The Prehistory of the Mind

To what extent can the material traces left by ancient hominins, in conjunction with other evidence, reveal the distinctive character of their minds? Archaeological opinion has varied on this subject, but there have been surprisingly regular flashes of optimism, even confidence (Wynn and Coolidge 2004, 2007); and even more (Lewis-Williams 2002). I say “surprising optimism” when one considers the struggles of cognitive science to identify the psychological mechanisms that support the behavior of living agents. Consider, for instance, the controversy over the supposed existence of imitation in nonhuman primates and (especially) great apes. While it has been evident for some time that information flowed socially between, in particular, chimps, it took innovative experiments to establish that chimps had some ability to learn the techniques for exploiting a resource by watching another agent using those skills (see, for example, Horner and Whiten 2005). In general, the inference from adaptive behavior to cognitive mechanism directing that behavior is uncertain. Many of the skills an agent can learn by imitation can be learned instead by emulation (plus some trial-and-error learning); that is, by attending to the object another agent is manipulating, rather than attending to those manipulations. The more tightly constrained the solution space—if there is one more obvious or easier way of (say) breaking open nuts—the population-level behaviors of the emulators will look very similar to those of a population of imitators, and the debris they leave behind would be indistinguishable (Whiten et al. 2003).

I shall suggest that some attempts at cognitive archaeology have been overoptimistic. But the overall message of the article will be tempered optimism. Perhaps we cannot distinguish imitators from emulators in the archaeological record. But we can distinguish populations that can reliably and accurately transmit technical information across the generations from populations that do so less reliably and accurately. We will not typically be able to specify the cognitive mechanism driving the behaviors of ancient hominins with fine-grained precision; we will not be able to identify the algorithms, the computational processes implemented in ancient hominin brains. But, quite often, we will be able to identify what these agents were able to do: to identify their capacities to plan, to coordinate; to transform materials in precise and unobvious ways; to resist destructive or disruptive impulses. Dietrich Stout, to take a particularly salient and impressive case, has analyzed in quite fine-grained detail the skill structure, the behavioral program, of Oldowan technology and compared those techniques to those of the Acheulian. Since these are both stoneworking techniques, the atomic units of action—angled, controlled blows—are comparable, and so Stout was able to show that Acheulian techniques depend on deep, more sustained control of action (Stout 2011; Stout et al. 2014, 2015). Putting the point in Dan Dennett’s helpful terminology (Dennett 1991), we can reasonably hope to identify the “performance specifications” of ancient hominin minds. So in the final sections of the article, I will return to a famous skeptic, and suggest a more optimistic way of reconceiving his hierarchy of increasing empirical intractability.

I begin, though, with a false start, and the lessons of that false start. In paleoanthropology’s recent past, there was considerable enthusiasm for a cognitive diagnosis of a recent, apparently sharp change in material culture, the Upper Paleolithic Transition (Klein 1995; Tattersall 1995; Mellars 1996). That supposed revolution was a broad-based surge in human material culture. The surge included, but was not limited to, the technologies of economic life: composite tools tipped with microliths; the use of bone and ivory (and by inference from awls and needles, fitted clothes). In the view of some, even more significant was the appearance of material symbols: shells and other items of personal adornment; the first musical instruments; (a little later) cave art. Homo sapiens as a species is roughly 200,000 years old but (on this view), until the Upper Paleolithic Transition, its material culture did not contrast sharply either with that of its European sister species, the Neandertals, or that of its immediate ancestor (perhaps Homo heidelbergensis).

Material culture did not just change. On the then-current story, there was a sharp upward shift in variety and complexity. This, together with the entry of material symbols into the archaeological record, was treated as a signal of a change in the cognitive capacities of those humans. A problem, though, was to specify and empirically support specific suggestions about this change. The leading candidates were: (1) Increased working memory and executive control. This suggestion maps naturally onto the appearance of complex, composite tools whose overall functionality depended on the precise complementarity of their parts; their production depends on a planned assembly of the right raw materials, with tool construction requiring sustained attention and care (Wynn and Coolidge 2007). (2) Language and/or enhanced theory of mind (Klein 2000; Klein and Edgar 2002). Some material symbols pose the same technical challenges as composite tools. Mammoth ivory flutes, for example, were much more challenging to make than the contemporaneous bird-bone flutes, since bird bones were already hollow. The mammoth tusk had to be split, hollowed out, then glued together to form an airtight seam. Likewise, and famously, some cave art (though significantly younger) is technically very impressive. But by late Pleistocene standards, making beads and pendants from shells or teeth was not technically demanding. These were not cutting-edge technologies of 40 kya. So why is their appearance in the material record so late? A focus on material symbols and the cultural complexity they hint at suggests that the changes were in representational capacities: in language and/or enhanced theory of mind (on some leading views, these are necessarily linked: no language without sophisticated theory of mind). For example, one might think that beads and other jewelry are attractive options only to agents who represent, and care about, how others see them. They are a technology that depends on perspective-taking.

