Words, Thoughts, and Theories

1. The Other Socratic Method

• The child treats the adult language as another source of information about the structure of the world, along with her own experience. While the child is inferring the structure of the outside world, she is also untangling the structure of language itself, and she may use solutions in one area as clues to solutions in the other.
• Ultimately, our motivation for watching and talking to children is the same as Socrates'. The most central questions in cognitive science are questions that only they can answer.

I: The Theory Theory

2. The Scientist as Child

• Science is cognitive almost by definition, insofar as cognition is about how minds arrive at veridical conceptions of the world.
• A cognitive view of science might provide a useful middle ground, even a bridge, between logical and sociological views of science.
• Quite typically, underlying any human cognitive activity is an abstract structure that is not at all apparent in superficial phenomenology or practice.
• From an evolutionary point of view, three of the most distinctive features of human beings are their long, protected immaturity, the plasticity of their behavior, and their ability to adapt to an extremely wide variety of environments. Equipping human children with particularly powerful and flexible cognitive devices might be an important part of this picture. Indeed, we might think of childhood as a period when many of the requirements for survival are suspended so that children can concentrate on acquiring a veridical picture of the particular physical and social world in which they find themselves.
• Once, as children, we have engaged in the theorizing necessary to specify the features of our world, most of us most of the time may simply go on to the central evolutionary business of feeding and reproducing. But, we suggest, these powerful theorizing abilities continue to allow all of us some of the time and some of us, namely professional scientists, much of the time to continue to discover more and more about the world around us.
• A characteristic lesson of the cognitive revolution is that human beings (or, for that matter, machines) can perform extremely complex feats of information processing without any phenomenology at all. It is rather characteristic of human cognition that it is largely inaccessible to conscious reflection.
• Children, no less than scientists, develop their understanding of the world in the context of a society that already has much knowledge of the world. they live in a rich social structure with much opportunity for contradiction, instruction, and the linguistic transmission of information.
• We might argue that much of the social structure of science is an attempt to replicate the privileged sociological conditions of infancy. Aside from the division of labor, the social hierarchy largely determines who will get the leisure and equipment to do cognitive work and who other scientists should listen to. The infant solves these problems without needing elaborate social arrangements.
• What happens when the theorizing mechanisms are faced with a causal problem but have no relevant evidence to operate on? What happens, we suggest, is magic, a combination of narrative, deference, and association.
• In the course of developing an account of the mind, children postulate such mental entities as perceptions, beliefs, and desires as a way of explaining ordinary human behavior.
• It is not that children are little scientists but that scientists are big children. Scientific progress is possible because scientists employ cognitive processes that are first seen in very young children.

Theories:

