Let me start with a somewhat obvious caveat. Given the overwhelming amount of feedback I received, my response will miss out on numerous topics worthy of further discussion. I regret not being able to fully honor the careful thought and work put into each commentary, and I sincerely hope my response does not distort matters too much.

Below I do four things. First, I defend my methodology against three points of critique (sect. R1). Second, I reassess the nine cognitive capacities of the target article in light of the open peer commentary (sects. R2–R8). I conclude that my original conclusion stands firm: Human tool use still reflects a profound discontinuity between us and chimps in matters of social and non-social wit. Third, I briefly take up a topic underplayed in the target article, namely, the evolutionary history of the cognitive traits reviewed (sect. R9). Fourth, I reconsider a topic I found underplayed in the commentaries, namely, the question of technological accumulation (sect. R10). I sketch how I am currently incorporating the cognitive assumptions made explicit in the target article into existing population-dynamic models of human cultural evolution; I sketch, thus, how I am making the necessary move from the individual level to the level where cumulative culture must be studied eventually, namely, that of the population.

R1. Methodological notes

R2. Hand-eye coordination

In her commentary, Dounskaia conjectures that differences in human and nonhuman primate motor control may contribute substantially to the uniqueness of human tool use. She offers compelling evidence for the idea that some limb movements require much more cognitive effort than others do. The ability to perform accurately even these more effortful movements, Dounskaia suggests, may have enabled humans to deploy much more sophisticated tools.

At present, however, she lacks proper comparative evidence. Although it seems true that the repertoire of human motor actions greatly exceeds that of nonhuman primates, Dounskaia still needs to establish that this difference is attributable to a difference in the ability to perform more complex gestures. Chimpanzees may have such an ability but not, say, the creativity to exploit it. In this respect, Dounskaia's argument may benefit from an observation made in the target article. Chimpanzees have less neural tissue devoted to their locomotor muscles (Walker Reference Walker2009), so that they must recruit larger numbers of motor units at once. This limits their ability for fine motor control, and arguably, the level of movement complexity they can achieve. Regardless, I certainly find Dounskaia's leading joint hypothesis promising enough to warrant further research, especially in a comparative setting.

Jacquet, Tessari, Binkofski, & Borghi (Jacquet et al.) argue that human tool use does not need to involve the high-level cognitive skills I discuss, as it may be based on much simpler detection systems. Their primary example is affordance perception: Humans are able to recognize manipulation opportunities, “matching the perceived physical features of objects and the agent's biomechanical architecture, goals, plans, values, beliefs, and past experiences” (italics added). The text in italics not only shows that Jacquet et al. deviate substantially from J. J. Gibson's original formulation of affordances, it threatens to undermine their argument.

J. J. Gibson (Reference Gibson1979) indeed introduced affordance perception as a fairly low-level process. Affordance perception referred to an animal's unreflective capacity to discern in the environment possibilities for action, only constrained by its own physical constitution (e.g., that a rake can be grasped; that it affords grasping). As such, affordance perception was a capacity also exhibited by animals that did not have goals, plans, values, or beliefs. Goals, plans, values and beliefs were added as constraints only in the work of Norman (Reference Norman1988), whose research primarily concerned humans. Norman's reformulation (which Jacquet et al. adopt) is not merely terminological. It implies a shift of focus from direct, low-level perception to indirect perception; that is, perception dependent on interpretation and background knowledge. With an example of Norman's, a knob on a refrigerator may be directly perceived as turnable (per Gibson), but one needs a Normanian conceptual model to perceive it as “to-lower-the-temperature-with–able.” Likewise, the “Delete” key on a keyboard may be directly perceived as pressable (per Gibson), but one needs a Normanian conceptual model to perceive it as “to-delete-a-character-with–able.”

In sum, Jacquet et al. face a dilemma. They either endorse a Normanian notion of affordance, thereby making affordance perception a conceptually rich, and fairly demanding, enterprise (as the target article's section on function representation suggested). Or they pursue a Gibsonian account, at the cost of being unable to explain the humanique ways in which humans navigate their humaniquely engineered environments.

R3. Body schema plasticity

Arbib and Longo & Serino find my conclusions regarding body schema plasticity uncompelling. According to these commentators, the question is not so much whether nonhuman primates can extend their body schema, but whether this happens as flexibly and rapidly as in humans. That question, Arbib and Longo & Serino believe, should be answered with a clear “no.”

