It is difficult to determine exactly when I began this book, as such. However, the research on latent inhibition was initiated in the late 1950s while I was still a graduate student at Cornell University. The next memorable date for measurement was the publication of an article reviewing the latent inhibition literature in the Psychological Bulletin (1973), which appeared in about half of its original length. The conditioned attention theory of latent inhibition was developed later in a series of articles in the Journal of Experimental Psychology: Animal Behavior Processes in the mid-1970s, and then presented more fully in a chapter in Progress in Learning and Motivation edited by G.H. Bower (1981). The idea of developing this material into a book emerged during a sabbatical year at Yale University, 1977–1978, while much of the writing itself was postponed until another sabbatical, 1981–1982, as an Israel-Canada Fellow at Concordia Unversity. To all of these institutions, and to the individuals who were responsible for inviting me, I express my sincerest gratitude, particularly to Allan Wagner and Eugene Rothman.

In addition, throughout the years I have been fortunate to have my research supported by a number of organizations: The National Institutes of Health were particularly encouraging, especially at the early stages of my career with a Career Development Award. Other support has come from the Scottish Rite Schizophrenia Research Program and, in Israel, from the Charles Smith Psychobiology Fund, the Israel National Academy of Science, Israel Research Trustees Foundation (Ford), and Tel-Aviv Unversity.

Similarity of preexposed stimulus and test stimulus

Studies of the influence of stimulus preexposure on subsequent learning usually assume that these effects are stimulus-specific. That is, preexposure to stimulus A should not retard the subsequent acquisition of an association between stimulus B and another event. This assumption is particularly critical if one maintains the idea that the subsequent decrement in the acquisition of the association between A and another event is a result of some previous associative learning during the preexposure phase. Associative learning, by definition, presumes some degree of stimulus specificity. Indeed, the apparent absence of such specificity in the learned helplessness effect, at least in rats (Maier & Seligman, 1976) in which preexposures of a motivationally significant stimulus is administered, by itself, might raise the question whether or not one is dealing with an associative learning phenomenon. For latent inhibition, across different paradigms, the results are quite clear. The decremental effects of preexposure of the to-be-conditioned stimulus on subsequent acquisition of a new association are, without doubt, stimulus-specific.

Such stimulus specificity may be demonstrated in several ways. (1) A within-subject experimental design may be employed whereby the animal is preexposed to stimulus A and tested on both stimulus A and stimulus B. When appropriately counterbalanced, slower learning to the familiar stimulus compared with the novel stimulus serves as evidence for the stimulus specificity of latent inhibition. Such a design was employed by Lubow and Moore (1959), Reiss and Wagner (1972), and Wickens et al. (1983).

Concepts

Risk

Risk is associated with the undesired outcomes of an action or absence of an action. Every choice involves risk because the consequences are never certain and inevitably some consequences are less desirable than others. Statistically, risk can be defined as the sum of the probabilities of the separate undesirable consequences of an activity. As Rowe (1986) put it, risk is the downside of a gamble. Alternatively the pursuit of any objective can be assessed in cost: value terms and risks are associated with the costs. Risk is often defined as a compound measure of probability and magnitude of adverse effect, e.g. Lowrance (1980), but this complicates even further an already complex concept because the magnitudes of adverse effects involve value judgments which are idiosyncratic. With this kind of definition there can be no agreed measure of a specific risk.

Risk has become a social and technical rather than a personal problem because technology at any level generates hazards which are not confined to one person pursuing his private affairs. Even when using a hand-tool a worker who suffers or causes suffering might readily imply that the risk was generated not by him but by the tool design, by lack of appropriate training or by inadequate instructions. With higher technology the situation becomes still more complicated. A passenger in any vehicle is subject to appreciable risks over which he has no control once he has entered the vehicle.

Concepts

One of the fundamental attributes of human behaviour is that it changes continuously along the time dimension. This fact is neglected by many experimental psychologists who are attempting to describe the behaviour of the human operator, the average man, the laboratory subject (actually labelled ‘S’ in some journals) or even the human organism (labelled ‘O’). This is only defensible provided that it is recognised that whatever properties are adduced they relate to a static abstraction. The properties of this abstraction are described in Chapter 5 but as a preliminary to such studies the set of concepts which express what is known about dynamic behaviour are needed.

The constructive changes in time are generally put into the category of learning already described in Chapter 2. The analogous negative changes which occur in parallel are described as fatigue (in the short term) and ageing (in the long term). Ageing is also dealt with in Chapter 2.

There are several concepts which centre on the facility to change level of activity either voluntarily or involuntarily. This is described as arousal. Within arousal studies two sets of phenomena have been identified and investigated in great detail, particularly over the past forty years. These are stress and vigilance. Stress was adopted as a popular behavioural research topic because it seemed to have possibilities of easier measurement than fatigue which had its heyday from 1920 to 1950.