• Are always constructed with reference to evidence.
• Include entities and laws that are postulated or recruited from elsewhere to explain evidence.
• The entities postulated by a theory are closely, "lawfully", interrelated with one another.
• Appeal to some underlying causal structure that we think is responsible for the superficial regularities in the data.
• An accepted theory is supposed to carve nature at its joints; the theoretical entities and laws are supposed to tell you what there is and what it must do.
• A test of theoreticity is the nature of our surprise at violations of the theory. If we are committed to the theory, such violations should strike us not only as surprising but as being impossible and unbelievable in an important and strong way.
• A theory, in contrast to a mere empirical generalization, makes predictions about a wide variety of evidence, including evidence that played no role in the theory's initial construction
• The ability to produce wide-ranging predictions is perhaps the most obvious pragmatic benefit of science, and it may also be the most important evolutionary benefit of developing theorizing abilities. The evolutionary value of a system that leads to accurate and wide-ranging predictions should be obvious. In fact, making accurate predictions about the behavior of the world and your fellow organisms is the sine qua non of cognition.
• Theories produce interpretations of evidence, not simply descriptions and typologies of it. Indeed, theories strongly influence which pieces of evidence we consider salient or important. Theory-driven interpretations help to solve what computer scientists call "the frame problem". Theories provide a way of deciding which evidence is relevant to a particular problem
• What we mean by saying that we've explained something is simply that we can give an abstract, coherent, causal account of it. We might say that explanation is to cognition as orgasm is to reproduction.
• We might speculate that we were designed with a theorizing drive and that explanation is a symptom of that drive in action.
• Theories may turn out to be inconsistent with the evidence, and because of this theories change. In fact a tenet of modern epistemology is that any aspect of a theory, even the most central ones, may change.
• The initial reaction of a theory to counter-evidence may be a kind of denial. The interpretive mechanisms of the theory may treat the counter-evidence as noise, mess, not worth attending to. At a slightly later stage, the theory may develop ad hoc auxiliary hypotheses designed to account specifically for such counter-evidence. The theory gets ugly and messy instead of being beautiful and simple. The preference for beautiful theories over ugly ones plays an additional major role in theory change.
• A next step requires an alternative model to the original theory. The theory may limp along for some time under the weight of its auxiliary hypotheses if no alternative way of making progress is available. The ability to fix on this alternative is the mysterious logic of discovery.
• The new idea is not simply the olde idea applied to a new domain, but rather the earlier idea is itself modified to fit its role in the new theory. Moreover the fertility of the alternative idea may not be immediately recognized. Initially it may be applied only to the problematic cases.
• A final important dynamic feature of theory formation is a period of intense experimentation and/or observation. This period might span both the crisis of the earlier theory (when anomalous results are being produced right and left) and the first stages of the new theory (when a whole range of new predictions become available).
• In the development of the heliocentric theory of the planets:
  • Auxiliary hypotheses involving more and more complex arrangements of epicycles were initially invoked to deal with counterevidence
  • Later heliocentrism was introduced by Copernicus, but it continued to rely on epicycles as an auxiliary hypothesis.
  • Tycho Brahe's account acknowledges many of the flaws of the Ptolemaic accounts and uses the ides of heliocentrism to deal with them, but fails to accept the central idea that the earth itself goes around the sun.
  • Only with Kepler is there a really coherent heliocentric account that deals both with the anomalies and with the earlier data itself. And while, strictly speaking, experimentation on the heavenly bodies was impossible, periods of intense and detailed observation in this transitional period provided a far richer empirical base than had been available before.
• This distinctive pattern of prediction, interpretation, and explanation is among the best indicators of a theoretical structure and the best ways of distinguishing the theory theory from its developmental competitors.
• Infants show a kind of motivation and satisfaction that goes well beyond any immediate functional or social reward. Infants seem to have cognitive orgasms.
• A person's theory is a system that assigns representations to inputs just as one's perceptual system assigns representations to visual input or one's syntactic system assigns representations to phonological input.
• The representations are operated on by rules that lead to new representations; for example, the theory generates predictions. There are also distinctive functional relations between the theoretical representations and the input to them; theories predict, interpret, and explain data.
• Exactly what is the input to our theory systems?
• There is no atheoretical, evidential, "ordinary knowledge" level of representation, at least not once we get past very low level perceptual processing. All representation will be theory-laden. Even apparently "ordinary" kind of knowledge, like our knowledge that this is a jar and that is a table, or perhaps even that these are two objects and one is on top of the other, will be the result of the application of everyday theories.
• The most important and distinctive thing about theories is the fact that the very patterns of representation that occur can alter the nature of the representational system itself.
• Moreover, this system may be dynamic at yet another level. It is also possible that the very rules that restructure the relations between inputs and representations may change as a result of new input. That is, as we learn more, we also learn new ways to learn. Certainly this seems to be true in science, and it may also be true in development.
• To borrow Neurath's philosophical metaphor, the theorizing view sees knowledge as a boat that we perpetually rebuild as we sail in it. At each point in our journey there may be only a limited and constrained set of alterations we can make to the boat to keep it seaworthy. In the end, however, we may end up with not a single plan or rivet from the original structure, and the process may go on indefinitely.
• Presumably, creatures who constructed representations in different ways in childhood, who did not assume underlying causal structure, did not search for the simples explanation, did not falsify hypotheses when there was counter-evidence, and so on, were at an evolutionary disadvantage. In adulthood, such creatures were less good at predicting which berries were members of a poisonous natural kind, which kinds of minerals would make the most seductive body paint, or when their babies were old enough to be left alone without danger. In this sense nature itself guarantees that the system gets to an understanding of nature.

3. Theories, Modules, and Empirical Generalizations

• We want to argue for a kind of developmental pluralism: there are many quite different mechanisms underlying cognitive developments, and theory formation is an important one of them.