I remain unconvinced. Let me start with Arbib. Arbib refers to a study by Arbib et al. (Reference Arbib, Bonaiuto, Jacobs and Frey2009) in support of his argument. Now, Arbib and colleagues observe the facility humans have in tool use, and infer from that fact that human body schema plasticity has unique properties (p. 458). But this does not follow. Tool use in nonhuman primates may be cumbersome due to numerous other reasons (limited grasp of causality, poor hand-eye coordination, and so forth). To be fair, this point is conceded a bit further in the text, when the authors suggest how future studies could establish the difference between humans and nonhuman primates as regards body schema plasticity. But in and of itself, the paper by Arbib et al. (Reference Arbib, Bonaiuto, Jacobs and Frey2009) does not seem to provide the necessary evidence.

Longo & Serino's comparative evidence is wanting, too. They refer to a study by Quallo et al. (Reference Quallo, Price, Ueno, Asamizuya, Cheng, Lemon and Iriki2009) in support of the idea that the body schema of monkeys is fairly rigid. Quallo and colleagues indeed demonstrate fairly persistent increases of gray matter in, among other places, the intraparietal sulcus of macaques that were trained to use a tool. Still, what Longo & Serino do not mention is that similar increases were observed in human volunteers learning to juggle (reported by Draganski & May Reference Draganski and May2008, as cited by Quallo et al. Reference Quallo, Price, Ueno, Asamizuya, Cheng, Lemon and Iriki2009).

R4. Causal reasoning

By and large, commentators propose three useful extensions to the target article's section on causal reasoning.

First, that the section would have benefited from discussions of experimental paradigms other than trap-tube tasks (Taylor & Clayton) and of experimental paradigms other than those presented in the target article's Figure 1 (Cachel). In light of the study by Seed et al. (Reference Seed, Call, Emery and Clayton2009), Taylor & Clayton remark, for example (as I do), that the reason for the modest performance of chimpanzees in trap-tube tasks may be demands posed by the tool aspect of the task; that is, that the extra cognitive load may block the animals' ability to properly assess the task's causal set-up. Other changes, such as allowing animals to push rather than pull the food item in the tube, may also yield different results (Mulcahy & Call Reference Mulcahy and Call2006a). In sum, one should not draw too strong conclusions about great ape causal cognition from only one, potentially confounded test. Perhaps Taylor & Clayton are right that trap-tubes received too much attention in my paper. On the other hand, I do shortly describe three other of Povinelli's (Reference Povinelli, Reaux, Theall and Giambrone2000) seminal experiments (see also Fig. 1): the flimsy-tool problem, the inverted-rake problem, and the table-trap problem. Chimpanzees performed poorly on these tests too; and, as Cachel is right to point out, Povinelli's book describes even more experiments, which together are at the very least suggestive of the fact that chimpanzees' grasp of causality is rather modest. There is little follow-up research based on these other paradigms, which is rather unfortunate indeed.

Second, Penn, Holyoak, & Povinelli (Penn et al.) miss a treatment of non-mechanism approaches to causal understanding. Indeed, in the target article I suggested that causal understanding requires the cognizer to infer a mechanism that relates the cause to the effect. This mechanism account is associated most prominently with Ahn and colleagues (Ahn & Kalish Reference Ahn, Kalish, Wilson and Keil2000; Ahn et al. Reference Ahn, Kalish, Medin and Gelman1995) and is fairly intellectualist:

Penn et al. are right that there are other, less-demanding accounts of causal understanding. For example, Waldmann and Holyoak (Reference Waldmann and Holyoak1992; Waldmann et al. Reference Waldmann, Holyoak and Fratianne1995) argue that human mental representations of cause-effect relations are organized into causal models. Basic causal models include representations of directionality (e.g., the causal arrow between A and B goes from A to B, not the reverse), strength (A impacts strongly/weakly on B), and polarity (A makes B happen versus A prevents B from happening); they typically do not refer to the mechanisms responsible for the said cause-effect relation. A cognizer may know that there exists a strong causal arrow from eating rotten food to diarrhea without appreciating the unobservable underlying mechanisms – say, how bacterial toxins derange the normal bowel flora.