So even when the Upper Paleolithic Transition seemed a secure feature of the archaeological record, there was always a problem about just what cognitive change was responsible for the change in material culture. However, this conception of cognitive archaeology and of reading the character of ancient minds from material traces turned out to be in trouble for more fundamental reasons. One was methodological. If the change in material culture was indeed the result of an upgrade in human cognitive capacity, the change had to be an improvement: a real increase in diversity and complexity. But counting tool types in Paleolithic archaeology is notoriously fraught, as it is difficult to identify differences that are the result of design for different tasks (“choppers” vs. “scrapers”) from those that are the results of raw material variation, or of different stages in a reduction sequence. Likewise, except in narrowly circumscribed cases, judgments of relative complexity are problematic. But the crucial problem was empirical. The archaeology turned out to be grossly oversimplified. In a landmark paper, Sally McBrearty and Allison Brooks showed that the technological and cultural innovations that seemed to appear suddenly and together in the European record had older African antecedents (McBrearty and Brooks 2000; McBrearty 2007). This did not just export the Upper Paleolithic Transition backwards in time and far to the south. The innovations of this supposed revolution were not synchronous; there was no single point of origin of these signs of complex behavior in either space or time. Nor were these signs of complex behavior continuous in the record once they originated. For example, there is quite persuasive evidence of bow-and-arrow technology at Still Bay and Howiesons Poort between (roughly) 75 and 60 kya (Lombard and Phillipson 2010), but not thereafter, for the next 10,000 years (Lombard and Parsons 2011). So if it was right to conceptualize these apparently new technologies as upward shifts in complexity, they were followed by downward shifts. The uncoordinated and unstable entrance of these new aspects of material culture into the material record undermines the idea that those innovations are the result of a genetically triggered, developmentally canalized cognitive change.

This empirical challenge exposed a tacit assumption behind the cognitive upgrade interpretation of the Upper Paleolithic Transition. That interpretation depended on a simple reflection thesis: the material culture of a group is the result of the intrinsic, individual ingenuity and insight of the individuals in that group. This perspective is explicit in (for example) Stephen Pinker’s work on “the cognitive niche.” In reflecting on the technical achievements of traditional societies, he agrees that humans are typically superbly adapted to their local physical and biological environment. That adaptation, he suggests, is mediated by purposeful innovation and error correction, and those successful innovations are taken up by onlookers through intelligent assessment of both the results of innovation and of the methods through which innovation is made (Pinker 2010). So the background assumption is that marked and enduring differences in material culture are explained by differences in individual insight in making and assessing innovations. In the standard narrative (see, for example, Foley and Lahr 2003), the history of hominin material culture shows marked and enduring differences in material culture. As it is normally told, Oldowan technology, with its simple choppers and flakes, originated about 3.3 mya; and about 1.7 mya, it was supplemented (widely, but much less in East Asia) by Acheulian biface technology. At roughly 300 kya, the Acheulian ended with the onset of the Mousterian; with new ways of working stone that gave knappers more control over form. The iconic products of the Mousterian are known as discoid bifaces, and these can be worked into further tools, including some blade-like tools (i.e., longer and thinner flakes with sharp edges). In turn, on the standard chronology, these techniques were supplemented or supplanted in the Upper Paleolithic, with its allegedly more diverse and complex material culture. The thought was that if we abstract away from local detail, there are identifiable transitions in material culture that are explained by transitions in hominin cognition. Both the history and the inference are dubious.

Environmental Drivers of Technical Repertoires

The inference to the emergence of a fully human mind at the Upper Paleolithic would not be sound, even if these new aspects of material culture really had a single, joint point of origin in the archaeological record. First: the material culture of a group is sensitive to economic, environmental, and social-demographic factors, not just cognitive ones. Life in physically challenging environments often selects for sophisticated technical responses: it is no accident that work on the power of cultural evolution to shape sophisticated adaptation regularly recycles Inuit examples (Richerson and Boyd 2005; Henrich 2016). Arctic survival depends on the technologies of fitted clothes; windproof dwellings; movement across ice and water; large-game hunting. Likewise, broad-spectrum foraging often selects for a diverse toolkit, as different resources require specialist tools (snares, nets, fish spears). On the other hand, if the campsite must be moved frequently (as in low-productivity arid and semi-arid zones) the cost of movement is high, and the weight of the total toolkit must be minimized. It is also true that toolkit diversity can be minimized if the ecology favors specialization on a small range of resources. This may explain Tasmanian Aboriginals’ notoriously limited toolkit (Hiscock 2008, Chap. 7).

