Innovation, life history and social networks in human evolution

Abstract

There is a famous puzzle about the first 3 million years of archaeologically visible human technological history. The pace of change, of innovation and its uptake, is extraordinarily slow. In particular, the famous handaxes of the Acheulian technological tradition first appeared about 1.7 Ma, and persisted with little change until about 800 ka, perhaps even longer. In this paper, I will offer an explanation of that stasis based in the life history and network characteristics that we infer (on phylogenetic grounds) to have characterized earlier human species. The core ideas are that (i) especially in earlier periods of hominin evolution, we are likely to find archaeological traces only of widespread and persisting technologies and practices; (ii) the record is not a record of the rate of innovation, but the rate of innovations establishing in a landscape; (iii) innovations are extremely vulnerable to stochastic loss while confined to the communities in which they are made and established; (iv) the export of innovation from the local group is sharply constrained if there is a general pattern of hostility and suspicion between groups, or even if there is just little contact between adults of adjoining groups. That pattern is typical of great apes and likely, therefore, to have characterized at least early hominin social lives. Innovations are unlikely to spread by adult-to-adult interactions across community boundaries. (v) Chimpanzees and bonobos are characterized by male philopatry and subadult female dispersal; that is, therefore, the most likely early hominin pattern. If so, the only innovations at all likely to expand beyond the point of origin are those acquired by subadult females, and ones that can be expressed by those females, at high enough frequency and salience for them to spread, in the bands that the females join. These are very serious filters on the spread of innovation.

This article is part of the theme issue ‘Life history and learning: how childhood, caregiving and old age shape cognition and culture in humans and other animals’.

1. Stasis in the early Acheulian: a problem?

There is a striking, puzzling and hence controversial pattern in the record of hominin technological evolution: one of a very long fuse of slow change for 2.5+ Myr, followed by an accelerating burst of both innovation and diversification. To a first approximation, the first stone tools (at Lomekwi) date to about 3.3 Ma [1], though see [2]. At about 2.5 Ma, Oldowan tools appear in the record [3]. From about 1.7 Ma, the famous Acheulian Large Cutting Tools are found in the record, and these tools dominate the record until about 250 ka, when the pace of innovation and diversification begins to pick up (for a splendid recent synthesis of this history, see [4]). At one stage, 20 years ago, that pattern seemed even more marked, with the period of acceleration more truncated, and with a final burst of innovation that seemed so abrupt and sharp as to be an ‘Upper Palaeolithic Revolution’ [5,6]. A landmark paper in 2000 showed that there was no such revolution [7]. The apparent innovation burst of the last 40 kyr of the European record had antecedents in Africa, dating back to 200 ka or more. In general, the most recent work has softened this picture of near-stasis followed by accelerating change, as the earliest dates for important innovations in the stone tool record are significantly earlier than they once were. Composite tools, as recently as 2013 dated to about 200 ka [8], may have been first made at almost 500 ka, and the same seems to have been true of Levallois stone-working techniques [9–11]. These techniques require highly skilled pre-shaping of a stone core, giving the knapper quite fine-grained control of the flakes struck from that core. Domesticated fire is difficult to securely identify in the archaeological record, but there seems to be clear signal of controlled fire at a 780 ka site (though control does not yet show the capacity to ignite at will) [12,13]. Moreover, it now seems clear that there was a major change in technical capacities within the Acheulian tradition, with the emergence perhaps around 900/800 ka of a capacity to produce thinner Large Cutting Tools by preparing an impact platform to strike a large thin flake from the proto-handaxe [4].¹ This reduces its thickness while maintaining its other dimensions.

However, even though the earliest dates for these more sophisticated technologies are being pushed deeper in time, they remain outliers. There are many younger sites in the record that show no signs of controlled fire or cooking (until about 400 ka [13]); tools produced with Levallois-like techniques do not seem to have become routine features of human toolkits until about 250 ka; composite tools do not seem to have been routine elements until about 100 ka. So, the pattern still seems to include a long initial period of very slow change: Lomekwi tools at about 3.3 Ma; Oldowan flakes and choppers from about 2.5 Ma (but rather patchily in the record until about 2 Ma); an initial version of Acheulian establishing at about 1.7 Ma; sophisticated Acheulian from about 800 ka; composite tools and Levallois flakes from about 500 ka; from about 120 ka increasingly rapid innovation and diversification. The pattern as seen now is much less marked than was once supposed, but it still seems true that the pace of change itself vastly changed.

