Cognitive Science - Introduction PDF

CHAPTER ONE INTRODUCTION Exploring Mental Space A BRAVE NEW WORLD...

CHAPTER ONE INTRODUCTION Exploring Mental Space A BRAVE NEW WORLD Learning Objectives We are in the midst of a scientific revolution. For centuries, science has made great strides in our understanding of the external observ- After reading this chapter, able world. Physics revealed the motion of the planets, chemistry you will be able to: discovered the fundamental elements of matter, and biology has told us how to understand and treat disease. But during much of this 1. List at least five time, there were still many unanswered questions about something disciplines that participate in perhaps even more important to us—the human mind. the field of cognitive science. What makes mind so difficult to study is that, unlike the phe- 2. Describe what a mental nomena described above, it is not something we can easily observe, representation is. measure, or manipulate. In addition, the mind is the most complex 3. Describe what mental entity in the known universe. To give you a sense of this complexity, computation is. consider the following. The human brain is estimated to contain 10 billion to 100 billion individual nerve cells or neurons. Each of 4. Define what these neurons can have as many as 10,000 connections to other neu- interdisciplinary means. rons. This vast web of neural tissue is the core engine of the mind and helps generate a wide range of amazing and difficult-to-under- stand mental phenomena, such as perception, memory, language, emotion, and social interaction. The past several decades have seen the introduction of new technologies and methodologies for studying this intriguing organ, and its relationship to the body, and to the environment. We have learned more about the mind in the past half-century than in all the time that came before that. This period of rapid discovery has coin- cided with an increase in the number of different disciplines—many of them entirely new—that study mind. Since then, a coordinated effort among the practitioners of these disciplines has come to pass. This diversely interdisciplinary approach has since become known as cognitive science. Unlike the sciences that came before, which were focused solely on the world of physical events in physical space, this new endeavor now turns its full attention to discovering the fascinating mental events that take place in mental space. 1  WHAT IS COGNITIVE SCIENCE? Cognitive science can be roughly summed up as the scientific interdisciplinary study of the mind. Its primary methodology is the scientific method—although, as we will see, many other methodologies also contribute. A hallmark of cognitive science is its interdisciplinary approach. It results from the efforts of researchers working in a wide array of fields. These include philosophy, psychology, linguistics, artificial intelligence (AI), robotics, and neuroscience, among others. Each field brings with it a unique set of tools and perspectives. One major goal of this book is to show that when it comes to studying something as complex as the mind, no single perspective is adequate. Instead, intercommunication and cooperation among the practitioners of these disciplines will tell us much more. The term cognitive science refers not so much to the sum of all these disciplines but to their intersection or converging work on specific problems. In this sense, cognitive science is not a unified field of study like each of the disciplines themselves but, rather, a collaborative effort among researchers working in the various fields. The glue that holds cognitive science together is the topic of mind and, for the most part, the use of scientific methods. In the concluding chapter, we talk more about the issue of how unified cogni- tive science really is. To understand what cognitive science is all about, we need to know what its theo- retical perspective on the mind is. This perspective began with the idea of computation, which may alternatively be called information processing. Cognitive scientists started out viewing the mind as an information processor. Information processors must both represent information and transform information. That is, a mind, according to this perspective, must incorporate some form of mental representation and processes that act on and manipulate that information. We will discuss these two ideas in greater detail later in this chapter. Cognitive science is often described as having been heavily influenced by the devel- opment of the digital computer. Computers are, of course, information processors. Think for a minute about a personal computer. It performs a variety of information- processing tasks. Information gets into the computer via input devices, such as a keyboard or modem. That information can then be stored on the computer—for example, on its hard drive—in the form of binary representations (coded as 0s and 1s). The information can then be processed and manipulated using software, such as a text editor. The results of this processing may next serve as output, either to a monitor or to a printer. In like fashion, we may think of people performing similar tasks. Information is “input” into our minds through perception—what we see or hear. It is stored in our memories and processed in the form of thought. Our thoughts can then serve as the basis of “outputs,” such as language or physical behavior. Of course, this analogy between the human mind and computers is highly abstract and imperfect. The actual physical way in which data are stored on a computer bears little resemblance to human memory formation. But both systems are characterized by some form of computation (e.g., binary computation in the case of computers and analog computation in the case of brains). In fact, it is not going too far to say that most 2 Cognitive SCienCe cognitive scientists view the mind as a machine or mechanism whose workings they are trying to understand. REPRESENTATION As mentioned before, representation has been seen as fundamental to cognitive science. But what is a representation? Briefly stated, a representation is something that stands for something else. Before listing the characteristics of a representation, it is helpful to describe briefly four categories of representation. (1) A concept stands for a single entity or group of entities. Single words are good examples of concepts. The word apple refers to the concept that represents that particular type of fruit. (2) Propositions are statements about the world and can be illustrated with sentences. The sentence “Mary has black hair” is a proposition that is itself made up of a few concepts. (3) Rules are yet another form of representation that can specify the relationships between propositions. For example, the rule “If it is raining, I will bring my umbrella” makes the second proposition contingent on the first. (4) An analogy helps us make comparisons between two similar situations. We will discuss all four of these representations in greater detail in the “Interdisciplinary Crossroads” section at the end of this chapter. There are four crucial aspects of any representation (Hartshorne, Weiss, & Burks, 1931–1958). First, a “representation bearer” such as a human or a computer must realize a representation. Second, a representation must have content—meaning it stands for one or more objects. The thing or things in the external world that a representation stands for are called referents. Third, a representation must also be “grounded.” That is, there must be some way in which the representation and its referent come to be related. Fourth, a representation must be interpretable by some interpreter, either the representation bearer or somebody else. These and other characteristics of representations are discussed next. The fact that a representation stands for something else means it is symbolic. We are all familiar with symbols. We know, for instance, that the dollar symbol ($) is used to stand for money. The symbol itself is not the money but, instead, is a surrogate that refers to its referent, which is actual money. In the case of mental representation, we say that there is some symbolic entity “in the mind” that stands for real money. Figure 1.1 shows a visual representation of money. Mental representations can stand for many different types of things and are by no means limited to simple conceptual ideas such as “money.” Research suggests that