2. Nature, nurture, culture

2.2. Cultural evolution

The importance of culture (sensu information inherited via social interaction) in shaping human behaviour has been emphasised by cultural evolutionists with increasing force since the 1980s (Boyd & Richerson Reference Boyd and Richerson1985; Campbell Reference Campbell1965; Cavalli-Sforza & Feldman Reference Cavalli-Sforza and Feldman1981; Henrich Reference Henrich2015; Morin Reference Morin2015; Sperber Reference Sperber1996). The idea of cultural evolution comes in three strengths: historical, populational, and selectionist (Fig. 2; Brusse Reference Brusse2017; Godfrey-Smith Reference Godfrey-Smith2009; Lewens Reference Lewens2015). When the term cultural evolution is used in the weakest historical sense, it means nothing more than change over time in some characteristic that varies between human groups. The stronger populational conception assumes that large-scale changes of this sort (e.g., changes in the distribution within a population of the use of particular technologies, or the consumption of certain foods) are the aggregate consequences of many episodes of social learning – of episodes in which individuals learn from others to use a particular technology or to eat a certain food. The strongest conception of cultural evolution, the selectionist view, shares the populational assumption and claims that the conditions necessary for Darwinian or natural selection are present in the cultural domain: There are mechanisms for introducing variation, selection processes, and mechanisms preserving selected variants (Campbell Reference Campbell and Schlipp1974). Cognitive Gadgets pursues the selectionist approach because this approach has the potential to explain the adaptive character of distinctively human cognition mechanisms – why they do their jobs reasonably well. It assumes that genetic evolution and cultural evolution are based on the same variation-and-selective-retention heuristic, and proposes that, rather than being on a short “genetic leash” (Lumsden & Wilson Reference Lumsden and Wilson2005, p. 144), cultural evolution is highly autonomous with respect to genetic evolution.

Figure 2.

Relations between purely historical, populational, and selectionist conceptions of cultural evolution.

3. Starter kit

Cognitive Gadgets suggests that the genetic starter kit for human cognition, although extensive, is very similar to the starter kits of other animals, including chimpanzees. In the course of hominin evolution, natural selection operating on genetic variants tweaked the mind in small but important ways. Genetic evolution has not given us programmes for the development of powerful domain-specific cognitive mechanisms, such as mindreading and language, but it has made us friendlier than our primate ancestors; enhanced our attentional biases towards other agents; and expanded our capacities for domain-general learning and executive control. These are the “Small Ordinary” gene-based changes that enable developing humans to upload “Big Special” cognitive mechanisms – cognitive gadgets – from their culture-soaked environments (Heyes Reference Heyes2018, pp. 52–53).

4. Cultural learning

Cultural evolutionary psychology is both a framework for research and a hypothesis. As a framework, it recognises that distinctively human cognitive mechanisms can be shaped by culturally inherited information, as well as by genetically inherited information and learning (chapter 2). As a hypothesis, it proposes that cultural inheritance has played the dominant role in shaping all or most distinctively human cognitive mechanisms. To advance the hypothesis, chapters 5–8 each look in detail at evidence relating to one type of distinctively human cognition: selective social learning, imitation, mindreading, and language. These are all varieties of cultural learning.

Cultural learning is especially important for two reasons. First, both evolutionary psychologists and cultural evolutionary theorists, although divided on many issues, are united in assuming that the mechanisms of cultural learning are genetically inherited. Therefore, cultural evolutionary psychology warrants pursuit as a descendant of evolutionary psychology and cultural evolutionary theory only if there are good reasons to challenge this consensus. Second, from the perspective of cultural evolutionary theory, which I broadly share, mechanisms of cultural learning play a crucial role in making human lives so different from those of other animals. Like other distinctively human faculties, cultural learning meets challenges that arise in an individual's lifetime, enabling each of us to navigate the world of people (cf. face processing) and things (cf. causal understanding). However, unlike other faculties, cultural learning also underwrites a whole new inheritance system: cultural evolution. It is a gift that goes on giving. Cultural learning enables each person and social group to benefit from the accumulated experience of innumerable other people, past and present, and thereby collectively to acquire knowledge and to develop skills that are way beyond those of other species.