The idea that this pattern represents a real trend in overall hominin evolution depends, of course, on the assumption that stone technology is a decent index of overall technical repertoire. That assumption is not arbitrary. For while one can use stone to work stone, and to work a range of other materials (like wood or hide), one cannot readily use wood to work wood, or soft materials to change the structural properties of other materials. Stone is a keystone technology. Even so, it is conceivable that the pattern in the record of traces does not reflect actual history: rather, it is a consequence of differential trace survival. As we go back in time, traces disappear; many organic materials disappear almost completely. That effect is magnified if more temporally distant hominin populations are also smaller, for then they would have produced absolutely fewer traces. Moreover, the tools and other elements of material culture that are produced in lower relative frequencies are more likely to vanish from the record altogether than regularly made, stock standard items. More complex or more intensively made elements of material culture were probably made, if they were made at all, at lower relative frequencies than simpler, less intensively engineered items. For they are more expensive. The archaeological record is likely to undercount the repertoire of smaller and more temporally distant groups, and that bias is accentuated with respect to the less regularly deployed elements of their repertoire. Less frequently made items may well be entirely unrepresented. Moreover, the impression of stasis depends on how we classify stone technologies, on the similarities and differences we take to be important. In analysing stone tool assemblages, classification is notoriously difficult because of the problem of distinguishing between functional differences in shape and differences that are the result of repair and re-use as tools wear and chip.

Even so, the archaeological consensus is that there is a genuine signal in this pattern. So Shea writes in 2017:

Why is Later Pleistocene and Holocene lithic variability so much more clearly patterned than its precursors? After hundreds of thousands of years of stability, around 0.2–0.3 Ma, the lithic record begins to vary divergently by region and cumulatively over time. [15, p. 201]

Let us suppose then that the pattern is robust. Is it puzzling? Arguably, not from Lomekwi to the emergence of the Acheulian. Throughout this period, hominins were probably restricted to Africa (and the stone-working hominins to East Africa), perhaps in small, fairly scattered populations (though for evidence for an earlier hominin presence in China, see [16]). The effects of demography on innovation rate remain controversial [17,18], but no-one would deny that all else equal, the larger the population of tool users, the higher the absolute number of innovations. Estimates of the population size of long-extinct groups are very conjectural. But shortly after the evolution of Homo erectus-grade hominins at about 1.8 Ma, erectines are found in Eurasia, East Asia and island South East Asia; the initial western European records are somewhat later [19]. Whatever the absolute numbers of this metapopulation, the extent and the speed of this geographical expansion shows that it was larger—probably much larger—than the East African populations of tool users. So, we might expect a demographically based contrast in innovation rates between erectines and their immediate predecessors, for there were simply more erectines than earlier and more geographically restricted hominins. Moreover, the role stone tools played in the lives of stone tool users probably changed over this very long period. Shea distinguishes between occasional, habitual and obligate users of stone technologies [15], and argues that in these earlier phases of stone tool-making (before 1.7 Ma), tools played less of a central, repeated role in hominin foraging economies. For even in East Africa, in the core part of the range of early stone tool-using hominins, despite intensive sampling, there are large temporal gaps in the early stone tool record. All else equal, the less regularly stone tools are used, the less we expect innovations. That is especially true to the extent that innovations depend on extra investment: more careful selection and curation of raw materials; more intensive shaping of the flake; caching raw materials at sites for later use. Shea suspects that hominins were not dependent on stone technologies until after the Oldowan. Finally, while there is a consensus that H. erectus-grade hominins were encephalized by comparison with great ape norms, the extent to which Pliocene and initial Pleistocene hominins were encephalized is less clear [20]. No doubt an individual capacity to innovate is not just a simple reflection of relative neocortical volume (or relative overall brain size). But given how expensive large brains are to build and run, there must be some positive correlation between innovation capacity and encephalization. So Pliocene and Early Pleistocene hominins were probably intrinsically less capable of innovation.

Collectively, these considerations suggest that if there is a puzzling period of hominin technical evolution, it is the period of near-stasis (perhaps excepting a partial and unstable domestication of fire) between about 1.7 Ma, with the emergence of Acheulian tools, and about 900/800 ka, with the appearance of much more sophisticated Acheulian technology, more reliable indicators of the perhaps still partial control of fire and the establishment of H. erectus-grade populations in more seasonal, more challenging environments. Erectines were significantly encephalized and widely distributed (hence their total population must have been substantial), their life-history features had shifted towards those characteristic of recent and modern hominins [21], and they were habitual users of stone tools. Indeed, very likely they were dependent on them.

Moreover, it is unlikely, especially given their wide range of habitats, that there is an ecological or economic explanation of the stability of their stone tool repertoire between 1.7 and 0.8 Ma. Economic considerations cannot be neglected when we consider the stone tool record. We cannot assume that if agents have the capacity to make a wide range of tools, they will make them. For increasing complexity or diversity adds costs. It is typically more expensive to have a kit of several items rather than make do with a single general purpose tool. Moreover, if tools are relatively enduring items of personal gear, a kit will generally be heavier and impose greater travel costs. The earliest tools were probably not part of personal gear, for they seem to have been made, used and discarded (there is not much evidence of resharpening). But as individual tools required more effort to make, they became worth keeping.² Complex tools are more expensive to make than simple ones. That is particularly obvious with composite tools. For then different raw materials are needed for the components, together with the glues and bindings required to assemble the parts into a complete tool. Moreover, the interdependence of the parts of a composite tool—even a relatively simple one, like a single-shaft stone-tipped spear—increases its failure rate. Such a spear will fail if any of the tip, binding, shaft or thrower fail. The failure rate of the composite tool is a product of the failure rate of each component, so if they are to have low overall failure rates, the components of a composite tool need to be engineered with greater precision and control than that required for a one-component tool. It follows that all else equal, economic considerations favour a simple toolkit of one or a few simple, multi-use items, items like an Acheulian handaxe. However, all else does not seem to be equal. Composite tools seem to be almost ubiquitous features of ethnographically known forager populations. That is true even of foragers with relatively simple toolkits, like those of the Australian Western Desert. They have high mobility costs as they move frequently over a large range. Even so, they invest in hafted tools and tipped spears. If ethnographic evidence is to be trusted, over a very wide range of environments, the mechanical advantages of hafting outweigh the extra costs of hafted tools.