This position clearly conflicts with the view of Ahn & Kalish (Reference Ahn, Kalish, Wilson and Keil2000; with their third point in the quote above in particular), but it can still be made to fit with the idea that chimpanzee causal understanding is limited. Using the terminology of Penn et al., chimpanzees may be able to represent “first-order” causal models, but not “higher-order” ones. That is, whereas chimpanzees may be able to reason about the causal relationships between observable contingencies, they do not generalize these principles into higher-order “intuitive theories” (Penn et al. Reference Penn, Holyoak and Povinelli2008), which typically refer to unobservable causal properties, such as gravity. Whereas for chimpanzees causal arrows between A and B remain on a perceptual level, no such limitations hold for the human case.

A third extension to my discussion of causal reasoning is offered by Orban & Rizzolatti, and it concerns a putative neuronal basis for the enhanced grasp of causality observed in humans. They refer to a study by Peeters et al. (Reference Peeters, Simone, Nelissen, Fabbri-Destro, Vanduffel, Rizzolatti and Orban2009), who found evidence that a specific sector of left inferior parietal lobule (i.e., anterior supramarginal gyrus, or aSMG) was activated in humans during the observation of tool use, but not in monkeys. Importantly, aSMG is not involved in understanding causal relationships in general; it codes tool actions in terms of the causal relationship between the intended use of the tool and the result obtained by using it. This study is interesting for at least two reasons. First, it may resolve some of the uncertainties regarding production-level representations of tool use skills (see Note 18). That is, aSMG may support larger motor repertoires, thereby supporting larger toolkits. Second, with the proviso that Peeters and colleagues studied rhesus monkeys and not chimpanzees, aSMG may perhaps explain why, as observed above, chimpanzees fail the trap-tube task when tools are implied. To wit, human aSMG would provide the computational power needed to overcome the additional demands posed by the tool aspect of the task.

R5. Function representation

Commentaries on the target article's section on function representation reveal some confusion as regards the notion of function. Several authors (i.e., Blitzer & Huebner; Osvath, Persson, & Gärdenfors [Osvath et al.]; Patterson & Mann; Penn et al.) argue that monkeys and apes are able to form functional representations, because these animals are capable of distinguishing between “functional” and “non-functional” tools (see e.g., Osvath & Osvath Reference Osvath and Osvath2008) and are able to distinguish between “functionally” relevant (e.g., the shape of a rake) and “functionally” irrelevant (e.g., the color of the rake) properties of a tool (see e.g., Santos et al. Reference Santos, Miller and Hauser2003). Where these authors refer to functionality, I would speak rather of causal efficacy: An ape may appreciate that a certain rake is causally efficacious for food retrieval, but this does not mean it attributes to the rake that function. For that to happen, the ape must somehow conceive the rake as being for the said purpose. To get a feeling for the distinction: I may appreciate that a cup is causally efficacious to be used as a paperweight without attributing to it that particular function.

How could we know whether nonhuman primates form such permanent function representations? One way is to see whether they stick to a tool when functionally equivalent alternatives become available. The target article referred to a study by Cummins-Sebree and Fragaszy (Reference Cummins-Sebree and Fragaszy2005) suggesting that they do not. Patterson & Mann, however, are right to point out that Whiten et al. (Reference Whiten, Horner and de Waal2005) may count as counter-evidence. In that study, chimpanzees continued to use a tool for its function even in the presence of functional equivalents.

Second, evidence of re-use of tools would support the idea of stable function attributions. I suggested that reports of tool re-use are scarce, with the exception of a study by Carvalho et al. (Reference Carvalho, Biro, McGrew and Matsuzawa2009). Thanks to Blitzer & Huebner, I can here add a study by Sanz and Morgan (Reference Sanz, Morgan, Lonsdorf, Ross and Matsuzawa2010).

Finally, observations of functional fixedness would indicate that tools are conceptualized as being for one particular purpose rather than another. The target article suggested that functional fixedness was a humanique phenomenon. Yet, a study that was not available at the time of writing the paper – performed by Hanus et al. (Reference Hanus, Mendes, Tennie and Call2011) and pointed out to me by Patterson & Mann and Rizzo – may prove me wrong. Hanus and colleagues indeed provide suggestive evidence for functional fixedness in chimps. What remains to be seen, however, is whether chimpanzees' fixedness attests to a conceptual system storing functional information (as in humans), rather than being the result of associative learning, where repeated exposure to a tool's function blocks alternative, more creative uses.

For all three diagnostic features, it appears, commentators have raised quite forceful counter-arguments. Contrary to what I stated in the target article, it may therefore well be that nonhuman primates attach particular functions to particular objects. Whether they hereby rely on a conceptual system storing functional knowledge remains uncertain, as well as the question of what difference that would make.