Cultural learning is typically understood to be a subset of processes known as social learning, and social learning processes are thought to overlap with those of asocial or individual learning (e.g., Henrich Reference Henrich2015). This way of thinking (shown in Fig. 3) has been shaped by the anthropologists, biologists, economists, and mathematicians who have pioneered research on cultural evolution, and it has done some good service. However, from a cognitive science perspective, the framework in Figure 3 has two significant problems:

1. 1. Cultural evolutionists tend to treat as processes phenomena that cognitive scientists would regard as effects, that is, as things to be explained rather than things that do the explaining. They ignore the cognitive and neurological processes that produce observable changes in behaviour (e.g., Whiten & Ham Reference Whiten and Ham1992).

2. 2. Cultural learning is understood to be a “sophisticated subclass of social learning” (Henrich Reference Henrich2015, p. 13), but there are no ground rules, empirical or conceptual, for deciding whether a particular type or example of social learning is or is not an example of cultural learning.

Figure 3.

The received view of relations between individual learning, social learning, and cultural learning.

To enable dialogue between cultural evolutionary theory and cognitive science (Heyes Reference Heyes2017b), I propose a subtly different way of situating cultural learning, shown in Figure 4. In this alternative framework, the superordinate category is learning – encoding for long-term storage information acquired through experience. When learning is assisted by contact with other agents, it is called social learning. When learning is not assisted by other agents, it is called asocial learning or individual learning. Cultural learning is a subset of social learning involving cognitive processes that are specialised for cultural evolution – for example, processes that enhance the fidelity with which information is passed from one agent to another. This framework does not allude to processes in distinguishing asocial from social learning, and therefore avoids the misleading impression that social learning is known to depend on different cognitive mechanisms from asocial learning. Furthermore, although it makes the conventional assumption that cultural learning involves processes specialised for cultural inheritance, it does not embody any assumptions about how or why these processes are specialised. Rather, it is a framework for investigation of three questions that cultural evolutionists rarely tackle:

1. 1. Cognition question: How do the mechanisms of cultural learning differ from those of social learning at the cognitive level?

2. 2. Specialisation question: How have genetic evolution and/or cultural evolution contributed to the specialisation of cultural learning?

3. 3. Contribution question: In what ways do the features that distinguish cultural learning from social learning contribute to cultural inheritance? For example, do they make “improved” cultural variants more likely than “unimproved” variants to be passed on?

Figure 4.

A framework for research on the relations between learning, social learning, and cultural learning, enabling dialogue between cognitive science and cultural evolutionary theory.

Rather than appealing to sophistication, implying that we already know what is distinctive about the mechanisms of cultural learning at the cognitive level (question 1 above), this framework defines cultural learning by ostension – by pointing at putative examples of cultural learning. The cultural learning box in Figure 4 lists the five categories of psychological phenomena (each containing behavioural effects and weakly specified cognitive processes) most commonly said by cultural evolutionists to be types of cultural learning:

1. 1. Selective social learning (also known as learning biases, transmission biases, social learning rules, and social learning strategies)

2. 2. Imitation (called true imitation when imitation is used as a synonym for social learning)

3. 3. Teaching (or pedagogy)

4. 4. Mindreading (also called theory of mind, mentalising, shared intentionality, folk psychology, and social understanding)

5. 5. Language (so good they named it once)

These five categories are a natural place to start asking the cognition, specialisation, and contribution questions about cultural learning.

