opposite of that seen in autism -they are hypersocial, and their unusual social skills in the face of impairments in non-social domains have been taken as further evidence for the modularity of social cognition 23 . Of particular interest will be comparisons among subjects with Williams syndrome, autism, and focal brain lesions, some of which are now underway. For example, a recent study found that subjects with Williams syndrome showed selective sparing in their ability to recognize other people's mental states from photographs of their eyes The amygdala: social judgment of faces We glean considerable social information from faces, and there is evidence to suggest that faces are processed in a relatively domain-specific fashion by neocortical sectors of the temporal lobe. For instance, visual processing in regions of Review Focal brain damage can result in impaired processing that is limited to highly specific categories. For instance, patients have been reported who are specifically unable to recognize, or to name, tools, animals, people, or a variety of other selective categories. There is thus very strong evidence that categories are, in some sense, mapped in the brain (but in a way that differs from aspects of objects that are a direct consequence of topography at the sensory epithelium). While initially surprising, the finding is in fact predicted from the assumption of a few, very simple, local rules that specify how brains represent stimuli (Ref. a). In essence, local rules for organizing neural tissue as a function of activity suffice to generate topographic representations of abstract stimulus categories. The categories that are abstracted emerge naturally out of the covariances of our interactions with certain classes of stimuli in the environment. Thus, we typically interact with members of the class of animals in a similar way; that is, the similarity is greater among animals than it is to how we typically interact with members of the class of tools, or members of the class of people. Similarity in sensorimotor interaction can thus translate into functional and anatomical similarity in the brain (Refs b,c). The above view suggests a strong component of experience and learning in such self-organized topographic maps. A different explanation comes from the view that there are innately specified modules in the brain for processing specific categories of knowledge. The evidence for this latter view is strongest from domains such as language, and it is the view that has historically been associated with the notion of 'modularity ' (Ref. d). As with many dichotomies, it is likely that both the above views are right, in the proper context, and recent interpretations suggest a softer version of 'modularity' that does not require a rigid set of criteria (Ref. e). It may well be that there are domainspecific modules for processing certain kinds of information that are ecologically highly relevant and that would benefit from a particular, idiosyncratic processing strategy that does not apply to other kinds of information. That is, one would expect the brain to provide problem-specific structures for processing information from those domains in which there is a premium on speed and survival. Within, and beyond, such a module there might also be topographic mapping of the same domain. It is likely that domain-specific processing draws upon innately specified modules, as well as upon self-organized maps that emerged as a consequence of experience with the world. Is social cognition modular? And if so, is it innate, or is it a consequence of learning? It is likely that both are true, and that whether or not social cognition is modular will depend both on one's notion of modularity and on the aspects of social cognition under consideration. Some rather basic attributes of stimuli, such as self-directed motion, bilateral symmetry, presence of eyes, and so forth, might be processed similarly by different primate species, by mechanisms that are largely innately specified. But there also seems little doubt that the class of social stimuli needs to be explored during development in order to be able to make more fine-grained distinctions -a developmental process that is likely to include parental behavior and pretend play as critical aspects. The most plausible scenario, then, would view social cognition as relying on a neural architecture in which there is interaction between components that are innately specified and others whose operation emerges through experience in the context of a specific culture. A similar answer would presumably obtain in regard to the broader question of cognition, not only with respect to the social world, but the animate world in general (see Ref. f for more extensive discussions of this topic). Future goals will be to provide a more detailed account of the relative contributions that innate and culturally acquired components make to social cognition, and to explore how such functional components might be subserved by specific neuroanatomical structures. Box 1. Social cognition, modularity and innateness the human fusiform gyrus appears to contribute disproportionately to the perception of faces 28 , and viewing dynamic information from faces that convey socially relevant information (such as eye or mouth movements) activates regions in the superior temporal sulcus 29 . Recent data suggest that it is a particular property of how we interact with faces that leads to the specific neuroanatomical processing seen, namely, that we need to become expert at distinguishing many exemplars that are visually extremely similar and yet socially highly distinctive 30 . High-level visual cortices in the temporal lobe project to the amygdala 31 , which has also received historical and recent interest in regard to its role in processing emotionally and socially salient information from faces. A small proportion of neurons within the amygdala show responses that are relatively selectively modulated by the sight of faces, compared with other visual stimuli There are two issues of additional interest: the specificity of the above impairment to faces, and its consequences for social behavior in the real world. In regard to specificity, follow-up studies revealed that bilateral amygdala damage also