As Stephen Kuhn has pointed out, to the extent that we think hominin cognitive capacities (individually or collectively) allow groups to adapt optimally to their environment, we should expect to see local and regional variation, as different trade-offs between the costs and benefits of technology select for more or less complex and diverse toolkits (Kuhn 2006). There will be variation as well because there may be equally good trade-offs among different mixes of social, ecological, and technical strategies.Footnote 1 The signature of increasing intelligence, increasing cognitive sophistication should be local variation rather than a linear increase in technical complexity. In framing these ideas, he borrows Sewell Wright’s metaphor of a rugged fitness landscape: there are many distinct local optima, not one global optimum.Footnote 2

Of course, optimization to local circumstances is optimization under constraint. Social life and social organization can help or hinder the accumulation of technical resources. There is persuasive modeling, and some empirical support, for the idea that small and isolated groups are vulnerable to the attrition of their informational resources: size and connection support redundancy as the more heads information is stored in, the less easily it is lost and the more easily it is retained. Moreover, size and connection make social learning more reliable: novices have access to more models, and larger groups can make some steps towards specialization, allowing especially skilled individuals to invest more of their time and effort into developing and passing on their skills. History matters too. It would, for example, be very difficult for ancient hominins to domesticate fire in regions without regular exposure to wildfire. Wildfires give agents many opportunities to learn about the risks and benefits of fire, and many opportunities to partially control fire; to exploit, maintain, regulate, and move fire, without yet being able to ignite at will. Partial control probably long preceded full control; but once ignition was under control, fire-dependent ways of life could spread to regions not naturally fire prone. The bottom line, though, is that we should expect the technical repertoire of a group to be shaped by many factors other than its members’ intrinsic cognitive competence. Moreover, the more competent those agents are, the more their material culture will respond to economic and ecological factors, and the more trade-offs they can explore amongst technical, social, and ecological practices.

Material Culture Makes Us Smarter

In their continuing defense of an innate cognitive difference between Neandertals and sapiens, as Wynn, Coolidge, and Overmann argue:

extant humans are a single species with surprisingly little genetic variation, despite a misleading phenotypic variability .… This means extant humans have all the same cognition, even if that cognition varies slightly in interaction with different material-cultural situations. (Wynn et al. 2016, p. 13)

In my view, that is exactly wrong, and that is the second problem with the simple reflection framework: the causal arrow runs from material and social culture to individual cognitive capacity, as well as from individual capacity to collective culture. Many cognitive capacities are not intrinsic; they are not strongly genetically canalized. Minds are powerfully shaped by material culture and social environment. Social learning does not just store information in our brains that we would not otherwise have; it stores skills, including cognitive skills (Heyes 2012; Heyes and Frith 2014; Henrich 2016). That should not be controversial, so let me make the point with an example that does not depend on the recent invention of scripts and mathematical notation systems. One important cognitive tool is a natural taxonomy or classification system. “Cat” names a natural taxonomic category, because cats are similar to one another in hidden respects, not just obvious ones; and unobserved cats are similar to ones we already know. So the concept is projectable; inductive and causal reasoning about cats is reasonably secure. The category “costs under $10.00” is not very natural; items that cost less than 10 dollars are very heterogeneous, and we can expect prices to vary dramatically over space and time. Woodworkers and other artisans acquire, as an aspect of their skill set, natural classification systems for the various woods on which they work, classifications that relate only very approximately to botanical categories. These help them choose raw materials for their various products: woods suitable for floors may not be ideal for making furniture or for decking; different woods shrink in different ways over time, especially important in the crafting of composite objects, where these variations can either bind components more securely together, or cause them to crack. What is true of wood is true of other craft skills. Artisans do not invent these natural taxonomies of raw materials for themselves, and their invention is far from trivial. In his Structure of Scientific Revolutions, Kuhn charts the route from protoscience to science in various branches of chemistry, and one of his most vivid and important points is the struggle for a natural taxonomy (Kuhn 1970). We do not know the raw materials taxonomy of ancient knappers, nor when those taxonomies became nuanced. But we can be confident that by the late Acheulian, when handaxe production sometimes showed an exquisite control over form, and when artisans were investing significant effort in transporting raw materials over tens of kilometers, that they had developed ways of identifying, evaluating, and probably communicating about different kinds of stone. Artisans with these natural classification systems could think about stone more efficiently than their predecessors without those cognitive tools. Culturally built resources can amplify individual cognitive capacity, and in my view that amplification has its roots deep in hominin history.

Rethinking Behavioral Modernity

The supposed phase shift in human cognition has often been described as the arrival of behaviorally modern humans, with sapiens the species having evolved before the final evolutionary tuning of the human mind (Sterelny 2011). As we have seen, the signal of an abrupt change in material culture was misleading. The apparent rapid acquisition of a suite of impressive new capacities was probably the signal of the arrival of humans who had acquired those capacities piecemeal, over considerable volumes of space and time. But even had the signal been veridical, it is by no means obvious that it would have been best explained by positing the final genetic tweaks that prompted the emergence of a fully modern mind. Wynn et al. (2016) offer a diagnosis: the project of cognitive archaeology is misconceived by focusing on a list of traits found at a place and time; some of these are informative; some are not; some elements on the list will be reliable, regular, uncontroversially central to the lives of those ancient humans; others rare and ambiguous.Footnote 3 They point out that while Neandertals clearly did produce material symbols in the form of shell beads, compared to sapiens they did so very rarely (less than ten unambiguous cases; some thousands found from Aurignacian sapiens). So they suggest a more explicitly comparative and hypothesis-driven approach; their hypothesis is that modern sapiens differ from Neandertals and from their own ancestors in having enhanced memory and executive control. If that hypothesis is right, they suggest we should see archaeological traces of lifeways that depend on deeper planning histories and their execution; of more sustained, complex, perhaps more error-intolerant behavioral sequences. Particular items of material culture—bows, heavy investment in material symbols—may well appear only in specific environmental circumstances. But it is plausible to suggest that the capacity to make and carry through planned sequences of behavior, while it will be manifest very differently in different environmental contexts, will be advantageous across the board. Living for the moment and acting on impulse are rarely even locally optimal.

