by Dan Sperber
The metacognitive experience of having an intuition is sometimes vivid and indeed salient, and sometimes little more than an elusive feeling of self-confidence in a judgment or a decision. There is a continuum from wholly unconscious inferences to inferences of which we have some partial awareness, and there is no clear way to draw a boundary on this continuum. Is there, on the other hand, a boundary between intuitions and the conclusions of reasoning? Is it that, in the case of reasoning, not just the conclusion but also the process of inference is conscious? Don’t bet on it yet. In reasoning, we will argue, the opacity of the inferential processes that is typical of intuitions is not eliminated; it is merely displaced. The metacognitive properties of intuitions will thus help us look at reasoning in a new perspective.
4
Modularity
Hume was right: humans and other animals perform inferences all the time. But are they, in doing so, using the same kind of mechanisms? Animals are born with instincts, whereas humans, the old story goes, having few and quite limited instincts, compensate with higher intellectual abilities, and acquire knowledge and skills through learning. But is this really an “either … or …” kind of issue, either instincts or higher intelligence and learning? Or could learning actually rely on specialized inferential mechanisms that may, to a variable degree, have some instinctual basis?
Between Instinct and Expertise
Comparative psychologists have shown that many animals, such as songbirds, corvids, or apes, acquire complex skills by observing and emulating others or even by discovering on their own new ways of solving problems. Developmental psychologists, on their side, have shown that humans have strong evolved dispositions that influence their cognitive processes from birth. To mention again just one example, from the day they are born (if not already in utero), infants pay special attention to speech sounds and start working on acquiring their mother’s tongue.1 This would already be enough to talk of a “language instinct” (an instinct that, Steven Pinker has famously argued, may do much more than merely focus infants’ attention on the sounds of speech).2
Bridging the gap between instinct and learning, the ethologist and songbird expert Peter Marler suggested that animals have “learning instincts.”3 Depending on how it is interpreted, the expression might be seen as a contradiction in terms—what is learned isn’t instinctive; what is instinctive isn’t learned—or on the contrary as an original way to make a rather trivial point: that some animals, humans in particular, have a biologically inherited disposition to learn. Marler, however, understood the expression he had coined in a more specific and interesting way. A learning instinct, as he meant it, isn’t an indiscriminate disposition to learn anything; it is an evolved disposition to acquire a given type of knowledge, such as songs (for birds) or language (for humans). A learning instinct not only targets a specific learning goal, it also provides the instinctive learner with appropriate perceptual and inferential mechanisms to extract the right kind of knowledge from the right kind of evidence.
Instincts could be seen as “natural expertises.” Expertises could be seen as “acquired instincts.” What makes Marler’s idea of a “learning instinct” a source of insight is that it suggests that, rather than a gap, there can be a continuum of cases between wholly evolved instincts at one end and wholly acquired expertises at the other end. Cognitive mechanisms may occupy various positions on this continuum. In particular, when psychologists study a human cognitive mechanism, the question shouldn’t be: Is it innate or is it acquired? It should be: How much and how is the development of this mechanism in each individual prepared by evolved capacities and dispositions (whether these are present at birth or mature later)? To what extent do some evolved learning capacities target specific learning goals? To what extent do other such capacities, on the contrary, facilitate learning in several domains that in spite of their differences happen to be best understood by using one and the same “mode of construal”4 (as when we spontaneously use psychological category to learn and think not only about people but also about groups and organizations)? How do learning instincts take advantage of experience to produce mature cognitive mechanisms?
Faces, Norms, and Written Words
There has been an ongoing debate on the mechanisms of face recognition. The ability to recognize other people’s faces plays a major role in human social life. Does this competence result just from a lot of practice starting in infancy, a practice strongly motivated by the social benefits of recognizing others and the costs of failing to do so? Or are there also evolved predispositions that drive the attention of infants to faces and provide them with an innate “face template” and with procedures to exploit sensory input in a way uniquely efficient for processing faces?
While nobody denies that experience plays an important role in acquiring the adult competence in the matter, much evidence seems to point to the existence and essential role of evolved predispositions. A small brain area (in the inferior temporal lobe of both hemispheres) named by the American neuropsychologist Nancy Kanwisher the “fusiform face area” is crucially involved in face recognition.5 Lesion to this area results in an inability to recognize faces. Face recognition has certain features not found (at least to the same degree) in the recognition of other visual stimuli of comparable complexity. For instance, face recognition is much less effective when the face is upside down, an effect that is much stronger for faces than for any other kind of stimuli. Still, some researchers have put forward evidence and arguments to suggest that there is no special skill dedicated to face recognition. It is just that we are much more experienced at visually recognizing faces than at recognizing almost anything else. We have in the matter, they suggest, an expertise that comes of habit.