• According to modularity theories, representations of the world are not constructed from evidence in the course of development. Instead, representations are produced by innate structures, modules, or constraints that have been constructed in the course of evolution. These structures may need to be triggered, but one they are triggered, they create mandatory representations of input
• On the starting-state view, the child is innately endowed with a particular set of representations of input and rules operating on the representations. According to this view, such initial structures, while innate, would be defeasible; any part of them could be, and indeed will be, altered by new evidence. We propose that there are innate theories that are later modified and revised. The process of theory change and replacement begins at birth.
• Certainly, this type of account seems more tractable than one in which theory-like conceptual structures are constructed from scratch from a disorganized flow of experience.
• Modularity nativism, on the other hand, implies a much stronger set of claims. For Fodor, for example, modules are not only innate; they are also encapsulated and the representations that are the outcome of modules cannot be overturned by new patterns of evidence.
• At least some of the structural and functional features of theories - their abstractness, coherence, and predictive and interpretive force - can also be found in modules.
• According to the theory theory, input is evidence. It radically alters the nature of theoretical concepts.
• A theorizing mechanism is a representational system that reorganizes itself in response to input. It is an inherently developmental system. This leads to a distinctive set of developmental predictions. The theory theory predicts a succession of different theories, each replacing the earlier theory
• Modular representations do not lead to predictions through some set of inductive or deductive generalizations or through a process of theory testing, confirmation and disconfirmation. They lead to predictions because they are specifically designed by evolution to do so.

• Some theories in cognitive science have proposed that knowledge consists of empirical generalizations. "Scripts" are an example.
• Like theories, but unlike modules, empirical generalizations, scripts, narratives, and nets are defeasible.
• When we want deeper explanatory adequacy or wider predictive power, we turn from scripts to theories, even when we are talking about restaurants.
• Modules must help to provide the input to theorizing processes. On any view, there are encapsulated atheoretical processes that take us from patterns of stimulation at the retina to patterns of representation in the visual cortex.
• How far up, as it were, do these modular process go? Where is the border between modules and theories, between the periphery and the center?
• A lesson of the cognitive revolution is that phenomenology does not, in general, recapitulate epistemology. On this view, many representations that are theoretical, in the sense of being revisable and defeasible in theory-like ways, might have much of the direct, vivid phenomenology of perception.
• Empirical generalizations often seem to play a crucial role in theory change. Almost by definition, they are the kind of knowledge that provides us with counter-evidence to the predictions of the theory.
• On our view, modules, theories, and empirical generalizations are all in play literally from birth. Innate theories will assign powerful, abstract, and predictive representations to the outputs of modules. New information coming into the cognitive system will be represented in terms of these theories. Some relations among the inputs will not, however, be predicted by the theories; they will constitute empirical generalizations. Eventually, some of these kinds of inputs will lead to a revision of the theory, the modular outputs will be assigned new representations by the new theory, and so on.
• In the same way that technological advances enable or facilitate change in science, so information processing advances might enable or facilitate theory change in children.
• The most significant alternative to modularity accounts in syntactic development is not the theory theory but rather various functionalist theories that emphasize the role of social interaction in language acquisition. Assimilating all cognitive development to the model of socialization is, however, a dreadful mistake.

II: Evidence for the Theory Theory

• We suggest that there is a continuous line of development from the understandings of object appearance and disappearance involved in infant "object permanence" tasks, to the early understanding of visual perception, to the later understanding of belief and representation reflected in theory-of-mind tasks.
• A less well researched domain that is nevertheless equally important from the point of view of young children is the understanding of desires, plans, and intentions. Here we will suggest a line of development from an early understanding of the relations between actions and goals, to an early nonrepresentational understanding of desire and intention, to a later fully fledged understanding of desire as a representational state.
• Similarly, we will argue that there is a continuous line of development that bridges early object manipulation, early naming, and later "natural kind" understanding.
• We will suggest that these sets of problems - the nature of appearances, actions, and kinds - are at the core of the child's cognitive and semantic development in the period between late infancy and early childhood. On our view, these capacities are available from the very start of development. Developing a particular set of inductions and deductions, however, depends on having a particular body of knowledge organized in a theorylike way.

