Matters of factual detail

Several commentaries (van Leeuwen & Roth, Krupnick, Shi) allude to the fact that orangutans are solitary. They are not. They certainly forage on their own (in most, but not all, habitats), but in actual fact, they are very social. This is why they are usually housed in groups in zoos; on a number of behavioural markers, they are as social as gorillas (Dunbar, Reference Dunbar2026). Orangs live in (dispersed) communities, just like chimpanzees do. This is because both species are living under ecologically challenging conditions on the limits of a much wider former biogeographic distribution – in the case of orangs, one that extended right up into mainland China. Orangs just take the enforced dispersal to its limit. Under further climate warming, both gorillas and chimpanzees will do the same (Korstjens, Lehmann, & Dunbar, Reference Korstjens, Lehmann and Dunbar2010; Lehmann, Korstjens, & Dunbar, Reference Lehmann, Korstjens and Dunbar2010). The community size for orangs, based on two reliable field studies, is in fact 14 (about what would be expected for their neocortex ratio, and very similar to gorillas). When we undertake comparative analyses, it is essential to compare like with like; we should not compare foraging group size in one species (orangs) with social group size in others (gorillas and chimpanzees).

Bshary & Triki and Shi imply that primate brain evolution may be better explained by the demands of foraging rather than sociality, citing DeCasien, Williams, and Higham (Reference DeCasien, Williams and Higham2017) as evidence. The apparent desperation to defend at all costs an exclusively foraging explanation for everything is beginning to read like a religious mantra, more especially given that, unlike the social brain hypothesis, the foraging hypothesis hasn’t actually moved on conceptually at all since the 1980s. We still have no idea what is involved, how cognition influences food intake, or whether individual differences in foraging skills influence either survival or fertility; in contrast, there is very extensive evidence on all three points in respect of social skills, not just for primates but also for humans and for other large-brained group-living mammals (summarised in Dunbar, Reference Dunbar2018b, Reference Dunbar2025).

The real problem, however, is that the DeCasien et al. analysis (and that of Powell, Isler, and Barton [Reference Powell, Isler and Barton2017], who made a similar claim) is statistically flawed beyond possibility of redemption. They used ordinary least squares regression when the data do not meet the minimum standards required and ignored grades (thereby falling doubly foul of the Yule-Simpson paradox, a version of the ecological fallacy, which, in the limit, can cause the sign of a slope to reverse). In addition, by reversing the causality of the axes in order to run a multiple regression, they inadvertently tested a developmental hypothesis when they thought they were testing a selection hypothesis (thereby falling foul of Dobzhansky’s Dictum) (see Dunbar & Shultz, Reference Dunbar and Shultz2021a, Reference Dunbar and Shultz2023; Shultz & Dunbar, Reference Shultz and Dunbar2022). Asking a different question will inevitably yield very different results, as Wartel, Lindenfors, and Lind (Reference Wartel, Lindenfors and Lind2019) pointed out. The only statistically correct way to test hypotheses of this complexity is by path analysis. Four independent path analyses of different primate datasets (including both the DeCasien et al. and Powell et al. datasets) make it quite clear that sociality is the correct answer (with foraging skills as a constraint, not a selection factor) (Shultz & Dunbar, Reference Shultz and Dunbar2022). Of ∼35 tests of the social brain using different datasets and a variety of methods, these are the only two that commit these errors and the only two to obtain contrary results (Dunbar & Shultz, Reference Dunbar and Shultz2023). Both claim that their analyses are novel in several ways, but in fact they are not: they use the same data and the same statistical methods as everyone else – they just ask a different question. I need say no more.

Shi reminds us that many small-brained species (insects, birds) live in large, complex social groups. This misses the point: the social brain hypothesis is actually about animals’ abilities to manage relationships, not about group size or organisation (Dunbar, Reference Dunbar1998). Group size is not a trait that natural selection can act on: it is an emergent property of animals’ social competencies (which natural selection can select for). For the record, we and others have shown, in many analyses, that the social brain hypothesis takes the form of a much simpler categorical relationship between large-brained monogamous species and small-brained promiscuous species in insectivores, carnivores, ruminant ungulates, bats, prosimians, and birds (with lifelong pairbonder species having bigger brains than annual pairbonders in the latter) (Shultz & Dunbar, Reference Shultz and Dunbar2007, Reference Shultz and Dunbar2010a,Reference Shultz and Dunbarb; Pérez-Barbería, Shultz, & Dunbar, Reference Pérez-Barbería, Shultz and Dunbar2007; Dunbar & Shultz, Reference Dunbar and Shultz2025a). It seems that pairbonded relationships are hard cognitive work (as, I am sure, most of us will be only too painfully aware). Anthropoid primates, and a very small number of other mammal taxa, extend the same cognition to a wider set of individuals.

