Introduction

Primates have unusually large brains for body size compared to all other vertebrates. The conventional explanation for this is known as the “social brain hypothesis,” which argues that primates need large brains because their form of sociality is much more complex than that of other species (Byrne & Whiten, 1988). This does not mean that they live in larger social groups than other species of animals (in fact, they don’t), but rather that their groups have a more complex structure. In exploring the nature of this unique kind of primate sociality, this article shall argue that, so far, social neuroscience has barely scratched the surface of what is actually involved in what it means to be social. To borrow an analogy, social neuroscience has devoted its time to examining the bricks and mortar in great detail but has so far overlooked the complexity of the building that lies at the real heart of primate (and human) sociality.

The original idea for the social brain dates back to the 1970s, when a number of primatologists suggested that primate intelligence might be related to the demands of their more complex social world (Jolly, 1969; Humphrey, 1976; Kummer, 1982), and the name itself was later coined by the neuroscientist Lesley Brothers (1990). The primary evidence in support of the social brain hypothesis comes from the fact that, across primates, there is a correlation between mean social group size and more or less any measure of brain size one cares to use (Fig. 1) (Dunbar, 1992, 1998; Barton, 1996; Barton & Dunbar, 1997; Dunbar & Shultz, 2007; Dunbar, 2011a), although the relationship improves as the measure of brain size is focused more toward the frontal lobes (Joffe, 1997; Dunbar, 2011a). In this respect, primates differ from almost all other mammals and birds: in most birds and nonprimate mammals, large brains are associated not with social group size but with a monogamous mating system (Shultz & Dunbar, 2007, 2010a, b; Pérez-Barbería et al., 2007). Note that in Figure 1 there appears to be an obvious grade difference between the apes and the monkeys. This suggests that apes require a proportionately larger brain than monkeys to deal with groups of the same size, implying that their form of sociality requires more computing power to handle. More careful analysis has since revealed that there are similar grade differences among the monkeys, with a clear distinction between more and less intensely social species. As indicated in Figure 1, extrapolating from the relationship between group size and neocortex size in apes predicts a natural social group size for humans of around 150 (Dunbar, 1993). There is considerable evidence for the existence of such a group size in terms of both natural human groupings (e.g., community sizes in small scale societies) and personal social networks (Dunbar, 2008, 2011b).

Open in new tab

Figure 1. Mean social group size plotted against relative neocortex volume (indexed as the ratio of neocortex volume to the volume of the subcortical brain) in anthropoid primates. Filled circles: apes (including humans); unfilled circles: monkeys. The regression lines indicate grades of increasing socio-cognitive complexity (indexed by the increasing density of the line). (Redrawn from Dunbar, 2014.)

Secondary support for the social brain hypothesis comes from neuroimaging studies, which have recently shown that the size of an individual’s living group (macaques: Sallet et al., 2011) or personal social network (humans: Lewis et al., 2011; Powell et al., 2012, 2014; Kanai et al., 2012) correlates with the size of core regions of the brain, mostly in the temporal and, especially, the frontal lobes. These regions turn out to be essentially those involved in the mentalizing, or theory of mind, neural network. This is an important finding because it demonstrates that the social brain hypothesis applies not just at the level of the species but also at the level of the individual. Individuals with more processing capacity in core brain units have proportionately larger social networks.

Historically, a number of alternative ecological and developmental hypotheses have been proposed for why primates have such large brains (for an overview, see Dunbar, 2012b). Among these, the importance of foraging skills, and especially the role of social learning of foraging skills, has attracted a great deal of interest (e.g., Reader et al., 2011). This is not the place to discuss the ensuing debates in detail, but some points of clarification are desirable. It is important to note at the outset that everyone agrees that foraging skills have played an important role in primate evolution; the critical question is whether these have been the main, or only, driver of increases in brain size or whether they are an evolutionary by-product of large brains evolving for other (perhaps mainly social) reasons because the same cognitive skills (causal reasoning, predictive reasoning, planning, etc.) underpin both kinds of behavioral outcomes.

In fact, ecology lies at the heart of all explanations for brain evolution, including the social brain hypothesis: the core differences between them are (1) whether animals solve their ecological problems by individual trial-and-error learning or do so socially and (2) which particular ecological problem (foraging or avoidance of predation) is the more fundamental selection pressure (i.e., the evolutionary driver). What makes the social brain hypothesis intrinsically social is that it claims that animals solve their ecological problems by first solving the problem of group cohesion and coordination. One reason for thinking this is that the primary ecological problem faced by primates (and probably most other animals) is the risk of predation (either by predator species or by conspecific raiders) rather than how to find food (as important as this is in the life of any animal). Primates, like most other animals, solve the predation risk problem by living in groups (van Schaik, 1983; Shultz et al., 2004; Shultz & Finlayson, 2010) and have opted to do so by evolving an unusual form of bonded sociality to maintain group coherence through time (Dunbar & Shultz 2010). In effect, primates solve the predation problem indirectly by first solving the problem of creating coherent, stable, coordinated social groups. The issue thus comes down to the task demands of foraging versus social coordination.

A second issue we need to clarify is that the social brain hypothesis has sometimes been seen as simply the quantitative relationship between social group size and brain size shown in Figure 1. In fact, it should properly be seen in systemic terms as a set of causally related functional behavioral relationships. Animals need to solve a variety of ecological problems in order to be able to survive and reproduce successfully, and primates solve these problems communally in a way that requires them to solve a number of social and physiological problems first. In effect, primates establish the means to solve the ecological problem (an alliance or coalition) ahead of its need, and the capacity to form coalitions in anticipation of their future need seems to be a unique feature of monkey and ape behaviour (Harcourt, 1992). It is this that gives rise to the unique form of primate sociality that we refer to as “bonded sociality”, in contrast to the more casual groupings found in most other species of birds and mammals where social groups (herds) can fragment and come together relatively easily (Dunbar & Shultz, 2010).