Of course, there is a significant technological gap between the one-piece tools characteristic of the Acheulian and the hafted tools of mobile Holocene foragers. To cross that gap, hominins needed a range of new skills, in particular abilities to make reliable glues and bindings. Since those could only be developed incrementally, perhaps there was a fitness trench between the simplicity of the early Acheulian and the lightweight composite tools of recent foragers. Poorly made hafted tools—ones whose bindings were liable to break or with glues liable to crack—were less reliable and more expensive than one-piece tools. So perhaps the generations of R&D required to reach thresholds of efficiency and reliability did not pay their way. While this may be true, it would make the Late Pleistocene flourishing of hafted tools hard to explain. In any case, it is hard to see how considerations of economy alone could explain the 800+ kyr gap between the emergence of the first Acheulian tools, and those made by bifacial thinning. It is true that these latter Acheulian tools require somewhat more investment of time to make (perhaps most significantly in learning how to make them). But, they are decidedly lighter to carry.

In the light of these economic considerations, the stasis of the lower Acheulian does seem genuinely puzzling. In principle, it could be explained if Middle Pleistocene hominins were incapable of innovation. But at a finer temporal and spatial scale, the Acheulian record shows quite a lot of innovation, as Hopkinson et al. [24] show. While it is plausible that the encephalization of Late Pleistocene hominins has increased the innovation rate, there is persuasive empirical evidence that erectine hominins were capable of at least a modest flow of innovation. An alternative suggestion fingers limited capacities for cultural learning, either because individual Pleistocene hominins were not adept cultural learners or because their social environments were less supportive of the forms of cultural learning needed for innovations to spread locally and then regionally. Innovations that fail to establish locally and spread regionally are likely to be archaeologically invisible. The remarkable acceleration of innovation in the last 100 kyr of the Pleistocene is typically taken to be the result of improved cultural learning, with a lively but still unresolved debate on the role of an increase in social scale in that improvement. So a natural thought is that the early stasis of the Acheulian is the result of inefficient social learning. But while inefficient social learning probably contributed to that stasis, the next section shows it is unlikely to be the full explanation. An additional factor comes from the metapopulation dynamics of erectine populations. Those dynamics contribute to the prospects for a regional establishment of an innovation (as distinct from its local establishment). I shall add to the case for the importance of these dynamics by highlighting their interaction with hominin community organization. The upshot is that, probably, no single factor caused stasis: innovations were less likely to appear and establish locally, and those that did establish locally were less likely to establish regionally.

2. Constraints on cultural learning

The boldest version of the idea that Acheulian stasis is explained by constraints on cultural learning is developed by Corbey et al. in [25]. As they see it, the hominins using Acheulian tools lacked the capacity to innovatively improve this technology because their use of the technology was not based on an intelligent, flexible understanding of the fracture properties of stone, and of how fracture lines can be manipulated through changing the overall shape of the cobble, and especially by changing the striking angle and local geometry of the impact point. Their production of these stone tools was under partial genetic control, and so was routinized, automatized and hence rigid [25]. If that were so, it would be no surprise that they cannot build on the capacities needed to innovate or to recognize and take up others' innovations. However, their analysis is not persuasive. The main consideration Corbey and colleagues advance is that the Acheulian tradition is too invariant for it to be the result of cultural learning by earlier hominins. If transmission were cultural, they argue, we would expect broader variation in space and time (here lumping together the earlier and the more developed Acheulian traditions, and perhaps under-stating the regional variation documented in [24]). However, if the invariance of the Acheulian is indeed a puzzle, appealing to genetic inheritance and genetic control fails to solve it. Corbey and colleagues themselves emphasize the fact that this technology was used by several species and over a very long period, so given that independently evolving lineages used Acheulian tools, and given the temporal depth of the Acheulian tradition, we would expect both drift and adaptation to local environments to generate variation in time and space.