A second common misperception (also repeated by Bshary & Triki) is that the social brain hypothesis is some kind of non-ecological alternative to the ecological hypotheses. The social brain is an ecological hypothesis (Dunbar & Shultz, Reference Dunbar and Shultz2017; Shultz & Dunbar, Reference Shultz and Dunbar2022). The issue is, and always has been, whether individuals solve their ecological and life history (Timbs & Manages, please note) problems individually by trial-and-error learning or socially as a group (in which case extra cognitive work is required to maintain group cohesion through time, given that this doesn’t come for free). The cultural intelligence hypothesis (van Schaik & Burkart, Reference Schaik van and Burkart2011), for example, is a subversion of the social brain hypothesis (it requires animals to live in stable groups so as to allow information exchange), not an alternative to it. The main difference between the two positions is simply whether energy intake (which doesn’t require group-living) or predation risk (which does) is the principal factor limiting an individual’s fitness.

A couple of minor points. Ostner & Schülke emphasise the importance of using phylogenetic methods (though without offering any reasons why). Phylogenetic methods are intended to guard against the risk of type I errors (rejecting the null hypothesis when it is true) due to the fact that closely related species are likely to inherit the same traits (the pseudo-replication problem). However, good statistical practice reminds us that they should only be used when there is evidence for a phylogenetic signal, since introducing an additional variable into a multiple regression inevitably reduces the degrees of freedom, and hence statistical power, thereby increasing the risk of type II errors (accepting the null hypothesis when it should be rejected). (This being so, we could in fact solve the statistical problem more easily by simply reducing the degrees of freedom.) In reality, contrary to their statement, the phylogenetic signal for group size and most behavioural and cognitive indices in primates is very low (Kamilar & Cooper, Reference Kamilar and Cooper2013), and at genus level, non-existent (Dunbar & Shultz, Reference Dunbar and Shultz2021a,Reference Dunbar and Shultzb). Unfortunately, it seems that phylogenetic methods have become something of a virtue-signalling emblem, and are hence applied de rigeur without any forethought, often resulting in a significant reduction in the power of the statistical test being used.

Mention was made of a role for oxytocin and serotonin (Zhiou et al.). Oxytocin certainly plays some role in sociality, but in fact it is limited to mother-infant bonding and romantic relationships (having been evolutionarily co-opted from its fish origins, where it manages homeostasis of water balance to manage lactation in mammals). Its main disadvantages are that it is very short-lived (its half-life in the venous system is on the scale of minutes because it is excreted very quickly) and that it affects only the actor (if you have the right gene, you act prosocially). Not only are endorphins much longer-lasting (with a half-life in the order of hours), but their real benefit is that, through social grooming, they allow you to make other individuals prosocial towards you (see Dunbar, Reference Dunbar2018b, Reference Dunbar2025; Bzdok & Dunbar, Reference Bzdok and Dunbar2022). At least in humans, we did not find evidence for any involvement of oxytocin at any social level other than romantic pairbonds, and even then endorphins were important (Pearce et al., Reference Pearce, Wlodarski, Machin and Dunbar2017a, Reference Pearce, Wlodarski, Machin and Dunbar2018). Indeed, it seems that even the original vole pairbonding claim is better explained by endorphins (Resendez et al., Reference Resendez, Kuhnmuench, Krzywosinski and Aragona2012, Reference Resendez, Keyes, Day, Hambro, Austin, Maina and Aragona2016; Inoue, Burkett, & Young, Reference Inoue, Burkett and Young2013). Similarly, serotonin may do many things physiologically, but it has only a very limited social signature (Pearce et al., Reference Pearce, Wlodarski, Machin and Dunbar2017a). The issue here is not whether different neurohormones play a role in regulating social interactions, but which one(s) are crucial for social bonding and network coherence. The answer seems to be only β-endorphin and its stable mate dopamine (the two seem to work in a push/pull tandem fashion: see also Loui & Margulis).

