5.2 Historical Overview

Most readers will already know that cognitive theory came into its own as an extension of (some would say a replacement of) behavioral theory (see 2.2. 1). However, many of the tenets of cognitive theory are not new and date back to the very beginnings of the autonomous discipline of psychology in the 19th century. We therefore begin with a brief discussion of introspection and of Gestalt theory before turning to the story of cognitive psychology's reaction to behaviorism.

5.2.1 Introspection

One of the major forces that helped psychology emerge as a distinct discipline at the end of the 19th century was the work of the German psychologist Wundt (Boring, 1950). Wundt made two significant contributions, one conceptual and the other methodological. First, he clarified the boundaries of the new discipline. Psychology was the study of the inner world, not the outer world, which was the domain of physics. And the study of the inner world was to be the study of thought, or mind, not of the physical body, which was the domain of physiology. At first glance, these two distinctions may strike us as somewhat naive. However, it is worth noting that a great deal of recent research in cognitive psychology has looked at the issue of how the physical world is mapped onto memory, and in some cases it is not always clear where the physical world ends and the mental world begins. Also, there is now a growing interest in neurophysiological explanations of perception and cognition. This interest is occurring at a time when philosophers and psychologists are questioning Cartesian dualism, which proposes that mind and body are separate and which has held sway in Western thought since the 17th century. The distinction between mind and brain is becoming blurred. Thus, today, physics and physiology are not necessarily cleanly separated from psychology.

Wundt's methodological contribution was the development of introspection as a means for studying the mind. Physics, and to a large extent physiology, deals with phenomena that are objectively present and therefore directly observable and measurable. Thought is both highly subjective and intangible. Therefore, Wundt proposed, the only access to it, if one was to study it directly, was for a person to examine his or her own thoughts. And the only way to do that was through introspection. Wundt developed a program of research that extended over many decades and attracted adherents from laboratories in many countries. Typically, his experimental tasks were simple: pressing buttons, watching displays. The data of greatest interest were the descriptions his subjects gave of what they were thinking about as they performed the tasks.

On the face of it, Wundt's approach was very sensible. You best learn about things by studying them directly. And the only direct route to thought is via a subject's description of his or her own thinking. The danger of introspection lies in the difficulty persons have thinking about their own thinking. Behaviorists would soon decry the lack of objectivity in the method. What is more, we have to ask whether the act of thinking about thinking interferes with and changes the thinking that one is interested in studying. Is there an "uncertainty principle" at work whereby the act of thinking about thought changes its very nature?

It is important to note that, in spite of criticism that led to its ultimate demise, introspection (the first psychology) was unashamedly cognitive. What is more, the same general access route to cognitive processes is used? today in think aloud protocols (Ericsson & Simon, 1984) obtained while subjects perform natural or experimental tasks. The method is respected, judged to be valid if properly applied, and essential to the study of thought and behavior in the real world or in simulations of it.

5.2.2 Gestalt Psychology

The word Gestalt is a German noun that

has two meanings: besides the connotation of "shape" or "form" as a property of things, it has the meaning of a concrete individual and characteristic entity, existing as some thing detached and having a shape or form as one of its attributes. Following this tradition, in Gestalt theory, the word Gestalt means any segregated whole ... (Hartmann, 1935). Thus, Gestalt psychology is the study of how people see and understand the relation of the whole to the parts that make up that whole.

Wertheimer (1924) stated that Gestalt psychology was not trying to find the meaning of each individual part at the expense of the whole. He stated:

Gestalt theory will not be satisfied with sham solutions suggested by a simple dichotomy of science and life. Instead, Gestalt theory is resolved to penetrate the problem itself by examining the fundamental assumptions of science. It has long seemed obvious-and is, in fact, the characteristic tone of European science-that "science" means breaking up complexes into their component elements. Isolate the elements, discover their laws, then reassemble them, and the problem is solved. All wholes are reduced to pieces and piecewise relations between pieces. The fundamental "formula" of Gestalt theory might be expressed this way: There are wholes, the behavior of which is not determined by that of their individual elements, but where the part-processes are themselves determined by the intrinsic nature of the whole. It is the hope of Gestalt theory to determine the nature of such wholes (Wertheimer, 1924).

Although the major features of this "new" psychology were developed by Wertheimer, his two prot6g6s, Kohler and Koffka, were responsible for the wide dissemination of this school of thought. This spread was assisted by the rise in Germany of the Nazi party in 1933. Hitler expelled Wertheimer, Levin, von Hornbostel, Stem, Werner, and other Gestalt scholars, ensuring the spread of the concept. Koffka was appointed a research professor at Smith College, and Kohler would soon be at Harvard. Both had been giving lecture tours explaining the principles and concepts of this new school.