Maintaining the coherence and cohesion of bonded social groups through time is very demanding because animals have to be able to override the natural tendency for the stresses of social life to drive them apart (Dunbar 2010a, 2012a), and the social brain hypothesis argues that this comes down to resolving various tensions and stresses in both dyadic relationships and the collective set of relationships formed within a social group. To do this, monkeys and apes require novel cognitive skills, and these cognitive skills in turn require appropriate hardware (or wetware in this case) to underpin them. Hence, the relationship between brain size and group size is indirect, and the real functional relationship in the social brain hypothesis is that between brain (or brain region) size and/or wiring and social cognitive abilities or competences that allow primates to manage relationships (Dunbar, 1998, 2011a, 2012a). In effect, group size is an emergent property of how well the animals solve the problems associated with living in close proximity.

In other words, in contrast to all the alternative ecological hypotheses that have been proposed (for overviews, see Dunbar 2011a, 2012b), the social brain hypothesis is a two-step explanation for the evolution of large brains in primates. In contrast to all alternative hypotheses, it explicitly claims that primates are doing something radically different to all other species of animals. The ultimate evolutionary driver is not simply the capacity to engage socially or live in large groups but the extent to which this allows the animals to solve the problems associated with successful survival and reproduction. The proximate mechanism involves solving the coordination problem that lies at the heart of maintaining cohesive social groups. To the extent that primates solve this second problem (group coordination), they also solve the first (predation risk).

What’s So Social About Primate Sociality?

All mammals and birds are, of course, social in some generic sense. The central premise of the social brain hypothesis is that sociality in anthropoid primates (and perhaps a very small number of other mammalian families, including elephants, the dolphin family, and maybe the camel family, that also live in complex, multi-level social systems: Hill et al., 2008, Shultz & Dunbar, 2010a) is a step up from this: it involves a more bonded form of sociality built around intense dyadic relationships (friendships) (Silk, 2002; Dunbar & Shultz, 2010; Massen et al., 2010). This form of bonded sociality is a response to the need to handle the stresses that arise when animals live in close proximity and cannot escape these pressures simply by leaving (i.e., by group fission). Living in groups creates significant stresses (mainly due to harassment from conspecifics) that radically affect female fitness (Dunbar, 1980, 1988; Hill et al., 2000; Smuts & Nicholson, 1989; Roberts & Cords, 2013) via an endocrinological mechanism that is now relatively well understood. Among other effects, social stress destabilizes the female menstrual endocrinology system and results in amenhorrea (temporary infertility) (Bowman et al., 1978; Abbott et al., 1986). Unless animals are able to find solutions that buffer them against these and other costs, group fission is inevitable because the cumulative costs for low-ranking females in terms of lost reproduction can become intense. These stresses are a linear function of group size: the more animals there are in the group, the more individuals one can be harassed by. Moreover, sociality itself is costly: for both primates (Dunbar, 1991; Lehmann et al., 2007) and humans (Roberts & Dunbar, 2011; Miritello et al., 2013), relationships require the investment of considerable quantities of time for their maintenance, and this time cost is more or less proportional to the number of individuals involved multiplied by relationship quality (Sutcliffe et al., 2012). This is partly because the mechanism involved in creating and servicing relationships involves the endorphin system: the more frequently this is activated, the stronger the relationship. We'll return to this later.

The endogenous stresses that the animals face from living in groups act as a constraint on group size because they create centrifugal forces that, if not defused, will eventually cause the group to break up. In species that do not have bonded social systems (most non-monogamous birds and mammals), these stresses are resolved by individuals simply leaving one group to join a smaller one on an ad hoc basis (the bees-around-a-honeypot model of sociality). This solution is not available to species that live in bonded social systems because of the resistance to individuals transferring between groups created by bonded relationships: members of a group do not tolerate strangers.

Anthropoid primates deal with these stresses by forming defensive alliances mediated by social grooming (Dunbar, 1980; Silk et al., 2003, 2009; Wittig et al., 2008), and this in turn gives rise to highly structured social networks (Dunbar, 2008, 2012b; Lehmann & Dunbar, 2009). It is this “decision” to use coalitions as a buffer for the stresses of group living that seems to create the complexity that is widely recognized as characteristic of primate societies. This social world is more complex to handle than the physical world, partly because it is dynamic and in a constant state of flux, and partly because it involves phenomena (other individuals’ mind states) that cannot be perceived directly but instead have to be inferred (Dunbar, 2010a, 2011a, 2012b). In effect, social systems of this kind are implicit social contracts. For a group to be stable through time, its members have to be willing to allow each other to have a fair (though not necessarily equal!) share of the benefits of sociality. Failure to hold back on prepotent actions that would offer immediate benefits to oneself (such as stealing someone else’s newly discovered food item or displacing someone from a safe roosting site) risks driving others away and destabilizing group cohesion.