Shi noted that many birds and insects live in complex societies, with some insects using pheromonal mechanisms to regulate sociality. The latter might be seen as paralleling the role that endorphins play as a form of social “glue” in primates. However, what these produce in insects is a highly regulated and regimented society – as is, in fact, also the case in mammals like mole rats who have a similar olfactory-based caste system. This does not make these societies complex. What makes some mammal and bird societies complex is that the animals’ behaviour is facultative and highly flexible in a way that allows them to adjust their behaviour and social arrangements to differing environmental and demographic conditions. They can form “friendships” with complete strangers, for example, or choose one family member in preference to another. No insect does that.

Papadopoulis, Michlich & Andrews are concerned that my argument is anthropomorphic in that it sets humans up as the ideal to which all species must aspire. Aside from being an oddly Lamarckian “Great Chain of Being” view (discredited as long ago as the 1860s: Dunbar Reference Dunbar2019), at no point do I suggest that what humans do is “better.” Humans are mentioned at all only to indicate that they fall within the same spectrum as the great apes and the very social monkeys. Herd formation (aggregations) and stable group formation (congregations) are alternative adaptive peaks that function best under different environmental and biological conditions, with the first being far more common in mammals as a whole (Sutcliffe, Dunbar, & Wang, Reference Sutcliffe, Dunbar and Wang2016). Having personally carried out field work on the behaviour and ecology of seven primate species, six ungulate species, two species of birds and one insect (dragonflies) (but, mea culpa, no fish or reptiles), my views are not coloured by what humans do, or by armchair preconceptions about what animals are presumed to do; they are coloured simply by what animals actually do.

Issues of theory and alternative mechanisms

Several of the commentaries offer interesting alternatives to grooming as the basis for the formation of stable groups: Wolf emphasises joint attention, van Heijst & Kret, mimicry and attention, Zhiou et al. focus on oxytocin, Pelagi et al. highlight play, Benítez-Burraco prefers aggression, and van Leeuwen & Roth and Zhiou et al., tolerance, while van Leeuwen & Roth and Shi favour the complexity of dominance hierarchies. All of these surely play a role – although, to be honest, it is difficult to see how mimicry, for example, can, in and of itself, influence the strength of social bonds in primates, given that even in humans the evidence is equivocal (Rauchbauer et al., Reference Rauchbauer, Jank, Dunbar and Lamm2023). Dominance rankings have always been considered an important feature of social organisation, but I would argue that any species that resorts to dominance hierarchies is likely to be one that relies on brute force and ignorance to get its way on a take-it-or-leave-it basis, rather than one that uses more sophisticated social strategies to negotiate a more equitable division of the spoils. By the same token, attention is really an outcome of having a relationship, not the determinant of the relationship (Dunbar & Shultz, Reference Dunbar and Shultz2010). Attention might indicate a desire to have a relationship, but it doesn’t create one: a stranger who stares constantly at you can make you nervous – in baboons as much as in humans. Important as all these behaviours are socially, we need to remember that they only describe the bricks and mortar of dyadic relationships; our concern should also be to understand the building (i.e. groups and networks) they are used to create, and the functions these are designed to subserve.

A number of commentators (Benítez-Burraco, van Leeuwen & Roth, Zhiou et al.) argue for the importance of tolerance in understanding social evolution. I agree: tolerance is one of the skills of diplomacy that makes group-living possible for both animals and humans. But, like attention, tolerance is a behavioural description, not an explanation. It must depend on some kind of brain activity, and the most plausible suggestion is inhibition (self-control). That aside, we have no idea what tolerance actually is, how it is distributed across species, or what brain processes it requires. In contrast, we know a great deal about inhibition: we can measure it reliably using suitable behavioural tests, we know it is only found in anthropoid primates, and we can show that it varies significantly across species in a way proportional to core brain regions (Shultz & Dunbar, Reference Shultz and Dunbar2010c). Integrating work on tolerance and inhibition would surely be helpful.

Kaufman, among others, seems to take the view that I argue for an exclusively dyadic view of relationships. I am not sure why. The whole emphasis of the target article is on how to build an alliance that consists of several individuals integrated into a network, not a single dyadic relationship. Networks are, by definition, full of third-party relationships that are part of the deal. I have been arguing (for the past 50 years!) that, if we really want to understand social evolution in species like primates where group-living has been the key to their evolutionary success, we need to take a group-level perspective and consider multilevel selection (see also Sueur & Deneubourg).

