Conceptualizing a possible discipline of human–computer interaction

 

John M. Carroll
Center for Human–Computer Interaction and College of Information Sciences and Technology, The Pennsylvania State University, University Park, PA 16802, USA

John Long's Comment 1 on this Paper

Abstract

This essay is a personal reflection on John Long’s keynote address at the BCS People and Computers meeting in Nottingham in the summer of 1989. I try to locate the paper’s purpose and significance within the history of human-computer interaction (HCI), both prior to 1989 and subsequently, and particularly with respect to the abiding questions of what sort of enterprise HCI is, and of what sorts of knowledge it uses and produces.

Comment 2

 

1. Introduction: HCI’s first three decades

Comment 3

The general position, here, is that, in agreement with Carroll, the HCI community has grown; but, in disagreement with him, the HCI discipline has not made desired progress.

Human–computer interaction (HCI) is an area of research and practice that emerged in the early 1980s, initially as a specialty area in computer science. HCI has expanded rapidly and steadily for three decades, attracting professionals from many other disciplines and incorporating diverse concepts and approaches. To a considerable extent, HCI now aggregates a collection of semi-distinct fields of research and practice in human-centered informatics. However, the continuing synthesis of disparate conceptions and approaches to science and practice in HCI has produced a dramatic example of how different epistemologies and paradigms can be reconciled and integrated.

Comment 4

 

Until the late 1970s, the only humans who interacted with computers were information technology professionals and dedicated hobbyists. This changed disruptively with the emergence of personal computing around 1980. Personal computing, including both personal software (productivity applications, such as text editors and spreadsheets, and interactive computer games) and personal computer platforms (operating systems, programming languages, and hardware), made everyone in the developed world a potential computer user, and vividly highlighted the deficiencies of computers with respect to usability for those who wanted to use computers as tools.

The challenge of personal computing became manifest at an opportune time. The broad project of cognitive science, which incorporated cognitive psychology, artificial intelligence, linguistics, cognitive anthropology, and the philosophy of mind, had formed at the end of the 1970s. Part of the programme of cognitive science was to articulate systematic and scientifically-informed applications to be known as ”cognitive engineering”. Thus, at just the point when personal computing presented the practical need for HCI, cognitive science presented people, concepts, skills, and a vision for addressing such needs. HCI was one of the first examples of cognitive engineering.

Other historically fortuitous developments contributed to the establishment of HCI. Software engineering, mired in unmanageable software complexity in the 1970s, was starting to focus on nonfunctional requirements, including usability and maintainability, and on non-linear software development processes that relied heavily on testing. Computer graphics and information retrieval had emerged in the 1970s, and rapidly came to recognize that interactive systems were the key to progressing beyond early achievements. All these threads of development in computer science pointed to the same conclusion: The way forward for computing entailed understanding and better empowering users.

Finally human factors engineering, which had developed many techniques for empirical analysis of human–system interactions in so-called control domains such as aviation and manufacturing, came to see HCI as a valuable and challenging domain in which human operators regularly exerted greater problem-solving discretion. These forces of need and opportunity converged around 1980, focusing a huge burst of human energy, and creating a highly visible interdisciplinary project.

One of the most significant achievements of HCI is its evolving model of the integration of science and practice. Initially this model was articulated as a reciprocal relation between cognitive science and cognitive engineering. Later, it ambitiously incorporated a diverse science foundation, notably Activity Theory, distributed cognition, and ethnomethodology, and a culturally embedded conception of human activity, including the activities of design and technology development. Currently, the model is incorporating design practices and research across a broad spectrum. In these developments, HCI provides a blueprint for a mutual relation between science and practice that is unprecedented.

Early HCI sought to develop synergies between cognitive science and cognitive engineering. During the 1980s a rich reciprocal relationship developed. In areas like user modeling, HCI directly applied key cognitive science theories to the design of command languages and information visualizations. In other cases, HCI provided guidance to cognitive science through embodied concepts like direct manipulation and user interface metaphor. Mutual reciprocity between underlying science and application is rare, but not unprecedented (the discovery of the transistor effect in physics emerged from applied research). For HCI, this starting point was an intellectually sophisticated and ambitious foundation for more radical possibilities.

HCI research and application provided a strong force for theoretical integration within cognitive science. The very first HCI theories were far more ambitious integrations than had been attempted in the basic science. For example, the model human processor (Card et al., 1983) integrated aspects of perception, attention, short-term memory operations, planning, and motor behavior in a single model, at a time when most cognitive science models addressed only isolated laboratory phenomena. Ironically, early models were criticized within HCI as too limited with respect to understanding and creating applications. This self-criticism promoted increasingly comprehensive modeling that has jointly driven the basic science and its applications. But more importantly, these early successes, and their deconstruction, further fueled paradigmatic aspirations in HCI.