5. Selective social learning

In both of the schemes shown in Figures 3 and 4, social learning names a rag bag of behavioural effects – from a snail following a slime trail, to a student reading about calculus – in which learning by one agent, the “observer,” is influenced in some way by contact of some sort with another agent, the “model” or “demonstrator.” Social learning is said to be selective primarily when the influence of the model varies with the circumstances of the encounter (“when” selectivity; e.g., greater influence when the observer's environment has recently changed, known as a copy when uncertain social learning strategy), or with some feature of the available models (“who” selectivity; e.g., greater influence by older than younger models, known as a copy older individuals social learning strategy).

Selective social learning has been a focus of cultural evolutionary studies since the 1980s, but it barely appears on the radar of cognitive scientists. Consequently, whereas chapters 6–8 bring cultural evolutionary theory to bear on problems in cognitive science, chapter 5 brings cognitive science into closer contact with cultural evolutionary theory. More specifically, chapter 5 tackles head-on the cognition, specialisation, and contribution questions.

Addressing the cognition question, I suggest that most social learning is (1) mediated by the same domain-general, associative processes as asocial learning, and (2) made selective by the same broadly attentional processes that make asocial learning selective. Four lines of evidence support the first of these hypotheses (Heyes Reference Heyes1994; Reference Heyes2012c):

1. 1. Social and asocial learning ability covary. Among birds and primates, species and individuals that perform well in tests of social learning tend also to perform well in tests of asocial learning (Boogert et al. Reference Boogert, Giraldeau and Lefebvre2008; Bouchard et al. Reference Bouchard, Goodyer and Lefebvre2007; Reader et al. Reference Reader, Hager and Laland2011).

2. 2. Solitary animals are capable of social learning. In laboratory tests, animals such as red-footed tortoises (Wilkinson et al. Reference Wilkinson, Kuenstner, Mueller and Huber2010), which lead solitary lives in the wild, prove themselves adept at learning from social cues.

3. 3. Social learning and asocial learning each come in the same three basic varieties: learning about single stimuli, about relationships among stimuli, and about relationships between stimuli and responses, or actions and outcomes (Heyes Reference Heyes1994; Reference Heyes2011). Each type of social and asocial learning has been found in a wide range of species, including humans (Dawson et al. Reference Dawson, Avarguès-Weber, Chittka and Leadbeater2013; Leadbeater et al. Reference Leadbeater2015).

4. 4. Social learning bears the footprints of associative learning. For example, studies of human decision-making combining mathematical modelling with functional brain imaging have found that the same computations, based on the calculation of prediction error, are involved in processing information from social partners (social learning) and personal experiences of reward (asocial learning) (Behrens et al. Reference Behrens, Hunt, Woolrich and Rushworth2008; Garvert et al. Reference Garvert, Moutoussis, Kurth-Nelson, Behrens and Dolan2015; Hill et al. Reference Hill, Boorman and Fried2016).

The second hypothesis suggests that, in most cases, social learning is selective by virtue of domain-general attentional processing, rather than domain-specific strategic processing. For example, when exposed to two potential models, observers attend more to one model than the other, and therefore learn more from one model than the other; they do not learn equally from both models and then, in a second stage of cognitive processing, decide which of the models they should trust to guide their own behaviour. Evidence consistent with this view comes from studies of selective social learning in children, adults, and nonhuman animals (Heyes Reference Heyes2016a; Reference Heyes2016d; Heyes & Pearce Reference Heyes and Pearce2015). However – and here's the crucial part of my answer to the cognition question – in adults and children older than four or five years, there is evidence that some selective social learning is truly strategic; the observer chooses to trust one model rather than another by applying an explicit, metacognitive rule, such as copy the boat builder with the biggest fleet or copy digital natives (Fleming et al. Reference Fleming, Dolan and Frith2012). In one such study, people used information from another agent – advice about which of two options to choose – to the extent that they believed the advisor to be motivated to help rather than to mislead them (Diaconescu et al. Reference Diaconescu, Mathys, Weber, Daunizeau, Kasper, Lomakina, Fehr and Stephan2014). These beliefs were explicitly stated, and the basic effect – covariation between the advisors’ incentives and the participants’ use of their advice – disappeared when participants were told that the advisors did not know which option they were recommending. Therefore, these results indicate that the participants used an explicitly metacognitive strategy such as copy when the model intends to help.