It is true that if populations are large enough, this tendency for variation to increase can be partially suppressed by selection. If, for example, handaxes occupy a local optimum in the trade-off between multi-functionality, portability, raw material availability and manufacturing costs, stabilizing selection can keep handaxe morphology near that optimum. However, if selection can stabilize a phenotype transmitted by genetic inheritance, it can stabilize a phenotype transmitted by cultural inheritance—perhaps more effectively, to the extent that cultural inheritance is oblique rather than strictly vertical, and is mediated by novices choosing the most expert models. It is true that Corbey and colleagues quote some modelling work suggesting that with cultural transmission (but not genetic transmission), noise in transmission will swamp selection. But these models are not realistic accounts of learning in a lithic landscape [26]. Apprentices have multiple opportunities to learn, and importantly, they can learn not just through imitation but through emulation and practice. If handaxes are made in social, communal settings, apprentices will have many opportunities to see and handle complete, broken and partially made handaxes (for an ethnographic example, see [27]). Still more importantly, this form of learning is hybrid: apprentices get to try out many handaxes, and get feedback that helps them control error—keeping the handaxes that they make within the space of functionality. True, transmission is never perfect. If it were indeed the case that stone-working skills were passed on down the generations through Chinese-whispers style transmission chains, error would indeed accumulate. But artisan skills are not passed on through such a transmission chain: feedback from existing templates, feedback from the world in manufacture and use, and feedback from peers all offer opportunities to identify and correct error.

While I am sceptical of their analysis of the Acheulian, I think Corbey and colleagues are right to argue that over such a long period of time, and with the manufacture and use of handaxes playing a pivotal role in the economy of these agents (as attested by the sheer numbers found, even if specific functional claims remain in dispute), we would expect some form of genetic assimilation: genetic changes that make handaxe-making more readily and efficiently learned. But genetic assimilation need not take the form of the canalization of a specific motor programme, rendering it relatively independent of specific environmental inputs. In the case of a capacity that needs to be sensitive to the particular quirks of local raw materials, selection is unlikely to favour that form of assimilation. Instead, genetically prepared learning can support cultural learning rather than replace it [28, pp. 330–333], and if the cognitive architecture of living humans is any guide, that seems to have happened. For it seems as if something like a ‘folk physics’ module is part of our cognitive organization: humans readily interpret their environment in relatively abstract, functional terms. In general, modular conceptions of human cognitive architecture are implausible, as the social and biological environments hominins must navigate vary at times and over time. But that is not true of the causal properties of mundane physical objects [29, pp. 229–230]. As Povinelli et al. have pointed out, great ape physical manipulation of the world is based much more on perceptual categories and fairly blind trial and error [30]. As a consequence, various tube, stick and trap tasks that are causally transparent to humans are much less so to great apes. We do not know, of course, whether erectines or their immediate successors had something like our naive physics, or whether their recognition of their environmental affordances was more tied to immediate perceptual properties. But if we do indeed have something like a folk physics module, very likely that is the result of a long evolutionary history of tool manufacture and use, an evolutionary history the erectines in part shared. If some more rudimentary version of that module dates back to the erectines, playing a role in the acquisition of lithic skills, the acquisition of Acheulian technology would depend both on genetically prepared learning and on appropriate cultural inputs. We can be genetically prepared to learn culturally, and this form of preparation would enhance hominin abilities to innovate and profit from innovation. We can accept Corbey and colleagues' view that the long history and central importance of the Acheulian helped shape the genetic evolution of Pleistocene hominins, without thereby accepting their view that it made those hominins less responsive to cultural inputs about technology and its uses.

While there is a general consensus that the Acheulian is indeed a tradition, that consensus is by no means complete. Tennie et al. [31–33] suggest that Lomekwian and Oldowan techniques lie within Early Pleistocene hominin's ‘zone of latent solutions (ZLS)’. That is, individual hominins would have a good chance of inventing these techniques by themselves, even if in practice those techniques were acquired as a result of some form of cultural stimulation. Tennie (and perhaps his co-workers) extends this claim to the early Acheulian (C. Tennie 2019, personal communication). In their view, early Acheulian tools do not show the ability to copy from others. Rather, these hominins had very limited capacities to learn culturally, and stasis is to be expected unless agents have the individual capacities, and live in social contexts, that make cumulative cultural learning possible. Without an ability to accurately copy from others, neither cumulative cultural learning nor escape from the ZLS is possible.