4. The Child's Theory of Appearances

• For adults, there are:
  • The laws governing the movement of objects
  • The properties of the objects themselves
  • The spatial relations between stationary objects
  • The laws that govern the perceptual relations between observers and objects.
• The theory of object appearances is an interesting bridge between "folk physics" and "folk psychology". It involves ideas about objects and the relations among them, but just as crucially, it involves ideas about the way that e and others perceive objects.
• The general principle of conjuring tricks is that they violate our deep ontological commitments about the nature of the world. When we are faced with these illusions, we use our higher-level theories to devise alternative explanations for the perceptual appearances. As we say, we can't believe our eyes. Much of the conjuror's art is devoted not just to producing the appearances but also to frustrating the explanations.
• The clown violates all our familiar scripts of behavior or events. But while he may evoke a kind of moral or social unease, he doesn't evoke the ontological unease that a conjuror does.
• If our knowledge was really modular in a strict sense, we should not be able to represent the event at all, as our perceptual system can't represent the reality in a perceptual illusion and children can't represent the syntax of a pidgin language. If the event violates a script or an empirical generalization, we might find it weird or surprising. However, only if it violates a theory, with all its ontological commitments, will we think it's magic.
• So our adult understanding of permanent objects seems to have some of the structural and functional characteristics of a theory. It involves a system of abstract causal entities and rules related to one another in coherent ways. It allows a wide variety of coherent predictions, and it leads to interpretations and explanations.
• When we eat thing, we do not assume that they will reappear at the location where they disappeared. Instead, we assume that they will be transformed or destroyed.
• Shadows violate our ordinary account of objects. Interestingly, infants initially seem to pay special attention to these violations. Eventually, of course, we learn that these violations are not really surprising. We modify our theory to account for them.
• Research suggests that infants are born assuming a world of 3D amodal objects. Their representation of the world are abstract and interrelated, and they license certain kinds of inferences and predictions. Even very young children seems to have a minimal notion of object permanence.
• The process of integrating features and movements and using features to individuate objects seems to develop gradually in the first year.
• Some studies show that 9-mont-olds can remember and act on events that took place as much as 24 hours earlier.
• As adults, we are not committed to the fact that objects will continue to move along the same trajectory.
• Many objects in the child's world have habitual locations. The rule is "The object will be where it appeared before."
• An 18-month-old seems to use its new theory of objects productively to make inferences and predictions in quite new situations.
• Even very young infants pay special attention to eyes. At about 18 months infants develop the ability to follow the point or gaze of another person toward an object and begin to use points to get someone else to look at an object. They also draw attention to objects in order to discover another person's attitude toward them. Faced with an ambiguous situation, infants look to their mother's face for information and may draw her gaze to the objects in question
• These behaviors imply a primitive but genuine understanding of visual perception. These infants seem to know that the direction of the eyes, or perhaps just of the head and body, indicates something about the person's visual contact with an object. There seem to be some connection in these children's minds between the visual display of another person's head and eyes and the phenomenal experience of vision.
• Infants who solve a task will often themselves hide the object in the complex way and indicate to the experimenter that she should search for it. It seem clear that the child is not primarily concerned with retrieving the object but with understanding the problem. We suggest that these activities are similar to experimentation in adult science.
• Between 15-18 months the infant develops a new theory of objects positing that all object disappearances are a consequence of movements, including invisible movements, either of objects or of the observer. This allows them to consolidate all the information about disappearances that has been accumulated up till then, including the earlier theory of visible movements and trajectories, the theory of invisible objects, new information about spatial relations and visual perception, and information captured by the various empirical generalization that have made in recent months.
• Infants very consistently use a word for disappearance, so as "gone" about when they first demonstrate this new object theory.
• By 18 months, infants seem to understand that they and others may share the perception of an object and that they can make objects invisible to others visible by pointing or showing.
• Perception and belief have many features in common, but perception may be easier to understand than belief. The understanding of perception may be the basic model on which a later understanding of belief is constructed.
• "See" typically emerged shortly after "gone" and its first uses applied to the child's own attempts to see objects. References to the sight of other people and to failures to see emerged only later.
• Belief words such as "know" and "think" appear during the fourth year of life, after "see" is well established. The developing theory of mind is apparent both in semantic change and in conceptual change.
• The onset of crawling may enormously increase opportunities for the observation of counterevidence. A baby who crawls will observe many more systematic appearances and disappearances that a baby who doesn't.