Thus, my answer to the cognition question is: The selective social learning mechanisms that are specialised for cultural inheritance, that constitute cultural learning, differ from other selective social learning mechanisms in being explicitly metacognitive; they represent who knows in the form of conscious, reportable, domain-specific rules. If this is correct, then research on the development of metacognitive rules showing that they are learned through social interaction (Bahrami et al. Reference Bahrami, Olsen, Bang, Roepstorff, Rees and Frith2012; Güss & Wiley Reference Güss and Wiley2007; Heine et al. Reference Heine, Kitayama, Lehman, Takata, Ide, Leung and Matsumoto2001; Hurks Reference Hurks2012; Li Reference Li2003; Mahmoodi et al. Reference Mahmoodi, Bang, Ahmadabadi and Bahrami2013; Mayer & Träuble Reference Mayer and Träuble2013) provides an answer to the specialisation question; it suggests that the selective social learning mechanisms that constitute cultural learning have been specialised by cultural evolution for cultural evolution. Consistent with this answer, there is a growing body of evidence of cross-cultural variation in the metacognitive social learning strategies used by adults (Efferson et al. Reference Efferson, Richerson, McElreath, Lubell, Edsten, Waring, Paciotti and Baum2007; Eriksson Reference Eriksson2012; Henrich & Broesch Reference Henrich and Broesch2011; Mesoudi et al. Reference Mesoudi, Chang, Murray and Lu2015; Toelch et al. Reference Toelch, Bruce, Newson, Richerson and Reader2014). For example, in contrast with Westerners, Fijians are less likely to seek advice from people with more formal education (Henrich & Broesch Reference Henrich and Broesch2011). Compared to Britons, people from mainland China engage in more social learning, and their social learning is less dependent on uncertainty (Mesoudi et al. Reference Mesoudi, Chang, Murray and Lu2015).

Finally, my answer to the contribution question comes in three steps:

1. 1. Metacognitive social learning strategies are able to focus social learning on knowledgeable agents with greater accuracy and precision because these strategies have been honed by cultural selection.

2. 2. When knowledgeable agents can be identified accurately, individuals and social groups can afford to invest in the development of cognitive mechanisms enabling high-fidelity cultural inheritance of skills.

3. 3. High-fidelity inheritance promotes cultural adaptation by reducing the number of models contributing to each new token of a cultural trait and the degree to which the model's influence is contaminated by asocial learning (Godfrey-Smith Reference Godfrey-Smith2012).

6. Imitation

Imitation is the longest-serving category of cultural learning. Scientists have been claiming for more than a century that imitation involves complex computations specialised by genetic evolution for high-fidelity cultural inheritance, and that this cognitive instinct plays a crucial role in allowing humans to make and use tools (Washburn Reference Washburn1908). Chapter 6 embraces the idea that imitation is special but argues that it is made possible by a culturally inherited mechanism. The selling point of chapter 6 is that it addresses head-on the question of how a new cognitive mechanism could be assembled in the course of ontogeny through social interaction.

Imitation occurs when observation of a model causes the observer to perform topographically similar behaviour, that is, behaviour in which parts of the observer's body move in the same way, relative to one another, as parts of the model's body. Thus, the boy in Figure 5 is imitating the men, not because he is wearing similar clothes and heading in the same direction, but because parts of the boy's body, his arms and torso, are configured – spatially related to one another – in the same way as those of the men. Imitation has been assumed to involve complex, dedicated computations because in many cases, like that in Figure 5, it solves a thorny correspondence problem. When the boy puts his hands behind his back, he doesn't see (or hear or feel) anything resembling what he sees (or hears or feels) when he looks at the men putting their hands behind their backs, and yet somehow the boy's cognitive system has produced an action that looks the same, that corresponds, from a third-party perspective.