There are a wide variety of potential signals of lives dependent on executive control, and which might lend themselves to comparative analysis. One is the volume and distance of raw material transport. Another is investment in the environment: there are impressive examples of forager niche construction in the form of fish traps, artificially maintained wetlands, fire stick farming. Wynn, Overmann, and Coolidge (2016) themselves mention traps, nets, and snares, as these require careful placement and patience in their use, not just in their construction. In general, composite technologies require executive control, as the raw materials need to be collected and the components constructed systematically, before the final assembly of the tool. There are degrees of difficulty here: there is persuasive argument that bow-and-arrow technology is particularly demanding, given the number of components of this weapons system as a whole, and given that some of the components are themselves challenging to make. The thought then is that while no particular trace is a smoking gun, if two hominin species vary in their capacities for executive control, over a range of different sites and circumstances we can reasonably expect that difference to be archaeologically visible. An argument with a similar form could be constructed by those who think that the sapiens/Neandertal difference is a difference in social intelligence and levels of intentionality (Gamble et al. 2014). That difference, the argument would go, will underpin a difference in group size and complexity (which might itself have knock-on ecological and technical consequences). In turn, over a range of sites and circumstances, such a difference in social complexity will be archaeologically visible, though not in the same way, and not everywhere.

This explicitly hypothesis-driven, comparative approach does mitigate the problem posed to cognitive anthropology by the sensitivity of material culture to local factors. But it depends on the assumption that within-species cognitive differences are trivial compared to those between species, and in the critical cases we have no reason to accept that assumption. Consider Wynn and Coolidge’s own hypothesis on variation in working memory and executive control. Such variation need not depend on intrinsic features of the agents in question, for planning and control problems can be simplified by agents organizing their workspace. Both David Kirsh and Andy Clark have written extensively on the ways agents simplify tasks by adapting their working environment rather than adapting to their working environment (Kirsh 1995, 1996, 2009; Clark 2003, 2008). Some of Kirsh’s work is specific to contemporary environments because it is focused on human/computer interactions. But much is not. He points out that artisans simplify problems by organizing their tools, raw materials, and components in advance. For example, tasks are easier when nails, nuts, bolts, washers, and the like are pre-sorted into types and into containers; those that will be needed are placed in easy reach and view. Components that are easily confused with one another can be color-coded or marked in some other way. The components from which structures will be built are often laid out in the order in which they will be needed; measured up, and preprocessed (for example, with holes drilled where they will be needed). The layout of the components in the workspace codes the organization of the task itself. The artisan does not have to remember, while carrying through some perhaps precise procedure, what comes next. What comes next is what is next to hand. Cooks often do the same thing: pre-preparing ingredients; making sauces and spice mixtures in advance so they are ready to use; laying ingredients out so they are easy to see and ready to hand. Many of these techniques for reducing the cognitive load of complex task organizations were available to our deep time ancestors: in particular, those of preparing an uncluttered workspace; of having those tools needed ready to hand and ready for use; pre-making those components whose properties are not time sensitive (so bindings can be pre-made; adhesives probably could not be); placing components and raw materials into the workspace so their spatial layout codes the temporal organization of the task.Footnote 4

Likewise, the social organization of work can reduce the cognitive burdenFootnote 5 of complex and temporally extended tasks. Most obviously, labor can be divided. Peter Hiscock’s example of Slippery and Billy working jointly on stone tool production illustrates both the physical and the cognitive division of labor (Hiscock 2004), and in a domain clearly relevant to ancient human activities. The social organization of work can ease the motivational burden of temporally extended tasks too. In collective action, the expectations of others can reduce the temptation to slack off or procrastinate, and as Michael Tomasello has emphasized in his work on collective intentionality (Tomasello 2014), humans find collective activity intrinsically rewarding. When we work together with friends, we are less tempted to slack off, as the rewards of the activity are not so purely instrumental. The upshot of these considerations is that agents’ capacities to carry through temporally deep, intricate, and demanding planned activity does not depend on their internal cognitive resources alone, let alone their genetically entrenched internal cognitive resources.Footnote 6 It depends as well on whether these groups and individuals have learned techniques of adapting their workspace to reduce the cognitive demands on individuals, and on the extent to which they have developed techniques of working together in efficient and mutually encouraging ways.