5. The Child's Theory of Actions

• Understanding action underlies the child's later development of a theory of mind:
  • It provides the basic evidence on which any theory of the mind must be based, since we infer mental states from actions
  • Many of the fundamental concepts of our theory of mind share important features with concepts of action
• Searle distinguishes between mental states with:
  • World-to-mind direction, like desire, where the mind alters the world to fit its representations.
    • The early understanding of actions, plans, and goals might be the child's first conception of these states.
    • These are among the earliest and most central concepts in the child's theory of mind from around age 2.
  • Mind-to-world direction, like belief, where the mind alters its representations to fit the world.
    • The early understanding of perception revealed in object permanence, social referencing, and visual-perspective taking is the child's first conception of these states.
• Our adult theory of action:
  • We believe that the world contains both psychological agents and inanimate physical objects.
  • Agents crucially include ourselves and other human beings, and we assume that we and others act in the same way.
  • Agents have internal mental states (desires, plan, intentions, and so on) that cause bodily movements which, in turn, cause things to occur in the world in particular and predictable ways.
  • The causal consequences of bodily movements will be different when we are interacting with other people and when we are acting on objects.
  • We influence people by communication and objects by contact.
  • We believe that the events in the world that result from our movements will "satisfy" our mental states - ie, the consequences of our actions will be the events we wanted to happen in the first place.
  • The practical syllogism: "If a psychological agent wants event y and believes that action x will cause event y, he will do x." This is believes to be the basic explanatory schema of folk psychology.
• Among the conjuror's most persuasive tricks are to appear to act on people without communication and to act on objects without contact.
• The events in a conjuror's show violate both our familiar scripts and any innately given predictions we might make. But, given our ordinary theory of action, these facts don't make the events utterly incomprehensible. On the contrary, they provide us with a stimulus to generate new predictions and explanations. The conjuror's technique is to provide counter-evidence to those predictions and to frustrate those explanations.
• In Star Trek, Q's powers call into question fundamental ideas about will, desire, morality, and time. The Borg calls into question ideas of personal identity, responsibility, and agency.
• From an extremely early age, infants distinguish between inanimate objects and people and respond in a distinctive way to human faces and voices.
• We claim that infants already know that they are like other people. We suggest that from birth, information about action that comes, literally, from inside ourselves is coded in the same way as information that comes from observing the behavior of others. There is a fundamental cross-modal representational system that connects self and other. This initial coding then becomes the basis for the child's further acquisition of a theory of action.
• In particular, we innately map the visually perceived motions of others onto our own kinesthetic sensations.
• To genuinely imitate another person sticking out his tongue, for instance, I have to recognize the equivalence between the visual spectacle of his face, an oval with a protruding moving cylinder in its lower portion, and the set of feelings that make up my own kinesthetic image of my invisible tongue. Yet imperfect imitations are observed in babies as young as 42 minutes old!
• Newborn imitation is nature's way of solving both the problem of other minds and the mind-body problem at one fell swoop.
• Very young infants gaze at the faces of significant (or even not so significant) others and vocalize and gesture in a way that seems "tuned" to the vocalizations and gestures of the other person.
• The mere physical appearance of certain facial expressions is itself enough to induce a parallel internal affective state.
• Infants can make inferences on the basis of their representations of others vocalizations and gestures. They can recognize which movements of their own faces and bodies will correspond to the observed movement of the other's face and body and then produce that movement.
• It is conceivable that young infants think that persons are rather like the Borg, fundamentally sharing the same psychological states. This rather Rousseauean view turns the problem of other minds on its head: we start out deeply identifying with our fellow human beings and need to develop an understanding of separation and conflict.
• Persons make objects do things.
• Infants begin by making overly general predictions about the connections between actions and events, but then begin to realize that only some subclasses of relations between actions and events are actually reliable over the long run.
• Around 9 months, children will being only to attempt solutions that obey the rules of spatial contact. This enables their problem solving to be both more constrained, since a wide array of actions are ruled out, and more general at the same time. They can see a new problem, like obtaining an object by pulling at the end of a horizontal string, and immediatley hit on one of several appropriate solutions.