there are more complex mental representations that can stand for rules—for example, knowing how to drive a car and analogies, which may enable us to solve certain problems or notice similarities (Thagard, 2000). Human mental representations, especially linguistic ones, are said to be semantic, which is to say that they have meaning. Exactly what constitutes meaning and how a repre- sentation can come to be meaningful are topics of debate. According to one view, a repre- sentation’s meaning is derived from the relationship between the representation and what it is about. The term that describes this relation is intentionality. Intentionality means “directed on an object.” Mental states and events are intentional. They refer to some actual thing or things in the world. If you think about your brother, then the thought of your brother is directed toward him—not toward your sister, a cloud, or some other object. Chapter one introduCtion 3  Figure 1.1 Different aspects of the symbolic representation of money. The Mind Representation (Symbolic) $ Intentionality Referent (Nonsymbolic) The World Source: PhotoObjects.net/Thinkstock. An important characteristic of intentionality has to do with the relationship between inputs and outputs to the world. An intentional representation must be triggered by its referent or things related to it. Consequently, activation of a representation (i.e., thinking about it) should cause behaviors or actions that are somehow related to the referent. For example, if your friend Sally told you about a cruise she took around the Caribbean last December, an image of a cruise ship would probably pop into mind. This might then cause you to ask her if the food onboard was good. Sally’s mention of the cruise was the stimulus input that activated the internal representation of the ship in your mind. Once it was activated, it caused the behavior of asking about the food. This relation between inputs and outputs is known as an appropriate causal relation. Symbols can be assembled into what are called physical symbol systems, or more simply as formal logical systems. In a formal logical system, symbols are combined into expressions. These expressions can then be manipulated using processes. The result of a process can be a new expression. For example, in formal logic, symbols can be words like animals or mammals, and expressions could be statements like “animals that nurse their young are mammals.” The processes would be the rules of deduction that allow us to derive true concluding expressions from known expressions. In this instance, we could start off with the two known expressions “animals that nurse their young are mammals,” and “whales nurse their young,” and generate the new concluding expression: “whales are mammals.” (There are, in fact, a few nonmammals that nurse their young, but that is 4 Cognitive SCienCe for an advanced biology textbook.) More on this below, where we discuss propositions and syllogisms. According to the physical symbol system hypothesis (PSSH), a formal logical system can allow for intelligence (Newell & Simon, 1976). Since we as humans appear to have representational and computational capacity, being able to use things that stand for things, we seem to be intelligent. Beyond this, we could also infer that machines are intelligent, since they too have this capacity, although of course, this is debated. Several critiques have been leveled against the PSSH (Nilsson, 2007). For example, it is argued that the symbols computers use have no meaning or semantic quality. To be meaningful, symbols have to be connected to the environment in some way. People and perhaps other animals seem to have meaning because we have bodies and can perceive things and act on them. This “grounds” the symbols and imbues them with semantic qual- ity. Computing machines that are not embodied with sensors (e.g., cameras, microphones) and effectors (e.g., limbs) cannot acquire meaning. This issue is known as the symbol grounding problem and is in effect a reexpression of the concept of intentionality. A counterargument to this is that computer systems do have the capability to designate. An expression can designate an object if it can affect the object itself or behave in ways that depend on the object. One could argue that a robot capable of per- ceiving an object like a coffee mug and able to pick it up could develop semantics toward it in the same way that a person might. Ergo, the robot could be intelligent. Also, there are examples of AI programs like expert systems that have no sensor or effector capability yet are able to produce intelligent and useful results. Some expert systems, like MYCIN, are able to more accurately diagnose certain medical disorders than are members of the Stanford University medical school (Cawsey, 1998). They can do this despite not being able to see or act in the world. Types of Representation The history of research in cognition suggests that there are numerous forms of mental representation. Paul Thagard (2000), in Mind: Introduction to Cognitive Science, proposes four: concepts, propositions, rules, and analogies. Although some of these have already been alluded to and are described elsewhere in the book, they are so central to many ideas in cognitive science that it is, therefore, useful to sketch out some of their major characteristics again here. A concept is perhaps the most basic form of mental representation. A concept is an idea that represents things we have grouped together. The concept “chair” does not refer to a specific chair, such as the one you are sitting in now, but it is more general than that. It refers to all possible chairs no matter what their colors, sizes, and shapes. Concepts need not refer to concrete items. They can stand for abstract ideas—for example, “justice” or “love.” Concepts can be related to one another in complex ways. They can be related in a hierarchical fashion, where a concept at one level of organization stands for all members of the class just below it. “Golden retrievers” belongs to the category of “dogs,” which in turn belongs to the category of “animals.” We discuss a hierarchical model of concept representation in the network approach chapter. The question of whether concepts are innate or learned is discussed in the philosophical approach chapter. Chapter one introduCtion 5  A proposition is a statement or assertion typically posed in the form of a simple sentence. An essential feature of a proposition is that it can be proved true or false. For instance, the statement “The moon is made out of cheese” is grammatically correct and may represent a belief that some people hold, but it is a false statement. We can apply the rules of formal logic to propositions to determine the validity of those propositions. One logical inference is called a syllogism. A syllogism consists of a series of propositions. The first two (or more) are premises, and the last is a conclusion. Take the following syllogism: Premise 1: All men like football. Premise 2: Charlie is a man. Conclusion: Charlie likes football. Obviously, the conclusion can be wrong if either of the two premises is not fully and completely true. If 99% of men like football, then it might not be true that Charlie likes football, even if he is a man. And if Charlie is not a man, then Charlie may or may not like football, even assuming all men like it. Logical conclusions are only as reliable as the premises on which they are based. In the artificial intelligence approach chapter, we will discuss how probabilities (like 99%) can be used in probabilistic reasoning (and even fuzzy logic) to work with premises that are not fully and completely true. You may have noticed that propositions are representations that incorporate con- cepts. The proposition “All men like football” incorporates the concepts “men” and “football.” Propositions are more sophisticated representations than concepts because they express relationships—sometimes very complex ones—between concepts. The rules of logic are best thought of as computational processes that can be applied to prop- ositions to determine their validity. However, logical relations between propositions may themselves be considered a separate type of representation. The evolutionary approach chapter provides an interesting account of why logical reasoning, which is difficult for many people, can be made easier under certain circumstances. Formal logic is at the core of a type