The changepoints we identified document trends in brain size but do not reveal underlying causes and mechanisms of encephalization. The expensive tissue hypothesis (Aiello and Wheeler, 1995) posits that a trade-off in allocated resources from one expensive tissue (the brain) to another (the gut) was made possible by shifts to a higher quality diet in Homo, enabled in part by enhanced technological skills (Lepre et al., 2011), exploitation of diverse resources (Braun et al., 2010), and the invention of cooked foods through controlled fire (Wrangham, 2009; Herculano-Houzel, 2016). In addition to evidence that energetic constraints on encephalization may have been released by dietary shifts in early Homo, others have posited that brain expansion in early Homo may have been driven by the need for enhanced social intelligence (Dunbar, 1998; Dunbar and Shultz, 2017). With growing complexity in social life, perhaps associated with increased group size, brain expansion occurred. Resource sharing within groups and allocare may have provided the energy surplus needed to support the increased energetic cost of a larger brain (Isler and van Schaik, 2012). These hypotheses are not mutually exclusive: a combination of social challenges and ecological pressures and increased dietary quality and breadth drove Pleistocene brain expansion in our ancestors, who were living in increasingly larger groups and likely benefitted from enhanced group-level cognitive abilities and greater cultural intelligence (van Schaik et al., 2012), which could in turn accelerate brain expansion via a feedback loop (Markov and Markov, 2020).

The cause of brain reduction in the Holocene is also unclear. One possibility is that it is associated with a corresponding decrease in body mass (Henneberg, 1988; Ruff et al., 1997). The early expansion of the brain in Homo has been explained as an increase in absolute body size (Ruff et al., 1997; Wood and Collard, 1999; McHenry and Coffing, 2000; Lee and Wolpoff, 2003) though Grabowski (2016) later employed quantitative genetics to suggest that selection had favored brain enlargement and body size increase was a by-product. Hawks (2011) contends that the absolute change in brain size from the Pleistocene to the Holocene is greater than expected based on changes in body mass in the same time period, emphasizing that given the correlation between human brain and body size (Holloway, 1980), the observed 5 kg decrease in body size in the Holocene would account for only a 22 mL decrease in brain volume (Hawks, 2011). However, the actual reduction is more than 5x greater, suggesting that body size alone cannot entirely explain the decrease in brain volume. We find here that body size reduction may have preceded brain size reduction by several millennia. Ruff et al. (1997) notes that the decrease in body size in late Pleistocene humans began around 50 kyr, whereas brain reduction appears to been a Holocene phenomenon.

Yet there may be other non-allometric explanations for the reduction in human brain size at the Pleistocene-Holocene transition due to energetic, nutritional (Wiercinski, 1979), and/or developmental constraints (Hawks, 2011). The brain size reduction in Holocene humans parallels that of domesticated animals, suggesting that humans have self-domesticated by deliberately removing highly aggressive individuals from breeding populations, leading to a reduction in intra-population (but not inter-population) aggression (Leach, 2003; Hare and Tomasello, 2005; Wrangham, 2018, 2019; Bruner and Gleeson, 2019; Hare and Woods, 2020). Brain reduction in this case would be a by-product (or cause) of docility, a phenomenon documented recently in domesticated cattle (Balcarcel et al., 2021). Groves (1999) suggests that domesticated dogs have sufficiently co-evolved with humans to have symbiotically become our external senses, thereby decreasing our reliance on brain centers that process sensory information. However, our finding here—that brain size reduction occurred in just within last 3,000 years—is temporally inconsistent with these prior explanations. Human self-domestication is argued to have occurred at the onset of species ∼300,000 years ago (Wrangham, 2019) or coincident with the evolution of what some have called “behavioral modernity” ∼80,000 years ago (Hare and Woods, 2020). Recent fossil and genetic evidence indicate that dogs were domesticated > 20,000 years ago (Perri et al., 2021).