Incidentally, Gainotti makes an intriguing remark about the conceptual system implied in human functional representation. He observes that tool concepts are typically represented unilaterally in a left-sided fronto-parietal network, because of their close link to actions, which are typically performed by the contra-lateral right hand. Living category concepts (e.g., about animals, plants), by contrast, rely more on visual data and are therefore stored in a bilateral network comprising rostral and ventral parts of the temporal lobes.

R6. Executive control

The target article subdivides executive control into mechanisms of monitoring online action, inhibition, foresight, and autocuing. Commentaries primarily take issue with my treatment of the latter two. Weiss, Chapman, Wark, & Rosenbaum (Weiss et al.) and Osvath et al. challenge my views concerning foresight; Stoet & Snyder add considerable refinement to my discussion of autocuing. Finally, the commentary by Beck, Chappell, Apperly, & Cutting (Beck et al.) sheds new light on the role that executive control plays in tool innovation. Let me consider each commentary in turn.

Weiss et al. describe research evidencing anticipatory effects in the reaching behaviors of lemurs, tamarins, and rhesus monkeys. These monkeys were shown to prefer non-canonical hand postures in preparation of a subsequent grasping task. Such behavior, the authors point out, indicates some form of planning ahead. I agree. Still, the behavioral evidence does not meet the standards of foresight set in the target article; that is, it does not involve the formation of long-term goals, nor the prospection of needs other than those experienced in the immediate present. I certainly do not mean to downplay the significance of more basic forms of foresight, as those described by Weiss and colleagues. I believe indeed that these may increase our understanding of the evolution of planning and goal maintenance. To press the issue, however, the target article focused on those types of foresight where discrepancies between humans and nonhuman primates might be most apparent. In a search for discontinuities, I think, such an approach is justified. Nonetheless, I admit that to do right to the short-span motor-planning abilities discussed by Weiss et al., one would need to start with a much finer grained subdivision of executive control than the fourfold subdivision I deployed.

According to Osvath et al., the target article misinterprets a study on great ape foresight by Osvath and Osvath (Reference Osvath and Osvath2008). They argue that I dismiss Osvath and Osvath's results too readily as a consequence of associative learning rather than as a consequence of foresight. In light of a paper by Osvath (Reference Osvath2010) that Osvath et al. refer me to, I am willing to concede (again) that the experiments by Osvath and Osvath properly control for associative learning. Yet, my other observation still holds: The results of Osvath and Osvath (Reference Osvath and Osvath2008) may be due to inhibitory strength rather than to forethought – at least if we evaluate their experiments by the standards they set themselves:

The rewards that Osvath and Osvath believe to be tapping “different desires” are a grape and half a liter of rosehip berry soup. Osvath et al. agree, and they justify Osvath and Osvath's assumption based on the idea that “eating and drinking are dissimilar activities, with different physiological outcomes.”' This may be a salient distinction when the comparison concerns, say, eating a grape and drinking water, but much less so when it concerns eating a grape and drinking rosehip berry soup. These latter activities have at least one target in common: a craving for fruity sugars. On this construal, subjects in the experiments of Osvath and Osvath may well have exercised inhibitory strength, but not have anticipated a drive different from the present one.

Stoet & Snyder refer to a set of recent and fascinating studies that demonstrate endogenous control – or as I called it, autocuing – in monkeys. These animals appear capable of letting internal representations act as cues for their behavior, rather than merely reacting on external stimuli. The reason why I believe autocuing to be relevant for tool use differs from that of Stoet & Snyder. My thought, which does not conflict with the observations of Stoet & Snyder, is that it allows deliberate practice, needed to achieve skill in complex tool use. Stoet & Snyder also see a link with skill complexity, but spell this out in terms of enhanced concentration in humans. Humans appear less flexible to switch rapidly between endogenously controlled task representations. That, in turn, supports concentration, a necessary component of long-lasting and complex tasks, such as developing skill in complex tool use. Together, Stoet & Snyder's and my proposal make plausible why humans, compared with other primates, seem to be capable of learning so much more intricate tasks-sets.