Space precludes a full discussion of this framework, but I am sceptical for both theoretical and practical reasons. (i) The idea of a ZLS is tacitly committed to an implausibly strong version of nativism. There is a well-defined ZLS only if agents' potential repertoires are largely independent of their developmental environment. If hominins’ capacities to learn are themselves variably shaped through development [34], then there is no species-typical ZLS. For then, general features of the socio-developmental environment will change the agent's sensitivity to specific, task-sensitive inputs. For instance, innovations often involve novel combinations or sequences of pre-existing elementary capacities. If growing up in a tool-using environment builds new elementary capacities to shape or manipulate, that agent will have a richer repertoire on which to draw. The developmental environment, including cultural input, shapes the space of the potential skills that agent might then acquire through individual learning. It is of course possible that cognitive plasticity is limited in great apes and Early Pleistocene hominins, becoming marked only later in the hominin lineage. However, Tennie et al. [33] implicitly concede the importance of plasticity even in great apes, in arguing that the atypical abilities of human-reared great apes make them irrelevant to characterizing the chimpanzee ZLS. (ii) It is probably true that Acheulian skills are not transmitted by high-fidelity copying. But that does not make it impossible for agents to acquire capacities they could not readily invent themselves. Chimpanzee nut-cracking seems to be transmitted through some mix of social support and individual practice, rather than high-fidelity copying. But nut-cracking is not readily re-invented. It is confined to a small fragment of the chimpanzee geographical distribution in West Africa [35]. So, cultural learning can stabilize skills that an individual would find it difficult to invent without copying. Moreover, emulation learning can support reliable skill transfer [33], especially if novices have access to the entire production history of an artefact, so they know what intermediate forms are like. In short, even if Pleistocene hominins could not copy accurately, cultural learning could still make it possible for them to learn a technique they could not discover for themselves. (iii) While Tennie et al. have plausible methods for identifying skills that fall within the ZLS of great apes, these do not extend to long-extinct hominin populations. One method depends on the contrast between experimental and control populations, obviously only an option for living species. Another depends on the distribution of skills across socially unconnected wild populations. We rarely know whether the hominins responsible for one site are informationally isolated from those responsible for others, for informational connection comes in degrees and can be mediated by archaeologically invisible intermediate populations, and by coming across the artefacts themselves, as they can persist in the environment indefinitely. Partial cultural information flow, mediated, say, by coming across a site where stone has been worked, can provide a crucial hint that makes the difference between a difficult problem and a tractable one. (iv) Finally, this viewpoint understates both the difficulty of re-inventing Acheulian technology and its costs. As Hiscock [26] points out, unguided trial and error learning of stone-working techniques is seriously dangerous. Very sharp fragments detach at speed. Moreover, experimental evidence suggests that real mastery takes tens or hundreds of hours of explicit instruction [36]. Furthermore, Acheulian tools were largely absent from non-African sites until about 800 ka [37], and that suggests that this technology was not easily re-invented without any cultural input.

As a consequence, I take it that Acheulian toolmakers learned to make their tools with the aid of information from their elders and their peers, probably accompanied by very occasional complete or partial re-invention of that same technology. That said, there is little doubt that erectines were less adept at cultural learning than later hominins—especially early erectines. For one thing, the local social environment profoundly influences the opportunities for cultural learning. Minimally, the skilled must tolerate close observation from the less skilled. As technology becomes more challenging, and this likely includes the Acheulian, active cooperation by the skilled is probably necessary [26]. Moreover, even with active cooperation, skill transfer requires frequent association, and the frequency of association depends on the social and economic organization of the band. For example, if mothers forage alone accompanied by their children, or even if they forage only with other women, those children's cultural learning opportunities are quite sharply limited. In particular, they will have quite limited opportunities to learn male-skewed skills. Thus, as Ron Planer has pointed out to me, a shift to bi-parental care and paternal investment in children has profound consequences for the cultural learning of juvenile hominins (R. Planer 2019, personal communication). Cultural learning within residential groups will be less efficient (i) if cultural learning is primarily vertical and (ii) if juveniles and/or adolescents primarily associate with their mothers and their female affiliates, and (iii) if there are sex-based differences in foraging targets and the skills associated with these targets. In addition, individual cognitive capacities matter, and these have surely changed quite markedly over evolutionary time. As Heyes [34] has recently argued at length, contemporary minds are stocked with a battery of cognitive gadgets that (inter alia) support cultural learning: theory of mind, imitation, language, semantic and episodic memory. No-one thinks Middle Pleistocene hominins had equivalent capacities in all those respects.

So children growing up in the Middle Pleistocene probably grew up in social environments that were less well adapted to cultural learning, and with minds less well adapted for cultural learning too. In addition, as Nowell & White [38] show, their life-history characteristics reduced their prospects for stable cumulative cultural evolution. While there was some shift in erectine life-history characteristics towards those typical of later hominins, erectine childhood and adolescence was compressed compared with later hominins: less time to learn reduced the prospects of innovations establishing locally and then regionally. Despite the fact that great ape data do not show a rigid connection between group size and encephalization, Nowell & White also accept Dunbar's inference from relative neocortical size to group size [20], and thus conclude that the typical erectine band was smaller than those of later hominins. In growing up as an erectine, the skills to make tools needed to be acquired more rapidly, but from fewer models and peers. These factors contribute to the long-run stasis of the Acheulian [24,38]. They conclude that ‘the constriction of those life history phases in which novelty, experiment and innovation are most readily embraced would thereby inhibit the likelihood of novel behaviours arising and being locally adopted’ [24, p. 68]. This effect might be especially strong if, as these authors suspect, local traditions of handaxe-making were symbolically loaded, as an aspect of local group identity.