Human social life has long been analyzed in reference to the social organization of ants. Imms (1946) review of Haskins (1939) Of Ants and Men described ants as “predominant” among the “very few living creatures whose social development at all parallels our own.” He emphasized the value of having “a nearly complete series of evolutionary forms among living ants, numerous ‘missing links’ and ‘living fossils”’ and a history “much more complete and much better preserved than that of man.” Comparisons made to identify commonalities in patterns of sociality between these diverse clades continue to be made today (Boomsma and Franks, 2006; Wilson, 2012; Crespi, 2014; D’Ettorre, 2017a,b; Friedman and Søvik, 2021; Gowdy, 2021). The benefits of ant models for the study of human social evolution noted by Imms are made more compelling by an additional 80 + years of new species discovery (roughly 5,000–6,000 circa 1,940 to > 14,000 today), an expanded phylogeny (6 to ∼20 subfamilies), and an eruption of integrative sociobiological and neurobiological research. Ants exhibit striking variation in colony size and demography, and individual and colony-level cognition. Diversity in social phenotypes creates the potential to provide general insights into human social structure and brain evolution.

Acknowledging the limitations of analogies, human and ant comparisons may reveal patterns in nature that broadly suggest social and ecological selective forces relevant to brain evolution. We recognize that the uniqueness of human cognition (Laland and Seed, 2021) cannot be overemphasized, and neurobiological parallels with ants are as constrained as sociobiological comparisons, if not more limited. Although ant and human brains have the common function of processing environmental information to adaptively respond to social signals and cues, they are structured very differently. The size, neuron number, and synaptic connectivity of an ant brain is a minute fraction of that of a human brain (Herculano-Houzel, 2016; Godfrey et al., 2021). However, the computational power of an ant brain is remarkable for its size and miniaturization does not appear to constrain behavioral performance and/or higher-order processing (Muscedere et al., 2014), social learning, or consciousness (Avarguès-Weber and Giurfa, 2013; Barron and Klein, 2016; Perry et al., 2017; Lihoreau et al., 2019; Perry and Chittka, 2019; Elek et al., 2020; D’Ettorre et al., 2021). Social organization in ants may “require relatively simple and computationally inexpensive forms of cognition” (Lihoreau et al., 2012), some extrinsic to brain operations, that could mitigate the need for advanced processing capability. Behavior in ants typically involves task routines and kinesthetic performance rather than sophisticated cognition. Unlike the human brain (Noonan et al., 2018), ant brains appear to lack executive function and the ability to mentalize, among other circuitry specialized for human social performance.

Relative needs for social information processing are strikingly different in ants and humans. Olfaction is dominant in ants and involves a relatively small number of chemicals to guide individual actions and organize colony-level behavior. Vision appears to have very limited social functions and mainly functions in navigation during foraging (Hölldobler and Wilson, 1990). The anatomy of ant brains thus largely reflects investment in brain centers such as the antennal lobes and optic lobes responsible for processing olfactory and visual stimuli, respectively. These inputs are mainly integrated in the mushroom bodies, brain compartments specialized in higher-order processing, learning, and memory, and to a lesser extent in the central complex. The mushroom bodies, deeply homologous to the vertebrate cortex (Tomer et al., 2010), have long been consider to be the neuroanatomical seat of intelligence in ants and other insects (Strausfeld et al., 1998), but the nature of ant intelligence is only very loosely comparable to that of humans. Mushroom body elaboration also preceded the evolution of eusociality (Farris and Schulmeister, 2011).