It seems to us—but we are not specialists—that the case for an evolved basis that guides the individual development of a domain-specific face-recognition mechanism is strong and, over the years, has become compelling. Be that as it may, our point in evoking this example is not to take a stand on the issue but to illustrate how today, research on specific inferential mechanisms involves finding out where they might fit on a continuum of possibilities from specialized cognitive “learning instincts” to a proficiency in using general mechanisms to handle specific types of information based on an expert’s level of experience. The very fact that different answers can be given implies that inference is a function that may well be carried out through quite diverse mechanisms.
Similar questions arise for types of learning and inference that are less closely linked to perceptual recognition and more conceptual in character. Parents, for instance, commonly note how eager their toddlers are to learn the “right way” of doing various things and how eager they are then to show their grasp of a norm they have just acquired. This raises a puzzle. When very young children observe actions performed around them, how are they to distinguish those that exemplify norms from other actions that are socially acceptable but neither positively nor negatively sanctioned? Perceptual cues are of very limited help here. Teaching of norms, moreover, varies greatly across cultures; it is much more often implicit than explicit; it never comes close to being exhaustive. So how do very young learners recognize which of the behaviors they observe exemplify norms?
The Hungarian psychologists Gergely Csibra and György Gergely have shown that infants are already disposed to treat information addressed to them “ostensively”—that is, addressed in an attention-arresting, manifestly intentional way—as information of general relevance in their community. When adults ostensively demonstrate some novel behavior, infants readily infer that such behavior exemplifies something like “the way we do things.”
In a famous study, infants were shown a dome-shaped table lamp that could be switched on by pressing directly on the dome. This was demonstrated by an adult who switches on the lamp by pressing on it not with her hand as would have been normal, but with her forehead. When this uncommon action has been ostensively demonstrated to them (as if it were some
kind of game or ritual), infants readily imitate it. If, on the other hand, infants witness the same head-touch action but this time not performed ostensively, then they don’t imitate it; rather, when it is their turn to manipulate the lamp, they switch it on by pressing it with their hand. This research6 reveals a disposition to selectively imitate actions that because they have been ostensively demonstrated are understood to exemplify the “proper way” to perform them.
Toddlers are able to infer that a way to act is normative from the fact that it is ostensively demonstrated. At a later age, they become able to also use characteristic properties of the action itself to infer its normative character. Psychologists Hannes Rakoczy, Marco Schmidt, Michael Tomasello, and Felix Warneken have shown how two- or three-year-olds spontaneously demonstrate their newly acquired understanding of a norm by trying to enforce it when another individual fails to obey it.7 All this suggests that there may well exist in humans a quasi-instinctive disposition to identify and acquire social norms.
Face recognition and norm-obeying behavior are universal features of human social life. It is quite plausible therefore that, in both cases, the cognitive mechanisms involved have been in good part shaped by biological evolution. On the other hand, natural selection is much too slow to have evolved the specialized competencies involved in recent practices such as surfing or computer programming. Even more ancient cultural skills such as reading or chess playing are still much too recent for something like a learning instinct to have evolved in order to help with their individual acquisition.
Until recently, reading was a skill possessed by a minority of experts: scholars and scribes. Even though reading is now quite widespread, it still is, from a cognitive point of view, an expertise, that is, a skill acquired through intense practice guided by organized teaching. Such expert skills stand in stark contrast to ordinary instincts, which need little or no learning at all. And yet neuropsychological studies of reading have shown that its neural basis is quite similar to that of more “instinctive” competencies like face recognition.
The French neuroscientists Stanislas Dehaene and Laurent Cohen have established that the recognition of written words in reading recruits a small and precise brain area they named the “visual word form area” that is next to the fusiform face area in the left hemisphere of the brain. This area recognizes letters and words in the script acquired by the individual independently of whether they are in upper- and lowercase, in handwriting or in printing fonts. There is evidence that the same area is involved in blind people reading in Braille with their fingers. Clearly, the information that this brain mechanism extracts is much more abstract than the visual or tactile stimuli from which it is inferred.
How can it be that a small brain area, the same across individuals, societies, and writing systems, should be recruited for reading? Dehaene and Cohen’s hypothesis is that the development of a dedicated area in readers is the result of a process of “neuronal recycling”:
On the one hand, reading acquisition should “encroach” on particular areas of the cortex—those that possess the appropriate receptive fields to recognize the small contrasted shapes that are used as characters, and the appropriate connections to send this information to temporal lobe language areas. On the other hand, the cultural form of writing systems must have evolved in accordance with the brain’s learnability constraints, converging progressively on a small set of symbol shapes that can be optimally learned by these particular visual areas.8
The visual word form area is situated, as this hypothesis suggests it might be, in a zone that includes several mechanisms dedicated to perceptual recognition of specific kinds of input (including faces). Moreover, the left hemisphere location is close and well connected to the language areas where, once read, written words must be interpreted.