of computing system that produces effects in the real world: production systems. Inside a production system, a production rule is a con- ditional statement of the following form: “If x, then y,” where x and y are propositions. In formal logic, the “if” part of the rule is called the antecedent, and the “then” part is called the consequent. In a production system, the “if” part of the rule is called the condition, and the “then” part is called the action. If the proposition that is contained in the condi- tion (x) is verified as true, then the action that is specified by the second proposition (y) should be carried out, according to the rule. The following rules help us drive our cars: If the light is red, then step on the brakes. If the light is green, then step on the accelerator. In fact, it is production rules like these that are used in the computational algorithms that run self-driving cars. Notice that, in the first rule, the two propositions are “the light is 6 Cognitive SCienCe red” and “step on the brakes.” We can also form more complex rules by linking proposi- tions with “and” and “or” statements: If the light is red or the light is yellow, then step on the brakes. If the light is green and nobody is in the crosswalk, then step on the accelerator. The or that links the two propositions in the first part of the rule specifies that when either proposition is true, the action should be carried out. By contrast, when an and links these two propositions, then the rule specifies that both must be true before the action can occur. Rules bring up the question of what knowledge really is. We usually think of knowl- edge as factual. Indeed, a proposition such as “Candy is sweet,” if validated, does pro- vide factual information. The proposition is then an example of declarative knowledge. Declarative knowledge is used to represent facts. It tells us what is and is demonstrated by verbal communication. Procedural knowledge, by comparison, refers to skills. It tells us how to do something and is demonstrated by action. If we say that World War II was fought during the period 1939 to 1945, we have demonstrated a fact learned in history class. If we ski down a snowy mountain slope in the winter, we have demon- strated that we possess a specific skill. It is, therefore, very important that information- processing systems have some way of representing actions if they are to help an organism or machine perform those actions. Rules are just one way of representing procedural knowledge. In the cognitive approach chapters, we discuss two cognitive, rule-based sys- tems: the atomic components of thought and SOAR (state, operator, and result) models. In the ecological embodied approach chapter, we discuss more sensorimotor ways of representing procedural knowledge (and even declarative knowledge). Another specific type of mental representation is the analogy—although, as is pointed out below, the analogy can also be classified as a form of reasoning. Thinking analogically involves applying one’s familiarity with an old situation to a new situation. Suppose you had never ridden on a train before but had taken buses numerous times. You could use your understanding of bus riding to help you figure out how to take a ride on a train. Applying knowledge that you already possess and that is relevant to both scenarios would enable you to accomplish this. Based on prior experience, you would already know that you have to first determine the schedule, perhaps decide between express and local service, purchase a ticket, wait in line, board, stow your luggage, find a seat, and so on. Analogies are a useful form of representation because they allow us to generalize our learning. Not every situation in life is entirely new. We can apply what we already have learned to similar situations without having to figure out everything all over again. Several models of analogical reasoning have been proposed (Forbus, Gentner, & Law, 1995; Holyoak & Thagard, 1995). COMPUTATION As mentioned earlier, representations are only the first key component of the traditional cognitive science view of mental processes. Representations by themselves are of little use Chapter one introduCtion 7  unless something can be done with them. Having the concept of money doesn’t do much for us unless we know how to calculate a tip or can give back the correct amount of change to someone. In this cognitive science view, the mind performs computations on represen- tations. It is, therefore, important to understand how these mental mechanisms operate. What sorts of mental operations does the mind perform? If we wanted to get details about it, the list would be endless. Take the example of mathematical ability. If there were a separate mental operation for each step in a mathematical process, we could say the mind adds, subtracts, divides, and so on. Likewise, with language, we could say that there are separate mental operations for making a noun plural, putting a verb into past tense, and so on. It is better, then, to think of mental operations as falling into broad categories. These categories can be defined by the type of operation that is performed or by the type of information acted on. An incomplete list of these operations would include sensation, perception, attention, memory, language, mathematical reasoning, logical rea- soning, decision making, and problem solving. Many of these categories may incorporate virtually identical or similar subprocesses—for example, scanning, matching, sorting, and retrieving. Figure 1.2 shows the kinds of mental processes that may be involved in solving a simple addition problem. The Tri-Level Hypothesis Any given information process can be described at several different levels. According to the tri-level hypothesis, biological or artificial information-processing events can be evaluated on at least three different levels (Marr, 1982). The highest or most abstract level of analysis is the computational level. At this level, one is concerned with two tasks. The first is a clear specification of what the problem is. Taking the problem as it may originally have been posed, in a vague manner perhaps, and breaking it down into its main constitu- ents or parts can bring about this clarity. It means describing the problem in a precise way such that the problem can be investigated using formal methods. It is like asking, What exactly is this problem? What does this problem entail? The second task one encounters at the computational level concerns the purpose or reason for the process. The second task consists of asking, Why is this process here in the first place? Inherent in this analysis is adaptiveness—the idea that biological mental processes are learned or have evolved to enable the organism to solve a problem it faces. This is the primary explanatory perspec- tive used in the evolutionary approach. We describe a number of cognitive processes and the putative reasons for their evolution in the evolution chapter. Stepping down one level of abstraction, we can next inquire about the way in which an information process is carried out. To do this, we need an algorithm, a formal pro- cedure or system that acts on informational representations. It is important to note that algorithms can be carried out regardless of a representation’s meaning; algorithms act on the form, not the meaning, of the symbols they transform. One way to think of algorithms is that they are “actions” used to manipulate and change representations. Algorithms are formal, meaning they are well defined. We know exactly what occurs at each step of an algorithm and how a particular step changes the information being acted on. A mathematical formula is a good example of an algorithm. A formula specifies how 8 Cognitive SCienCe  Figure 1.2 Some of the computational steps involved in solving an addition problem. 36 Computational Steps 1. 6 + 7 = 13 Add right column + 47 2. 3 3. 1 4. 3 + 4 = 7 Store three Carry one Add left column 83 5. 7 + 1 = 8 6. 8 Add one Store eight 7. 