One explanation for the grade structure observed in Figure 1 is that this reflects a step-change in the complexity of primate social relationships and the behaviors that underpin them as neocortex volume increases. Indeed, across primates, neocortex size correlates with increasing use of sophisticated mating strategies, larger grooming cliques, higher frequencies of tactical deception, and the formation of coalitions (Pawłowski et al., 1998; Kudo & Dunbar, 2001; Byrne & Corp, 2004; Dunbar & Shultz, 2007), as well as increasing complexity of both visual (Dobson, 2009) and vocal (McComb & Semple, 2005) communication repertoires. One example of this is that cognitively more advanced species like macaques are aware of third-party relationships and refrain from attacking or exploiting another animal when they know that individual has powerful allies, even when those allies are not physically present (Datta, 1983). Computational models suggest that managing third-party relationships is more demanding in terms of information processing time than managing simple dyadic relationships (Dávid-Barrett & Dunbar, 2013). Similarly, playback experiments have demonstrated that baboons (another cerocpithecine) can integrate at least two different relationship dimensions (kinship and dominance) simultaneously, an ability that may be beyond cognitively less well-endowed species (Bergman et al., 2003).

There has been a near-universal tendency to assume that the social groups of all animals are “of a kind.” However, in anthropoid primates, grooming networks become increasingly substructured as the number of individuals in the group increases, especially so in species that have larger neocortices (Kudo & Dunbar, 2001; Hill et al., 2008; Lehmann et al., 2009; Lehmann & Dunbar, 2009b). In effect, these species are able to maintain two qualitatively distinct kinds of relationship simultaneously: intimate relationships with principal grooming partners (allies) and weaker ones with other group members. In this respect, monkey and ape relationships resemble the two-tier structure of human social relationships, where parallel distinctions are drawn between weak and strong “ties” (Granovetter, 1973; Sutcliffe et al., 2012) and, cutting across the weak/strong divide, between family and friends (Curry et al., 2013; Roberts & Dunbar, 2011; Roberts et al., 2014). This gives the social systems of anthropoid primates (and those of a small number of other mammals) a layered structure (Hill et al., 2008) similar to that found in humans (Zhou et al., 2005; Hamilton et al., 2007; Dunbar et al., 2015). While in both humans and primates an individual’s relationships with the other members of their social group can be ranked on a simple continuum based on frequency of interaction (or emotional closeness: Sutcliffe et al., 2012; Roberts et al., 2014), these nonetheless cluster into quite discrete layers of very distinctive size, as shown in Figure 2. The numerical sizes of these grouping layers seem to be common to both human social networks and the structure of primate social groups (Hill et al., 2008), and one explanation for the differences in social complexity between species may be the number of layers that can be maintained as a coherent, stable system.

Open in new tab

Figure 2. The circles of acquaintanceship for normal human adults. Ego indicates the subject of the network. Normal adult humans are embedded in a series of hierarchically inclusive layers of friendship, with each successive layer enclosing a larger number of individuals at a progressively lower level of emotional closeness. The layers have very distinct sizes, with a scaling ratio that approximates three (each layer is three times the size of the layer immediately inside it). The average sizes of each layer are indicated by the numbers against each circle in Figure 2, although there is considerable individual variation. The circle of ~150 corresponds to the number of individuals with whom one has reciprocated relationships of trust, obligation, and reciprocity. Beyond the 150 layer there are at least two further layers: the layer of acquaintances (totaling ~500 individuals) and the number of faces one can put names to (~1500 individuals). While the two innermost layers (at 5 and 15) tend to be densely interconnected and constitute a single subnetwork, the remaining layers typically consist of more isolated sets of subnetworks (work colleagues, different sets of hobby club friends, church friends, distant family, etc.) for whom the only connection is via Ego. Each of the four innermost layers is typically split between extended family members and unrelated friends, with an overall ratio of about 50:50 (Sutcliffe et al., 2012).

In humans at least, there is evidence suggesting that the size of an individual’s personal social network correlates with their mentalizing competences, indexed as the ability to solve multiple-individual false belief tasks (Stiller & Dunbar, 2007; Lewis et al., 2011; Powell et al., 2012). Mentalizing, perhaps the archetypal form of social cognition, is the ability to handle other individuals’ mind states simultaneously and forms a naturally recursive sequence from first order intentionality (I know my own mind state) through second order (I know that A knows something—otherwise known as formal theory of mind) to a maximum of around fifth order (I know that A knows that B knows that C knows that D knows something) in most normal adult humans (Stiller & Dunbar, 2007). Since mentalizing competences (the number of different mind states one can have in mind at the same time) correlate with the volume of core areas in the frontal lobes (Lewis et al., 2011; Powell et al., 2012, 2014), it follows that maintaining larger social groups is more demanding in terms of the need to allocate neural resources to those regions of the brain implicated in this task.

Further evidence that social cognition is likely to impose limits on social group size comes from an agent-based model that used processor time to assess the cognitive demands of different levels of information processing associated with managing relationships: this demonstrated not only that more complex information processing is more demanding but, more importantly, that this in turn sets limits on the size of group that can be maintained (Dàvid-Barrett & Dunbar, 2013; see also McNally et al., 2012, Moreira et al., 2013). It may be no coincidence, then, that the social brain graph in fact consists of a series of socio-cognitive grades (Dunbar, 1993, 2011a; Lehmann et al., 2007).