impaired judgments for the preferences of non-social visual stimuli, such as color patterns or landscapes, although the effect was not as large. In one such study, subjects with amygdala damage liked pictures of non-social stimuli more than did controls The second question, concerning the social impairments following amygdala damage in real life, is more difficult to investigate. However, observation of patients with complete bilateral amygdala damage suggests a common aspect to their social behavior: they tend to be unusually friendly towards others, consistent with the idea that they lack the normal mechanisms for detecting individuals that should be avoided. Similar changes in behavior are seen in non-human primates with selective bilateral amygdala damage 8,9 . On the other hand, the human patients do not appear to be as severely impaired in their social behavior as do monkeys with similar brain damage. It may be that humans with amygdala damage, unlike other animals, possess additional mechanisms for social reasoning and decision making, and are able also to draw substantially on declarative knowledge encoded in language, resulting in partial compensation for their impairment 47 . One would also like to extend the above line of investigations to additional types of stimuli, and to additional types of social information that can be gleaned from such stimuli. We have begun such an investigation, using visual motion cues to provide information about biological and psychological categories. In one experiment, subjects were shown a short video that depicts three geometric shapes moving on a plain, white background 48 . Although visual motion is the only available cue in this experiment, normal subjects have no difficulty interpreting the motion of the shapes in terms of social categories: the shapes are attributed psychological states, such as goals, beliefs, desires and emotions, on the basis of their relative motion. By contrast, a subject with selective bilateral amygdala damage did not spontaneously make such attributions A final important consideration concerns the amygdala's role beyond recognition and judgment, to encompass such A d o l p h s -H u m a n s o c i a l c o g n i t i o n 473 T r e n d s i n C o g n i t i v e S c i e n c e s -V o l . 3 , N o . 1 2 , D e c e m b e r 1 9 9 9 Review Subjects were shown a short movie of simple geometric shapes in motion (a still from the movie is shown in Normal control subject 'I saw a box, like a room, that had an opening to it. There was a large triangle chasing around a smaller triangle, and a circle…got into the box, or the room, and hid. And then the big triangle chased the little triangle around. Finally he went in, got inside the box to go after the circle, and the circle was scared of him…but manoeuvred its way around and was able to get out the opening, and they shut it on him. And the little circle and the little triangle were happy that they got that, the big one, caught. And they went off on their way, and the big triangle got upset and started breaking the box open.' Subject SM 'OK, so, a rectangle, two triangles, and a small circle. Let's see, the triangle and the circle went inside the rectangle, and then the other triangle went in, and then the triangle and the circle went out and took off, left one triangle there. And then the two parts of the rectangle made like an upside-down V, and that was it.' processes as attention and memory. It is clear from studies in animals that the amygdala contributes importantly to these processes 50 , and that its role extends well beyond a function restricted to recognizing potential threat or danger; but such a possible role in humans is just beginning to be explored. For instance, emotionally 51 or socially 52 salient stimuli are remembered better by normal individuals, an effect that correlates with activation of the amygdala in functional imaging studies An active program of research has explored why it might be adaptive to make certain social judgments about faces with certain properties. For instance, average faces are perceived to be highly attractive 56 , but very slight deviations from the average may be considered even more attractive 57 . A possible evolutionary explanation of this effect proposes that averageness, symmetry, or slight deviations from it, are correlated with fitness; consequently, one would predict that such features could have signal value, and one would predict the evolution of perceptual mechanisms for their detection. However, these interpretations are very contentious (see Ref. 58 for a review). Contrary to prediction, some recent data suggest that people with attractive faces are not more healthy While the studies reviewed above strongly implicate the amygdala in several of the processes that are important for normal social cognition, they are problematic from an anatomical point of view: they are both too macroscopic and too microscopic. They are too macroscopic because it is clear that different nuclei within the amygdala subserve different functions 11 , an issue that is addressed in animal studies by lesioning specific nuclei rather than the entire amygdala. Functional imaging studies using fMRI with high field strengths, as well as rare studies of human patients with chronically implanted depth electrodes for monitoring seizures Of equal importance, lesion studies of the amygdala are too microscopic in that it is important to consider the amygdala as one component of a distributed neural system for social cognition. In particular, amygdala and prefrontal cortex appear to function together in processing the rewarding contingencies of emotionally salient stimuli The Ventromedial (VM) prefrontal cortex: social reasoning and decision making Decision making: the somatic marker hypothesis The frontal lobes have a long history in social behavior, going back to the story of Phineas Gage discussed above. More recently, it has become clear that the frontal lobes, specifically their ventromedial sectors, are critical in linking perceptual representations of stimuli with representations of their emotional and social significance The role of the human ventromedial prefrontal cortex in decision making has been explored in a series of studies that used a task in which subjects had to