As the cases of face recognition and reading jointly illustrate, both evolved and expert cognitive skills exploit quite specific brain areas, and when these areas are injured, these skills are impaired. In important respects, evolved and expert skills work in a similar manner: much of their operations are fast, quasi-automatic, and tailored to their specific task. Evolved and expert mechanisms are remarkably efficient in their specific domain of competence. When these mechanisms are presented with inputs that do not belong to their proper domain but that nevertheless stimulate them (the way a pattern in a cloud or a smiley may stimulate face recognition), then their performance may be poor or result in “cognitive illusions.”
Modules
All these mechanisms on the instinct-expertise continuum are what in biology (or in engineering) might typically be called modules: they are autonomous mechanisms with a history, a function, and procedures appropriate to this function. They should be viewed as components of larger systems to which they each make a distinct contribution. Conversely, the capacities of a modular system cannot be well explained without identifying its modular components and the way they work together.
The biological notion of a module is a broad one. It allows for big and small modules, for sub- and sub-submodules (such as the various components of the visual system), and for teams of modules that are themselves modular (such as the combination of nervous system submechanisms and features of the human hand that permits both the “power grip” and the “precision grip”). The whole brain is a biological module, and so is a single neuron. A biological module may be an anatomical feature of all the members of a species such as the elephant’s proboscis, or a behavior such as rumination in cows. A biological module may also be an anatomical trait or a behavioral disposition that manifests itself only in certain environmental circumstances such as a callus growing on skin exposed to friction or the collective behavior known as stampede in herd animals.
There are deep reasons why organisms are, to a large extent, modular systems, that is, articulations of relatively autonomous mechanisms that may have distinct evolutionary and developmental trajectories. Individual modules are each relatively rigid, but the articulation of different modules provides complex organisms with adaptive flexibility. Could a nonmodular organism achieve a comparable kind of flexibility? It is not quite clear how. Modular systems, moreover, are more likely to overcome a local malfunction or to adjust to a changing environment. Most importantly, modular systems have a greater—some have argued a unique—ability to evolve.9
In psychology, the mind had long been viewed as a unitary general intelligence with an integrated memory, and connected to the world through various sensory and motor organs. Today, evidence and arguments from neuroscience, developmental psychology, and evolutionary psychology has favored a view of the mind as an articulation of a much greater variety of autonomous mechanisms. Identifying and describing these mechanisms have become a central task of cognitive science.
In philosophy of mind and in psychology, however, talk of modules or of modularity is, for historical reasons, controversial. The notion that the mind might include several autonomous modules was famously defended by the philosopher Jerry Fodor in his groundbreaking 1983 book The Modularity of Mind.10 Fodor argued that the mind’s input systems (perception and language) are modular, while its central processes, reasoning in particular, are not. His stringent and, in retrospect, somewhat arbitrary definition of a module and his ideas about the limited modularity of the mind were a source of inspiration but also of endless polemics. To avoid these polemics, a number of psychologists have resorted to talking of “mechanisms” or, like Stanislas Dehaene, of “processors.” The terminological price they pay in giving up “module” is that they thereby forsake the use of “modular,” “modularization,” and “modularity,” notions that are useful in asking, for instance, how modular is a given mechanism, under what conditions an explicitly taught skill may modularize, or what is the role of modularity in evolution.
Here, we are wholly eschewing the somewhat stale debates raised by the idiosyncratic Fodorian notion of modularity by simply defining cognitive modules as biological modules having a cognitive
function. The notion of a biological module and that of a cognitive function are understood, if not perfectly, at least well enough to allow us to combine the two and move ahead. Contrary to a common misinterpretation of the idea of biological modularity, this does not imply that cognitive modules must be “innate.” Reading, we saw, is a perfect illustration of a cognitive module realized in brain tissues, the biologically evolved properties of which lent themselves to being recycled for a novel, culturally evolved function.
Another common misinterpretation of modularity is to assume that a modular mind must be quite rigid. It makes no sense to compare the relative rigidity of individual modules to the flexibility of cognitive systems as a whole (or to the plasticity of the brain as a whole). In biology, flexibility and plasticity are typically explained in terms of a modular organization. This, we suggest, should also be the case in psychology.
We acknowledge that, given the way research is fast progressing in the domain and how much remains to be done, our current understanding of the organization of the human mind is likely to improve and change considerably in the near future. The idea that the mind consists in an articulation of modules provides, when properly understood, a challenging working hypothesis to contribute to this improvement.11 To move ahead, it is crucial to improve our grasp not just of what inferential modules do but of how they do it. How can individually fairly dumb micromodules combine into smarter but still quite limited larger modules that jointly provide for the kind of superior intelligence that we humans feel confident in attributing to ourselves?