83 Record result the data are to be transformed, what the steps are, and what the order of steps is. This type of description is put together at the algorithmic level, sometimes also called the programming level. It is equivalent to asking, What information-processing steps are being used to solve the problem? To draw an analogy with computers, the algorithmic level is like software because software contains instructions for the processing of data. The most specific and concrete type of description is formulated at the implemen- tational level. Here we ask, What is the information processor made of? What types of physical or material changes underlie changes in the processing of the information? This level is sometimes referred to as the hardware level, since in computer parlance, the hardware is the physical “stuff” the computer is made of. This would include its various parts—a hard drive, a screen, keyboard, and so on. At a smaller scale, computer hardware consists of multiple circuit boards and even the flow of electrons through the circuits. The hardware in biological cognition is the brain and body and, on a smaller scale, the neurons and activities of those neurons. At this point, one might wonder, Why do we even need an algorithmic or formal level of analysis? Why not just map the physical processes at the implementational level onto a computational description of the problem or, alternatively, onto the behaviors or actions of the organism or device? This seems simpler, and we need not resort to the idea of information and representation. The reason is that the algorithmic level tells us how a particular system performs a computation. Not all computational systems solve a problem in the same way. Both computers and humans can perform addition but do so in drastically different fashions. This is true at the implementational level, obviously, but understanding the difference formally tells us much about alternative problem-solving approaches. It also gives us insights into how these systems might compute solutions to other novel problems that we might not understand. Chapter one introduCtion 9  This partitioning of the analysis of information-processing events into three levels has been criticized as being fundamentally simplistic since the levels clearly interact with one another, and each level can, in turn, be further subdivided into its own sublevels (Churchland, Koch, & Sejnowski, 1990). Figure 1.3 depicts one possible organization of the many structural levels of analysis in the nervous system. Starting at the top, we might consider the brain as one organizational unit, brain regions as corresponding to another organizational unit one step down in spatial scale, then neural networks, individual neu- rons, and so on. Similarly, we could divide algorithmic steps into different substeps and problems into subproblems. To compound all this, it is not entirely clear how to map one level of analysis onto another. We may be able to specify clearly how an algorithm executes but be at a loss to say exactly where or how this is achieved with respect to the nervous system. David Marr’s separation of the computational, algorithmic, and imple- mentational levels can be a useful tool for studying cognition. But, like any tool, there may be parts of the task where it doesn’t apply. (You don’t want to find yourself using a hammer for a job that requires a screwdriver, or a laser scalpel.) Differing Views of Representation and Computation Before finishing our discussion of computation, it is important to differentiate between several different conceptions of what it is. So far, we have mostly been talking about computation as being based on the formal systems notion. In this view, a computer is a formal symbol manipulator. Let’s break this definition down into its component parts. A system is formal if it is syntactic or rule governed. In general, we use the word syntax to refer to the set of rules that govern any symbol system. The rules of language and mathematics are formal systems because they specify which types of allowable changes can be made to symbols. Formal systems also operate on representations independent of the content of those representations. In other words, a process can be applied to a symbol regardless of its meaning or semantic content. A symbol, as we have already indicated, is a type of representation and can assume a wide variety of forms and can undergo a wide variety of manipulations. Manipulation here implies that computation is an active, phys- ical process that takes place over time. That is, manipulations are actions, and they occur physically on some type of computing medium or substrate (e.g., neurons or circuits). And they take some time to occur (i.e., they don’t happen instantaneously). But this is not the only conception of what computation is. The connectionist or network approach to computation differs from the classical formal systems approach of cognitive science in several ways. In the classical view, knowledge is represented locally— in the form of symbols. In the connectionist view, knowledge is represented as a pattern of neural activation, or a pattern of synaptic strengths, that is distributed throughout a network and so is more global than a single symbol. Processing style is also different in this approach. The classical view has processing occurring in discrete sequential stages, whereas in connectionism, processing occurs in parallel through the simultaneous acti- vation of many nodes or elements in the network. However, some cognitive scientists downplay these differences, arguing that information processing occurs in both systems 10 Cognitive SCienCe Figure 1.3 Structural levels of analysis in the nervous system. Brain Brain regions Neural networks Neurons Synapses Molecules Chapter one introduCtion 11  and that the tri-level hypothesis can be applied equally to both (Dawson, 1998). We further compare and contrast the classical and connectionist views at the beginning of the network approach chapter. What happens to representations once they get established? In some versions of the symbolic and connectionist approaches, representations remain forever fixed and unchanging. For example, when you learned about the concept of a car, a particular symbol standing for “car” was formed. The physical form of this symbol could then be matched against perceptual inputs to allow you to recognize a car when you see one or used in thought processes to enable you to reason about cars when you need to. Such processes can be simplified if the symbol stays the same over time. This, in fact, is exactly how a computer deals with representation. Each of the numbers or letters of the alphabet has a unique and unchanging ASCII (American Standard Code for Information Inter- change) code. Similarly, in artificial neural network simulations of mind, the networks are flexible and plastic during learning, but once learning is complete, they must remain the same. If not, the representations will be rewritten when new things are learned. A third view of representation comes from the dynamical perspective in cognitive science (Casasanto & Lupyan, 2015; Friedenberg, 2009; Spivey, 2007). According to this view, the mind is constantly changing as it adapts to new information. A representation formed when we first learn a concept is altered each time we think about that concept or experience information that is in some way related to it. For example, let’s say a child sees a car for the first time and it is red. The child may then think that all cars are red. The next time he or she sees a car, it is blue. The child’s concept would then be changed to accommodate the new color. Even after we are very familiar with a concept, the context in which we experience it is always different. You may hear a friend discussing how fast one type of car is, or you may find yourself driving a car you have never driven before. In each of these cases, the internal representation is modified to account for the new experience. So it is unlikely that biological representations of the sort we see in humans will ever stay the same. We talk more about the dynamical systems approach and how it can unify differing perspectives on representation and computation in the ecological embodied chapter at the end of this book. THE INTERDISCIPLINARY PERSPECTIVE There is an old fable about five blind men who stumble on an elephant (see Figure 1.4). Not knowing what it is, they start to feel the animal. One man feels only the elephant’s tusk and thinks he is feeling a giant carrot. A second man feels the ears and believes that the object is a big fan. The third feels the trunk and proclaims that it is a pestle, while a fourth touching only the leg believes that it is a mortar. The fifth man, touching the tail, has yet another opinion: He believes it to be a rope. Obviously, all five men are wrong in their conclusions