Ants have nevertheless emerged as important models for understanding the role of sociality (Ilies et al., 2015; Kamhi et al., 2016, 2019; Godfrey and Gronenberg, 2019) and behavioral performance and cognition (Muratore and Traniello, 2020; Muratore et al., 2021) in brain evolution. Analyses of sociality and brain size scaling (e.g., Godfrey and Gronenberg, 2019; Muratore and Traniello, 2020) and information processing through social interaction (e.g., Davidson et al., 2016) can contribute to broadly understanding how social biology may have influenced general aspects of human brain evolution. The allometric scaling of functionally specialized brain centers in ants allows the adaptiveness of brain mosaics to be explored (e.g., Muratore et al., 2021) and the metabolic costs of neural tissues that are highly significant to brain evolution (Aiello and Wheeler, 1995) can be recorded in individual ant brains (Coto and Traniello, 2021). Studies of the influences of diet, which is considered significant in primate (and human) brain evolution (Wrangham, 2009; DeCasien et al., 2017), are facilitated by the diverse feeding ecologies of ants. Ant nutritional socioecology spans highly specialized as well as generalist predators that vary in colony size and degree of complexity of social organization, allowing the relative roles of diet and social complexity to be separated because specialist and generalist species that each form small and large colonies (Azorsa et al., in preparation). Ants also have culture in the form of tool use (Zhou et al., 2020 and references therein), and although they do not process food by cooking, they may preserve prey (Maschwitz et al., 1979).

The application of theories of vertebrate brain evolution to social insects has been debated (Lihoreau et al., 2012; Farris, 2016) and studies describe variable relationships between sociality, brain and brain compartment size, body size, and group size in eusocial insects (Farris and Schulmeister, 2011; Seid et al., 2011; Riveros et al., 2012). Workers of ant species characterized by larger colony size have larger brains, suggesting greater social interaction selects for increased brain size (Wehner et al., 2007). Similarly, comparisons of socially basic and social complex ant species suggest larger colony size and collective actions have selected for larger brains and mushroom bodies (Kamhi et al., 2016). In contrast, in monomorphic fungus-growing ants, which have agrarian habits, larger colony size is associated with decreased brain size (Riveros et al., 2012). This latter comparison among agricultural ant species provides insight into how increased group size and sociality in human populations may have reduced brain size (Bailey and Geary, 2009) due to a high level of emergent complexity.

Worker behavior undergoes age-related development in ants. Like human brains, ant worker brains exhibit age-related synaptic remodeling, suggesting synaptic pruning in association with an increasingly diverse, flexible, and efficient behavioral repertoire (Seid et al., 2005; Seid and Traniello, 2006; Muscedere et al., 2009). Age and behavioral development can be associated with increased brain volume (Muscedere and Traniello, 2012) and declines in the density of mushroom body microglomeruli (Gordon and Traniello, 2018)—“microprocessor”-like synaptic structures that underlie plasticity in sensory processing capability and behavior (Groh and Rössler, 2011; Groh et al., 2014). Elements of human brain development are thus mirrored in ant brain ontogeny and social behavior. In mole rats, eusocial mammals convergent in social structure with eusocial insects, there is no clear association between social system and relative brain size: eusocial species do not have smaller or larger brains than social or solitary species (Kverková et al., 2018). Although there is an apparent trend between social system and forebrain neuron number, this is likely an artifact of the extreme metabolic adaptations to subterranean life found in H. glaber (Kverková et al., 2018; Browe et al., 2020). Ants appear to provide more useful models than eusocial mammals.

Ponce de León et al. (2008) and Hawks (2011) suggested that reduction of human brain size—without any evidence for intellectual diminishment—implies selection for brain efficiency. In ants, there is evidence of selection for metabolic efficiency in the brain. Kamhi et al. (2016) used cytochrome oxidase (COX) activity, a proxy for neuron metabolism, to contrast brain evolution and social evolution in the weaver ant Oecophylla smaragdina and the garden ant Formica subsericea, two sister clades whose workers are equivalent in body size but differ strongly in social organization and collective intelligence. Increased social complexity in weaver ants—reflected in division of labor by worker physical castes, large colony size, and remarkable decentralized group action including cooperative nest building—was associated with larger mushroom bodies, implying greater needs for higher-order information processing. Weaver ant worker mushroom bodies have reduced COX activity. Therefore, increased brain size in socially complex species may be associated with reduced brain operation costs, contrasting with the assumption that increased brain size increases metabolic costs (Aiello and Wheeler, 1995; Isler and van Schaik, 2009).