Lastly, the commentary of Beck et al. targets one of the outstanding questions formulated at the end of the target article. There (sect. 12.2.1) I observed that executive control appears critical for innovative tasks, such as solving Tower of London problems, and I asked whether the same would hold for other innovative acts, especially those involving tools. Beck et al. report on evidence that tentatively supports my suggestion. The authors tested human children on a tool innovation task based on Weir et al.'s (Reference Weir, Chappell and Kacelnik2002) wire-bending problem. Children up to 5 years old had great difficulties fashioning a straight piece of wire to make a hook for retrieving a bucket from a vertical tube. Given that the children displayed a proper causal understanding of the task, Beck et al. suggest that the children's poor performance was due to the immaturity of their executive system. It is unclear, however, how much executive control is really needed for solving the wire-bending problem. Prototypical tests of executive function involve multi-step actions (e.g., the Tower of London task, the Six Element Test), where a solution must be planned ahead and kept in mind during each step of the task. No such goal maintenance is implied, it seems, in the single-step wire-bending problem, where the ultimate solution of the task and its execution can run almost in parallel. Future research on the performance of dysexecutive patients on similar single-step and open-ended tasks could perhaps corroborate the hypothesis of Beck et al.

R7. Social learning, teaching, social intelligence

Surprisingly few commentators seem to disagree with my presentation of primate social skills (social learning, teaching, and social intelligence). Osvath et al. find my treatment of theory of mind too short – I agree, but referred the reader to the much more detailed discussions by Penn and Povinelli (Reference Penn and Povinelli2007b) and Call and Tomasello (Reference Call and Tomasello2008). Moerman points to the enormous impact of new kinds of social organization on recent technological developments – I fully agree, and consider this topic more fully in section R10. Finally, Tennie & Over believe that I too quickly reject explanations based on a small number of social traits. In particular, they make the following two claims: (1) Humanique forms of social learning and teaching are sufficient to explain cumulative culture; and (2) cumulative culture positively impacts on cognition, giving rise to many of the non-social cognitive skills discussed in the target article.

In the target article, I provided two arguments that undermine Tennie & Over's first claim: the problem of the Acheulean, and the fact that non-social skills are part and parcel of sophisticated forms of social learning. This may not have convinced Tennie & Over. Therefore, let me provide an additional argument, which I draw, quite ironically, from a study referred to by Tennie & Over themselves, namely, Enquist et al. (Reference Enquist, Ghirlanda, Jarrick and Wachtmeister2008).

Tennie & Over invoke that paper in support of their second claim. Indeed, Enquist and colleagues show that exponential cultural accumulation requires feed-forward loops between culture and creativity (or “intelligence,” as Tennie & Over call it). Whereas genetically evolved creativity may produce accumulation at a constant rate, only culturally evolved creativity has the power to yield accelerating accumulation. But what Tennie & Over omit to mention is that, according to the very same models of Enquist and colleagues, the process of accumulation can get started only once genetically evolved creativity has evolved. Genetically evolved creativity, not cultural transmission, is the primary genetic bottleneck for cumulative culture:

Put differently, for Enquist and colleagues, cultural transmission is insufficient for sustaining processes of cumulative culture – pace Tennie & Over. Incidentally, Enquist and colleagues black-box the cognitive skills that make up genetically evolved creativity. In the target article I discerned at least two contenders: a capacity for causal reasoning (sect. 12.1) and enhanced executive control (sect. 12.2; see also Beck et al.).

Let me turn to Tennie & Over's second claim. Here the idea is that cultural environments are responsible for qualitative changes in cognitive skills. Tennie & Over write: “[A]t least some of the factors that Vaesen identifies as causes of human tool use are, in fact, effects of growing up in rich cultural environments.” This may be right. To have bite, however, Tennie & Over need to specify which traits are implied. And evidently, they need to show for every single trait on the list that it is culturally acquired rather than innate. I am prepared to go for either option; but at present, especially in the face of a disheartening scarcity of cross-cultural data, I think it is more honest to admit that the science is not settled yet.

R8. Language

IJzerman & Foroni provide an argument that is structurally similar to the one of Tennie & Over. What social learning is for Tennie & Over, language is for IJzerman & Foroni. That is, IJzerman & Foroni argue that I underestimate the role of language in supersizing humans' cognitive toolkit, and that I thereby overestimate the cognitive discontinuity between chimps and humans.