However, were erectine capacities to learn from elders and peers so limited as to make it very unlikely that innovations could establish locally? I doubt it. One reason is the wide geographical and ecological dispersal of the erectines though this period of apparent stasis. While they still seemed unable to cross sea barriers of any distance, or to exploit far northern latitudes (above 55° N), they established over much of Eurasia and South East Asia [39, pp. 160–165]. Moreover, I am persuaded by Henry Bunn's and Travis Pickering's view that these hominins were successful ambush hunters, able to select and take prime-age male medium and large game [40–42]. If so, cultural learning could not have been so sharply constrained. For as Lewis Binford points out, ambush hunting with simple weapons (like wooden spears) can be successful only if hunters have rich and accurate natural history knowledge. Hunting animals larger than oneself poses special challenges, especially as wooden spears are unlikely to be immediately lethal. So, hunters need both to be close and to take advantage of the target's intrinsic limitations, or to attack in situations of disadvantage, like swamps or box canyons [43, pp. 196–203]. Indeed, this seems to be true of hunting even with Upper Palaeolithic toolkits [44]. Hunters need to know the routines of their targets, their vulnerabilities, their response to attack and their patterns of flight. Expertise of this kind requires cultural learning. Bunn, Gurtov and Pickering's view that early erectines hunted fairly systemically is still controversial. But while the significance of cut-marks on bones is difficult to assess, these authors also infer age profiles from the bone residues. Those profiles suggest a bias towards males in their prime, a pattern inconsistent with access through scavenging [40,41]. Moreover, while claims about H. erectus hunting early in the Acheulian are controversial, by about 1 Ma, they are much less controversial [45]. Furthermore, as Pickering in particular argues, ambush hunting is a planned activity that demands a high level of executive control, not just in conceiving of a plan but in resisting boredom, distraction and discomfort [42]. It is true that the more advanced Acheulian technology found after 900 ka required deeper hierarchical control than the simpler earlier forms, for it depended on the preparation of the striking platform, as this allowed long thinning flakes to be detached. Perhaps that was an impassable barrier for the cultural learning capacities of erectines. But if they really were successful ambush hunters, with the executive control and cultural learning capacities that that lifeway implies, this seems unlikely.

In sum: while very likely cultural learning in erectine residential groups was much less efficient than in later hominin social life, it seems unlikely that this is the full explanation of the stasis of the early Acheulian.

3. Metapopulation dynamics

While the individual capacities of Middle Pleistocene agents and the social organization of their residential groups both influence the prospects of an innovation establishing locally, for innovations to persist over time and to be more archaeologically visible, they must establish regionally. That makes the metagroup dynamics of these populations causally relevant to the prospects of innovations persisting. Premo & Kuhn [46] point out that residential groups are demographically fragile, vulnerable to many kinds of bad luck and trouble, so that an innovation that establishes locally is almost certain to be lost unless it spreads beyond its group of origin. Hopkinson et al. [24] take up this idea, combining it with their view of limited cultural learning within residential groups to offer a hybrid account of Acheulian stasis. They develop the insights of Premo & Kuhn about the role of metapopulation dynamics by focusing on the factors that influence the expected lifespan of residential groups. The more truncated that lifespan, the less chance an innovation will spread before it disappears with its carriers. In their view, that matters because Middle Pleistocene residential groups probably had truncated lifespans compared with those of later hominins, flickering in and out of existence more rapidly. (i) Residential groups' expected lifespans are shorter if they are smaller: the smaller the group, the more vulnerable it is to moderately unlucky runs of accident or disease. (ii) Likewise, groups are more vulnerable if they have relatively small ranges, for then a quite local ecological disturbance can have catastrophic impacts on their resource bases. (iii) Finally, vulnerability is increased if groups are thinly scattered across a landscape. Groups are somewhat demographically buffered by migration, but the more thinly groups are scattered, the riskier movement from one to another becomes, constricting migration. Nowell and co-authors propose that Middle Pleistocene groups have all these risk factors: (i) overall low population densities suggest a thin scatter; (ii) restricted raw material movement indicates relatively small range sizes, and (iii) Dunbarian arguments about the relationship between encephalization and group size indicate that erectine residential groups were smaller than those of later hominins [24,38].

In rough outline, I think this analysis is correct, though I will offer a different analysis of the residential group size and migration. I doubt that the size of residential groups changed for much of the Pleistocene. But at some stage, perhaps around 800 ka, residential groups became nested in larger communities. If earlier Pleistocene residential patterns were similar to those of Pan species, early hominin dispersal patterns and the character of intergroup relations significantly constrained the spread of an innovation from its point of origin to adjoining residential groups. If great ape metapopulation networks are any guide to those of Pliocene and Early Pleistocene hominins, hominin metapopulation structure was transformed over the Pleistocene. The residential groups of ethnographically known foragers, and the residential groups of great apes, and, in particular, chimpanzees and bonobos, are very different. Forager residential groups are socially open, in two important respects. First, they are nested in larger social entities (communities or ethnolinguistic groups). This nesting is manifest in regular interaction variously through kinship ties (in humans, much more elaborate than those of great apes and not dependent on co-residence [47,48]), through acknowledged relations of reciprocation (of which the San Hxaro system is a famous exemplar [49]), or through shared ritual connection [50]. These various forms of relationship establish and stabilize horizontal affiliative connections between agents who live in different residential groups. It is manifest in the fact that in most forager cultures, in favourable circumstances, residential bands aggregate, often supposedly for shared ritual activities [51,52]. It is manifest in an explicitly shared set of symbols: modes of dress and decoration; distinctive ritual practices which often include some form of a narrative of common origins and connection; often a shared language or dialect, as in Papua New Guinea wantoks (= ‘one talk’ in Tok Pisin). These links bind residential groups into larger communities.