There is evidence that social cognition is itself significantly more demanding than more conventional forms of cognition. We have shown, using both reaction time experiments and fMRI in humans, that mentalizing tasks (those that involve modeling the mental states of other individuals [for more details, see below]) are cognitively more demanding than equivalent non-mentalizing (i.e., purely factual memory) tasks and involve the recruitment of more neural circuitry, and that the magnitude of this difference increases with the complexity of the proposition being processed (Lewis et al., forthcoming). One reflection of the fact that social cognition may be very costly is that it seems to develop much more slowly than more conventional instrumental cognition. In humans, emotional cue recognition (Deeley et al., 2008) and aspects of social cognition such as theory of mind (Blakemore & Choudhury, 2006; Henzi et al., 2007) can take as long as two decades to mature: their developmental progress seems to map onto the slow process of myelinization in the frontal lobes, which in humans is not completed until well into the third decade (Sowell et al., 2001, 2003; Gogtay et al., 2004). Socialization seems to play an important role in this: Joffe (1997) showed that, across primates, the best predictor of the non-V1 neocortex volume is the length of the period of socialization (the period between weaning and puberty), suggesting that a considerable amount of practice over a lengthy period is required to develop the skills that underpin the social brain. These findings suggest that social skills require conscious thought in frontal lobe units before they eventually become automated and localized elsewhere in the cortex or subcortical regions (in humans, as late as the mid-20s). In other words, merely having a big computer (i.e., brain) is not enough: the hardware requires programming, and this is in large part dependent on extensive social experience. This is social learning on a dramatic scale and may explain why social learning appears to be so important in primates (Reader et al., 2011). A useful by-product of this is that the cognition that underpins social learning in this context then becomes available for the exchange of factual information about foraging among adults. Although this has sometimes been interpreted as the driver of brain evolution on the basis of correlational evidence (Reader & Laland, 2002; Reader et al., 2011; Pasquaretta et al., 2015), it could, in fact, just as easily be a consequence rather than the cause of brain evolution—a possibility that, surprisingly perhaps, never seems to have been considered.

Neuropsychology and the Social Brain

In primates, the neocortex accounts for a very large proportion of total brain volume (50–80%, compared to 10–40% in all other mammals) (Finlay et al., 2001). This probably explains why even total brain volume on its own gives a reasonable correlation with group size and other social variables in primates—subject to some error variance introduced by species like the gorilla and orangutan that have unusually large cerebella and relatively small neocortices and for whom neocortex size gives a significantly better prediction of community size than does total brain size (Dunbar, 1992, 2011a). The fit is improved by excluding striate cortex (the primary visual area, V1, in the occipital lobe: see Fig. 3) (Joffe & Dunbar, 1997), and it is improved still further by narrowing the focus down to the frontal lobes (Dunbar, 2011a), implying that the automated processing of incoming perceptual stimuli is not itself a major component of the social brain processes—and why would it be, given that it is the meaning attached to these percepts rather than the percepts themselves that lies at the heart of complex sociality? Since the successive visual processing areas (V1 through V5/MT) scale isometrically with each other up through the occipital and parietal lobes (Dougherty et al., 2003; Yan et al., 2009), it is likely that the fit would be improved still further by excluding these and other basic perceptual processing regions in the brain (i.e., by focusing mainly on the social cognition circuits in the frontal and temporal lobes). Nonetheless, the fact that the brain acts as a distributed processing network may explain why many of the comparative analyses reveal respectable correlations between social behavior and relatively large brain regions like the neocortex.

Open in new tab

Figure 3. The main brain regions involved in mentalizing (the “theory of mind network”). PFC, prefrontal cortex; ACC, anterior cingulate cortex (buried within the cortex); TPJ, temporoparietal junction; STS, superior temporal sulcus; V1, primary visual cortex (striate cortex) in occipital lobe. Dashed arrows indicate the principal connections of the “theory of mind” network.

A number of analyses have shown that executive function skills also increase with brain (or brain region) volume (Dunbar et al., 2005; Deaner et al., 2006; Shultz & Dunbar, 2010b; Reader et al., 2011). Inevitably, these analyses rely on extremely coarse anatomical resolutions and so have not allowed us to narrow down the cortical circuits involved in any detail (although the availability of more sophisticated imaging techniques may offer new opportunities in this respect; see Mars et al., 2014). In the only serious attempt to address this issue to date, Passingham and Wise (2012) concluded that some brain regions (notably the dorsal prefrontal cortex and the frontal pole [Brodman area 10 at the very center of the forehead]; Fig. 3) are crucial for causal evaluation and strategic planning in anthropoid primates. However, their analysis was inevitably based on a very small sample of species. That said, the question as to what function(s) these competences subserve remains open: they may well be generic skills required for all forms of decision-making. All the experimental tests on which these studies are based (“odd-one out” problems, mapping tasks, analogical reasoning, causal reasoning) involve tasks that are essentially instrumental (mainly foraging tasks) rather than social ones. The problem for comparative psychology has always been that genuinely social tasks are not easy to devise: they tend to have long time delays to their outcomes (sometimes on the scale of a lifetime; see Silk et al., 2003, 2009), and experimentalists require an immediately measurable outcome. This has been compounded by a long-held and widespread assumption that, in the wild, animals do very little other than sleep and search for food. Historically, there has been no incentive to devise more complex tasks.

If the different social and ecological uses to which primates put their brains depend on essentially the same cognitive mechanism (and, in particular, the same second-order cognitive processes such as causal reasoning, one-trial learning, analogical reasoning, comparison between two or more alternative projections into the future; Passingham & Wise, 2012), it may not be too surprising that there is evidence to support both the instrumental and the social hypotheses. However, a task analysis suggests that, while certain kinds of cognition are likely to be common to all the functional hypotheses for primate brain evolution, there is a natural asymmetry among the hypotheses. The kinds of cognition required to support bonded relationships may allow social (i.e., cultural) transmission of information or novel foraging behaviors, but the reverse is probably not the case; similarly, the kinds of cognition required to support social transmission of foraging skills would likely allow individual trial-and-error learning of foraging behavior, but the reverse is not the case. This is especially likely to be true to the extent that the real complexity of social relationships depends on the need to model other individuals’ minds and behavior in a virtual mental state space, something that seems to be cognitively very demanding even for humans (Lewis et al., forthcoming). Some evidence to support this suggestion is provided by one of the few experimental studies to compare social and instrumental cognitive skills across primate species directly: Herrmann et al. (2007) found striking differences between humans and great apes in performance on social tasks but much less so on instrumental tasks.