De Dreu et al., Loui & Margolis, Pelagi, Caruana and Burghardt, Ritov and Jacob & Engelman raise matters that relate specifically to humans. I thank them for drawing attention to this genuinely interesting species and placing it in its proper context as part of a continuum with the other primates. Humans were not part of my remit for this target article, though I did include brief mention of them in order to point to some of the many continuities between humans and other primates. I have published experimental studies on the role of laughter, singing, dancing, music-making, rituals, feasting, and storytelling as crucial mechanisms that humans use for social bonding, both at the dyadic and the group level, and on the way all of them trigger the endorphin system (confirmed in several cases by positron emission tomography brain imaging or chemical deactivation of the brain’s endorphin system) (for a summary, see Dunbar, Reference Dunbar2021). Many of these studies were cited in the target article.

It is true that I did not consider play (Pelagi et al.), but that was because play is an activity that, outside of humans, is generally found only in juveniles, where it plays a vital role in the learning processes whereby mentalising, inhibition, and other socially crucial cognitive skills are fine-tuned. For reasons that elude me, the social brain often seems to be viewed as implying that animals are born capable of sustaining groups of a certain size. The social brain provides only capabilities; considerable practice is needed to be able to use these skills effectively in social contexts (as I have made clear right from the earliest papers: see Dunbar, Reference Dunbar1998). This practice has to be done in face-to-face social contexts (and not online…!). This is why, as we have shown, the length of the period of socialisation (i.e. weaning to puberty – but not the period of parental investment from conception to weaning!) correlates, across primate species, with the size of both the neocortex (not the whole brain!) and group size (Joffe, Reference Joffe1997). Without play, there would be no social brain, and hence no groups.

Several commentators (Bshary & Triki, Lee & Strier, Ostner & Schülke, Timbs & Menanges) emphasise the primacy of environmental quality and ecological competition (with passing mention of life history and behavioural ecology more generally) as the principal determinant of variation in both fertility and group size. With respect to fertility, the evidence is clear: both environmental quality and the number of females in the group are determinants, but they act independently (Hill, Lycett, & Dunbar, Reference Hill, Lycett and Dunbar2000; Dunbar, MacCarron, & Shultz, Reference Dunbar, MacCarron and Shultz2018a; Dunbar, Reference Dunbar2019 Reference Dunbarb; Dunbar & Shultz, Reference Dunbar and Shultz2021a). Environmental effects are random, and hence, in and of themselves, unlikely to influence the course of evolution – although they may have significant demographic consequences (e.g. by driving groups or populations to extinction). More importantly, the evidence is clear that environmental effects only start to influence fertility when animals have lost >15% of their body mass (i.e. are on the verge of starvation); in contrast social stress can cause females to become infertile even when they are fat and have access to unlimited food (Dunbar & Shultz, Reference Dunbar and Shultz2021a). In any case, the real cost of reproduction to female mammals is lactation, when females can lose up to 10% of their body mass (Dunbar & Dunbar, Reference Dunbar and Dunbar1988; Dunbar, Anderson, & Knight, Reference Dunbar, Anderson and Knight2025; Dunbar, Reference Dunbar2026).

I should add that we have explored the environmental determinants of group size (and variability) in a series of genus-specific, population-level time budget models (based on a linear programming approach) that have explicitly studied this in all the major Old World primate genera. Individual models have been built for Colobus, Piliocolobus, Cercopithecus, Chlorocebus, Theropithecus, Papio, the hylobatids, Pongo, Gorilla, and Pan (with a model for Macaca in progress), as well as some platyrrhines (Ateles, someone else was having a go at Alouatta) and an ungulate (with another antelope genus in progress) (for an early overview, see Dunbar, Korstjens, & Lehmann, Reference Dunbar, Korstjens and Lehmann2009). These models are more accurate in predicting species’ biogeographical distributions than the climate envelope models currently used by conservation biologists. In addition, they tell us which aspect of a species’ biology constrains its distribution, as well as how much ecological stress individual populations are under (neither of which conventional biogeographic models do). One thing that emerges from these models is that, except on the margins of their biogeographical distributions, all species typically live in significantly smaller groups than their ecologies would allow. That deficit can be attributed to the infertility trap, but it also provides a buffer for the group to expand into during the lead up to group fission (Dunbar et al., Reference Dunbar, Korstjens and Lehmann2009; Dunbar, MacCarron, & Robertson, Reference Dunbar, MacCarron and Robertson2018b).