In the latter 1980s and early 1990s, HCI assimilated ideas from Activity Theory, distributed cognition, and ethnomethodology. This comprised a fundamental epistemological realignment. For example, the representational theory of mind, a cornerstone of cognitive science, is no longer axiomatic for HCI science. Information processing psychology and laboratory user studies, once the kernel of HCI research, became important, but niche areas. Field studies became typical, and eventually dominant as an empirical paradigm. Collaborative interactions, that is, groups of people working together through and around computer systems (in contrast to the early 1980s user-at-PC situation) have become the default unit of analysis. The contemporary theory-base of HCI draws broadly upon social, cognitive and computation science, and strongly emphasizes design research, pragmatics, and aesthetics. It is remarkable that such fundamental realignments were so easily assimilated by the HCI community.

Comment 5

Although HCI was always conceived of as a design science, this was construed at first as a boundary, with HCI providing guidance to system design and development. Throughout the 1990s, however, HCI directly assimilated, and eventually itself spawned, a series of design communities. This engagement with design communities coincided with substantial advances in user interface technologies that shifted much of the potential proprietary value of user interfaces into graphical design. Somewhat ironically, designers were welcomed into the HCI community just in time to help remake it as a design discipline. A large part of this transformation was the creation of design disciplines that did not exist before. For example, user experience design and interaction design were not imported into HCI, but rather were among the first exports from HCI to the design world. Design is currently the facet of HCI in most rapid flux. It seems likely that more new design proto-disciplines will emerge from HCI during the next decade.

Conceptions of how underlying science informs and is informed by the worlds of practice and activity have evolved continually in HCI since its inception. Throughout the history of HCI, paradigmchanging scientific and epistemological revisions were deliberately embraced by a field that was, by any measure, succeeding intellectually and practically. The result has been an increasingly fragmented and complex field that has continued to succeed even more. This example contradicts the Kuhnian view of how intellectual projects develop through paradigms that are eventually overthrown.

Comment 6

 

The continuing success of the HCI community in moving its meta-project forward thus has profound implications, not only for human-centered informatics, but for epistemology. (Other sketches of the history of HCI are Carroll, 1997; Myers, 1998; Grudin, 2005.). In this paper, I will elaborate the foregoing interpretation of the emergence of HCI, particularly with respect to conceptions of HCI as a discipline, and of what sorts of theory are possible and appropriate in HCI. My touchstone for this is the contribution of John Long to the development of HCI, especially during the latter 1980s, an extremely formative period. I will place these contributions into a context, reconstructed through the benefits of hindsight, and the latitudes of an essay format. In my view, there is not one true narrative for HCI (or for anything else of even reasonable complexity). Nevertheless, I think my view is grounded and valid, and more importantly, it leads to prospective interpretations of HCI, and of Long’s contributions to it. We do not want to, and, in any case, cannot relive the history of HCI, but we surely ought to try to learn what we can from it.

2. Long’s conception of an engineering discipline of HCI

I met Long in 1989. We met at a conference in Nottingham, at which we were both invited speakers. We were both wrestling, in our separate ways, with what some have called the mid-1980s ”theory crisis” in HCI. As briefly sketched above, HCI had been born at the beginning of the 1980s as a paradigm case of cognitive engineering. But what exactly was cognitive engineering? Most early discussions were vague. I first encountered the programmatic notion of cognitive engineering in discussions at the inaugural meting of the Cognitive Science Society in La Jolla, California, and subsequently in a talk by Norman (1982) at the early HCI conference held at the US Bureau of Standards in Gaithersburg, Maryland. The kernel of the idea was that domains strongly shape cognition, and that studying and supporting cognition in real and complex domains is salutary, if not essential, for developing a science of cognition and, of course, for applying it to real problems.

Comment 7

 

There is no doubt that this conception was transformative with respect to a range of cognitive science activities in the 1980s and subsequently, among them HCI. But likewise there is no doubt that this conception leaves out many critical details. For example, what does it mean to apply cognitive science? How exactly would that work? It would necessarily involve generalizing from principles and results that originated in narrow and contrived laboratory tasks. There would have to be some perilous inductive leaps, at the least. And even if some of the leaps worked out, would applying cognitive science in some number of cases be enough to warrant the claim that cognitive engineering had arrived?