This suggests (1) that the cognitive demands of instrumental tasks are significantly less than those of social tasks and (2) that the ability to manage social tasks depends crucially on frontal lobe volume (in particular). It would seem that only the social hypotheses would naturally provide for the other hypotheses as emergent by-products. This is not to say that cognitive evolution did not begin with solving simple ecological problems like food-finding (it almost certainly did), but rather to suggest that the demands of social cognition have resulted in additional more sophisticated cognitive competences being added to this mix and that these have, in turn, then allowed more sophisticated food-finding strategies.

In the previous section, it was suggested that mentalizing may be central to complex sociality in humans because it allows individuals to work with virtual representations of other individuals in a mental state space. Meta-analyses of a large number of neuroimaging studies of theory of mind in humans have identified the medial and/or orbitofrontal prefrontal cortex (PFC) as being differentially activated during mentalizing tasks in more than 90% of studies, the temporoparietal junction in 58%, the anterior cingulate cortex in 55%, and superior temporal sulcus (STS) in 50%; other regions that were less commonly activated included the amygdala and the insula (13% of studies in both cases) (Carrington & Bailey, 2009; see also Gallagher & Frith, 2003; van Overwalle, 2009; Apperly, 2012). Figure 3 shows the relative locations of these regions in the brain. It is well known that lesions in the prefrontal cortex specifically disrupt social skills, whereas those elsewhere typically do not (Kolb & Wishaw, 1996), while the role of the prefrontal cortex and the temporoparietal areas in managing false belief tasks (the benchmark for theory of mind) has been confirmed experimentally using transcranial magnetic stimulation to knock these regions out during experimental tasks (Costa et al., 2008). Recently, Makinodan et al. (2012) reported that mice that had been socially isolated immediately after weaning exhibited irrecoverable functional deficits in both the prefrontal cortex and its myelination, indicating that there may be a critical period that is vital for neurotypical development in a region that is crucial for normal adult social behavior.

This network also appears to be present in at least the catarrhine primates (Rushworth et al., 2013), although it is unlikely that it is capable of producing fully functional theory of mind sensu stricto in these species. What it probably does allow is perspective-taking, and that may be an important evolutionary and developmental precursor for full-blown theory of mind as well as being functionally essential for much of what is involved in the social interactions of nonhuman primates. There is considerable evidence that great apes, at least, are able to take others’ perspective into account (Hare et al., 2000, 2001), and perspective-taking is probably crucial to managing monogamous pair-bonded relationships, since monogamy requires close coordination between the pair in a way that is not as necessary in the more fluid social systems that characterize most birds and mammals. Perspective-taking may thus have provided the initial step that started the evolutionary process that eventually gave rise to the evolution of full-blown mentalizing (Dunbar, 2011b). This would explain why large brains are associated with monogamous mating systems rather than with group size in birds and non-primate mammals (Shultz & Dunbar, 2007; Pérez-Barbería et al., 2007).

In humans, damage to these prefrontal regions is associated with dramatic (and usually catastrophic) changes in personality and empathy, commonly resulting in socially inappropriate behavior (Adolphs, 1999) as well as more directly utilitarian responses on emotionally salient moral dilemmas such as the “trolley task” (Koenigs et al., 2007). More broadly, there is evidence from clinical studies that lesions in the prefrontal cortex tend to disrupt the processing (manipulation) of knowledge as well as social skills, whereas lesions in the temporal cortex tend to disrupt factual knowledge but leave the processing of social knowledge unaffected (Roca et al., 2010; Woolgar et al., 2010). Low densities of gray matter in the prefrontal cortex have also been linked to socially dysfunctional conditions such as schizophrenia (Lee et al., 2004; Yamada et al., 2007). More importantly for present purposes, individual differences in mentalizing competences in normal human adults correlate with the volume of neural matter in the key regions of the theory of mind network, especially those in the frontal lobes (Lewis et al., 2011; Powell et al., 2010, 2014).

Seeley et al. (2007) have suggested that the regions associated with mentalizing constitute two distinct functional networks: an “executive control” network (involving mainly the dorsolateral prefrontal cortex and parietal areas) and an “emotional salience” network (involving mainly the anterior insular cortex and the anterior cingulate cortex, the amygdala and the hypothalamus), although the former may be specifically associated with rational thinking (“fluid IQ”) rather than social cognition per se (Woolgar et al., 2010). Nonetheless, emotion and cognition are not entirely independent of each other: the anterior insula and the medial prefrontal cortex are included in both networks, suggesting some level of interaction between the two networks (Craig, 2009).

The prefrontal cortex seems to be crucially involved in the management of social relationships in both humans (Powell et al., 2010, 2012, 2014; Lewis et al., 2011; Kanai et al., 2012) and macaques (Sallet et al., 2013). More importantly, perhaps, Powell et al. (2012) have shown, using path analysis, that there is a clear causal sequence here: individual differences in orbitofrontal cortex volume determine mentalizing competences (how well individuals do on multi-level/multi-individual false belief tasks), and mentalizing competences in turn determine the individual’s social network size. In humans, the medial and mid-prefrontal cortex is also associated with moral judgment, critical assessment, and core executive functions related to self-control, deception, and lying (MacDonald et al., 2000; Karton & Bachmann, 2011), all of which are associated with both social skills in general and theory of mind in particular.