Shi suggests that the coordination problem may not be such a serious problem as I make out – after all, starling flocks produce the most spectacular aerobatic displays using a very simple rule (do what your immediate neighbour does). Primates certainly exploit the same mechanism, but this does not mean that there is nothing more to what primates do. Mathematical modelling has shown that a set of coupled oscillators (i.e. foraging animals) will inevitably drift out of behavioural synchrony during the course of foraging (hence leading to the group breaking up) unless there is some kind of gravitational force (or social “glue”) to maintain synchrony (Schaposnik, Hsu, & Dunbar, Reference Schaposnik, Hsu and Dunbar2025). We should note, by the way, that starling murmurations only occur just before nesting in the evening, in perfect visibility conditions when everyone is free to do the same action, and normally only in winter; they never occur during the day when the birds are foraging or travelling to and from feeding areas. In other words, monkeys can do something starlings cannot, namely, maintain group coordination in poor visibility conditions, while foraging, all year round. If you think primate foraging is just a less complex two-dimensional version of a starling murmuration, try spending a day with monkeys foraging in dense vegetation. You will end up on your own; they won’t.

Several commentators (Ritov et al., De Dreu et al.) emphasised the importance of cooperation, as opposed to coordination. The evolution of cooperation has been a matter of longstanding interest in evolutionary biology, psychology, anthropology, and economics. It has proved difficult to find a general explanation for the evolution of such behaviour – yet humans, and some animals, clearly do it. One reason for this conundrum is that cooperation is actually much rarer in real life than we tend to imagine. More importantly, the substantive issue is that there is a general tendency, mainly due to economists, to assume that social life consists of collaborations with strangers. It does not – not even in humans. Almost all our collaborations (other than rather specialised economic and financial transactions) are with members of our personal social network of ∼150 family and friends. Within this very limited circle, we collaborate freely, but once beyond it, our relationships become strictly transactional (I’ll help you out, but only if….[you pay me in some way]) (Dunbar, Clark, & Hurst, Reference Dunbar, Clark and Hurst1995; Madsen et al., Reference Madsen, Tunney, Fieldman, Plotkin, Dunbar, Richardson and McFarland2007). Even in hunter-gatherer societies, cooperation and altruism are focused on members of the group of known individuals (the clan of 100–200) – often widely misinterpreted (as De Dreu et al. do) as cooperation with strangers, on the mistaken assumption that the band (an unstable ecological unit) is the social grouping. It is not.

In other words, maybe we have been looking down the wrong end of the telescope. What comes first is living in groups – a pure coordination problem that does not involve a public good dilemma because everyone pays the same price by virtue of living in the group. Cooperation then comes for free because these groups are based on trusted relationships, thanks to repeated interactions (Dávid-Barrett & Dunbar, Reference Dávid-Barrett and Dunbar2012).

A more important question touched on by several commentators (Kaufman; Papadopoulos et al.; Shi; Sueur & Deneubourg) is how to define complex sociality. Sueur & Deneubourg argue that it is all about multilevel selection and the emergent properties of networks. My response is: welcome to the club, but where has everyone else been all this time? Nonetheless, we need to be a little more nuanced than they seem to assume. A top-down network analysis is not an explanation: it is a description. We still need a bottom-up approach to explain why the network has the form it does (i.e. the outcome of individual psychology). Similarly, multilevel selection is simply a description of how fitness is sourced; it is not an explanation as such. It becomes interesting in the context of multilevel social systems only when we can apportion different components of fitness to different social layers. This, by the way, was precisely what Hamilton [Reference Hamilton1964] was getting at with his neighbourhood-modulated fitnesses, though the point of his original analysis is invariably lost by the conventional focus on his simplified formula and his rephrasing of this in terms of “inclusive fitness.”

Bshary & Triki remind us that if fish can do anything primates do, then, if we want to claim that primates are different, we need to think about exactly how they are different and why. The answer is precisely what the target article is all about. We must be careful not to be seduced by superficialities in two separate respects. First, much confusion has been sown in the literature by naïve assumptions about the cognitive abilities that different tests index (Dunbar & Shultz, Reference Dunbar and Shultz2025b). This much has become apparent in the rather overblown claims that birds (and even cuttlefish!) have the capacity for self-control. They don’t: the tests that have been used to measure self-control actually measure causal understanding, not strategic self-control (Dunbar & Shultz, Reference Dunbar and Shultz2025b). Second, it is one thing to claim that fish have many, or even all, of the basic cognitive abilities of primates, but quite another thing to conclude from this that there are no differences between them when we haven’t tested for any of the more advanced forms of meta-cognition. Most forms of basic cognition (memory, learning, etc) are necessary for life and everyone has them. Not everyone has those forms of cognition like one-trial learning, inhibition, and mentalising that, in anthropoid primates, are built up out of these elemental components – and are dependent on specific brain regions (Brodman Area BA10) that only anthropoid primates have (Passingham & Wise, Reference Passingham and Wise2012), and hence even other mammals and birds don’t possess (Dunbar & Shultz, Reference Dunbar and Shultz2025b).