Second, forager residence is fluid to such an extent that Marlowe, in writing about the Hadza, argued that the standard concepts of male or female philopatry simply do not apply to forager bands, given their fluidity of residence [53]. Individuals and their families move in and out of these bands fairly freely and regularly, and without prejudice to opportunities to return. Mothers will move in or out to visit their daughters and daughters-in-law (especially after the birth of a grandchild). Conflicts are often settled by moving away. In general, residence is open-textured and negotiable. This residential fluidity co-occurs with (and perhaps depends on) the nesting of residential groups within larger communities. In short, in ethnographically documented forager lives, residential groups are not closed social worlds, both in the sense that movement in and out is an established practice, and because friendly social connections are maintained across bands. Pan groups are much more closed, in both these senses. In both species, the male is philopatric. Most adolescent females disperse. After an initial period in which they probe their reception in a neighbouring residential group, they do not return, and in almost all chimpanzee residential groups that have been extensively studied,³ civil social interaction takes place only within the group [35]. When chimpanzee groups meet one another at the fringes of their respective territories, there is mutual cool suspicion, or worse. That is much less true of bonobos. There is evidence that when bonobo communities encounter one another, females quite often initiate affiliative interactions across the community boundary, while the males remain mutually but passively suspicious [54]. Female dominance, built on the fact that only female bonobos readily form coalitions, seems to keep male wariness from escalating into overt hostility. If this difference between chimpanzees and bonobos in their intergroup relations is due to the fact that male chimpanzees form coalitions and male bonobos do not, the ancestral hominin condition was very likely chimpanzee-like, for large and medium game hunting almost certainly involved male coalitions. The high-velocity projectile weapons that make solitary and small party hunting possible are not part of the record until about 100 ka or later [55].

Caution is needed in inferring ancestral hominin residential patterns from comparative data. As one reader of this paper pointed out, primate residential patterns are both varied and evolutionarily fluid. Moreover, mid-Pleistocene hominins are closer in time to Homo sapiens than to the last common Pan/hominin ancestor, let alone to living great apes. In particular, the intensity of active intergroup hostility might be characteristic only of chimpanzee social lives. But while great ape residential groups vary in their character, none of them is nested in a larger community with the open texture of recent human forager bands (unlike some baboon species).⁴ So, the least improbable inference is that the ancestral social organization of hominin society was Pan-like, with residential groups having very little social connection with neighbours, probably with male philopatry and adolescent female dispersal. Layton and colleagues [57,58] have argued on the basis of raw material transport distances that this ancestral social organization persisted into the Pleistocene, until the evolution of the Heidelbergensians. If they are right, that would help explain the Acheulian stasis of about 1.7 Ma to about 800 ka, and the slow fade-out thereafter. First, it supports the expectation that erectine residential groups had more restricted lifespans, for the emergence of multi-level societies reduces the cultural and demographic fragility of residential groups. Regular interaction helps compensate for the loss of expertise through untimely deaths. Moreover, residential fluidity means that sex ratio and other imbalances are eased by incentives to move in and out. Likewise, the existence of reciprocal ties gives residential groups options in the face of local hard times: they can often negotiate temporary movement into others' territories, on the expectation that they will reciprocate when necessary. Kelly calls these forms of negotiation the ‘social defense’ of territory, and suggests that it is the most typical form of mobile forager territoriality [59, ch. 6].

In short, to the extent that residential groups are nested in a larger community, they are much less vulnerable. In effect, their exposure to demographic risk depends as much on the size and territorial reach of the community as on the size of residential groups. That matters. Despite Dunbarian arguments to the contrary, the residential group size of mobile foragers did probably not increase much over the Pleistocene. To the extent that Late Pleistocene and Holocene residential groups are less apt to disappear, it is because of these changes in connectivity rather than because they are larger. Kelly argues that once hominins developed a lifeway in which hunting was important, residential group size would have been fairly constant at somewhere between 18 and 30 adults. That size balances the need to reduce daily variation in resource intake, the need to minimize the rate of depletion, and the costs of mobility [59, p. 274]. Given the high cost of growing and running a large brain, earlier hominins might have needed a little less than very recent hominins, but they were constrained by mobility and the costs of resource depletion in similar ways. So, Middle Pleistocene residential groups were probably about the same size as those of ethnographically known foragers. But if, as Layton's analysis [58] suggests, these residential groups were not part of a larger social whole, their exposure to risk depended only on their own small size and small territory; their risk was not moderated by links to a larger community. Likewise, without reciprocation, they would not have had similar rescue options in the face of local ecological disturbances. Thus, Hopkinson et al. [24] are right to think that the expected lifespan of closed residential groups was less than that of the more open groups of hominin life, with those more open groups probably beginning to appear with the evolution of the Heidelbergensians.

The upshot is that the survival time of innovations at their point of innovation was probably limited, compared with the prospects of an innovation in the more networked residential groups of later hominins. That said, residential groups and their traditions can persist for hundreds or even thousands of years. Nut-cracking by West African chimpanzees does not seem to have spread much [35, pp. 158–160], but archaeological evidence from marks on hammerstones suggests that the tradition is thousands (but not tens of thousands) of years old ([60], pp. 160–162). However, this form of social organization also sharply constrains the diffusion of innovation, as information about a local innovation is mostly carried by dispersing subadult females. While social worlds are still largely closed, even if adjoining groups meet without open violence, they are unlikely to be expressing their innovations in technology and technique. Perhaps some information about innovation leaks sideways through discarded tools and evidence of their uses. Perhaps, for example, the fact that fire can be domesticated might spread this way. Even so, information about what a focal group can do is mostly carried to adjoining groups by this kind of dispersal, and that has important consequences.