In response, let me repeat what I did and did not do in the target article. I compared humans and chimps with respect to nine tool-related cognitive skills (including linguistic ability, for that matter), and I found that humans excelled in almost all of them. Thereby, I deliberately bracketed questions of implementation. Our excellence may be hardwired, culturally acquired (as Tennie & Over propose), a side effect of our linguistic ability (as IJzerman & Foroni propose), or a bit of all three (see also sections R1.2, R7, and R9). In my opinion, IJzerman & Foroni overestimate how much we know about the impact of language on our cognitive toolkit to be able to adjudicate among these four scenarios, but I do not want to press that point. Instead, let me formulate two further critical remarks.

First, to be able to make their argument, IJzerman & Foroni must rely on a comparative assessment of the sort that the target article gives. For example, IJzerman & Foroni believe that language supersizes human planning abilities and executive control (Blitzer & Huebner, by the way, make a similar observation in passing). Such a claim makes sense only if humans have superior planning ability and superior executive control to begin with – indeed, precisely what the target article attempted to show. More generally, one does not need to prove a trait's independence from linguistic ability to be able to judge whether humans have it and how good they are at it.

Second, there is something in IJzerman & Foroni's charges that I cannot help but perceive as a plain inconsistency. The claim that language supersizes the human cognitive toolkit at the very least suggests a profound cognitive discontinuity between us and chimps; yet, IJzerman & Foroni charge me with overestimating the cognitive discontinuity between humans and chimps.

The second strand of comments concerning language comes from Holloway, Arbib, and Barceló-Coblijn & Gomila. These commentators point out, either implicitly or explicitly, that I have neglected the possibility that tool use and language co-evolved. And indeed, it is rather unfortunate that the target article examined only accounts according to which the evolution of tool use played a causal role in the subsequent evolution of language.

Holloway observes striking similarities between human language and toolmaking. He refers to his seminal paper “Culture: A Human Domain” (Reference Holloway1969), where he described the similarities as follows:

Holloway's co-evolutionary thesis appears a bit further on:

These early ideas clearly resonate in the more recent accounts of Arbib and Barceló-Coblijn & Gomila. According to Arbib, the evolution of complex forms of imitation underwrites the co-evolution of language and toolmaking. Complex imitation, here, involves increased capacities for recognizing and imitating hierarchically structured processes, needed for assembling both words (in the case of language) and actions (in the case of toolmaking) into superordinate structures.

Also Barceló-Coblijn & Gomila are keen to point out the profound similarities between toolmaking and language. In knotting, in particular, they see a formal structure of similar complexity to a context-sensitive grammar. Tying knots in nets and basketry, for example, cannot be specified as an iterable sequence of steps at the service of a higher-level constructive plan, because “each single operation [e.g., tying one of the knots of the net] is conditional on the state of the rest of the fabric and the physical forces the knot is supposed to resist.”

There is much to be said in favor of the accounts of Holloway, Arbib, and Barceló-Coblijn & Gomila. Still, I have one worry, which is not sufficiently stressed in the target article. Attempts at determining structural commonalities between language and toolmaking are easy prey for charges of arbitrariness. Take Barceló-Coblijn & Gomila's claim that context-sensitive procedures emerged very recently, only with the advent of knotting in Homo sapiens. Now, compare this with Holloway's (Reference Holloway1969) interpretation of Acheulean toolmaking:

As far as I can tell, Holloway here interprets Acheulean toolmaking as a context-sensitive procedure, in which each single blow is conditional on past and future blows. In the absence of a rigorous method for making similarity judgments, it is hard to decide whose interpretation is correct, Holloway's or that of Barceló-Coblijn & Gomila. Even an analysis of hierarchical complexity in stone toolmaking as detailed and systematic as that of Stout (Reference Stout2011, referred to by Arbib) contains a fair amount of arbitrariness (as Stout himself admits, p. 1057); attempts to mirror his approach onto (proto)language would only add more of it. Presumably, similarity will keep residing in the eye of the beholder.

R9. Evolutionary issues

Even if one accepts my description of human tool-related cognitive abilities, how did all these abilities evolve? I am glad that so many commentators took up that pertinent question in my stead. Broadly speaking, their hypotheses fall into three groups.

First, Crabb endorses the view that human technological ingenuity emerged in response to a process of technological selection. He argues that hominids, unlike other tool-using species, depended on tools for their survival. The increasingly dry and open landscapes made our ancestors extremely vulnerable to attacks by predators; the use of weapons for protection would clearly confer fitness advantage. Subsequent elaborations on these early tools would have provided even more survival benefits, and as such, favor even more cognitive sophistication.