gamble in order to win money. As with gambling in real life, the task involved probabilistic contingencies that required subjects to make choices based on incomplete information. Normal subjects learn to maximize their profits on the task by building a representation of the statistical contingencies gleaned from prior experiences: certain choices tend to pay off better than others, in the long run. The key ingredient that distinguishes this task from other tasks of probabilistic reasoning is that subjects discriminate choices by feeling; they develop hunches that certain choices are better than others, and these hunches can be measured both by asking subjects verbally, and by measuring autonomic correlates of emotional arousal, such as skin conductance response. Subjects with damage to the ventromedial frontal cortex fail this task 70 , and they fail it precisely because they are unable to represent choice bias in the form of an emotional hunch Review A d o l p h s -H u m a n s o c i a l c o g n i t i o n 474 T r e n d s i n C o g n i t i v e S c i e n c e s -V o l . 3 , N o . 1 2 , D e c e m b e r 1 9 9 9 frontal damage make poor choices on the task, they also acquire neither any subjective feeling regarding their choices 71 , nor any anticipatory autonomic changes 72 . These findings are consonant with prior reports that subjects with VM frontal lobe damage do not trigger a normal emotional response to stimuli, including socially relevant stimuli Reasoning: the Wason Selection Task The ventromedial frontal cortex appears to play a key role in a second domain of high relevance to social cognition: social reasoning. Human reasoning strategies have been intensively investigated using the Wason selection task, the most popular experimental design for probing deductive reasoning 75 . The Wason selection task consists of a conditional statement ('if P then Q'), often presented in some context (e.g. 'If you are drinking beer, then you must be over the age of 18'), and subjects must use deductive reasoning in order to decide its truth. Typically, the proportion of logically correct choices made by normal subjects on this task is facilitated by conditionals about social rules, threats, and promises (see Ref. 76 for a review). Cosmides and her colleagues have argued that these data provide evidence for evolved mechanisms for reasoning about social exchange. Specifically, the findings from the Wason selection task support the hypothesis of an evolved skill to detect deception in the context of social contracts (cheating), because an ability to rapidly and reliably detect such deception would have been adaptive 77 (although there is considerable debate regarding the interpretation of the data, and alternative models have been proposed). We investigated the role of the VM frontal cortex in such deductive reasoning, using three groups of subjects: patients with damage centered on the VM frontal cortex, patients with damage centered on the dorsolateral frontal cortex (specifically excluding the VM frontal cortex), and patients with damage outside the frontal cortex. Subjects with bilateral damage to the VM frontal cortex were disproportionately impaired in normal reasoning about social and familiar scenarios, whereas they showed no abnormality when reasoning about more abstract material 78 The above findings from humans can be related to a large number of studies from non-human primates Review A d o l p h s -H u m a n s o c i a l c o g n i t i o n 476 T r e n d s i n C o g n i t i v e S c i e n c e s -V o l . 3 , N o . 1 2 , D e c e m b e r 1 9 9 9 Primates appear to be highly skilled at predicting other individuals' behavior, but there is vigorous debate about how to interpret such an ability. The mechanisms by which we represent and predict other people's behavior have been viewed from two different theoretical perspectives. The two main camps argue either for a 'theory of mind', or for a set of processes that permits 'simulation' of other minds. The 'theory'-theory has been floated for some time in philosophy of mind as a possible explanation of what is commonly called 'folk psychology': our commonsense understanding of other people's behavior in terms of intervening mental states, such as beliefs, desires and intentions, on the basis of which people act. The other camp, however, views our ability to recognize and reason about other people's states of mind as an example of experience projection; in essence, we know other minds by empathy, or by simulation. It is likely that both these views have some truth to them, depending on the circumstances (see Ref. a for examples of both sides of the debate). The theoryview might afford greater economy and generalizability of prediction, or might be particularly suited to information that can be lexically encoded; but simulation may be the only option in cases that are sufficiently idiosyncratic, or in cases where the information is not easily encoded into language. In the latter situation, it could be that the only way to predict what another person will do is to run in one's own brain the processes that the other person is running in theirs. If this possibility is taken seriously, it suggests a role for conscious experience in social cognition: to obtain information about another person's internal mental state, it may be necessary to imagine what it would be like to be the other person via direct simulation. Simulation might find its developmental origins in infants' ability to mimic facial expressions spontaneously (Ref. b), and it has found some recent neurophysiological support from the finding of so-called 'mirror neurons', which appear to participate in simulating the actions of other individuals (Ref. c). Research into how we represent other minds began with a question about whether or not chimpanzees might possess a theory of mind (Ref. d), a question that is still unanswered (Ref. e). In humans, the theory-of-mind question was posed concretely in terms of the ability to attribute beliefs, specifically false beliefs, to other individuals. It has been shown that this ability begins to emerge around age four or possibly earlier (Refs f,g). The abilities that constitute a theory of mind have been fractionated into several distinct components, such as the ability