because each has examined only one aspect of the elephant. If the five men had gotten together and shared their findings, they may easily have pieced together what kind of creature it was. This story serves as a nice metaphor for cognitive science. We can think of the elephant as the mind and the blind men as researchers in different disciplines in cognitive science. Each individual discipline may make great strides in understanding its particular subject matter but, if it cannot compare its results to those 12 Cognitive SCienCe  Figure 1.4 If you were the blind man, would you know it is an elephant? of other related disciplines, may miss out on understanding the real nature of what is being investigated. The key, then, to figuring out something as mysterious and complex as mind is communication and cooperation among disciplines. This is what is meant when one talks about cognitive science—not the sum of each of the disciplines or approaches but, rather, their interaction. The best advances in our understanding have come from this kind of interdisciplinary cooperation. A number of major universities have established interdis- ciplinary cognitive science centers, where researchers in such diverse areas as philosophy, linguistics, neuroscience, computer science, and cognitive psychology are encouraged to work together on common problems. Each area can then contribute its unique strength to the phenomena under study. The consequent exchange of results and ideas then leads to fruitful synergies between these disciplines, accelerating progress with respect to find- ing solutions to the problem and yielding insights into other research questions. We have alluded to some of the different approaches in cognitive science. Some of these approaches are longstanding disciplines in and of themselves, such as philosophy and psychology. But some of these approaches are perhaps better described as subdis- ciplines or research areas, such as the emotional approach to cognitive science and the embodied ecological approach to cognitive science. Because this book is about explain- ing each approach and its major theoretical contributions, it is worth mentioning each now in terms of its perspective, history, and methodology. In the following sections, we will also provide a brief preview of the issues addressed by each approach. In the rest of the book, we have devoted a chapter to each approach. Chapter one introduCtion 13  The Philosophical Approach Philosophy is the oldest of all the disciplines in cognitive science. It traces its roots back to the ancient Greeks. Philosophers have been active throughout much of recorded human history, attempting to formulate and answer basic questions about the universe. This approach is free to study virtually any sort of important question on virtually any subject, ranging from the nature of existence to the acquisition of knowledge to politics, ethics, and beauty. Philosophers of mind narrow their focus to specific problems con- cerning the nature and characteristics of mind. They might ask questions such as, What is mind? How do we come to know things? How is mental knowledge organized? The primary method of philosophical inquiry is reasoning, both deductive and inductive. Deductive reasoning involves the application of the rules of logic to state- ments about the world. Given an initial set of statements assumed to be true, philos- ophers can derive other statements that logically must be correct. For example, if the statement “College students study 3 hours every night” is true and the statement “Mary is a college student” is true, we can conclude that “Mary will study 3 hours every night.” Philosophers also engage in inductive reasoning. They make observations about spe- cific instances in the world, notice commonalities among them, and draw general con- clusions. An example of inductive reasoning would be the following: “Whiskers the cat has four legs,” “Scruffy the cat has four legs,” and, therefore, “All cats have four legs.” However, philosophers do not tend to use a systematic form of induction known as the scientific method. That is employed within the other cognitive science disciplines. In Chapter 2, we summarize several of the fundamental issues facing philosophers of mind. With respect to the mind–body problem, philosophers wrangle over what exactly a mind is. Is the mind something physical like a rock or a chair, or is it nonphys- ical? Can minds exist only in brains, or can they emerge from the operation of other complex entities, such as computers? The knowledge acquisition problem deals with how we come to know things. Is knowledge a product of one’s genetic endowment, or does it arise through one’s interaction with the environment? How much does each of these factors contribute to any given mental ability? We also look into one of the most fascinating and enigmatic mysteries of mind—that of consciousness. In this case, we can ask, What is consciousness? Are we really conscious at all? INTERDISCIPLINARY CROSSROADS: SCIENCE AND PHILOSOPHY Philosophers have often been accused of reflect about the world. The critique is that they doing “armchair” philosophy, meaning they do ignore empirical evidence, information based nothing but sit back in comfortable chairs and on observation of the world, usually under 14 Cognitive SCienCe  controlled circumstances as is the case in the Logically, there should be no difference in sciences. This may be changing, however, with perceived intentionality between both condi- the development of a new trend known as tions. In each case, there was a desire to make experimental philosophy. profit and a failure to care one way or the other Experimental philosophy uses empirical about the environment. But clearly, the way methods, typically in the form of surveys that we see intentionality involves moral evalua- assess people’s understanding of constructed tions, the way people judge what is good or bad. scenarios, to help answer philosophical ques- This conclusion, what is now called the “Knobe tions. Followers of the movement call themselves effect,” could not necessarily have been deter- “X-Philes” and have even adopted a burning arm- mined a priori using pure reasoning. The term chair and accompanying song as their motto, a a priori relates to what can be known through an video of which is viewable on YouTube. understanding of how certain things work rather One of the more prominent X-Philers, than by observation. Joshua Knobe (2003), presented participants Experimental philosophy has addressed a with a hypothetical situation. In the situation, wide variety of issues. These include cultural dif- a vice president of a company approaches the ferences in how people know or merely believe chairman of the board and asks about starting (Weinberg, Nichols, & Stich, 2001), whether a a new program. The program will increase prof- person can be morally responsible if his or her its, but it will also harm the environment. In one actions are determined (Nichols & Knobe, 2007), version of the survey, the chairman responds the ways in which people understand conscious- that he or she doesn’t care at all about harm- ness (Huebner, 2010), and even how philosoph- ing the environment but just wants to make as ical intuitions are related to personality traits much profit as possible and so wants to begin (Feltz & Cokely, 2009). the new program. Of course, this new field has been critiqued. When participants were asked whether the Much of experimental philosophy has merely chairman harmed the environment intention- used questionnaires that tap into the intuitions ally, 82% responded “yes.” In a different version of experimental participants (much like social of the scenario, everything was identical except psychology already does). Williamson (2008) that “harming” was now replaced with “help- argues that philosophical evidence should not ing.” The program had the side effect of helping just rely on people’s intuitions. Another obvious the environment but the chairman again stated issue is that experimental philosophy is not phi- that he or she didn’t care at all about help- losophy at all, but science. In other words, if it is ing the environment and wanted to go ahead adopting the use of survey methodology, statis- with it. This time when asked if the chairman tics, and so on, then it really becomes psychol- intentionally helped the environment, only 23% ogy or a branch of the social sciences rather than