Comment 8

HCI in the 1980s was strongly invested in the cognitive engineering vision, though for the most part the larger framework remained unarticulated. Card et al. (1983), and more pointedly Newell and Card (1985), made the most comprehensive early effort toward articulating a framework for science and engineering in HCI. Their work embraced a simplistic view of the relation between science and engineering, emphasizing approximation but not boundary conditions on or scope of applicability, and it did not produce the paradigmatic consensus they had hoped for, indeed it evoked sharp criticism (Carroll, 2006; Carroll and Campbell, 1986).

Comment 9

 

But the Card, Moran and Newell work did provide a touchstone and focus for other theory-based work during the 1980s; it clearly raised the possibility of a theory-based paradigm, and it attracted many other voices to the debate.
When I met Long, we were both engaged in trying to work out what kind of a discipline or project HCI was, or could be seen as, and what kinds of knowledge or theory it used or could use. We were trying to describe how cognitive science knowledge might emerge from and be applied to HCI design work.

Comment 10

We were pursuing different answers, the nature of which I will address presently.

2.1. Defining the HCI discipline

Long’s keynote address at the Nottingham conference, the paper was written with John Dowell, contrasted three conceptions of HCI as a discipline – craft, applied science, and engineering (Long and Dowell, 1989). Long and Dowell define discipline as a particular body of knowledge supporting specific practices directed at solving a general problem. I like very much the approach of trying to be explicit as possible about definitions. This work bluntly proposes a boldly minimalist schematization of discipline.

Comment 11

Long and Dowell define the general HCI problem as ”the design of humans and computers interacting to perform work effectively” (p. 13). They characterize and analyze three disciplinary conceptions of HCI. First, they suggest that HCI can be pursued as a craft practice, relying on implicit, informal, and experiential knowledge. They discuss the example projects of Prestel videotex (Buckley, 1989) and of the Ded display editor (Bornat and Thimbleby, 1989) as paradigmatic. They argue that a craft practice of HCI cannot be effective: Because its knowledge is implicit, informal, and experiential, its knowledge cannot be operationalized: ”it cannot be directly applied by those who are not associated with the generation of the heuristics or exposed to their use” (p. 18). Moreover, because craft knowledge is heuristic, ”there is no guarantee that practice applying HCI craft knowledge will have the consequences intended” (p. 119). In other words, craft practice is ineffable and unreliable.

Comment 12

Second, they consider a disciplinary view of HCI as applied science, relying on knowledge in the form of theories, models, and principles used to formulate and investigate hypotheses, predictions, and explanations. They concede that sciences, like psychology, can be applied to HCI, giving the example of the role of confirmatory feedback in guiding sequences of behavior. However, they argue, such general scientific principles cannot be directly and deductively applied in a specific design because they do not ”prescribe the feedback required . . . to achieve effective performance of work” (p. 20). In other words, the context in which the scientific principles were formulated and developed are necessarily different than the ones that arise for a particular design application, and specifically so with respect to supporting effective work outcomes in the context of the design.

Comment 13

Long and Dowell also consider the related strategy of constructing guidelines/principles which are themselves grounded in scientific theory. They discuss Hammond and Allinson’s (1988) computer assisted informal learning system with respect to the theory-based design principle ”provide distinctive multiple forms of representation” as an example. Long and Dowell argue that such a principle cannot directly guide design, since neither it nor the theories that underwrite it are defined, operationalized, or generalized with respect to effective performance of work activity.

More generally, Long and Dowell argue that although HCI as an applied science describes knowledge more explicitly and more generally, and supports derivation of theory-based guidelines, it ultimately fails in the same way craft practice fails as a disciplinary model for HCI: applied science does not describe how to support particular work activity effectively. As a consequence, the use of science in design must always be empirically mediated by implementation, evaluation, and iteration.

Long and Dowell turn finally to their preferred disciplinary model for HCI: engineering. They state that engineering distinctively solves design problems ”by the specification of designs before their implementation” (p. 24). They describe engineering knowledge as principles that allow ”designs to be prescriptively specified for artifacts, or systems which when implemented, demonstrate a prescribed and assured performance” (p. 24). Finally, they state that engineering can deal systematically with complexity: ”Designs specified at a general level of description may be systematically decomposed until their specification is possible at a level of description of their complete implementation” (p. 24).
Long and Dowell are optimistic about the engineering conception, but provide few details. Indeed, they concede that engineering principles of the sort they require do not exist (as of 1989). They give two examples that they consider promising: Dix and Harrison (1987) and Dowell and Long (1989). Curiously, they do not mention the engineering models of Card et al. (1983). Further, they suggest – actually quite like Card et al. (1983) – that the most promising niches for this disciplinary model to be realized would be in highly practiced, expert performance of relatively low-level tasks domains in which ”human behavior can be usefully deterministic” (p. 27).