First, it means that skills and artefacts that are mostly associated with males will not readily disperse (or vice versa, if females are philopatric and males disperse). If hunting and bully scavenging were important components of the erectine foraging economy, there are likely to be sex-based differences. Wrangham has argued persuasively that the acquisition of medium and large game through either of these strategies is risky, with many failures [13]. So, it would not be sustainable without being paired with a fallback strategy. A division of labour between a high-risk high-reward option, and a reliable but less rewarding option, need not be sexual. It might be by age, with the younger and the older gathering. But Kelly is surely right that it is at best inefficient to combine hunting with care for young children [59, p. 274]. Some form of sex-based difference in foraging skills is not improbable. This difference would be exacerbated if subadults, before they disperse, have spent most of their time with their mothers and their mothers' affiliates.

Second, Gurven and colleagues [60–62] have argued that ethnographically known foragers do not develop their full range of skills until well into adulthood. This is somewhat controversial, especially as a claim about the evolutionary causes of modern human life histories. But if there are skills that require learning and practice into one's adult years, innovations that depend on those peak skills—innovations that depend on genuine adult expertise—cannot be exported into neighbouring residential groups by subadult dispersal, and shifting to a group in which that innovation is not yet established at the very least makes it less likely that the dispersing subadult will in the fullness of developmental time arrive at that expertise. Chimpanzee nut-cracking might be a case like this. It seems to be surprisingly hard for chimpanzees to master the skill, requiring years, and its limited spread may be due to the fact that females disperse before they are fully proficient.

Third, it is one thing to have a skill, another to be able to express it. Nowell & White [38] suspect that erectines were conservative and conformist, so a female joining a new group would face social costs in expressing new behaviours. Even if that conjecture is mistaken, and conservatism as such is not an issue, the low social prestige of a newly arrived subadult would reduce her chances of establishing any innovation she is carrying in her new group. Her practices will not be as salient to others. To the extent that rank or prestige influences the attention of others, she will be relatively invisible. Moreover, to the extent that foraging is collective, she will have little influence over the direction or target of travel. Female chimpanzees and bonobos often forage individually, but human female foragers mostly do not [59], probably because they are more vulnerable in open woodland environments. If migrant foraging choices are constrained, her opportunities to express her skills will be limited, and many skills are use-it or lose-it. Unexpressed capacities are likely to fade.

Finally, for reasons also deriving from an initially marginal status in the group, innovations whose pay-off requires active cooperation from others are less likely to be carried by subadult dispersal. Net fishing, for example, in many of its forms is collective rather than the individual. The same is true of hunts assisted by drive lines to corner targets in blind wadis or stampede them over hidden cliffs. Even if communication is possible, recent incomers are unlikely to have the trust and prestige required to persuade others to invest effort in unfamiliar and untested practices.

To sum up this thread: these are all soft constraints. None makes the export of innovation impossible. But while and to the extent that Pliocene and Middle Pleistocene hominins had Pan-like residential patterns, with cool, stand-offish relations with their neighbours (or worse), and with subadult single-sex dispersal, the probability of innovations spreading from their point of origin was low to very low. However, the difference between the closed social worlds of chimpanzees and the open social worlds of known foragers is graded, one of degree, rather than a dichotomy. Tolerance can increase, improving the prospects for peaceful adult-to-adult interactions in boundary zones. Residence can become less rigid, allowing return visits to natal groups. Kinship recognition can become less tied to co-residence. Given that, if this factor were causally important, we would expect to see, and do see, a slow increase in the rate of change and variation from about 800 ka.

4. Concluding discussion

This paper has argued that (i) the first half of the Acheulian, from about 1.7 Ma to about 0.8 Ma, was a period of technological stasis, though the possibility that this is an artefact of preservation bias cannot be completely excluded. (ii) If that stasis is real, it is not the result of the very low innovation potential of Middle Pleistocene hominins. (iii) Since the establishment of an innovation to the point of archaeological visibility depends on cultural learning, the cultural learning potentials of Middle Pleistocene hominins are relevant. Those potentials were probably markedly lower than those of later hominins, but that is unlikely to be the full explanation of stasis, for other features of their lifeways show significant cultural learning abilities. (iv) Archaeological visibility depends on innovations spreading regionally, and the prospects of such spread depend both on the character of local groups, including their expected long-term viability, and on the network structure of these groups. (v) In particular, to the extent that Middle Pleistocene hominin social life resembled the closed social lives of chimpanzees (especially), bonobos and other great apes, the chances of an innovation spreading were poor to very poor.

Data accessibility

This article has no additional data.

Competing interests

I declare I have no competing interests.

Funding

This study was funded by the Australian Research Council.