responded “yes.” philosophy. The Psychological Approach Compared with philosophy, psychology is a relatively young discipline. It can be consid- ered old, though, particularly when compared with some of the more recent newcomers to the cognitive science scene—for example, AI and robotics. Psychology as a science Chapter one introduCtion 15  arose in the late 19th century and was the first discipline in which the scientific method was applied exclusively to the study of mental phenomena. Early psychologists estab- lished experimental laboratories that would enable them to catalog mental ideas and investigate various mental capacities, such as vision and memory. Psychologists apply the scientific method to both mind and behavior. That is, they attempt to understand not just internal mental phenomena, such as thoughts, but also the external behaviors that these internal phenomena can give rise to. The scientific method is a way of getting hold of valid knowledge about the world. One starts with a hypothesis or idea about how the world works and then designs an experiment to see if the hypothesis has validity. In an experiment, one essentially makes observations under a set of controlled conditions. The resulting data then either support or fail to support the hypothesis. This procedure, employed within psychology, and cog- nitive science in general, is described more fully at the start of Chapter 3. The field of psychology is broad and encompasses many subdisciplines, each one having its unique theoretical orientations. Each discipline has a different take on what mind is. The earliest psychologists—that is, the voluntarists and structuralists—viewed the mind as a kind of test tube in which chemical reactions between mental elements took place. In contrast, functionalism viewed mind not according to its constituent parts but, rather, according to what its operations were—what it could do. The Gestaltists again went back to a vision of mind as composed of parts but emphasized that it was the combination and interaction of the parts, which gave rise to new wholes, that was important. Psychoanalytic psychology conceives of mind as a collection of differing and competing mind-like subsystems, while behaviorism sees it as something that maps stim- uli onto responses. The Cognitive Approach Starting in the 1960s, a new form of psychology arrived on the scene. Known as cognitive psychology, it came into being, in part, as a backlash against the behaviorist movement and its profound emphasis on behavior (and its neglect of mental computations). Cogni- tive psychologists placed renewed emphasis on the study of internal mental operations. They adopted the computer as a metaphor for mind and described mental functioning in terms of representation and computation. They believed that the mind, like a computer, could be understood in terms of information processing. The cognitive approach was also better able to explain phenomena such as lan- guage acquisition for which behaviorists did not have good accounts. At around the same time, new technologies that allowed better measurement of mental activity were being developed. This promoted a movement away from the behaviorist’s emphasis solely on external observable responses and toward the cognitive scientist’s emphasis on internal functions, as these could, for the first time, be observed with reasonable precision. Inherent early on in the cognitive approach was the idea of modularity. Modules are functionally independent units that receive inputs from other modules, perform a specific processing task, and pass the results of their computation onto other modules. The influence of the modular approach can be seen in the use of process models or flow 16 Cognitive SCienCe diagrams. These depict a given mental activity via the use of boxes and arrows, where boxes depict modules and arrows depict the flow of information among them. The tech- niques used in this approach are the experimental method and computational model- ing. Computational modeling involves carrying out a formal (typically software-based) implementation of a proposed cognitive process. Researchers can run the modeling process so as to simulate how the process might operate in a human mind. They can then alter various parameters of the model or change its structure in an effort to achieve results as close as possible to those obtained in human experiments. The model can also produce new results that then inspire new experiments. This use of modeling and com- parison with experimental data is a key strength in cognitive psychology and is also used in the AI and artificial network approaches. Cognitive psychologists have studied a wide variety of mental processes, includ- ing pattern recognition, attention, memory, imagery, and problem solving. Theoret- ical accounts and processing models for each of these are given in Chapters 4 and 5. Language is also within the purview of cognitive psychology, but because the approach to language is so multidisciplinary, we describe it separately in Chapter 9. The Neuroscience Approach Brain anatomy and physiology have been studied for centuries. It is only recently, how- ever, that we have seen tremendous advances in our understanding of the brain, espe- cially in terms of how neuronal processes can account for cognitive phenomena. The study of the brain and endocrine system and how these account for mental states and behavior is called neuroscience. The attempt to explain cognitive processes in terms of underlying brain mechanisms is known as cognitive neuroscience. Neuroscience, first and foremost, provides a description of mental events at the implementational level. It attempts to describe the biological “hardware” on which men- tal “software” supposedly runs. However, as discussed above, there are many levels of scale when it comes to describing the brain, and it is not always clear which level provides the best explanation for any given cognitive process. Neuroscientists, however, investi- gate at each of these levels. They study the chemistry of neurotransmitters, the cell biol- ogy of individual neurons, the process of neuron-to-neuron synaptic transmission, the patterns of activity in local cell populations, and the interrelations of larger brain areas. A reason for many of the recent developments in neuroscience is, again, the devel- opment of new technologies. Neuroscientists employ a wide variety of methods to measure the performance of the brain at work. These include electroencephalography (EEG) that detects changes in the electric fields generated by large groups of highly active neurons, magnetoencephalography (MEG) that detects changes in the magnetic fields generated by groups of highly active neurons, and functional magnetic resonance imaging (fMRI) that detects changes in the magnetic properties of oxygenated blood being delivered to small groups of highly active neurons, as well as many others. Stud- ies that use these procedures have participants perform a cognitive task, and the brain activity that is concurrent with the performance of the task is recorded. For example, a participant may be asked to form a mental image of a particular object with their Chapter one introduCtion 17  eyes closed. The researchers can then determine which parts of the brain become active during visual imagery and in what order. Neuroscientists use other techniques as well. They study brain-damaged patients and the effects of lesions in laboratory animals, and they use single- and multiple-cell recording techniques. The Network Approach The network approach is at least partially derived from neuroscience. In this perspective, mind is seen as a collection of computing units. These units are connected to one another and mutually influence one another’s activity via their connections, although each of the units is believed to perform a relatively simple computation—for example, a neuron can either “fire” by initiating an action potential that sends an electrochemical signal to other neurons or not “fire” and fail to initiate an action potential. In these networks, the connec- tivity among many units can give rise to representational and computational complexity. Chapter 7, which outlines the network approach, has two parts. The first involves the construction of artificial neural networks. Most artificial neural networks are com- puter software simulations that have been designed to mimic the way actual brain net- works