Cognitive Archaeology Meets Revisionist Cognitive Science

Wynn and Coolidge are now somewhat atypical in cognitive archaeology in framing questions about cognition within a classical cognitive science framework. Perhaps under the influence of Colin Renfrew, within cognitive archaeology there has been a significant uptake of ideas from revisionary cognitive science; in particular, from “4E Cognitive Science”: human cognition is extended, embodied, enactive, embedded (many of these issues are well reviewed in Menary 2010). This view of cognitive science advertises itself as: (1) Recognizing the importance for cognition of external physical prompts and supports. These most obviously include external representational media, perhaps literally as external memory and external working memory, but much more as well. This is extended cognition. (2) It recognizes the importance of know-how, and hence the importance of practice in developing automatized yet still nuanced and flexible skills. Intelligent action does not require explicit, conscious thought and planning. This is embodied cognition.Footnote 7 (3) It rejects (or minimizes) the importance of mental representation; skeptical of the idea that cognition and rational action require agents to represent the world as it is (or how they estimate it to be), and how they intend it to be. Enactive theories of the mind reject the representational theory of mind. (4) This approach recognizes the importance of the cognitive division of labor. Agents acting and communicating collectively are cognitively more than the sum of their parts. For example, John Sutton has reported on elegant case studies showing that long-established couples remember jointly what neither remember individually, because each can trigger recall in the other (Sutton et al. 2010; Harris et al. 2011). This is distributed or embedded cognition.

We need to ask two questions of this conception of human cognition. First: how plausible is it, as a conception of the cognitive lives of living and/or ancient humans? Second: to the extent that it is plausible, how does it affect the project of identifying the cognitive capacities of ancient humans? Reasons of space require me to be brisk and brutal with respect to the first question; I will explore the second in a little more detail.

As I have interpreted them above, the core ideas of distributed and embodied cognition seem to me to be right. One fundamental difference between humans and other great apes is the substantial, intensive development in our lineage—through practice, demonstration, error correction, and advice—of skill and even expertise. High skill might not require (the probably apocryphal) 10,000 hours of practice, but it does require a serious commitment of time, effort, and, often, social support. The result is not just increased competence. That competence is often automatized, allowing agents to exercise their skills while engaged in other activities. Skilled action need not require explicit thought and conscious control.Footnote 8 Great apes are extractive foragers, but they acquire their extractive techniques through trial-and-error attempts to extract, perhaps supplemented by some social learning, not by practice, error diagnosis, or teaching. Skill is so important, and has long been so important, in our lineage that, arguably, a novel form of social hierarchy—a status hierarchy of prestige and esteem—has evolved to incentivize its acquisition and transmission. The centrality of skill both depends on and selects for neural plasticity.Footnote 9 As Lambros Malafouris details in his review, sustained practice reconfigures neural organization: attracting new resources to the task domain; increasing the efficiency with which those resources are used; making their operation more modular, more under local control (Malafouris 2010).

Likewise, whilst the importance of the division of cognitive labor will no doubt vary widely on a case-by-case basis, there is good reason to think that it is an ancient feature of the hominin social landscape. For one thing, lithic technical skills may well have depended, at least since the Acheulian, on socially supported social learning: agents had these stoneworking skills because they were embedded in a social world in which others had them, and facilitated the transmission of their own skills to others. In addition, very likely hominins anciently lived and foraged in a fission–fusion social environment,Footnote 10 ranging over large home ranges (much larger than chimp home ranges, in part because they were bipedal, in part as they shifted into a predator/bully scavenger niche) (Layton et al. 2012). Fission–fusion foraging over large home ranges creates a steep informational gradient, as agents have observed different recent samples of their joint territory, and in a cooperative world, steep information gradients select for information pooling. The same is true as hominins’ foraging came to involve not just collective action but teamwork, where different agents have different roles and hence different perspectives on their common task. Travis Pickering argues that such teamwork, in the form of ambush hunting, dates back to the erectines (Pickering 2013). Even when individual foragers have high levels of skill, there is a payoff for information pooling: Joseph Henrich describes how the highly skilled trackers of the Kalahari pool information, discussing the significance of spores and traces, as they follow game across the desert (Henrich 2016, pp. 76–77). In sum, then, I agree that humans often confront cognitive challenges jointly, and with resources they inherit from their group, and that this matters.

The extended mind theorists, also, are onto an important insight. Humans make specialist information-storage and information-processing artifacts—maps, signs, trail markers, scripts, notation systems, labels, instruments—and their utilitarian equipment often has information-carrying capacities as well. As Dan Dennett notes, a wheel carries the idea of a wheel. Artifacts can carry information a long way: Peter Hiscock points out that the durability of stone artifacts, and the fact that they afford some potential for reverse engineering, mean that they can carry information about themselves for thousands of years (Hiscock 2014). Unfortunately, the extended mind literature, both in its original home in cognitive science and as it has been taken up in archaeology, has framed these genuine insights in an unproductive way: as the claim that these external scaffolds are part of an agent’s cognitive system, literally part of their mind (Malafouris 2008, 2010; Roberts 2015). Thus one famous early thought experiment described Otto, a memory-impaired individual who keeps information about his daily plans in his notebook, which he consults regularly. The contents of the notebook, the suggestion goes, are literally part of Otto’s memory (Clark and Chambers 1998). Lambros Malafouris, one of the cheerleaders of this view in cognitive archaeology, argues that that the regularly used stick of a blind man is part of him.