This relationship between mentalizing competences and the volume of the frontal lobe in humans seems to be mirrored in the comparative evidence from primates. It is generally accepted that monkeys do not have theory of mind (second order intentionality) and are thus effectively first order intentional (they are aware of their own mental states, but not those of other individuals). In contrast, there is some evidence to suggest that great apes do understand others’ mind states (chimpanzees: O’Connell & Dunbar, 2003; Hare et al., 2000, 2001; Crockford et al., 2012; orangutans: Cartmill & Byrne, 2007): although they are certainly not as good at formal theory of mind (the ability to pass false belief tests) as 6-year-old children (almost all of whom are fully expert on the task), they are about as good as 4-year-olds (most of whom are on the verge of acquiring this skill). By contrast, normal adult humans have been repeatedly shown to cope with fifth order intentionality (Kinderman et al., 1998; Stiller & Dunbar, 2007; Powell et al., 2010). For the limited data available, these competency levels turn out to map linearly against frontal lobe volume (Fig. 4). Figure 4 also plots the putative position of other monkey and great ape species for whom frontal lobe volume data are available on the assumption that their mentalizing competences are the same as those of the other members of their respective taxa. Notice how all these points cluster very tightly around the regression line: no monkey has a frontal lobe volume large enough to move it up to second order, and no great ape has one small enough to move it down to first order or large enough to move it up to third order. These data seem to tell us that mentalizing competences (whatever they actually are) are a function of the absolute volume of the frontal lobes (and most likely gray matter regions within the prefrontal cortex). It is important to appreciate that we still do not really understand what theory of mind (or mentalizing, more generally) actually involves cognitively (Roth & Leslie, 1998). Nonetheless, it seems to provide us with a convenient and reliable natural scale of social cognitive abilities, whatever the actual cognitive mechanisms involved may be.

Open in new tab

Figure 4. Mentalizing competences (indexed as the maximum achievable order of intentionality) of six Old World monkey and four great ape species, plotted against frontal lobe volume. Monkeys are generally assumed to be first order intentional; experimental evidence suggests that chimpanzees and orangs are just about second order intentional, whereas adult humans are fifth order intentional. Species for whom mentalizing competences have been estimated experimentally (left to right: chimpanzees, orangutans, and humans) are indicated by solid symbols; species for whom mentalizing competences are not known but who are assumed to have the same mentalizing competences as other members of their taxonomic family are indicated by open symbols. (Redrawn from Dunbar, 2009. Frontal lobe volume data from Bush & Allman, 2004.)

Indeed, the conventional mentalizing (or intentionality) scale essentially treats all competences below full theory of mind (i.e., second order intentionality) as a homogeneous set. This is almost certainly a radical oversimplification. Sperber & Wilson (1986) argued that there is a series of finer scale gradations at the lower end of this scale (see also Gärdenfors, 2012). This makes sense in the light of the fact that, behaviorally, some species of animals (baboons, macaques, spider monkeys) seem to be socially and cognitively more complex than other species (e.g., colobines, howlers, antelope) (Deaner et al., 2006, 2007; Shultz & Dunbar, 2010b), despite the fact that on the conventional scale all would be regarded as first order intentional. Unpacking the lower end of the scale may allow us to evaluate better the cognitive differences between the different nonhuman species.

Passingham and Wise (2012) have pointed out that anthropoid primates are characterized by the evolution of entirely new regions in the prefrontal cortex (in particular Brodman area 10, the frontal pole; Fig. 3) that are not present in prosimians or other mammals (see also Sallet et al., 2013). They argue that these new regions allowed monkeys and apes to engage in cognitive strategies that other mammals (including prosimian primates) are unable to master. These include one-trial learning (as opposed to more laborious forms of association learning), propositional reasoning, and the capacity to compare the future consequences of two or more alternative behavioral strategies (Passingham & Wise, 2012). Among the anthropoid primates, it seems that only the callitrichids (marmosets and tamarins) lack area 10—which might account for this taxon’s unusually labile social system, which can flip rapidly between monogamy, polygamy, polygynandry, and polyandry (Dunbar, 1995a,b; Opie et al., 2013), and the fact that their neocortex:group size ratio is completely out of line with those of all obligately monogamous primates (Dunbar, 2010b).

So far in this section, we have focused in a rather conventional way on the neuroanatomy of sociality. There is, however, an important aspect of the neurobiology of primate sociality that we need to consider, and this has to do with the role played by neuroendocrines. Much fuss has been made of the role of oxytocin in social relationships (Insel & Young, 2001); this mechanism is certainly widely distributed among mammals and has been shown to correlate with some aspects of social behavior in both chimpanzees (Crockford et al., 2013, 2014) and humans (Kosfeld et al., 2005) (for an overview, see Dunbar 2010c). However, the oxytocin system habituates very quickly (Dunbar, 2010c). More importantly, it is an endogenous response that appears to be insensitive to relationship quality or quantity: it causes individuals to act more or less affiliatively depending on the expression of the relevant gene, but it does not allow them to influence the responses of the individuals with whom they interact. It has been argued that the very unusual kind of bonded social relationships that are found in anthropoid primates (Silk, 2002; Shultz & Dunbar, 2010a; Massen et al., 2010) necessitated a more robust bonding mechanism, and this involved exploiting the endorphin system (van Wimersma Greidanus et al., 1988; Panksepp et al., 1997; Depue & Morrone-Strupinsky, 2005; Curley & Keverne, 2005; Broad et al., 2006; Barr et al., 2008; Dunbar, 2010b; Machin & Dunbar, 2011; Resendez et al., 2013).