Figure 5.

An example of imitation.

In the late 1970s, it was reported that newborn human babies can imitate a range of facial expressions and hand movements (Meltzoff & Moore Reference Meltzoff and Moore1977). The reliability and validity of these findings have been questioned repeatedly (Anisfeld Reference Anisfeld1979; Reference Anisfeld, Hurley and Chater2005; Jacobson & Kagan Reference Jacobson and Kagan1979; Jones Reference Jones2006; Reference Jones2007; Reference Jones2009; Koepke et al. Reference Koepke, Hamm, Legerstee and Russell1983; Masters Reference Masters1979; McKenzie & Over Reference McKenzie and Over1983; Meltzoff & Moore Reference Meltzoff and Moore1979). However, replicated and extended in some laboratories, they have led to widespread acceptance of a theory suggesting that the correspondence problem is solved by a black box delivered by the genes. This cognitive instinct theory suggests that humans have an innate device that detects “equivalences between observed and executed acts,” both encoded “supramodally” as “organ relations,” but does not propose computations that would allow organ relations to be derived from observed body movements or cashed out as executed actions (Meltzoff & Moore Reference Meltzoff and Moore1997). Thus, the cognitive instinct theory of imitation says there is a genetically inherited thing that solves the correspondence problem, but it does not say how the thing works. Identifying the thing with mirror neurons (Lepage & Theoret Reference Lepage and Théoret2007) creates another black box. The question “How do people imitate?” becomes the question “How do mirror neurons imitate?”

The alternative, “Associative Sequence Learning” or cognitive gadget, theory of imitation suggests that the correspondence problem is solved by “matching vertical associations” – bidirectional excitatory links between sensory and motor representations of the same action, forged by associative learning during self-observation and specified types of sociocultural interactions (see Fig. 6). This theory offers a mechanistic explanation for the imitation of both familiar actions (sometimes called mimicry) and novel actions (sometimes called true imitation or observational learning). In the latter case, it proposes that, via associative learning, matching vertical associations create a new cognitive mechanism by connecting two domain-general processes that normally operate independently. Matching vertical associations gear perceptual sequence learning, processes that encode the serial order of external stimuli, to motor sequence learning, processes that normally operate only when the agent is learning a new skill, such as riding a bike, through practice (Catmur et al. Reference Catmur, Walsh and Heyes2009; Heyes & Ray Reference Heyes and Ray2000).

Figure 6.

Matching vertical associations are acquired through sensorimotor learning. In the simplest case, self-observation (A), activation of a motor representation contributes to performance of an action (e.g., grasping; dotted arrow), and observation of the performed action produces correlated activation of a corresponding visual representation (dashed arrow). Correlated activation strengthens the excitatory link between the sensory and motor representations, establishing a matching vertical association (solid vertical line). Synchronous activities (B), being imitated by others (C), and optical mirrors (D) provide correlated sensorimotor experience for perceptually opaque actions, such as facial gestures and whole-body movements.

The cognitive instinct theory was recently undermined by a large-scale, longitudinal study of imitation in newborns, which reported negative results for all 11 gestures tested (Oostenbroek et al. Reference Oostenbroek, Suddendorf, Nielsen, Redshaw, Kennedy-Costantini, Davis, Clark and Slaughter2016). In contrast, the gadget theory is supported by evidence of two kinds (for reviews, see Catmur et al. Reference Catmur, Walsh and Heyes2009; Reference Catmur, Press, Heyes, Murphy and Honey2016; Cook et al. Reference Cook, Bird, Catmur, Press and Heyes2014): Training studies involving adults, infants, and nonhuman animals show that imitation – measured behaviourally and via mirror responses in the brain – can be enhanced, abolished, and reversed by novel sensorimotor experience. For example, adults usually do not imitate the actions of inanimate systems, such as robots, but after a brief period of training in which robotic movements are paired with topographically similar body movements performed by the observer, people imitate robots as much as they imitate other people (Press et al. Reference Press, Bird, Flach and Heyes2005). The second kind of evidence indicates that imitation, although flexible and adaptive, has the signature limits (Butterfill & Apperly Reference Butterfill and Apperly2013) one would expect if it is controlled by matching vertical associations. For example, imitation learning is effector-dependent; it does not readily generalise across parts of the body. People who have observed a complex sequence of key-pressing movements can reproduce the sequence when their fingers are in the same keyboard positions as the fingers of the model, but they cannot imitate the sequence when their hands are crossed on the keyboard (right hand operates left keys, and vice versa), or when they are asked to use their thumbs rather than their fingers to press the keys (Bird & Heyes Reference Bird and Heyes2005; Leighton & Heyes Reference Leighton and Heyes2010).