Concepts

Social factors

Activity involving others

It is rare for a human being to do anything in isolation for very long. It is possible to think of examples such as the tractor driver who may be alone for a working day, and there are unusual tasks such as single-handed yachting where the person may be alone for several weeks. Nevertheless, the obvious rarity of these illustrations emphasises the point that people at work are usually interacting with other people.

This may involve the interdependence of tasks, in which case there is a team or crew involved or it may be that people happen to be working in the same location and the interaction is social rather than operational. Many, perhaps most, tasks are of an intermediate kind where there is some common purpose linking the tasks, but nevertheless individuals are working more or less independently and their interaction is mainly social. They may chat at intervals, they share rest periods and meals and they may travel to and from work together. This is the common situation in industry and commerce. The social interaction is very highly prized, it is often mentioned as the main reason for enjoying going to work. A different kind of interaction occurs when one person is providing a service for a second person, for example the shop assistant or the dentist.

Concepts

Information acceptance

The processes on the input side of the human operator are, in serial order: sensation, perception and cognition. These are functional terms not neatly separated and not unambiguously related to physiologically identifiable mechanisms. Even the order is complicated by interaction and feedback processes.

Sensation

The sense organs are separately identifiable in that they contain specialised transducers: cells which are particularly reactive to light, pressure, chemicals or temperature. This separation is not the same as the classical five senses in that it merges the auditory system with the somaesthetic system within pressure detectors, and taste and smell within chemical detectors. The somaesthetic system provides experience of pressure discontinuities at the surface and in the movement of the body, it detects movement about joints (the kinaesthetic system), angular and linear acceleration (the vestibular system) and skin pressure (the tactile system) (Fig. 5.1). Its importance is underestimated because, by contrast with the distant senses, it functions with minimal conscious awareness. Taste is not important at work except for specialised tasks related to catering. In addition to its close relationship with taste, smell is useful in fault detection of high temperatures and leaks. Essentially the work of the human operator is guided by three main sensing devices: vision, hearing and somaesthesia. Sensation implies a generalised awareness of changes in light, sound or bodily position and posture. Thus the brain is involved as well as the specialised sense organs.

Concepts

This is one of the shorter chapters because, with the exception of road vehicles, the research effort is considerable but the practical application is limited and shows no signs of expanding rapidly in the near future. Nevertheless it is worthwhile to introduce the topic because at least potentially there is a considerable Human Factors contribution to the issues.

The technology-based society

The issue in general terms is that, throughout the world, changes in the environment, work and leisure are now based on advances in technology. If these changes are left to the physical scientists and engineers the consequences are not entirely satisfactory to the customer, the ordinary citizen. In his role as the people's representative in technology design teams the Human Factors specialist ought to take some responsibility for improving this state of affairs. There are a number of difficulties:

–the ultimate responsibility is with the politicians elected to government, but in terms of intellect, education and training, politicians are not well equipped to assimilate and weight the conflicting advice they receive, and national constitutions were designed for a different era. For example, most countries have a governmental Department of Transport, but no country has an integrated transport policy. There are too many vested interests, the interactions are too complicated, the consequences of decisions extend well beyond the life cycle of any one government – in general the problem is too big. Some impact is possible at regional and district levels, but not on a national scale.

Concepts

Developmental aspects

Most laboratory and field studies of human behaviour involve taking a situational snap-shot at a given time in a given place. It is easy to overlook the continuously changing nature both of people and of work situations.

When the skilled manager encounters what, on the face of it, is an intolerable set of work practices, attitudes and performance he does not necessarily take drastic action, he identifies the natural processes of change and accelerates them. In a remarkably short time changes in procedures, products and workload, promotions, resignations, retirements and so on can transform any situation.

The changes within an individual are even more inexorably continuous. Again it is easy to develop the illusion that a worker is a constant factor in a changing situation particularly for a person in the middle period of working life. This was encouraged by the traditional view of a skilled man as one who learned his trade by the age of 21 and thereafter practised it for more than forty years until he retired at 65. This was never the case and in the current working world it is even less so because of developing technology which has its impact on almost all jobs and also on the social context of work. For example, because of employment levels there is currently an emphasis on earlier retirement from and later entry into the working world.

Concepts

The human contribution

The human role as a system controller

The business of designing machines, processes and systems can be pursued more or less independently of the properties of people. Nevertheless people are always involved, the designer himself is a human being and his product will shape the behaviour of many workers and other users. More fundamentally, the design activity will be meaningless unless it is directed towards serving some human need. In spite of all this, the design process itself is often thought about and executed without any formal considerations about people. Inevitably the engineer, architect or other designer devotes most of his attention and expertise to devising mechanisms, buildings and so on which support some human activity more effectively than those currently available. The new machine or system must not be very different from the old one for a variety of reasons. The old one did its job, not perfectly but well enough to justify its existence. The new one is usually designed on the basis of copying the old one but removing as many as possible of the faults. There are other reasons such as commonality of components and, of course, shortage of imagination which lead to most design and development being a progressive iterative process. This happens to suit the human operators because most of their skills will transfer along the line of development of the machines and systems.