One of the best illustrations of the whole being different from the sum of the parts is provided by Ehrenfels in a musical example. If a melody is played on an instrument, it is recognizable. If the melody is played again, but this time in another key, it is still recognizable. However, if the same notes, in the same key, were played in a different sequence, the listener will not recognize any similarity between the first and the second melody. As an example, if the sequence of notes for the first melody was e e f g g f e d c c d e e d d, and the second melody played was b b c d d c b a g g a b b a a, the listener would recognize the melody immediately as being the same even though different notes are involved. But if the second sequence used the same notes but in a different order, e e g g f f c c d d e e e d d, the similarity would not be recognized unless, of course, the listener understood the precise way in which the melody has been transformed. Based on this difficulty, and the ability of a person to recognize and even reproduce a melody in a key different from the original one,

Ehrenfels concludes that the resemblance between spatial and tonal patterns rests upon something other than a similarity of their accompanying elements. The totals themselves, then, must be different entities than the sums of their parts.

The central tenet of Gestalt theory-that our perception and understanding of objects and events in the world depends on the appearance and actions of whole objects, not of their individual parts-has had some influence on the evolution of research in educational technology. The key to that influence are the well-known Gestalt laws of perceptual organization, codified by Wertheimer (1938). These include the principles of "good figure," "figure-ground separation," and "continuity." These laws formed the basis for a considerable number of message design principles (see 26.2) (Fleming & Levie, 1978), in which Gestalt theory about bow we perceive and organize information that we see is used in prescriptive recommendations about how to present information on the page, or screen. A similar approach to what we hear is taken by Hereford and Winn (1994).

More broadly, the influence of Gestalt theory is evident in much of what has been written about visual literacy (see 16.4). In this regard, Arnheim's book Visual Thinking (1969) is a key work. It was widely read and cited by scholars of visual literacy and proved influential in the development of that movement.

Finally, it is important to note the recent renewal of interest in Gestalt theory (Henle, 1987; Epstein, 1988). The Gestalt psychologists provided little empirical evidence for their laws of perceptual organization beyond everyday experience of their effects. Recently, perceptual psychologists (Pomerantz, 1986; Rock, 1986) have provided explanations for how perceptual organization works from the findings of controlled experiments. The effects of such stimulus features as symmetry on perceptual organization has been explained in terms of the "emergent properties" (Rock, 1986) of what we see in the world around us. We see a triangle as a triangle, not as three lines and three angles. Emergent properties, of course, are the same as the Gestaltist's "whole" that has features all its own that are, indeed, greater than the sum of the parts.

5.2.3 The Rise of Cognitive Psychology

Behavioral theory is described in detail elsewhere in this handbook (see 2.2). Suffice it to say that behaviorism embodies two of the key principles of positivism: that our knowledge of the world can only evolve from the observation of objective facts and phenomena; and that theory can only be built by applying this observation in experiments where only one or two factors are allowed to vary as a function of an experimenter's manipulation or control of other related factors. The first of these principles therefore banned from behavioral psychology unobservable mental states, images, insights, and Gestalts. The second principle banned research methods that involved the subjective techniques of introspection, phenomenology, and the drawing of inferences from observation rather than from objective measurement. Ryle's (1949) relegation of the concept of "mind" to the status of "the ghost in the machine," both unbidden and unnecessary for a scientific account of human activity, captures the behaviorist ethos exceptionally well.

Behaviorism's reaction against the suspect subjectivity of introspection was necessary at the time if psychology were to become a scientific discipline. However, the imposition of the rigid standards of objectivism (see 7.3) and positivism excluded from accounts of human behavior many of those experiences with which we are extremely familiar. We all experience mental images, feelings, insight, and a whole host of other unobservable and immeasurable phenomena. To deny their importance is to deny much of what it means to be human (Searle, 1992). Cognitive psychology has been somewhat cautious in acknowledging the ability or even the need to study such phenomena, often dismissing them as "folk psychology" (Bruner, 1990). Only recently, this time as a reaction against the inadequacies of cognitive rather than behavioral theory, do we find serious consideration of subjective experiences. (These are discussed in Bruner, 1991; Clancey, 1993; Edelman, 1992; Searle, 1992; and Varela, Thompson & Rosch, 1991, among others. They are also touched on elsewhere in this handbook.)