These claims focus debate on the wrong questions. First, they have triggered unproductive and obscure excursions into metaphysics and the “mark of the mental.” We do not have to ask whether the blind man’s stick is really, literally, part of his embodied mind. The critical question is actually: “how, and to what extent, do these external resources amplify our cognitive powers?” This question can be asked and investigated without haggling about boundaries. Once the importance of the stick is recognized; once we realize how the agent’s neural resources and physical routines have been reconfigured to make his probing the environment with his stick fast, seamless, reliable, fluent, and automatic, there is then no added value in then asking whether his stick is really part of him. Second, the idea that external resources are literally part of an agent’s mind can be argued with some plausibility only for a special class of external scaffolds. The blind man uses his stick very regularly; so regularly, in fact, that there has been neural adjustment to its use; its informational role is automatized. The stick is his; he has exclusive access to it, when and as he chooses. The stick is individualized; either customized in weight, length, and balance (as a test batsman customizes his bat) or he has picked it from a range to suit his individual preferences. It is part of a system in which each component has adjusted to the others, improving the efficiency of the system as a whole. So perhaps there is some plausibility in saying that the stick is part of him, though even here there is plenty of room for skepticism: it is more detachable than other components. But more importantly, there are external scaffolds that lack these features, that are not plausibly part of an agent’s mind, yet still amplify our cognitive powers. In London, I relied on a map of the underground to navigate my way, and navigated prices in cafes using Arabic numerals and English scripts. I have followed marked routes through the bush by using yellow tape and reflectors on trees. These are all public informational resources, used by me on an occasional basis. They are not integrated into my cognitive economy in the way Otto navigates his day with his notebook; and nor have they been individualized, specifically configured to the idiosyncrasies of my mind. But they allow me to think thoughts and carry through plans that would otherwise be impossible. Arabic numerals and positional notation make most of us far better quantitative thinkers than we otherwise would be. It is a mistake to focus exclusively on special cases, and the standard formulations of extended mind ideas in archaeology encourage that mistaken focus (these arguments are developed further in Sterelny 2010).

A word on the final E: the “Enactive Mind.” In rejecting the claim that the mind is a representation-forming, information-processing system, this is a wrong turn. Enactivists substitute obscure metaphors for a mechanistic hypothesis about intelligent behavior; for example “the hypothesis of enactive signification … explores the nature of the material sign not as a representational mechanism but as a semiotic conflation and co-habitation through matter that enacts and brings forth the world” (Malafouris 2013, p. 51). Now you know. Indeed, I incline to the opposite conclusion: the emphasis on practice, know-how, and plasticity; the importance of external scaffolds; the importance of information pooling and the division of cognitive labor, all help explain how we are able to assemble information, organize it in tractable, user-friendly ways, and exploit it. In sum: two-and-a-half cheers for revisionist cognitive science: cognition is partly embodied, distributed, and scaffolded by external resources. What does this do to the projects of cognitive archaeology?

The Archaeological Signals of a Scaffolded Mind?

At first glance, if 2.5E cognitive science is on the right track, the problem of identifying the cognitive capacities of ancient hominins is even less tractable. For it downgrades the value of one of the most regularly used evidential streams: information about the size and organization of ancient brains, as reconstructed from fossil skulls. Wynn and colleagues, for example, labor mightily to establish small but systematic differences in the organization of sapiens versus Neandertal minds, as part of their case for an intrinsic difference between the two cognitive systems. This looks to be a forlorn enterprise if, first, neuroplasticity is profound and pervasive in both lineages; and if, as argued earlier, executive control and working memory can be scaffolded through the material and social environment. At the end of their careful tour of neuroanatomical differences between anatomically modern humans (AMHs) and Neandertals, the authors grudgingly concede the limits of their case:

Paleoneurology is not yet in a position to argue from gross differences in brain size and shape to specific differences in cognitive function, and may never be in such a position. However, differences in gross brain anatomy do imply that differences in cognitive function were almost certainly in place. Neanderthal and AMH brains differed from one another in significant ways. AMH brains, especially, deviated from the typical primate and hominin pattern. It is not unreasonable to hypothesize that these differences had cognitive consequences. (Wynn et al. 2016, p. 7)

Their case is even less persuasive if the brain is hyperplastic, and if cognition is scaffolded and distributed.Footnote 11 An even more striking case is Robin Dunbar’s decades-long project of connecting group size/social complexity to relative neocortical volume (Dunbar 1998; Gowlett et al. 2012; Dunbar et al. 2014; Gamble et al. 2014). The core idea is that social complexity is constrained by social intelligence, for conflict can only be managed if (most) members of the group have an effective and accurate map of their social world and its probable dynamics. An expansion in group size/complexity requires an expansion of socially directed cognitive resources. This in turn requires an expanded neocortex. One result of this project is “Dunbar’s number”: a graph that specifies the maximum size of a networked group for a given relative neocortical volume. Of course many assumptions feed into Dunbar’s argument; the conflict management problem, for example, is sensitive to many features in addition to group size. The motivational psychology of the agents,Footnote 12 the breeding system, and the foraging ecology can all magnify or mute the stresses of social life. But most strikingly, Dunbar and his collaborators defend the connection between neocortical volume and the upper bound on social intelligence, while also defending the extended mind hypothesis. Yet a map of one’s social world can clearly be scaffolded by external resources (badges of rank and role), and through the division of cognitive labor. Gossip is one such resource. The attitudes of third parties to a focal individual is another. Is she treated with respect and deference? When she is engaged in some foraging or craft task, do others pay exact attention to what she is doing; do parents encourage their children to watch? Do others keep out of his way? Are his children well-fed and confident in their interactions with others? And so on. Dunbar and his colleagues do not defend a consistent package. One cannot both defend a relatively robust constraint imposed by neocortical volume on maximal group complexity, and accept an extended and distributed model of cognition.