Apperley questions the multilevel nature of mentalising. Mentalising is, by its very nature, infinitely recursive without limit. The only question is the empirical one of whether (and where) the limit is in humans, or any other species. Developmental psychologists have never been interested in anything beyond second-order intentionality (formal theory of mind: I believe that you think….) since this marks the first major transition in five-year-old children. Our data consistently show that the upper limit for normal adult humans is at about fifth-order intentionality (involving five mental states – four other people’s plus your own). It seems that the mistake Apperley makes is to assume all normal adults can manage that. But an average means what it says, and that means some perform at lower and a few at higher levels (in fact, the range in normal adults is approximately 3–6 levels of intentionality, based on half a dozen studies). The value Apperly quotes for success on fifth-order intentionality tasks in his study is similar to that in our data, so his results don’t actually “disprove” anything.

That said, I agree that most everyday interactions rarely require more than third-order intentionality (I believe that you think that Susan intends….), which is why third-order adults function just fine in society. Nonetheless, it is demonstrably the case that fiction (Carney, Wlodarski & Dunbar, Reference Carney, Wlodarski and Dunbar2014), jokes (Dunbar, Launay & Curry, Reference Dunbar, Launay and Curry2016), and cartoons (Dunbar & Stirling-Middleton, Reference Dunbar and Stirling-Middleton2024) are all rated as more enjoyable when their content requires higher orders of mentalising (providing it doesn’t exceed fifth-order intentionality). That said, however, third-order adults prefer third-order stories, whereas fifth-order adults prefer fifth-order stories – in both cases, stories that challenge but don’t exceed their individual mentalising competencies (Carney et al., Reference Carney, Wlodarski and Dunbar2014), probably because individual differences in mentalising correlate significantly with individual differences in syntactic understanding (the ability to unpack clauses) (Oesch & Dunbar, Reference Oesch and Dunbar2017). In addition, fifth-order mentalising sets the limiting size on conversation groups in Shakespeare’s plays and modern films (Stiller, Nettle, & Dunbar, Reference Stiller, Nettle and Dunbar2003; Krems & Dunbar, Reference Krems and Dunbar2013), as well as everyday conversations (Krems, Neuberg, & Dunbar, Reference Krems, Neuberg and Dunbar2016). In half a dozen samples (including one with N>1400: Street et al., Reference Street, Siy, Keeling, Baranes, Barnett, Mckibben, Kanyere, Lentz, Aguera y Arcas and Dunbar2025), five is the consistent upper limit. Individual differences in this limit correlate with the size of core brain regions (Lewis et al., Reference Lewis, Rezaie, Browne, Roberts and Dunbar2011, Reference Lewis, Birch, Hall and Dunbar2017; Powell et al., Reference Powell, Lewis, Dunbar, García-Fiñana and Roberts2010, Reference Powell, Kemp, Dunbar, Roberts, Sluming and García-Fiñana2014), as well as the number of friends an individual has (Powell et al., Reference Powell, Lewis, Roberts, García-Fiñana and Dunbar2012). So it seems that even if we don’t have fifth-order intentionality, we apparently still need it.

It is important to bear in mind that, among the primates, only the three great apes have formal theory of mind (second-order intentionality). Monkeys in general, as Devaine et al. (Reference Devaine, San-Galli, Trapanese, Bardino, Hano, Saint Jalme, Bouret, Masi and Daunizeau2017) showed, operate at first order, as do most other mammals with large brains. What we do know is that social cognition becomes exponentially more demanding in terms of neural processing with successive intentionality levels compared to simple causal reasoning or memory (see Lewis et al., Reference Lewis, Birch, Hall and Dunbar2017). Indeed, in primates, species’ mentalising competences (in so far as we can determine quantitative differences in performance between species) correlate with neocortex (and especially frontal lobe) volume (Dunbar, Reference Dunbar, Botha and Knight2009, Reference Dunbar2026; Devaine et al., Reference Devaine, San-Galli, Trapanese, Bardino, Hano, Saint Jalme, Bouret, Masi and Daunizeau2017).