Cognitive psychology's reaction against the inability of behaviorism to account for much human activity arose mainly from a concern that the link between a stimulus and a response was not straightforward, that there were mechanisms that intervened to reduce the predictability of a response to a given stimulus, and that stimulus-response accounts of complex behavior unique to humans, like the acquisition and use of language, were extremely complex and contrived. (Chomsky's [19641 review of Skinner's [1957] S-R account of language acquisition is a classic example of this point of view and is still well worth reading.) Cognitive psychology therefore focuses on mental processes that operate on stimuli presented to the perceptual and cognitive systems, and which usually contribute significantly to whether or not a response is made, when it is made, and what it is. Whereas behaviorists claim that such processes cannot be studied because they are not directly observable and measurable, cognitive psychologists claim that they must be studied because they alone can explain how people think and act the way they do.

Let me give two examples of the transition from behavioral to cognitive theory. The first concerns memory, the second mental imagery.

Behavioral accounts of how we remember lists of items are usually associationist. Memory in such cases is accomplished by learning S-R associations among pairs of items in a set and is improved through practice (Gagn6, 1965; Underwood, 1964). However, we now know that this is not the whole story and that mechanisms intervene between the stimulus and the response that affect how well we remember. The first of these is the collapsing of items to be remembered into a single "chunk." Chunking is imposed by the limits of short-term memory to roughly seven items (Miller, 1956). Without chunking, we would never be able to remember more than seven things at once. When we have to remember more than this limited number of items, we tend to learn them in groups that are manageable in short-term memory, and then to store each group as a single and out of images of familiar objects and found that the unit. At recall, we "unpack' (Anderson, 1983) each chunk and retrieve what is inside. Chunking is more effective if the items in each chunk have something in common, or form a spatial(, (McNamara 1986; McNamara, Hardy & Hirtle, 1989) or temporal (Winn, 1986) group.

A second mechanism that intervenes between a stimulus and response to promote memory for items is interactive of items and recall is cued with one item of the pair, performance is improved if they form a mental image in which the two items appear to interact (Paivio, 1971, 1983; Bower, 1970). For example, it is easier for you to remember the pair Whale Cigar if you imagine a whale smoking a cigar. The use of interactive imagery to facilitate memory has been developed into a sophisticated instructional technique by Levin and his colleagues (Morrison & Levin, 1987; Peters & Levin, 1986). The considerable literature on the role of imaginary in paired associate and other kinds of learning is summarized by Paivio (1971, 1983; Clark & Paivio, 1991).

The importance of these memory mechanisms to the development of cognitive psychology is that, once understood, they make it very clear that a person's ability to remember items is improved if the items are meaningfully related to each other or to the person's existing knowledge. The key word here is "meaningful". For now, we shall simply assert that what is meaningful to a person is determined by what they can remember of what they have already learned. This implies a circular relationship among learning, meaning and memory -- that what we learn is affected by how meaningful it is, that meaning is determined by what we remember, and that memory is affected by what we learn. However, this circle is not a vicious one. The reciprocal relationship between learning and memory, between environment and knowledge, is the driving force behind established theories of cognitive development (Piaget, 1968) and of cognition generally (Neisser, 1976) as we shall see in our examination of schema theory. It is also worth noting that Ausubel's (1963) important book on meaningful verbal learning proposed that learning is most effective when memory structures appropriate to what is about to be learned are created or activated through advance organizers. More generally, then, cognitive psychology is concerned with meaning, or semantics, while behavioral psychology is not.

Mental imagery provides another interesting example of the differences between behavioral and cognitive psychology. Imagery was so far beyond the behaviorist pale that Mandler's article which re-introduced the topic in was subtitled "The return of the ostracized". Images were, of course, central to Gestalt theory, as we have seen. But because they could not be observed, and because the only route to them was through introspection and self-report, they had no place in behavioral theory.

Yet we can all, to some degree, conjure up mental images. We can also deliberately manipulate them. Kosslyn, Ball & Reiser (1978) trained their subjects to "zoom" in and out of images of familiar objects and found that the "distance" between the subject and the imagined object constrained the subject's ability to describe the object. To discover the number of claws on an imaged cat, for example, the subject had to move closer to it in the mind's eye.

This ability to manipulate images is useful in some kinds of learning. The method of "Loci" (Kosslyn, 1985; Yates, 1966), for example, requires a person to create a mental image of a familiar place in the mind's eye and to place in that location images of objects that are to be remembered. Recall consists of mentally walking through the place and describing the objects you find. The effectiveness of this technique, which was known to the orators of ancient Greece, has been demonstrated empirically (Cornoldi & De Beni, 1991; De Beni & Cornoldi, 1985).