The final section of chapter 6 addresses five objections to the cognitive gadget theory of imitation, emphasising the following points.

1. 1. Intervention versus development. Like most scientific evidence, the results of training studies – and related studies of expertise (e.g., Calvo-Merino et al. Reference Calvo-Merino, Grèzes, Glaser, Passingham and Haggard2006) – support inference to the best explanation, not deduction (Lipton Reference Lipton2003). They favour the gadget theory over the instinct theory because they are exactly what the gadget theory predicts but difficult for the instinct theory to accommodate.

2. 2. Homo imitans. Humans are more skilled and prodigious imitators than other animals, not primarily because they have better resources “on the inside” (e.g., higher capacity mechanisms of associative learning), but because they have superior resources “on the outside,” cultural artefacts and practices that support the acquisition of matching vertical associations.

3. 3. Intentionality. A matching vertical association for an action, x, makes it possible, not obligatory, to imitate x.

4. 4. Overimitation. Children's propensity to imitate instrumentally superfluous features of action (Lyons et al. Reference Lyons, Young and Keil2007) raises questions about the motivation, rather than ability, to imitate. Although the gadget theory is concerned with ability rather than motivation, it is consistent with evidence that overimitation is due to reinforcement learning (Baer & Sherman Reference Baer and Sherman1964; Garcia et al. Reference Garcia, Baer and Firestone1971; Grusec & Abramovitch Reference Grusec and Abramovitch1982; Young et al. Reference Young, Krantz, McClannahan and Poulson1994).

5. 5. What's the use? The gadget model raises the possibility that evolutionists have overlooked the most important function of imitation: high-fidelity cultural inheritance, not of object-directed actions, but of communicative and gestural skills (Heyes Reference Heyes, Sterelny, Joyce, Calcott and Fraser2013).

7. Mindreading

Mindreading, the ascription of mental states, is classed as a form of cultural learning because it is likely to be the special ingredient of human teaching. Effective teaching involves many other cognitive and motivational ingredients, including social tolerance and attentiveness, but mindreading stands out as the most likely candidate for a human-specific cognitive adaptation for teaching.

The idea that mindreading is a genetic adaptation, a cognitive instinct, begins to be less compelling when one compares mindreading with print reading (literacy), a distinctively human cognitive mechanism that is known to be a product of cultural evolution (Heyes & Frith Reference Heyes and Frith2014; see Sect. 1). For example, studies of neural specialization (Van Overwalle Reference Van Overwalle2009), cultural variation (Shahaeian et al. Reference Shahaeian, Peterson, Slaughter and Wellman2011), and genetically heritable developmental disorders (autism; Frith Reference Frith2001), have all been treated as evidence that mindreading is a cognitive instinct, and yet print reading shows comparable degrees of neural specialisation (Dehaene & Cohen Reference Dehaene and Cohen2011) and cultural variation (Changizi et al. Reference Changizi, Zhang, Ye and Shimojo2006), and is associated with genetically heritable developmental disorders of its own (dyslexias; Paracchini et al. Reference Paracchini, Scerri and Monaco2007).