Comment 14

One of the key constructs that differentiates the engineering disciplinary model of HCI from the applied science and craft models is that the former incorporates an explicit model of the application domain. The craft practice and applied science disciplinary models have no notion of boundary conditions, applicability, or context built into them. Nevertheless, Long and Dowell are confident that engineering principles of the sort they imagine for HCI would be generalizable knowledge, that application of the principles would be direct and indeed specifiable, and effective.

Comment 15

2.2. Problems with Long and Dowell’s conception of discipline

The best thing about the Long and Dowell paper is that it clearly and forcefully sets forth definitions, makes sharp distinctions, and reasons toward a clear programmatic conclusion and recommendation for how HCI ought to organize itself as a discipline. I will return to this point in closing, but let me say now that I regard all that as very important and constructive. HCI is still much in need of clarifying its foundations, and making progress on that will require definitions, distinctions, and argumentation. I believe that Long and Dowell’s paper provides intellectual scaffolding to construct such a debate. Although, as I will presently make clearer, I feel that the paper did not achieve it goals, I think the goals are valuable and that the general versions of the questions raised are still quite open to debate and in need of investigation.

Comment 16

Long and Dowell’s specific argument, however, fails in important ways. First, its general conception of discipline is unwieldy, and the authors do little to ease this concern. They present their conception of discipline in the most general terms, and then hurry onwards without actually describing the HCI discipline in any empirical detail. This omission matters, because the second failure of their paper is that the scoping of HCI they presume is more narrow than the reality of HCI. Moreover, the manifold ways in which HCI has broadened since 1989 make it ever more difficult to see how to extend the Long and Dowell analysis to contemporary HCI. Third, the characterizations and arguments directed at the craft and applied science models for an HCI discipline are dismissive and inadequate. They just do not make the case that these paradigms are inappropriate or necessarily ineffective. Finally, the argument for an engineering disciplinary model of HCI rests on an academic idealization of what engineering is like. No wonder they could not find examples of it.

Comment 17

As is often true, the rub comes with unpacking the details. Surely disciplines codify and use knowledge through practices addressing disciplinary problems. So one would have to say far more to have said anything. For example, we might assume that ”practices” is a union of typical and/or critical workflows, information flows, roles and divisions of labor, and other social arrangements in and around work activity. Indeed, this seems a rather minimal notion of practices. But even this is more than encyclopedic in scope. Such an amorphous and expansive conception of practices is a burden for all would be framework makers, and not just for Long and Dowell. However, the challenge Long and Dowell more uniquely inflict on themselves is that they insist that the job of HCI is to render practices in an explicit specification. They do not address this explicitly, but it seems to me that they would have to require that the practices be made explicit through some kind of hierarchical task analysis. In 1989, this would have been seen as reasonable, perhaps even as too obvious to belabor. But this is no longer true. Comprehensive task analysis, outside of restricted safety critical interactions, is moribund (Carroll, 2002). Even its most enthusiastic proponents concede that it is not used (Diaper, 2002).

Comment 18

More pointedly, such a conception of practices rendered explicit through tedious recursive decomposition is empirically wrongheaded. Studies of technical practices in general (Latour, 1987; Latour and Woolgar, 1986; Orr 1996) and of practices in HCI settings specifically (Bentley et al., 1992; Heath and Luff, 2000; Suchman, 1987) refute the possibility that actual domain practices could ever be meaningfully specified in this manner. Practices cannot be specified a priori because at any reasonable level of complexity they depend on the improvisations of people and the local culture of groups. Neither of these can be analyzed a priori or generally. Any programme that requires this level of specification has stumbled on the starting blocks.

Comment 19

Finally, it is important to emphasize that discarding brittle and a priori notions of discipline, practice, and even knowledge does not leave us paradigmatically impaired. Quite to the contrary, empirical approaches have already taught us much. If we want to develop a set of models for disciplines, we should study what scientists, engineers, designers, and other technical persons do, and how they do it (Latour, 1987; Latour and Woolgar, 1986). It is true that empirical approaches do not easily lead to simple type contrasts, such as Long and Dowell’s three paradigms for disciplinary projects, but they do produce exemplars that can be used as models.