In primates, endorphin activation is triggered by social grooming (Keverne et al., 1989), and we have been able to show, using positron emission tomography (PET), that light stroking of precisely the kind that so characteristically defines social grooming in primates also triggers endorphin activation in the human brain, and frontal lobe in particular (Nummenmaa et al., under review). It seems likely that this mechanism is mediated by the afferent c-tactile neurons, a unique set of unmyelinated (hence slow) neurons that respond only to slow stroking and which are not associated with a return motor loop from the brain (Olausson et al., 2010; Morrison, 2012; Vrontou et al., 2013). The significance of this is that the endorphin system responds exogenously (i.e., it is triggered in the recipients of grooming by their social partners, rather than merely endogenously in the groomer as is the case for oxytocin) and so is more responsive to both the quantity of time invested in a relationship and the number of social partners. An endorphin agonist, such as morphine, increases the attractiveness ratings of faces as well as the motivation for continuing to view them, whereas antagonists like naltrexone decrease both (Chelnokova et al., 2014). Similarly, PET studies reveal that the density of μ‎-receptors (the opioid receptors that have a particular affinity for β‎-endorphins) in core areas of the brain correlate with both the size of personal social network (Nummenmaa et al., under review) and an individual’s attachment style (Nummenmaa et al., in press). These findings suggest a central role for endorphins in the processes that underpin social relationships.

Building a close relationship with someone requires time, and there is a strong correlation between time devoted to socializing with an individual and willingness to support or offer help to that individual in both monkeys (Dunbar, 1980, 2012a) and humans (Roberts & Dunbar, 2011; Curry & Dunbar, 2013; Sutcliffe et al., 2012; Curry et al., 2013). By triggering endorphin activiation, time spent interacting—grooming in the case of primates, engaging in laughter (Dunbar et al., 2012b) and perhaps other activities as well as affective touch (Nummenmaa et al., in press) in the case of humans—probably sets up an emotional attachment that allows a very rapid response based on a quantitative index of the quality of the relationship.

In sum, primate social bonding seems to involve a two-process mechanism. In effect, the endorphin system is used to create an internal psychopharmacological platform that enables the individuals to develop a more cognitive long-term relationship that involves reciprocity, obligation, and trust (Sutcliffe et al., 2012). The latter, of course, is where the social brain comes in, but it is important to appreciate that beneath the simple group–brain size correlation there is a more complex neurobiological story as well as a more complex behavioral superstructure that is supported by these neurological mechanisms.

Neuropsychological research offers considerable potential for understanding both the processing demands of different kinds of cognition and how these relate to neurological pathways in the brain, and hence to the volumetric demands on different brain units and their interconnections (see also Mars et al., 2014). Although there has been considerable interest in social cognition in the recent neuroimaging literature, much of it has typically been concerned with judgments of trustworthiness or with reward and punishment in simple dyadic contexts (e.g., Knoch et al., 2006; Behrens et al., 2008; Lebreton et al., 2009). While this clearly provides valuable insight into how such judgments are made, it does not really capture the richness of the social world in which humans and other primates live. Nor does it engage with the question of just how and why humans differ from other primates, or why anthropoid primates differ from other mammals not just in cognitive abilities but also in their social style. It is these issues that need to be addressed, and so far they have been conspicuous by their absence from the literature on brain evolution.

Social Cognition and Human Evolution

Human evolution has always been viewed through the lens of anatomy and archaeology, with a clear focus on the “stones and bones” of the archaeological record. While this has spawned an interest in the cognitive aspects of human evolution (sometimes referred to as cognitive archaeology; Renfrew & Zubrow, 1994), in practice the focus has been on task analyses of the demands of tool-making (e.g., Gowlett, 2006). More recently attempts have been made to relate these to mentalizing abilities (Barham, 2010; Cole, 2012). However, Gamble et al. (2011) and Gowlett et al. (2012) remind us that the processes of evolution, and human evolution in particular, do not proceed through material culture as such but through the behavior and minds of the people who made the material culture. Here, social cognition is likely to play an especially important role, and, difficult as this may be to study, it needs to be given much more attention.

Although archaeologists have shied away from grappling with social and cognitive evolution, our growing knowledge of the finer details of the cognitive differences between both human and other primates and, at the level of individual differences, within humans offers the possibility of a more principled approach. Given the explicit quantitative relationships between social and cognitive traits and brain (or brain region) volumes, it may, for example, be possible to make more informed inferences about human cognitive evolution. We do not, of course, have access to soft tissue morphology from fossil species, but there has been a long tradition within paleoanthropology of making inferences about brain composition from the impressions created on the inside of the skull by the brain (Bruner, 2010; Bruner et al., 2003). More importantly, perhaps, the tight allometric scaling between brain regions in living primates allows us to make inferences about the sizes of these units in fossil specimens, given observed cranial volumes. It is, of course, necessary to be cautious in interpreting individual cases, given that there are well-known exceptions to these allometric relationships in living primates (e.g., the large cerebella and small neocortices of the gorilla and orangutan that we noted earlier). Other exceptions include the impact that latitude has on the size of the visual system in both modern humans (Pearce & Dunbar, 2012; Pearce & Bridge, 2013) and Neanderthals (Pearce et al., 2013), which in the latter case at least results in a smaller neocortex than would be predicted on the basis of cranial volume. Nonetheless, such extrapolations from general equations can tell us something about the overall pattern of evolution. What is important here is that these trajectories are not open-ended: we know roughly where the trajectory started (essentially, the brain composition and cognition of great apes) and where it ended (those of modern humans); our problem is to infer how the changes that must have occurred are strung out between these two endpoints. This will be illustrated here with just two contrasting examples.