Crabb's hypothesis is reminiscent of, but interestingly different from, earlier technological intelligence hypotheses (for an elegant discussion, see Byrne Reference Byrne, Whiten and Byrne1997). According to these, tool use skills are favored whenever there is a premium on gains in efficiency with respect to (extractive) foraging; on Crabb's account, in contrast, the premium would initially be on gains in efficiency with respect to protection. What puzzles me, however, is how Crabb's account would accommodate the fact that the earliest known tools (i.e., Oldowan flakes) offer little protection against animal attacks. In this respect, earlier versions of the technological intelligence hypothesis seem to fare much better.

Second, several commentaries endorse some kind of cultural intelligence hypothesis. Tennie & Over, as discussed earlier, argue that the evolution of humanique forms of social learning and teaching subsequently drove the cultural evolution of other tool-related cognitive skills. Nielsen expresses a similar view, but adds quite a forceful argument in its favor. He observes that humans are the only species to have a childhood as a life stage, which provides ample opportunities for the acquisition of complex skills – including cognitive skills related to tool use. Finally, the examples of niche construction given by Blitzer & Huebner, Jeffares, Nonaka, and Arbib seem consistent with a cultural intelligence hypothesis, although not necessarily of the ontogenetic kind (as the one of Tennie & Over).

Third, there is the view that technical and sociocultural cognitive traits co-evolved, in concert with increasing brain size and reflecting a general cognitive ability. On this account – endorsed by Reader & Hrotic, Gibson, and perhaps Penn et al. – neither social nor ecological challenges alone account for human cognitive and brain evolution. In support of this view, Reader & Hrotic point to a very recent study by Reader et al. (Reference Reader, Hager and Laland2011), which I find particularly compelling. Reader and colleagues compiled cognitive measures from multiple domains (social, technical, ecological), examined their interrelations (for 62 primate species), and found strong cross-species associations. Rather than that each trait evolved in response to species-specific social and ecological demands, it therefore seems more likely that social, technical, and ecological traits evolved in concert, as part of a highly correlated cognitive suite.

R10. From individual cognition to population-level culture

My primary reason for examining primate social and non-social wit was the belief that doing so would help us to explain why technological accumulation evolved so markedly in us, and so modestly in chimps. To be sure, I was fully aware that an examination of cognitive capabilities alone would offer only half an explanation; that for the other part, one would need to study how these abilities play out at the aggregate level.

Therefore, I am in agreement with Ragir & Brooks that human cultural evolution cannot be properly understood if population and group dynamics are ignored. But the reverse holds as well: One needs accurate micro-level data to be able to give meaningful descriptions at the macro level. Consider, for example, Ragir & Brooks' contention that “changes in population density result in the specialization of labor and knowledge,” and that “as communities increase in size, functional institutions appear.” Without a proper micro-level foundation, these explanations are highly unsatisfactory. Increasing population densities will favor specialization and functional institutions only in animals that meet certain cognitive requirements; otherwise, the animal kingdom would have been replete with species as cooperative and institutionalized as humans are.

Moerman appreciates this complementarity well. He finds my characterization of human tool-related cognitive abilities exemplary but insufficient to account for such astonishing achievements as cathedrals, iPhones, and symphony orchestras. To explain these, Moerman argues, one also needs to consider the novel ways in which humans tend to organize themselves, acting collectively towards otherwise impossible outcomes. Although the target article described a set of micro-level mechanisms that enable these forms of cooperation (see e.g., sect. 12.3.1 and 12.3.2), I agree that their impact remained somewhat elusive.

Therefore, as a natural follow-up, I already started developing an agent-based model to assess the impact of collective learning on cumulative culture. Preliminary results indicate that at certain levels of technological complexity, default mechanisms of individual and social learning are unable to sustain further accumulation; and that at that point, only collective learning is able to reboot the cumulative process. The model is also used to examine the effects of certain population characteristics; for example, how isolation and interconnectedness of subpopulations play out at higher levels of aggregation.

In the present version of the model, complexity is defined just in terms of the number of components that a technology has. In a later stage, however, complexity will also be a measure of the number of interactions between components. Based on a paper by Rivkin (Reference Rivkin2000), the prediction now is that, even given collective learning, cumulation levels off at a critical level of complexity; and that the process can recover once the causal relationships between components are understood. Thereby, the macro-level impact of another favored micro-level trait, namely, causal reasoning, would have been addressed.