operate. They attempt to simulate the functioning of neural cell populations. Artificial neural networks that can perform arithmetic, recognize faces, learn concepts, play competitive backgammon, and read out loud now exist. A wide variety of network architectures has developed over the past 30 years. The second part of the network chapter is more theoretical and focuses on knowl- edge representation—on how meaningful information may be mentally encoded and processed. In semantic networks, nodes standing for concepts are connected to one another in such a way that activation of one node causes activation of other related nodes. Semantic networks have been constructed to explain how conceptual information in memory is organized and recalled. They are often used to predict and explain data obtained from experiments with human participants in cognitive psychology. In the chapter on networks, we will finish off with a discussion of network science. This is a new interdisciplinary field, much like cognitive science itself. However, in this field, researchers focus on the structure and function of networks. The term network is meant in a very broad sense here to include not just artificial or natural neural networks but also telephone and wireless networks, electrical power networks, ecosystem net- works, and human social networks. Surprisingly, we will see that there are commonalities among these different types of networks and that they share some organizational and operational features. We will examine these features and apply them particularly to the brain and cognition. The Evolutionary Approach The theory of natural selection proposed by Charles Darwin in 1859 revolutionized our way of thinking about biology. Natural selection holds that adaptive features enable the animals that possess them to survive and pass these features on to future generations. The environment, in this view, is seen as selecting from among a variety of traits those that serve a functional purpose. 18 Cognitive SCienCe  The field of evolutionary psychology applies the theory of natural selection to account for human mental processes. It attempts to elucidate the selection forces that acted on our ancestors and how those forces gave rise to the cognitive structures we now possess. Evolutionary psychologists tend to adopt a modular approach to mind. In this case, the proposed modules would be individual mechanisms in the brain that carry out various distinct cognitive abilities that were successful at solving certain problems, thus helping our ancestors contribute their genes to the next generation. Evolutionary theories have been proposed to account for experimental results across a wide range of capacities, from categorization to memory to logical and probabilistic reasoning, lan- guage, and cognitive differences between the sexes. They also have been proposed to account for how we reason about money and other resources—a new field of study known as behavioral economics. Also in this chapter, we examine comparative cognition. This is the study of animal intelligence. We look at the cognitive capacities of a number of different species and discuss some of the problems that arise in comparing animals with one another and with humans. The Linguistic Approach Linguistics is an area that focuses exclusively on the domain of language. It is concerned with all questions concerning language ability, such as, What is language made of? How do we acquire language? What parts of the brain underlie language use? As we have seen, language is also a topic studied within other disciplines—for example, cognitive psychol- ogy and neuroscience. Because so many different researchers in different disciplines have taken on the problem of language, we consider it here as a separate discipline, united more by topic than by perspective or methodology. Part of the difficulty in studying language is the fact that language itself is so complex. Much research has been devoted to understanding its nature. This work looks at the properties all languages share, the elements of language, and how those elements are used during communication. Other foci of linguistic investigation center on lan- guage acquisition, deficits in language acquisition caused by early sensory deprivation or brain damage, the relationship between language and thought, language use by nonhu- man primates, and the development of automated speech recognition systems. Linguistics, perhaps more than any other perspective discussed here, adopts a very eclectic interdisciplinary methodological approach. Language researchers employ experiments and computer models, study brain-damaged patients, track how language ability changes during development, and compare diverse languages. The Emotion Approach As you may have surmised, humans don’t just think—we also feel. Our conscious experi- ence consists of emotions, such as happiness, sadness, and anger. Recent work in cogni- tive psychology and other fields has produced a wealth of data on emotions and how they influence thoughts. In Chapter 10, we start out by examining what emotions are and how they differ from moods. We examine several different theories of emotion and describe Chapter one introduCtion 19  how they influence perception, attention, memory, and decision making. Following this, we look at the neuroscience underlying emotions and the role that evolutionary forces played in their formation. AI investigators have formulated models of how computers can “compute” and display emotional behavior. There are even robots capable of inter- acting with people in an emotionally interactive manner. The Social Approach Much of cognition happens inside individuals. But if some of it is happening in between an individual and his or her environment, then the cognitive relations between individuals clearly constitute an important aspect of cognition. The field of social cog- nition explores how people make sense of both themselves and others. We will see that thinking about people often differs from thinking about objects and that different parts of our brains are used when thinking socially. Neuroscience has revealed that in laboratory animals, specialized cells, called mirror neurons, are active both when an animal performs some action and when it watches another animal perform that same action. Later in Chapter 11, we introduce the concept of a theory of mind. This is an ability to understand and appreciate other people’s states of mind. This capacity may be lacking in people suffering from autism. We conclude by summarizing work on specific social cognitive phenomena: attitudes, impressions, attributions, stereotypes, and prejudice. The Artificial Intelligence Approach Researchers have been building devices that attempt to mimic human and animal func- tion for many centuries. But it is only in the past few decades that computer scientists have seriously attempted to build devices that mimic complex thought processes. This area is now known as artificial intelligence. Researchers in AI are concerned with get- ting computers to perform tasks that have heretofore required human intelligence. As such, they construct programs to do the sorts of things that require complex reasoning on our part. AI programs have been developed that can diagnose medical disorders, use language, and play chess. AI also gives us insights into the function of human mental operations. Designing a computer program that can visually recognize an object often proves useful in under- standing how we may perform the same task ourselves. An even more exciting outcome of AI research is that someday, we may be able to create an artificial person who will pos- sess all or many of the features that we consider uniquely human, such as consciousness, the ability to make decisions, and so on (Friedenberg, 2008). The methods employed in the AI perspective include the development and testing of computer algorithms, their comparison with empirical data or performance stan- dards, and their subsequent modification. Researchers have employed a wide range of approaches. An early attempt at getting computers to reason involved the application of logical rules to propositional statements. Later, other techniques were used. Chapters 12 and 13 give detailed descriptions of these techniques. 