Comment 20

Finally, it is important to emphasize that discarding brittle and a priori notions of discipline, practice, and even knowledge does not leave us paradigmatically impaired. Quite to the contrary, empirical approaches have already taught us much. If we want to develop a set of models for disciplines, we should study what scientists, engineers, designers, and other technical persons do, and how they do it (Latour, 1987; Latour and Woolgar, 1986). It is true that empirical approaches do not easily lead to simple type contrasts, such as Long and Dowell’s three paradigms for disciplinary projects, but they do produce exemplars that can be used as models.

Comment 21

2.3. Problems with Long and Dowell’s scoping of HCI

Long and Dowell spend quite a bit of discussion on the importance of identifying general disciplinary problems. They assert that each discipline has one, though it may be decomposable into subproblems with corresponding sub-disciplines. The general problem of HCI is stated as ”humans and computers interacting to perform work effectively” (p. 13). In their text, it is clear that Long and Dowell recognized that this was a very general general problem, essentially a cross-product of all topics involving humans and their organizations, computers – including embedded and networked devices, and all work activities in which the former employ the latter.

Comment 22

Long and Dowell explicitly identify HCI as a design domain. They give an alternate wording of their own general HCI problem highlighting this: ”the design of humans and computers interacting to perform work effectively” (p. 13). The general problem is
decomposed into ”the design of humans interacting with computers” and ”the design of computers interacting with humans”; they associate the former general problem with Human Factors or Ergonomics, and the latter with Software Engineering. All of these restatements and decompositions of the general problem focus on the effective performance of work activity; the phrase ”to perform work effectively” ends each version of the general HCI problem.

Long and Dowell’s conception, sweeping though it may be, is far too narrow. HCI in the reality of its practice addresses many activities that are not work. Indeed, it is rarely pointed out, but obviously true, that the importance of play, leisure, education, and myriad other non-work activities to HCI undermines many of the 1980s conceptions of HCI that came from ergonomics and human factors, but also from Activity Theory, work psychology, and other continental frameworks that are singularly focused on the workplace. Work is important, but all work and no play would have made HCI a far more dull enterprise. Fortunately, it didn’t turn out that way.

Comment 23

In their scoping of HCI, Long and Dowell seem to over focus on a particular methodological challenge of the 1980s. During this early period, designers and developers were often construed as customers or recipients of HCI methods and techniques. And it was generally believed that these counterparts wanted and needed HCI methods that were so well specified that they could be effectively put into practice by people not trained or experienced in HCI itself, namely, the designers and developers. This kind of boundary–mediated relationship still exists in some consulting arrangements, but it is no longer an appropriate view of HCI or its goals. Some of these boundaries have dissolved as HCI has come to directly include a greater diversity of professionals, and to produce professionals with more diverse skills. But perhaps more fundamentally, the goal of codifying formal methods that could be applied ”knowledgefree” to crank out good systems is misguided. We return to this point below.

Comment 24 The issue of rescoping the general problem of HCI to include non-work activity has a more specific problematic consequence for Long and Dowell. Once we broaden the definition of the general problem of HCI – appropriately – to include play, leisure, education, and so on, the very important qualifier effectively becomes much more difficult to understand. Yet, as we will see, Long’s conception of HCI, and his position regarding the question of what approaches to HCI could be appropriate, had everything to do with understanding and operationalizing effectiveness.

Comment 25

2.4. Problems with Long and Dowell’s conception of craft

The defining characteristic of a craft discipline is that craft knowledge is implicit, informal and acquired from experience. Long and Dowell did not give an example of traditional craft practice, but this would have been useful. Craft practice is the most developed paradigm for technology development in human history and there are many examples of it.

Comment 26 An excellent example is George Sturt’s (1923) book ”The Wheelwright’s Shop”. Sturt inherited a shop, and decided to learn why the wheelwrights made wheels and carts as they did. One of the design features for which he sought rationale was that traditional English carts have slightly bowl-shaped wheels, mounted so that the portion below the axel is perpendicular to the ground. Sturt was surprised to find that he got several different answers from his expert wheelwrights. He was told by various master wheelwrights that such wheels better accommodate the cooling of iron tires, that they have a smaller turning radius, that they better tolerate sideward swaying of a cart, and that they allow the cart body to be wider at the top, and thereby allow larger loads. This example illustrates how implicit, informal, experiential knowledge can guide a practice.