Thus evidence for plasticity and neural reuse, on both the ontogenetic timescales of individual life history, and on phylogenetic timescales (though not those of very deep phylogeny; hominin evolution is only a few million years deep: see Anderson 2014, and thanks to John Zerilli for alerting me to this literature) threatens inferences from the size and organization of hominin brains to conclusions about the cognitive and social capacities of these brains. But in general, seeing cognitive capacities as tightly coupled to features of the physical and social environment, and to regular patterns in physical activity makes minds less inscrutable (Jeffares 2010, 2014). If thinking depends on doing, and on the world in which the agent is embedded, ancient thinking is more tightly linked to ancient activity. Much, perhaps most, knowledge is know-how, and know-how is manifest in actions that leave physical traces. Moreover, in some respects we have a quite rich picture of the context of human action. We have direct information about the technology, the geographic distribution, the ecological breadth, and the foraging behavior of ancient hominins. We have good reason to believe that hominin range sizes were much larger than those of the living chimp species; they were bipedal, and their greater dependence on meat made their position in the food web (and hence in regional ecology) much more like that of a top predator than a larger vegetarian omnivore. Raw-material transport distances tell us about planning depth and (later) about social networks (Marwick 2003); there seems to be a reasonably clear signal that these transport distances increased over time (though as usual in a patchy, inconsistent way). Morphology encodes information about physical capacities (gait, dexterity, power); diet; life history; mating patterns (through relatively modest levels of hominin sexual dimorphism); and perhaps social complexity. Evidence of recovery from serious injury, for example, is evidence of social support. Morphology cannot tell us when language originated, for it might have originated in gesture. But it can tell us when hominins evolved the capacity for complex control of vocalization (perhaps language; perhaps song and music; perhaps some third, hybrid possibility); the Heidelbergensians seem to have had modern levels of control of their vocalization. Remarkably, there is even some direct evidence of quite specific cognitive capacities: a fluke trace fossil—the Laetoli footprints—is a trace of impressive executive control, for it shows that 3.6 mya Laetolians could track footprints, and consistently and precisely step into those prints (Shaw-Williams 2014).

Of course, there are very serious gaps. The most serious is the lack of clear evidence of band size, and of the size and complexity of metaband organization. There are a few cave sites that give some hint of the size of Neandertal bands, but there is very little clear signal. Soft material technology is largely lost. Equally seriously, since the older the site, the more likely it is to be lost, there are gaps in the record that might be a signal, or might be the erosion of a signal. I shall shortly be discussing the cognitive implications of fire, and there is a gap in the fire signal. There are no clear traces of domestic fire between 700 and 400 kya (Gowlett and Wrangham 2013; Shimelmitz et al. 2014). Did hominins lose the capacity to make fire entirely? Was that capacity retained only by a few groups? Is this just an artifact of the chancy nature of preservation? These are all genuine possibilities.

Even so, while there are gaps, we have a quite rich picture of the social and foraging lives of earlier hominins; of how they were embedded in their world, and this gives cognitive archaeology an important lever. Consider fire. Terrence Twomey has argued, to my mind persuasively, that the domestication of fire is both cognitively demanding and cognitively informative (Twomey 2013). Fire requires planning: fuel needs must be anticipated (perhaps over some time, if a group is choosing a camp site at which they expect to be based for some time). A site must be prepared (especially if a hearth is used), and the fire must be tended and controlled. If a site is poorly chosen, sparks or a change of wind can allow it to escape control. Exploitation is not especially demanding if the fire is used just for warmth and light; control is much more demanding if fire is used for controlled cooking, making adhesives, hardening wood, or improving stone quality. Most obviously demanding are ignition skills, and/or the ability to protect a fire well enough to move it over distance. Moreover, if the fire is made and used collectively, it imposes motivational demands as well. So when we know a hominin group has domesticated fire, we know that they do not just live in the moment, responding to the demands of the now. In addition, they are probably cooperative enough, and socially tolerant enough, to live, prepare food, and eat in close proximity, likely being in close proximity for some hours.