At 5 years of age, monozygotic twins are no more alike than dizygotic twins in their mindreading ability (Hughes et al. Reference Hughes, Jaffee, Happé, Taylor, Caspi and Moffitt2005). This suggests negligible genetic influence and a powerful role for learning in the development of individual differences in mindreading, but it does not tell us what kind of learning is important. In principle, it could be the kind of introspection-based learning emphasised by simulation theory; the science-like learning postulated by theory-theory, in which the child tests her self-generated hypotheses against a database of observed behaviour; or, as gadget theory suggests, a form of cultural inheritance in which mindreading experts – parents and others – instruct children about the mind, in conversation and by structuring developmental environments. In chapter 7 of Cognitive Gadgets, I argue that evidence from natural experiments (Mayer & Träuble Reference Mayer and Träuble2013; Pyers & Senghas Reference Pyers and Senghas2009), observational studies (Meins Reference Meins, Siegal and Surian2012; Meristo et al. Reference Meristo, Hjelmquist and Morgan2012; O'Brien et al. Reference O'Brien, Slaughter and Peterson2011; Slaughter & Peterson Reference Slaughter, Peterson, Siegal and Surian2012; Taumoepeau & Ruffman Reference Taumoepeau and Ruffman2006; Reference Taumoepeau and Ruffman2008), and traditional experiments (de Villiers & de Villiers Reference De Villiers and de Villiers2012; Lohmann & Tomasello Reference Lohmann and Tomasello2003) favours the third of their possibilities. For example, in a natural experiment, deaf people who had been deprived of conversation about the mind because they learned Nicaraguan Sign Language (NSL) when it included very few mental-state terms were less likely to pass a false-belief test than a second cohort who had learned NSL later, when it contained a wider range of mental-state terms (Pyers & Senghas Reference Pyers and Senghas2009). The first cohort was 10 years older than the second cohort; they had had 10 more years in which to introspect and test hypotheses. Therefore, if introspection or science-like learning, rather than conversation, were crucial for the development of mindreading, one would expect the first cohort to be better, not worse, at ascribing false belief.

Studies of implicit mindreading using eye-movement indices of behaviour prediction imply that nonhuman apes (Krupenye et al. Reference Krupenye, Kano, Hirata, Call and Tomasello2016) and prelinguistic infants (Kovács et al. Reference Kovács, Téglás and Endress2010; Onishi & Baillargeon Reference Onishi and Baillargeon2005) are capable of ascribing false beliefs. According to the continuity interpretation, implicit mindreading is mediated by the same, specialized cognitive mechanisms that mediate explicit mindreading in deliberating adults (Baillargeon et al. Reference Baillargeon, Scott and He2010). If the continuity interpretation is correct, mindreading could not be a cognitive gadget because it develops without (apes) and before (infants) conversation about the mind. Two other interpretations of implicit mindreading are compatible with the cognitive gadget theory, however. According to the “two-systems” (Apperly Reference Apperly2010; Perner Reference Perner, Frensch and Schwarzer2010) and “submentalizing” (Heyes Reference Heyes2014a; Reference Heyes2014b; Reference Heyes2015; Reference Heyes2017b) interpretations, implicit and explicit mindreading depend on different cognitive mechanisms. The two-systems view proposes that the mechanisms mediating implicit mindreading are specialised for fast and efficient representation of mental states, while the submentalizing view suggests that they are domain-general mechanisms, representing relatively low-level features of action stimuli – such as colour, shape, and movement – rather than mental states. Evidence that concurrent demands on executive function interfere with explicit mindreading (Bull et al. Reference Bull, Phillips and Conway2008), but not with implicit mindreading (Qureshi et al. Reference Qureshi, Apperly and Samson2010), and that people with autism can engage in explicit mindreading despite impairments in implicit mindreading favour the two-systems and submentalizing hypotheses over the continuity hypothesis (Senju et al. Reference Senju, Southgate, White and Frith2009).