20 Cognitive SCienCe The Robotics Approach Finally, we consider robotics. Robotics may be considered a familial relation to AI and has appeared on the scene as a formal discipline only recently. Whereas AI workers build devices that “think,” robotics researchers build machines that must also “act.” In fact, the embodied ecological approach has assisted roboticists in discovering that adding a robot body to an AI system can actually help the AI system think better and learn faster. Investigators in this field build autonomous or semiautonomous mechanical devices that have been designed to perform a physical task in a real-world environment. Examples of things that robots can do presently include walking through obstacle courses, competing in miniature soccer games, driving through city streets, welding or manipulating parts on an assembly line, performing search and rescue in hazardous environments, defusing bombs, vacuuming your apartment, and destroying each other on television. The robotics approach has much to contribute to cognitive science and to theories of mind. Robots, like people and animals, must demonstrate successful goal-oriented behaviors under complex, changing, and uncertain environmental conditions. Robot- ics, therefore, helps us think about the kinds of minds that underlie and produce such behaviors. In Chapter 13, we outline different paradigms in robotics. Some of these approaches differ radically from one another. The hierarchical paradigm offers a “top-down” per- spective, according to which a robot is programmed with knowledge about the world. The robot then uses this model or internal representation to guide its actions. The reactive paradigm, on the other hand, is “bottom up.” Robots that use this architecture respond in a simple way to environmental stimuli: They react reflexively to a stimulus input, and there is little in the way of intervening knowledge. The Embodied Ecological Approach Inspired in part by the reactive paradigm in robotics, cognitive scientists have been dis- covering that the body itself plays an important role in cognition. In the embodied cog- nition approach, your body does some of your thinking for you, not just your brain. The particular ways in which the body interfaces with the environment are partly responsible for the cognition that emerges. For example, while part of how a person recognizes an apple may involve mental representations of its visual features (e.g., curved contours and reddish hue), another part of how that apple is recognized is by partially activat- ing the actions that the hand and mouth would typically perform on it. So if you had a different body (e.g., different height, different strength, different number of limbs, different kind of eyes), the way you think would be a little different. In fact, even a tool in your well-trained hand can be treated as part of who you are. These insights about non-brain-based cognitive processes are couched well in the context of a perspective on perception-and-action called the ecological approach. In the ecological approach, intelligent behavior emerges not only from information inside the brain, and inside the body, but also from information inherent in the relationship between the body and the environment. For instance, when a person is standing with their eyes open, light reflects off of surfaces around them and projects a pattern of information on their retinas. When Chapter one introduCtion 21  that person begins walking, that pattern of retinal information changes in a systematic fashion due specifically to the way the person is walking. Rather than passively receiving that sensory input like a computational information processor, the active observer is in charge of how that information changes over time. If she chooses to walk fast, the retinal information changes quickly. If she chooses to walk slow, the retinal information changes slowly. If she chooses to walk with a bounce, the retinal information changes differently still. These insights on cognition are obtained only by focusing on the organism’s rela- tionship to its ecological environment, not by focusing on the organism in isolation. And there’s more than just apples, tools, and sidewalks in your ecological environment that become part of your cognition. There’s other people too. In Chapter 14, the embodied ecological approach is shown to draw its insights from dynamical systems theory, philos- ophy, psychology, neuroscience, linguistics, and even anthropology. Integrating Approaches Many of the approaches we have just listed inform one another. For instance, the fields of AI and robotics are in some cases inseparable. AI programmers often write computer programs that serve as the “brains” of robots, telling them what to do and providing them with instructions on how to perform various tasks like object recognition and manipulation. In recent years, the cognitive and neuroscience approaches have come closer together, with cognitive psychologists providing information-processing models of specific brain processes. For example, there are cognitive models of hippocampal function that specify how this brain region helps the brain encode memories based on our understanding of the neural substrates. In an even more interdisciplinary leap, there is the new area of social cognitive neuroscience in which social contexts are attached to cognitive–neural models. At the end of Chapter 14, we provide some insights into the ways in which these different approaches are and can be integrated. The integration of these different disciplines and approaches may point toward new ways to understand how cognition works. SUMMING UP: A REVIEW OF CHAPTER 1 1. Cognitive science is the scientific embodied cognition and ecological perception, interdisciplinary study of mind and sees the scientific study of emotions and social contributions from multiple fields, including behavior, AI, robotics, and more. philosophy (along with its newest offshoot, 2. Mind can be considered an information experimental philosophy), psychology, cognitive processor. At least some mental operations psychology, neuroscience, the connectionist bear some similarity to the way information is or network approach, evolution, linguistics, processed in a computer. 22 Cognitive SCienCe 3. Information processing requires that some environment, then it too may be able to ground aspect of the world be represented and then its symbols. operated on or computed. A representation 6. Examples of representations are concepts, is symbolic if it stands for something else. propositions, rules, and analogies. The thing a symbol stands for in the world Representations are realized by an information is called its referent. The fact that symbols bearer, have content, are grounded, and need to are “about” these things is called be interpreted. intentionality. 7. Computations are processes that act on or 4. A formal system is made of symbols— transform representations. According to the collections of symbols that form expressions tri-level hypothesis, there may be at least three and processes that act on those expressions levels of analysis for information processing to form new expressions. Formal logic is an systems: computational, algorithmic, and example of a formal system. implementational. 5. According to the physical symbol system 8. Several different schools of thought differ hypothesis, formal systems can be said to be in the way they view representation and intelligent, implying that computers may be computation. In the classical cognitive science intelligent. A problem for this is the symbol view, representations are fixed symbols grounding problem, which states that symbols and information processing is serial. In in people are grounded because we have the connectionist view, representations are bodies, can perceive objects, and can act on distributed and processing is parallel. According them. Therefore, if a computer were to be to the dynamical view, representations are embodied with sensors and effectors that allow constantly changing, being altered with each it to actively gather information from the new experience. SUGGESTED READINGS Friedenberg, J. (2009). Dynamical psychology: Complexity, Sobel, C. P. (2001). The cognitive sciences: An interdisciplinary self-organization, and mind. Charlotte, NC: Emergent. approach. Mountain View, CA: Mayfield. Harnish, R. M. (2002). Minds, brains, computers: An intro- Stainton, R. J. (2006). Contemporary debates in cognitive sci- duction to the foundations of cognitive science. Malden, MA: ence. Malden, MA: Blackwell. Blackwell. Chapter one introduCtion 23