Mental imagery will be discussed in more detail in the section on representation (5.3). For now, we will draw attention to two methodological issues that are raised by its study. First, some studies of imagery are symptomatic of a conservative color to some cognitive research. As Anderson (1978) has commented, any conclusions about the existence and nature of images can only be inferred from observable behavior. You can only really tell if the Loci method has worked if a person can name items in the set to be remembered. On this view, the behaviorists were right. Objectively observable behavior is all even cognitive researchers have to go on. This means that cognitive psychology has to study mental representation and processes indirectly and to draw conclusions about them by inference rather than from direct measurement. (This will doubtless change as techniques for directly observing brain functions during cognitive activity become available and reliable. See Farah,1989.)

The second methodological issue is exemplified by Kosslyn's (1985) use of introspection and self-report by subjects to obtain his data on mental images. The scientific tradition that established the methodology of behavioral psychology considered subjective data to be biased, tainted and therefore unreliable. This precept has carried over into the mainstream of cognitive research. Yet, in his invited address to the 1976 AERA conference, the sociologist Uri Bronfenbrenner (1976) expressed surprise, indeed dismay, that educational researchers do not ask subjects their opinions about the experimental tasks they carry out, nor about whether they performed the tasks as instructed or in some other way. Certainly, this stricture has eased in much of the educational research that has been conducted since 1976, and non-experimental methodology, ranging from ethnography to participant observation to a variety of phenomenologically-based approaches to inquiry are the norm for certain types of educational research (see, for example, the many articles that appeared in the mid-'eighties, among them Baker, 1984; Eisner, 1984; Howe, 1983; Phillips, 1983). Nonetheless, strict cognitive psychology still tends to adhere to experimental methodology, based on positivism, which makes research such as Kosslyn's on imagery somewhat suspect.

5.2.4 Cognitive Science

Inevitably, cognitive psychology has come face to face with the computer. 'Ibis is not merely a result of the times in which the discipline has developed but also emerges from the intractability of many of the problems cognitive psychologists seek to solve. The necessity for cognitive researchers to build theory by inference rather than from direct measurement has always been problematic. And it seems that it will remain so until such time as the direct measurement of brain activity is possible on a large scale.

One way around this problem is to build theoretical models of cognitive activity, to write computer simulations that predict what behaviors are likely to occur if the model is an accurate instantiation of cognitive activity, and to compare the behavior predicted by the model-the output from the program-to the behavior observed in subjects. A good example of this approach is found in the work of David Marr (1982) on vision.

Marr began with the assumption that the mechanisms of human vision are too complex to understand at the neurological level. Instead, he set out to describe the functions that these mechanisms need to perform as what is seen by the eye as it moves from the retina to the visual cortex and is interpreted by the viewer. The functions Marr developed were mathematical models of such processes as edge detection, the perception of shapes at different scales and stereopsis (Marr & Nishihara, 1978). The observed electrical activity of certain types of cell in the visual system matched the activity predicted by the model almost exactly (Marr &Ullman, 198 1).

Marr's work has had implications that go far beyond his important work on vision, and as such serves as a paradigmatic case of cognitive science. Cognitive science is not called that because of its close association with the computer but because it adopts the functional or computational approach to psychology that is so much in evidence in Marr's work. By "functional" (see Pylyshyn, 1984), we mean that it is concerned with the functions the cognitive system must perform, not with the devices through which cognitive processes, are implemented. A commonly used analogy is that cognitive science is concerned with cognitive software, not hardware. By "computational" (Arbib & Hanson, 1987; Richards, 1988), we mean that the models of cognitive science take information that a learner encounters, perform logical or mathematical operations on it, and describe the outcomes of those operations. The computer is the tool that allows the functions to be tested, the computations to be performed.

The tendency in cognitive science to create theory around computational rather than biological mechanisms points to another characteristic of the discipline. Cognitive scientists conceive of cognitive theory at different levels of description. The level that comes closest to the brain mechanisms that create cognitive activity is obviously biological. However, as Marr presumed, this level is virtually inaccessible to cognitive researchers, consequently requiring the construction of more abstract functional models. The number, nature, and names of the levels of cognitive theory vary from theory to theory and from researcher to researcher. Anderson (1990, Chapter 1) provides a useful discussion of levels, including those of Chomsky (1965), Pylyshyn (1984), Rumelhart and McClelland (1986), and Newell (1982), in addition to Marr's and his own. In spite of their differences, each of these approaches to levels of cognitive theory implies that if we cannot explain cognition in terms of the mechanisms through which it is actually realized, we can explain it in terms of more abstract mechanisms that we can profitably explore. In other words, the different levels of cognitive theory are really different metaphors for the actual processes that take place in the brain.