Reduction in brain size may not compromise cognitive performance if intelligence is an attribute of the society rather than the individual. Galton (1907) first described that the accuracy of decision making by human groups could exceed that of any individual group member. This concept of collective intelligence has since been elaborated in studies ranging from insects to humans (Woolley et al., 2010; Ward et al., 2011; Sasaki and Pratt, 2018; Almaatouq et al., 2020). If brain production, maintenance, and operation costs are metabolically significant (Aiello and Wheeler, 1995; Leonard et al., 2003; Isler and van Shaik, 2006; Chittka and Niven, 2009; Navarrete et al., 2011; Kuzawa et al., 2014; Pontzer et al., 2016; Herculano-Houzel, 2017), then collective intelligence may reduce demands for neural tissue to support individual cognitive capabilities. Using a multidisciplinary modeling approach, Reséndiz-Benhumea et al. (2021) indeed demonstrated that agents with relatively small brains can through social interaction achieve a level of behavioral performance comparable to those of larger-brained but solitary agents. Further testing of this idea, however, will require a better understanding of whether Holocene human brains reduced isometrically or if specific regions reduced in size.

We suggest that group cognition lowered the demands for neural architechtures required to support some aspects of individual intelligence and decision making (Bailey and Geary, 2009). This effect may have become even more pronounced with the advent of writing ca. 5000 years ago (Schmandt-Besserat, 2010), which falls within the estimated 95% CI for the pronounced reduction in Holocene human brain size (Figure 1). During human history, social groups became larger, social interactions more frequent, social networks more complex, and tracking relationships more demanding (Bailey and Geary, 2009; Foley and Gamble, 2009). A rise in sociocultural complexity was not due to particular individuals becoming more intelligent and culturally skilled, but because of the emergence of collective intelligence resulting from a growing population of interconnected humans and interacting human groups. As group size increases, interactions with a dynamic and exceedingly complex social landscapes result in increased demands on the brain (Bickart et al., 2011; Kanai et al., 2012). However, because of the metabolic demands of the brain (Pontzer et al., 2016), there may be limits to feedback loops between social network size and brain structure. If group decision-making generated adaptive group responses exceeding the cognitive accuracy and speed of individual decisions and had a fitness consequence, then human brain size may have decreased as a consequence of metabolic cost savings.

Population size expanded dramatically with the advent of agriculture, beginning ∼10 kyr and grew exponentially from an estimated five million to over 100 million by 3000 years ago (Goldewijk et al., 2011). This increase in population coincided with deterioration in individual health (Armelagos et al., 1991; Milner, 2019) and increases in infection rate (Eshed et al., 2010), pathogenic load (Page et al., 2016), and virulence (Menneret et al., 2010). It remains possible, then, that the high energetic cost of a heightened immune response (Wells and Stock, 2020), might have been a factor in Holocene brain reduction. In fact, Crabtree (2013a; 2013b) proposed this immunity-for-intelligence trade-off in his controversial “Idiocracy Hypothesis,” though this idea has been criticized on the basis of flawed assumptions (Kalinka et al., 2013; Mitchell, 2013).

Gowdy and Krall (2013) draw parallels between the ultrasocial human superorganism, complete with division of labor and “economic organization around surplus” that arose in the Holocene and the sociobiology of agricultural eusocial insects, including some ants and termites. Brain size reduction occurred in traditional hunter-gatherer human populations (Wiercinski, 1979) that never adopted sedentary agricultural practices, but have complex social networks (Apicella et al., 2012). Foster and Ratnieks (2005) suggest that the presence of post-reproductive female helpers (grandmothers) universally present in human societies is sufficient to characterize humans as a “new eusocial vertebrate,” offering additional support for the value of broad comparisons across unrelated taxa.