However, while the domestication of fire is demanding, its use helps create an environment that supports the very capacities on which fire depends. Hearths, fireplaces, stored fuel (containers, if there is any container-based cooking), all remind agents of the resources they will need, and set up social defaults of social interaction and eating in company. The ideas of Kirsh and Clark about the organization of workspaces, and the transformation of memory problems into recognition, apply to fire keeping. The previous infrastructure of fire serves both as a reminder, cueing agents of the need to collect fuel and keep the fire alive, and as the path of least resistance for regular patterns of interaction. Once a site has been chosen and prepared, the natural default is to keep using it, and so previous patterns of association and interaction tend to be reinforced. When a fire is used for warmth and light, the economics and physics of fire keeping encourage fire sharing, so domestic fire is likely to both require tolerance of close proximity and to train agents in those capacities; being around a fire in company can become associated with warmth, safety, comfort. These constraints are soft. In principle, fires could be started in a new place every day, and in principle, everyone could have their own fire, and manage their own ignition and maintenance (as families often do, when a small fire for cooking is all that is needed). But defaults of reuse and sharing will save time and trouble, without creating large incentives for free riding; especially if fuel collection and fire maintenance can be offloaded to the young and the old, those not up to the demands of full-on foraging. So as Twomey notes, the control of fire is socially and cognitively informative; perhaps more so, when we see these capacities are nested, supported, and practiced in an environment that is partly built through the regular use of fire; influencing the choice of campsites and their modification, and the social and economic use of space at that site (as firelight eases time budgets by increasing the number of usable hours).

Let me finish by recalling a famous skeptic about inferences from material culture to systems of belief, Christopher Hawkes (1954). My point will be to reinforce these considerations about the evidential importance of the ways cognition is embedded in daily life. As is well-known, Hawkes conceived the inferential problem of studying cognition as a hierarchy of increasing difficulty: we can know what our ancestors did to eat, find shelter, survive. But not what they thought, or how they organized themselves ideologically.

As he saw it, the “ladder of increasing intractability” looked like this:

  • Processes that create sites and materials

  • Subsistence economics (modes of production/environment)

  • Social/political institutions and dynamics (political economy)

  • Religious/spiritual institutions and thought life (ideology)

In contemplating his ladder, he drew an unflinchingly gloomy conclusion:

And now there is worse to come. If material techniques are easy to infer to, subsistence-economics fairly easy, communal organization harder, and spiritual life hardest of all, you have there a climax of four degrees of difficulty in reasoning. What is this climax? It is a climax leading up from the more generically animal in man to the more specifically human. Human techniques, logically speaking, differ from animal only in the use of extracorporeal limbs, namely tools, instead of corporeal ones only; human subsistence-economics differ from animal more obviously, but only (again logically speaking) in the amount and degree of forethought which they involve; human communal institutions next transcend the animal level very considerably; and human spiritual life transcends it altogether. So the result appears to be that the more specifically human are men’s activities, the harder they are to infer by this sort of archaeology. What it seems to offer us is positively an anticlimax: the more human, the less intelligible. (1954, p. 162)

I am no more optimistic than Hawkes, if the project is to reconstruct the specific content of the material symbols of long-vanished social worlds. But we have a reasonable chance of estimating the point and payoff of these investments in ideological infrastructure. Ideology and politics are embedded in agents’ workaday worlds, and both shape and are shaped by those worlds. Thus the ideological life of humans became much more readily visible in the archaeological record at the same time that forager economies became more complex, more spatially and temporally extended, imposing new stresses on their cooperative practices. In my view, this coincidence is no accident (Sterelny 2014). Hawkes took the behavioral ecology of past peoples—as manifest in site formation, material products, and economic activity—to be identifiable independently of their political and ideological life. That is why their political and ideological life is so inscrutable: it was not robustly and informatively coupled to subsistence. Likewise, it is what made the subsistence lives of past peoples scrutable: we do not have to solve the impossible problem of deciphering their ideologies to understand their economies. Neither the pessimistic thought nor the optimistic thought is right: hominin economic and reproductive lives have long involved the solution to cooperation and collective action problems. So, for example, Travis Pickering’s erectine ambush hunters of 1.6 mya depended not just on their technology and natural history expertise; they also depended on their politics. The division of the spoils of collective hunting had to incentivize further cooperation. So the inferential problem is not a ladder: we cannot work out the subsistence economics and the organization of reproduction and then use that as a lever to tell us something about hominins’ normative life and social organization. Rather, these elements of ancient lives constrain one another: hypotheses about material production, the organization of foraging, the distribution of its benefits, intergenerational recruitment, intergroup relations, all constrain, and are constrained by, the others. So, for example, our picture of the lives and life cycles of past humans—growing up in the Mid-Pleistocene—is constrained by our model of material technology (and hence the skills agents needed to acquire) and by our model of their social organization (Nowell and White 2010). Similarly, the Grandmother Hypothesis presupposes that daughters stay in their natal bands, and it presupposes respect for possession (otherwise grandmothers would risk losing valuable resources when taking them back to campsites for processing).

The idea, then, is that to the extent we can reconstruct the social, technical, and ecological lifeways of ancient hominins (admittedly, very partially) we can identify their cognitive and motivational capacities, as their lifeways are not just effects of hidden internal cognitive processes; they are causes, supports, and scaffolds of those processes. The organization of campsites, and the organization of material production trains, rewards, and reinforces patterns of association and social tolerance. Embedded cognition is by no means fully recorded in the material record of the past, but nor is it wholly hidden behind a curtain of equifinality.