Comment 27

Long and Dowell take a skeptically rationalist view that such codification of knowledge is problematic, perhaps because it cannot be conveyed in written form. But it is arguable that this is a superior form of technical practice. Sturt’s account suggests how multiple converging rationales might make practices more robust as they are passed from expert to apprentice. The wheelwright craft that Sturt investigated had been maintained successfully for more than century. It was manifestly effective.

Comment 28

Another perspective on alternative HCI paradigms is to ask what sorts of specific affordances a craft practice entails that applied science and engineering do not. One answer is that knowledge is objectified less and proceduralized more in a craft practice. Sturt’s wheelwrights had one another; they worked collectively. They were less in need of engineering standards or explicit theories and guidelines. Their practice had evolved through generations to a level of design refinement not considered by Long and Dowell. The features of the cart wheels could not be derived linearly from single guidelines or sources of rationale, they were over-determined.

Comment 29

Indeed, this line of thinking seems more relevant than ever to the contemporary configuration of HCI. Many more designers participate in HCI than did in the late 1980s. The paradigm of design work is not evolutionary like wagon wheels, but it is a craft practice. It is true that design is typically taught in a studio paradigm, through participation and enactment as opposed to lecture and discussion. And it is true that design knowledge consists in heuristic concepts and techniques rather than deductive principles and laws. Long and Dowell take this as a crippling epistemological limitation, but they argue from an a priori conception of knowledge and the use of knowledge that requires specification and logical derivation.

Comment 30

Such a requirement would only make any sense if we could be assured that such an option exists. Although Long and Dowell hope it might exist, they are unable to give even one example.

Comment 31

And in any case, even if we could define a positivistic programme for design along these lines, why would we want to, given that a successful paradigm for the design profession already exists and is already contributing broadly to HCI design?

Comment 32

I will suggest later that it might make sense to take cognizance of actual HCI design practices in conceptualizing models for a discipline of HCI.

Long and Dowell’s specific critical analysis of craft is directed most specifically at Bornat and Thimbleby’s (1989) development of the early display editor Ded. In contemporary terminology, Bornat and Thimbleby employed evolutionary prototyping, a method in which designers create a running system and then successively revise their design to respond to user experience. In hindsight, this example is unfortunate with respect to Long and Dowell’s case that craft practice cannot produce generalizable knowledge. For indeed, Ded is an early instance of a design paradigm for text editors that became utterly pervasive throughout the world. It would be difficult to find a better example of design knowledge that proved generalizable, applicable, and effective.

Comment 33

Long and Dowell assume that if a system design is iteratively developed that ipso facto it can only make use of implicit, informal, and experiential knowledge, and cannot be based at all on explicit science or principles. This is bizarre, and all the more bizarre because Bornat and Thimbleby clearly label some of the ideas they articulated and refined in this project as ”theories”. We should take them at their word rather than insist, with Long and Dowell, on a false dichotomy between designs that have an iterative process and designs that embody explicit general knowledge (or for that matter heuristic knowledge). This dichotomy is not consistent with design practice. Indeed, it sets an impossibly high standard for the successful use of knowledge in design, namely that the knowledge must be applied a priori through logical derivation and never be wrong, never need to be adjusted. It is important to keep the severity of this standard in mind because it seems likely that no use of knowledge in the history of HCI, and perhaps any complex design domain, has ever attained this standard.

Comment 34

2.5. Problems with Long and Dowell’s conception of applied science

Long and Dowell argue that the general problem of scientific disciplines is to predict and explain phenomena, not to specify designs that support working effectively. Thus, scientific knowledge can help us predict and explain, but not prescribe designs. This seems an unwarrantedly limited view of the bounds of knowledge and creativity.

Comment 35

If by prescribing designs, Long and Dowell really do mean logically derive designs, then one would have to say that scientific knowledge also cannot serve prediction or explanation, since these applications of scientific knowledge, even safely restricted to basic science discourses, are invariably creative, and not purely mechanical endeavors.

Comment 36

Moreover, as I will emphasize in the immediately following section, there is no known way to prescribe designs in this limited sense, thus failing to prescribe designs must be seen as a vacuous failure.

I am of course aware that the ascription of positivism is now regarded as discourteous, but I do think that Long and Dowell are venturing into the hoary traditions of positivism. Applying knowledge is frequently a creative endeavor, in science or anywhere else. Knowledge does not come with rules of application, rather these are argued for and constructed as knowledge is put into use. In contemporary epistemology, we alter our conceptions about knowledge and its application when confronted with insights and successes. We should not turn away from insights and successes because they fail to follow a priori rules. We should alter the rules.