I have suggested that the costliness of social cognition stems from the need to create a mental model of the mindstates/intentions of the other individuals involved. Hu, Xia & Feldman-Hall make a similar point in suggesting that part of the computational demand of large groups stems from the need to create a mental model of the network. In fact, West et al. (Reference West, Massari, Culbreth, Failla, Bologna, Dunbar and Grigolini2020, Reference West, Dunbar, Culbreth and Grigolini2023) argue that primate and human social networks involve a dual mapping exercise in which the animal has to cross-map the network as it sees it with the network as it imagines it (based on past experience), without the two leaking into each other (with the latter requiring cognitive inhibition). Note that this is not a memory problem, but a computational demand problem (the amount of work the brain has to do evaluating relationships that can change dynamically by the week, while factoring in third-party relationships) (see Dávid-Barrett & Dunbar, Reference Dávid-Barrett and Dunbar2013).

Promising directions

Hamilton mentions Lossy compression as a way humans might create mega-communities, and a similar suggestion is made by Moffett in respect of the ability to categorise. The capacity to classify people into categories (kin/non-kin, my tribe, types of roles, honest/dishonest, etc) is precisely what allows humans to create mega-communities from tribes to nation states (Forgas & Dunbar, Reference Forgas and Dunbar2026), with these subject to the same scaling law as friends and family (Dunbar, Reference Dunbar2020; Sandeford, Reference Sandeford2018; Hamilton et al., Reference Hamilton, Walker, Buchanan and Sandeford2020). How far this capacity permeates down the taxonomic scale is an open question. There is, as I noted in the article, evidence that the upper grade genera (mainly cercopithecines and apes) can factor kinship as well as dominance into their social calculations. Primates probably cannot go much beyond the social group, but culture allows humans to use labels (dialect, clothing, and knowledge) to do so on a very large, anonymous scale (my tribe, my religion, my political party, and my football fan club) (Dunbar, Reference Dunbar2020; Forgas & Dunbar, Reference Forgas and Dunbar2026).

Both Waller et al. and Roberts & Roberts remind us that gestural communication in its many forms (including vocal metacommunication: see Dunbar et al., Reference Dunbar, Robledo del Canto, Tamarit, Cross and Smith2022) plays an important role in signalling relationship quality, both within a dyad and to third-party onlookers. It is surely significant that, across primates, the size of both vocal and facial gesture repertoires correlate with group size and brain size (Dunbar, Reference Dunbar2012), suggesting that communication channels become more complex in order to exploit redundancy so as to reduce risk of misunderstandings. All these are important to know about, and not well enough studied.

Hobaiter & Dominy and Lee & Strier make the cogent point that studying within-species variation provides a way of testing the hypotheses. In fact, the infertility trap analyses are based on within-taxon (species or genus) analyses. The data to test this for the structural strategies that form the centrepiece of the target article are available in online Dataset-S1: each genus is based, on average, on a sample of 4.3 groups (2.4 groups per species, range 1–13). However, it would be naïve to assume that the three strategic modes appear by magic at different group size thresholds within species: cognition is, after all, a species-specific characteristic. Nonetheless, we might expect network structure and the balance between the different strategic modes to change in subtle ways with group size. Comparisons of grooming networks across group sizes for individual, groups of Papio hamadryas (Kummer, Reference Kummer1968), Theropithecus gelada (Dunbar & Dunbar, Reference Dunbar and Dunbar1975; Dunbar, Reference Dunbar1984, Dunbar et al., Reference Dunbar, Arnaboldi, Conti and Passarella2015), Macaca fuscata (Kudo & Dunbar 2000), and Colobus guereza (Dunbar et al., Reference Dunbar, Arnaboldi, Conti and Passarella2015) have shown that network structure changes with group size within species, as well as pointing to marked differences between species.

Krupnick calls for more neuroimaging studies on solitary versus social species. I am not sure whether this would help with understanding primates, because there aren’t any solitary primates (those that forage alone, which once caused them to be classified as solitary, turn out to nest together quite contentedly). That said, a number of recent neuroimaging studies have compared lemurs, macaques, and humans, revealing striking differences in the presence and size of core brain regions important for sociality (with qualitative differences in brain organisation between lemur and the two anthropoids, and quantitative differences between the two anthropoids) (Roumazeilles et al., Reference Roumazeilles, Lange, Benn, Andersson, Bertelsen, Manger and Mars2022). More studies on a wider range of species are desperately needed, not least because our knowledge of comparative brain organisation outside the primates is still very poor.