Few would disagree that analogy is an important tool in the acquisition of new knowledge. Indeed, work in cognitive science and educational psychology in the last dozen years provides ample evidence of the usefulness of analogy in learning and has substantially advanced our understanding of the psychological mechanisms responsible for that utility (e.g., Burstein, 1986; Carbonell, 1986; Collins & Gentner, 1987; Gentner, 1983; Gentner & Gentner, 1983; Gick & Holyoak, 1980; Rumelhart & Norman, 1981; Vosniadou & Ortony, 1983). Yet, as this chapter will demonstrate, the use of analogies in learning is far from straightforward and, surprisingly, often results in deeply held erroneous knowledge.

Our intention is to offer a more temporized and cautionary alternative to the general enthusiasm for learning by analogy, especially in its most common form: the use of a single mapping between a source and a target concept (the “topic”) – what we shall refer to as a single analogy. (For exceptions that address more complex uses of analogy, see Burstein, 1986; Collins & Gentner, 1987). We argue that simple analogies that help novices to gain a preliminary grasp of difficult, complex concepts may later become serious impediments to fuller and more correct understandings. Specifically, although simple analogies rarely if ever form the basis for a full understanding of a newly encountered concept, there is nevertheless a powerful tendency for learners to continue to limit their understanding to just those aspects of the new concept covered by its mapping from the old one. Analogies seduce learners into reducing complex concepts to a simpler and more familiar analogical core.

The contributions included in Part III focus on developmental and instructional aspects of similarity and analogy. In her chapter, Ann Brown argues that young children can transfer knowledge across analogous domains and that findings that they can not are an artifact of the paradigms used to assess transfer. Stella Vosniadou also argues that children are capable of using analogical reasoning to acquire new knowledge and that what develops is not analogical reasoning per se but the content and organization of the knowledge base to which analogical reasoning is applied.

Brian Ross discusses the role that remindings play in acquiring knowledge about unfamiliar domains and suggests using surface similarity as a way to enhance analogical access. In their contribution, John Bransford, Jeffery Franks, Nancy Vye, and Robert Sherwood offer several proposals about how to make the knowledge base more flexible. They suggest teaching people in problem-oriented environments, rather than teaching facts, and they mention various instructional techniques for promoting the noticing of similarities and differences. Rand Spiro, Paul Feltovich, Richard Coulson, and Daniel Anderson discuss the advantages and disadvantages of using instructional analogies to acquire knowledge in ill-structured domains. Finally, William Brewer offers a commentary on the contributions to Part III, focusing on the role of analogical reasoning in knowledge acquisition.

The contributions to this volume are extensively revised versions of papers delivered at a Workshop on Similarity and Analogy held at the Allerton House of the University of Illinois, Urbana-Champaign, in June 1986. The purpose of the workshop was to bring together scientists working on similarity and analogy to explore current theoretical developments in this area and to consider the practical implications of this work for learning and instruction. The group was interdisciplinary in character and included scientists looking at similarity and analogy from psychological, computational, and educational points of view. The workshop was exciting, enjoyable, and rewarding, and we would like to take this opportunity to thank all the participants for helping to make it so.

Much of the workshop's original structure survived in the transition to this edited volume. The contributions in the first part deal with the issue of similarity. The second part includes the contributions dealing with analogical reasoning. Because analogies are fundamentally concerned with similarity at the level of representational structure, the chapters in the second part provide a theoretical context for those dealing with analogical reasoning by examining a number of questions about the nature of similarity and its relation to conceptual structure. The contributions in the third part discuss analogical reasoning in relation to learning and instruction. All three parts end with one or more chapters that offer commentaries, providing quite detailed discussions of most, although not all, of the other chapters in the book.

For the past several years my colleagues and I have been analyzing what we call parallel distributed processing (PDP) systems and looking at what we call the microstructure of cognition (cf. McClelland, Rumelhart, & the PDP Research Group, 1986; Rumelhart, McClelland, & the PDP Research Group, 1986). In this work we developed computational models of cognitive processes based on principles of “brainstyle” processing. The major focus of this work has been in perception, memory retrieval, and learning. The question remains as to how this work extends to the domains of “higher mental processes.” We have made one attempt to show how our PDP models can be used to account for schemalike effects (Rumelhart, Smolensky, McClelland, & Hinton, 1986). This chapter is designed to push those ideas further and to sketch an account of reasoning from a PDP perspective. I will proceed by first describing the basic theoretical structure of the PDP approach. I will then give a brief account of the reasoning process and finally show how it can be seen as resulting from a parallel distributed processing system.

Parallel distributed processing

Cognitive psychology/information processing has become the dominant approach to the understanding of higher mental processes over the past 25 years or so. The computer has provided, among other things, the primary conceptual tools that have allowed cognitive psychology to succeed. These tools have been powerful and have offered a conceptualization of mind that has proven both more rigorous and more powerful than any that have preceded it.