to attribute desires, to recognize objects of shared attention, and to monitor others' direction of gaze. All these different components appear at distinct developmental stages in humans, and there is evidence that some of them may be selectively impaired in subjects with autism, a disorder that exhibits marked difficulties in social behavior (Ref. h). Several lesion and functional imaging studies have investigated the neural structures by which subjects generate knowledge about other people's mental states. In addition to a large literature demonstrating the involvement of amygdala, orbitofrontal cortices, and right hemisphere cortices in more general processing of emotion, including recognition of emotion in others, some studies have explicitly investigated attribution of higher-order mental states, such as beliefs and intentions. A recent study by Stone et al. found that subjects with bilateral damage to the orbitofrontal cortex were specifically impaired in their ability to attribute higher-order mental states to other people from stories (Ref. i). In particular, they were unable to detect a faux pas, something that subjects with high-functioning autism (Asperger syndrome) also fail. A functional imaging study that compared brain activation during theory-of-mind tasks between normal and high-functioning autistic subjects found evidence that sectors of left medial prefrontal cortex were also important to reason about other people's mental states (Ref. j), a finding consistent with earlier studies that showed that processing words for mental states This idea might help to explain why emotion and social cognition are closely related, not only in terms of shared processing strategies, but in fact in terms of their neural substrates: most structures important to social cognition are also important to normal emotional functioning. The common ingredient may be what we commonly call 'feeling': the representation of emotional body states, either in regard to one's own emotional reaction, or in regard to the empathy for, or simulation of, another person's internal state. In addition to the amygdala and ventromedial frontal cortices, which can trigger emotional responses to socially relevant stimuli, there is evidence for a third important structure that contributes directly to our ability to construct representations of other individuals. In a study of subjects with focal brain lesions, we found that recognition of emotions from other people's facial expressions critically relied on the integrity of somatosensory-related cortices in the right hemisphere (including S-I, S-II, and insula Conclusions Social cognition draws upon a vast set of abilities. Some of these are quite specific to the social domain, and others may be more general in their application. Some classes of emotions, such as guilt, shame, embarrassment and jealousy, only make sense in a social context and may have evolved to subserve very specific roles in social communication. Other social signals, and other types of social judgments, draw upon systems that subserve emotional processing in general, systems that permit us to build models of other individuals through simulation, and a vast network of structures that contribute to reasoning, inference and language. Three structures have been highlighted in this review: amygdala, ventromedial frontal cortex, and right somatosensory-related cortex. Normally, in a typical, complex, emotionally salient situation in real life, all three component structures will operate in parallel: the amygdala will provide a quick and automatic bias with respect to those aspects of the response that pertain to evaluating the potentially threatening nature of the situation, or with respect to allocating processing resources to those stimuli that are potentially important but ambiguous; ventromedial frontal cortex will associate elements of the situation with elements of previously encountered situations, and trigger a re-enactment of the corresponding emotional state; and right somatosensory-related cortices will be called upon to the extent that a detailed, comprehensive representation of the body state associated with emotional or social behavior needs to be made available. All of these components would be important to guide social behavior in a typical situation in real life, and all of them emphasize the close link between emotion and social cognition. There is no doubt that humans differ from other animals in their social skills, in that they are able to form higher-order representations of the social environment, and to manipulate those representations in reasoning that can be quite flexible. On the other hand, there is also good evidence that our reasoning is biased in domain-specific ways, and that our judgment of other individuals, and our behavioral responses Review Outstanding questions • Most of the neural structures known to be important to social cognition are also important to emotion, and to associating stimuli with reward and punishment. What is the relation between social behavior, emotion, and reward/punishment? Can social cognition be thought of as an elaboration on reward mechanisms? • What aspects of social cognition are truly unique to humans? • What aspects of social behavior are innately specified, and what aspects are acquired through experience? Related to that, what aspects of social behavior are invariant across different cultures? • How critical is language to social cognition? Is it possible that language evolved primarily to subserve social behavior? • If the evolution of the human mind was driven in large part by the need for cognitive mechanisms that are socially adaptive, then it becomes interesting to consider the constraints that our social cognition might have on cognition in general. Are we limited in what we can think about, and in how we can think, by a design that has optimized human cognition for social behavior? towards them, are strongly influenced by mechanisms that we share in common with other animals. The challenge for the future will be to offer a more precise account of the interplay between all these different processes as a function of the detailed specification of the performance demands required by a given experimental task, or by a given situation in real life.