The easiest and most secure extrapolation is that for social group size, since the social brain relationship is robust and empirically well substantiated. Using standard allometric equations to interpolate from cranial volume to neocortex volume, we can estimate the community sizes for individual fossil specimens of the main hominin species (Fig. 5). The community sizes for living chimpanzees are shown on the left side of the graph for comparison. Two things may be noted. First, for most of early human evolution (the australopithecine phase, represented by the genus Australopithecus and its allies) predicted community sizes do not differ from those observed in living chimpanzees. In effect, early hominins were just ordinary great apes. Second, community size undergoes a rapid increase with the appearance of the genus Homo at around 2 million years ago, stabilizes for about a million and a half years, and then increases rapidly and exponentially through archaic humans (Homo heidelbergensis and allies) into modern humans. To the extent that community size represents the outcome of the cognitive processes that underpin the social brain, these data reflect the pattern of change in cognition over time.

Open in new tab

Figure 5. Median (±50% and 95% ranges) social group for the main hominin species, in temporal order of appearance. Social group is estimated by interpolating through a series of equations from cranial volume, via brain size and neocortex size, to group size (using the relationship shown for apes in Fig. 1). The equations are given in Aiello & Dunbar (1993) and Gowlett et al. (2012). The values are for individual fossil specimens. The equivalent values for individual chimpanzee populations, based on actual community sizes, are shown on the left. (After Gowlett et al., 2012.)

We can, however, go one step further by considering cognition directly in the form of mentalizing competences, bearing in mind that these are almost certainly simply an emergent index of more conventional forms of cognition. Given that these appear to correlate linearly with the size of the frontal lobe, and, in general, brain units all correlate with total brain volume, it is in principle a simple matter of interpolating through a series of equations from cranial volume to mentalizing abilities. These are plotted in the same way for all major hominin species in Figure 6. The values for Neanderthals are corrected to take account of their larger occipital lobes and smaller frontal lobes, reflecting their relatively larger visual system (Pearce et al., 2013). Once again, our benchmarks are provided by great apes at level 2 intentionality and modern humans at level 5, and our problem is simply to decide the pattern of change between these two fixed points.

Open in new tab

Figure 6. Median (±50% and 95% ranges) mentalizing competences, indexed as the maximum achievable level of intentionality, for the main hominin species, in temporal order of appearance. Mentalizing competences are estimated by interpolating through a series of equations from cranial volume, via brain size and frontal lobe volume, to intentionality level (using the relationship for Fig. 5, and the equation for mentalizing competences from Dunbar 2010a). The values are for individual fossil specimens. (After Dunbar, 2015.)

Two points may be noted from this graph. First, once again, australopithecines were simply jobbing great apes, with no particular pretensions to advanced cognition. Second, all fossil anatomically modern humans (i.e., members of our own species) typically achieve level 5 intentionality, but no archaic humans (including Neanderthals) were likely to have done so. To be sure, all of these would have made level 4 intentionality, which, in the grand scheme of things, is itself pretty impressive: they would not have been intellectual slouches by any means. In cognitive terms, they would have been in the same bracket as the lower end of the normal distribution for modern human adults, and at about the same intellectual level as young teenagers. However, this key difference between archaic and modern humans would have had crucial implications in respect to their capacities for both language and culture.

In normal adult humans, individual differences in the ability to manage complex multi-clause sentences correlates one-to-one with mentalizing competences (Oesch & Dunbar, under review). In other words, mentalizing competences seem to determine how complex our language can be, and this would have had inevitable consequences both for the length of the propositional chains that Neanderthals could have managed and, hence, for the complexity of the stories they told. It may also have had implications for the complexity of the culture that these species would have been able to produce, and this at least seems to be borne out by the archaeological evidence. Attempts to claim that Neanderthal culture was as complex as that of contemporary anatomical modern humans (e.g., Zilhão et al., 2010) notwithstanding, the fact is that neither the Neanderthals nor the other archaic humans produced cultural artefacts that were nearly as sophisticated as those of contemporary anatomically modern humans (Klein, 1999). Neanderthal tools lacked both the technical sophistication of those developed by modern humans (multi-component tools like bows and arrows or spear-throwers) and the capacity to miniaturize (fine bone and flint points that functioned as arrowheads, buttons, awls, needles), and there is no evidence at all to suggest that they ever produced the kinds of “frivolous” material culture (Venus figurines, toys, cave paintings) that modern humans began to produce in abundance around the time the Neanderthals went extinct (Dunbar, 2015). This may be associated with the fact that several genes associated with both brain enlargement and neural efficiency in humans show evidence for strong recent selection (Burki & Kaessmann, 2004; Evans et al., 2005; Mekel-Bobrov et al., 2005; Uddin et al., 2008; Wang et al., 2008). This does not, of course, mean that Neanderthals were, as a result, in any sense intellectually primitive: it simply means they were not yet quite in the same league as modern humans, and this necessarily has consequences for what they could accomplish in social, cultural, and ecological terms.

On a more general note, human evolution provides a framework within which modern human behavior and cognition can be understood. It can tell us why we ended up the way we are, and so provide insights into the design, and perhaps flexibility, of the human mind. The importance of this historical framework is frequently overlooked in psychology, with its emphasis on the mechanisms and development of behavior in the here and now. Asking how and why we got to be the way we are can tell us a great deal about those mechanisms, especially when seen against a background of primate cognitive and social evolution. And it should remind us, above all, that human social evolution, like that of all primates, is not simply about individual traits but about how these traits enable us to live in an extensive, complex, highly dynamic social world.