The cognitive gadget theory implies that children learn to read minds through language and therefore appears to be in direct opposition to the Gricean view that ascription of mental states is a precondition for linguistic communication (Bloom Reference Bloom2000; Sperber & Wilson Reference Sperber and Wilson1995; Tomasello Reference Tomasello1999). For two reasons, however, I suspect that the cognitive gadget theory and the Gricean view of language are reconcilable. First, Grice offered a rational reconstruction, rather than a psychologically realistic account, of what is happening when people talk to one another (Sperber Reference Sperber and Aunger2000). Second, Moore (Reference Moore2016; Reference Moore2017) has argued persuasively that Gricean communication can get off the ground – in evolutionary and developmental time – with minimal mindreading; all that is needed is “a basic understanding of others’ purposive activities and desires [which I would characterise as knowledge of action-outcome relationships], operating in conjunction with some tracking what others had or had not seen [or viewed]” (p. 19).

Thus, advancing an alternative to simulation theory and theory-theory, chapter 7 argues that mindreading is culturally inherited – a cognitive gadget. Expert mindreaders communicate mental state concepts, and ways of representing those concepts, to novices. As the present generation of novices become expert, they pass on the knowledge and skill of mindreading to the next cultural generation.

8. Language

I have been thinking about social learning, imitation, and mindreading for a long time, but I write about language as an outsider. While developing the ideas in Cognitive Gadgets, I immersed myself for the first time in research on the origins of language, expecting to find clear evidence that language is a cognitive instinct – one on which gadgets are built. Instead, I found a wealth of evidence that language is itself a gadget and a divide between genetic and cultural evolutionists that no longer appears to be resolvable by empirical means.

Chapter 8 begins by contrasting a gradualist genetic theory of language evolution (Culicover & Jackendoff Reference Culicover and Jackendoff2005; Pinker Reference Pinker1994; Pinker & Bloom Reference Pinker and Bloom1990) with a cultural theory of language evolution (Christiansen & Chater Reference Christiansen and Chater2016), and then discusses evidence that should, or is widely thought to, support one of these theories over the other. The evidence relates to linguistic universals, critical periods, neural localisation, domain-general sequence learning, and social shaping.

9. Cultural evolutionary psychology

The final chapter of Cognitive Gadgets returns to some of the evolutionary questions in chapter 2, now with concrete examples from the case studies, and discusses the prospects for a cultural evolutionary psychology.

9.1. Cultural group selection

Attempting to make the cognitive gadgets hypothesis as clear as possible, I try to spell out who benefits from the cultural selection of cognitive mechanisms and the nature of the benefit. This analysis allows two types of multilevel selection (CGS 1 and CGS 2; Damuth & Heisler Reference Damuth and Heisler1988; Okasha Reference Okasha2005) and uses imitation as an example (see Fig. 7): Imagine a human population divided into two social groups, X and Y, defined geographically or culturally, not by genes. Each person has an imitation mechanism, gearing motor sequence learning to perceptual sequence learning via matching vertical associations (Ch. 6). There are two versions of this mechanism: M and M′; M′ is less common in X than Y. The M′ version has a richer repertoire of matching vertical associations for whole-body movements than the M version, enabling people with M′ more accurately to imitate actions involved in ritual (e.g., dance), hunting (e.g., stalking), and combat (e.g., spear throwing). As a consequence, bearers of M′ are better able than bearers of M to cooperate in a range of tasks (Heyes Reference Heyes, Sterelny, Joyce, Calcott and Fraser2013; Tarr et al. Reference Tarr, Launay, Cohen and Dunbar2015; Tunçgenç & Cohen Reference Tunçgenç and Cohen2016), and to sustain the cultural inheritance of techniques that enhance success in hunting and intergroup combat. These advantages lead groups in which the M′ mechanism predominates to acquire greater numbers of new members (CGS 1 in Fig. 7), or to produce more descendent groups (CGS 2 in Fig. 7), than groups in which M predominates.

Figure 7.

Two types of fitness in cultural group selection.