The computer has assumed two additional roles in cognitive science beyond that of a tool for testing models. First, some have concluded that, because computer programs written to test cognitive theory accurately predict observable behavior that results from cognitive activity, cognitive activity must itself be computerlike (see 19.2.3.1). Cognitive scientists have proposed numerous theories of cognition that embody the information-processing principles and even the mechanisms of computer science (Boden, 1988; Johnson-Laird, 1988).'Thus we find reference in the cognitive science literature to input and output, data structures, information processing, production systems, and so on. More significantly, we find descriptions of cognition in terms of the logical processing of symbols (Larkin & Simon, 1987; Salomon, 1979; Winn, 1982).

Second, cognitive science has provided both the theory and the impetus to create computer programs that "think" just as we do. Research in artificial intelligence blossomed during the 80s, and was particularly successful when it produced intelligent tutoring systems (see 19.3; Anderson & Reiser, 1985; Anderson, Boyle & Yost, 1985; Wenger, 1987) and expert systems (see 24.8; Forsyth, 1984). The former are characterized by the ability to understand and react to the progress a student makes working through a computer-based tutorial program. The latter are smart "consultants," usually to professionals whose jobs require them to make complicated decisions from large amounts of data.

Its successes notwithstanding, Al has shown up the weaknesses of many of the assumptions that underlie cognitive science, especially the assumption that cognition consists in the logical mental manipulation of symbols. Recently, scholars (Clancey, 1993; Dreyfus, 1979; Dreyfus & Dreyfus, 1986; Edelman, 1992; Searle, 1992) have been vigorous in their criticism of this and other assumptions of cognitive science, as well as of computational theory and, more basically, functionalism. The critics imply that cognitive scientists have lost sight of the metaphorical origins of the levels of cognitive theory and have assumed that the brain really does compute the answer to problems by symbol manipulation. Searle's comment sets the tone: "If you are tempted to functionalism, we believe you do not need refutation, you need help" (1992, p. 9). As we shall see in the last section of this chapter, cognitive science is at the point behavioral theory was in the early 60s-facing criticism from proponents of a new paradigm for psychology.

5.2.5 Section Summary

Although many of the ideas in this section will be developed in what follows, we think it is useful at this point to provide a short summary of the ideas presented so far. We have seen that cognitive psychology returned to center stage largely because stimulus-response theory did not adequately or efficiently account for many aspects of human behavior that we all observe from day to day. The research on memory and mental imagery that we briefly described indicated that psychological processes and prior knowledge intervene between the stimulus and the response, making the latter less predictable by behavioral theory. We have also seen that nonexperimental and nonobjective methodology is now deemed appropriate for certain types of research. However, it is possible to detect a degree of conservatism in mainstream cognitive psychology that still insists on the objectivity and quantifiability of data.

Cognitive science, emerging from the confluence of cognitive psychology and computer science, has developed its own set of assumptions, not least among which are computer models of cognition. These have served well, at different levels of abstraction, to guide cognitive research, leading to such applications as intelligent tutors and expert systems. However, the computational theory and functionalism that underlie these assumptions have been the source of considerable recent criticism and point perhaps to the closing of the current chapter in the history of psychology.

The implications of all of this for research and practice in educational technology will be dealt with in section 5.5. We would nonetheless like to anticipate three aspects of that discussion. First, educational technology research, and particularly mainstream instructional design practice, needs to catch up with cognitive theory. As we have suggested elsewhere (Winn, 1989), it is not sufficient simply to substitute cognitive objectives for behavioral objectives and to tweak our assessment techniques to gain access to knowledge schemata rather than just to observable behaviors. More fundamental changes are required.

Second, shifts in the technology itself away from rather prosaic and ponderous computer-assisted programmed instruction to highly interactive multimedia environments permit educational technologists to develop serious alternatives to didactic instruction. We can now use technology to do more than direct teaching. We can use it to help students construct meaning for themselves through experience in ways proposed by constructivist theory and practice described elsewhere in this handbook (see 7.4, 20.3, 20.4, 23.4, 24.6) and by Duffy and Jonassen (1992), Duffy, Jonassen, and Lowyck (1993), and others.

Third, the proposed alternatives to computer models of cognition-which explain first-person experience, nonsymbolic thinking and learning, and reflection-free cognition--lay the conceptual foundation for educational developments of virtual realities (see Chapter 15; Winn, 1993). The full realization of these new concepts and technologies lies in the future. However, we need to get ahead of the game and prepare for when these eventualities become a reality.