Comment 37

In their particular analysis of Hammond and Allinson’s theorybased design of a computer-aided learning system, Long and Dowell rather freely admit that the sophisticated use of psychological theory in the design ”might have been expected to modify learning behavior towards that of the easier recall of materials” (p. 22). This seems to contradict their general stance that science cannot prescribe designs, but nevertheless I agree with them. However, they go on to make an interesting distinction. They say that the theories Hammond and Allinson appealed to do not directly ”address the problem of the design of effective learning” (p. 22), and that resultingly Hammond and Allinson’s design might support more effective recall (the specific consequence that the theories did address), but still fail to support effective learning. This is cutting things pretty finely. Most psychologists would take better recall to be a learning achievement, though it is true that better recall is not necessarily indicative of better learning in every sense, or more specifically, of effective learning with respect to particular criteria or learning objectives. Accordingly, Long and Dowell conclude, Hammond and Allinson’s design work would necessarily have to progress via prototyping, evaluation, and iteration.

Comment 38

This argument is exceedingly peculiar. By granting that the theory-based design approach could reasonably expect to realize certain specific consequences for users (enhanced recall), Long and Dowell are essentially granting exactly what Hammond and Allinson claimed. Namely, they are granting that science can be applied in design; that designs can embody principles derived from scientific theory, and that consequences for users of the design can be anticipated from the theory. What they are balking at is the claim that scientific theory could completely specify a design, including all of its detailed consequences for users (cf. ”the design of effective learning”), a rather bold claim that, to my knowledge, no one ever made.

The shortcoming of Hammond and Allinson’s theory-based approach, namely, not being able to precisely specify the design of effective learning, entails that they must augment theory-based guidance with direct empirical approaches – prototyping, evaluation and iteration. But this is precisely how theory-based design in HCI has always worked (Johnson et al., 1989). Given that there is no way to prescribe designs, and that all design of any complexity must be empirical in just this sense, then Long and Dowell’s critique of the disciplinary model of applied science vis-à-vis HCI design is an argument about a straw man.

Comment 39

2.6. Problems with Long and Dowell’s conception of engineering

A touchstone example of software engineering in the real world is Brooks’s (1975) ”Mythical Man–Month”, essays written primarily about one of the largest software engineering projects in history, IBM Operating System 360. On Long and Dowell’s view of engineering we might expect to read about how the OS 360 design was successively decomposed until its subsystems could be completely specified, about how the system was explicitly and completely specified before it was implemented, and about how engineering principles were used to ”prescribe and assure” its performance before it was designed and implemented. But as everyone knows, the design of OS 360 did not work like that.

Comment 40

The complex and iterative nature of actual engineering processes is over-determined, as Brooks describes and as many subsequent studies have elaborated. For example, in a top–down framework, requirements guide the development of early designs. However, in practice, early designs typically cause requirements to be further developed, altered, and even abandoned. Design also helps to identify new requirements that were not part of the original project mandate, but which may be quite essential once they are identified. Brooks emphasizes that these iterative relationships obtain not merely because requirements happened not to be noticed, but because they cannot be identified until the early design enables their discovery. Today, none of this is shocking. Linear waterfall models of design and development have been succeeded by models including feedback and iteration.

Comment 41 Again turning the clock back to the mid-1980s, the apparent disconnect between Long and Dowell’s view of engineering, and Brooks’ case study of engineering practice in a large software development project, becomes more comprehensible. Early methodological conceptions of HCI were frequently oriented to system development processes that had traditionally overlooked consideration of users and other human stakeholders. Ironically, because HCI was not integrated into these processes, and often had no standing in the organizations practicing these processes, HCI methods were developed with an inadequate understanding of their ultimate context of use. Brooks’ view of OS 360 is an insider’s view, and in many respects an exposé. Part of the reason that Brooks’ book is a classic is that when published it was so iconoclastic, refuting core conceptions about software engineering. Long and Dowell were addressing engineering from an external view. They were trying to conceptualize an HCI discipline that could effectively contribute to system engineering processes as officially described.

Comment 42

There is, of course, an engineering paradigm for HCI. It developed in the 1990s, and it is interesting to observe how it differs from the conception of Long and Dowell. Usability engineering is one of the core clusters of method and process in HCI (Nielsen, 1994; Rosson and Carroll, 2002). Notably the sense of engineering in this stream of activity is systematic but it is fundamentally empirical. Its primary focus is methods to directly involve users via participatory design and analysis, and to assess user experiences through surveys and interviews, many kinds of scenario exercises, and direct evaluations, including thinking aloud studies, throughout every stage of the system development process, from
requirements identification through to documentation design. It uses models, for example like GOMS, but in relatively limited ways.