Both Jörntell and Krupnick raise the possibility that my findings may be relevant to psychiatry, and I agree that this is very likely. The number and quality of close friends is the single best predictor of both psychological and physical health and wellbeing, as well as future longevity (Dunbar, Reference Dunbar2025). Understanding individual differences in relationship management may well be relevant to understanding some psychiatric conditions, and perhaps even how they should be treated (see also Charles, Dunbar, & Farias, Reference Charles, Dunbar and Farias2020).

Avenues for future research

A number of directions for future research emerge out of this discussion. These include, but are not limited to: (1) testing whether the structural and behavioural changes within the same species also occur at group level; (2) conducting brain imaging studies on a much wider range of primate and especially non-primate species to obtain data on brain structural organisation and, more importantly, connectivity; (3) obtaining better cognition data on a wider range of species using more trait-appropriate tasks; (4) getting a better handle on mentalising and its demands in humans (about which we actually know very little); (5) determining the extent to which the capacity to categorise individuals into explicitly (i.e. linguistically) defined types (own clan, own tribe, own religion, etc) reduces cognitive load in humans, and whether, in attenuated implicit (i.e. non-linguistic) form, this also applies to any primates; (6) determining the group sizes at which ecological competition actually becomes an issue for animals in terms of gross energy intake (not simply whether or not they displace each other); (7) examining the group coordination (behavioural synchrony) issue in more species to determine at what group sizes this becomes a problem; (8) investigating the role of chorusing in social bonding; and (9) analysing the distributions of group size in a wider taxonomic sample to determine who does, and who does not, exhibit a fractal pattern. Most of these will be obvious from the earlier discussion, but let me briefly comment on three.

We still don’t know what kinship means to animals or how they judge it – although familiarity at some early sensitive period in life is likely, since this seems to be true of humans (Madsen et al., Reference Madsen, Tunney, Fieldman, Plotkin, Dunbar, Richardson and McFarland2007). We know that, in humans at least, kin and friend relationships are processed in different parts of the brain (Wlodarski & Dunbar, Reference Wlodarski and Dunbar2016; Roseman, Dunbar, & Arzy, Reference Roseman, Dunbar and Arzy2024).

Group chorusing occurs in many species of primates as well as in some canids (wolves) and cetaceans (humpback whales). We have shown that, in humans, group singing upregulates the endorphin system as well as creating a deep sense of community bonding (Weinstein et al., Reference Weinstein, Launay, Pearce, Dunbar and Stewart2016; Pearce et al., Reference Pearce, Launay, MacCarron and Dunbar2017b). Since it has recently been shown that vocal chorusing (“flock talk”) upregulates the endorphin system in birds in a way that suggests a group-bonding function (Riters, Kelm-Nelson, & Spool, Reference Riters, Kelm-Nelson and Spool2019), this raises the question as to whether similar chorusing behaviour in primates (notably howler monkeys, some colobines, and gibbons) might actually be for group bonding rather than territorial advertising as has conventionally been assumed. The same might also be true of some forms of contact calling that have a similar chorusing component (e.g. gelada baboons).

Multilevel sociality has been the subject of considerable interest in recent years. However, species that have a multilevel form exhibit two very different forms of modularity: irregular, casual modularity, and fractal modularity (Dunbar, Reference Dunbar2026). The first are really herds: they are unstable, irregular in size, require little in the way of sophisticated cognition, and may not even offer a relevant fitness payoff (e.g. when animals simply converge on rich feeding grounds). Examples include the occasional very large groups of Colobus angolensis and Rhinopithecus (snubnosed monkeys), perhaps even the mixed species groups of tamarins and guenons (Korstjens, Lehmann, & Dunbar, Reference Korstjens, Lehmann and Dunbar2018). In species that have fractal modularity (orcas, elephants, baboons, great apes, and humans), the distribution of group sizes has a highly regular pattern, with each grouping layer being two or three times the size of the layer immediately inside it (Dunbar, Reference Dunbar2026). The two kinds of multilevel grouping should not be confused. In order to understand the taxonomic distribution of fractal modularity and how and why it exists, we need more careful analyses of the patterning in the distribution of group sizes for a much wider range of species.