Comment 43

3. A hundred flowers that blossomed

In closing their paper, Long and Dowell (1989) considered whether craft practice, applied science and engineering could function together as a mutually supportive ensemble of disciplinary models. They point out that the three paradigms use and produce knowledge and results that are not always mutually intelligible. However, they counter-argue that the three paradigms together would better exploit what is known and what is practiced in HCI, and that integrating the three disciplinary models could encourage an HCI community ”superordinate to any single discipline conception” (p. 30). While I have tried to scrutinize many of the technical points and arguments in the paper, I think Long and Dowell ultimately were led to a larger truth, one that is strongly evidenced in the HCI we see today.

Comment 44 During the 20 years since Nottingham, HCI has changed in many ways. The lens of the Long and Dowell paper highlights three vectors of change that are intriguing and challenging, and that help to sound echoes of the Long and Dowell paper. First, it has become ever clearer that craft is the primary source of innovation in HCI. The primary role of science in HCI is to help us understand these innovations after they occur. Second, applied science in HCI provides explicit foundation for engineering models. These first two points emphasize how we need to link up the disciplinary models for HCI, instead of examining and evaluating them separately as competitive paradigms.

Comment 45

 

Third, the science foundation for HCI is incredibly rich and fragmented, more than anyone expected in the mid-1980s, and perhaps with more to come. We live in a time when the nature of science itself has been deeply questioned. Some of the debates occurring now in HCI with regard to its proper footing in science make the mid-1980s theory crisis seem mild indeed.

Comment 46

3.1. Craft innovations drive HCI science

The original vision for HCI was that cognitive science theory would produce or guide cognitive engineering of better systems. This programme was affirmed and pursued zealously throughout the 1980s, and through to the present. Good examples of this paradigm do exist, I believe. The Hammond and Allinson paper that Long and Dowell deconstructed is a wonderful example. However, it is also easy to be skeptical of ”science based design” as a general disciplinary model for HCI. Long and Dowell were articulating that skepticism.

Comment 47

Throughout the history of HCI the truly game-changing innovations have tended to be craft based. The pivotal design concept of direct manipulation, and the early scientific accounts of it are a case in point (Shneiderman, 1983; Hutchins et al., 1985). Direct manipulation is the style of computer interaction in which a person manipulates data and functions through gestures with display objects, for example, pointing and clicking in windows with a mouse — as contrasted with referring to data and functions by name in typed command strings. The principle object of these analyses, the point-and-click graphical user interface was developed years before the original scientific accounts, and indeed, direct manipulation is still being theorized and further developed (e.g., Plouznikoff et al., 2005). The graphical user interface as a design concept was developed pretty much through craft innovations during the decade from the mid-1960s through the mid-1970s (e.g., Buxton et al., 2005; Myers, 1998).

Comment 48

It is interesting that even the early theories of direct manipulation emerged long after the technology innovation itself. Thus, far from determining or even guiding the technology development and user interface design, the theories served the purpose of interpreting, consolidating, and abstracting the lessons from craft innovation to help move HCI research and development work forward in a more explicit and deliberate manner. This is very valuable; it allows for an engineering practice to be codified from the more implicit craftwork.

Comment 49

I describe the direct manipulation example because it is and was so central to the development of HCI. However, similar patterns can be seen in the development of HCI science and theory for other key concepts and techniques in HCI design. For example, Blackwell (2006) provides a vivid exegesis of how craft and science have shaped the use of metaphor in HCI through the course of three decades. The history of HCI science is one of technology innovations mysteriously popping out of craftwork, and then eventually being noticed, analyzed, and codified in models and theories.

Comment 50

3.2. Applied science provides explicit foundation for engineering models

The most ambitious articulation of the relationship between cognitive science and its application is that of a reciprocal relationship at the level of models (Norman, 1982). This is to make a distinction between specific (one-off) design applications of cognitive science, like the Hammond and Allinson computer assisted learning system, and systemic applications in which the science is a foundation for an engineering model that can be more generally applied. Card et al. (1983) development the Model Human Processor (MHP) and Goals, Operators, Methods and Selection rules (GOMS) model for analyzing routine human–computer interactions is a very early example of applied cognitive science theory in HCI that also provided direct guidance for a set of engineering models.

Comment 51

As I briefly noted above, these models were somewhat narrowly scoped, but seen in the context of cognitive science ca. 1980, this was some of the most comprehensive applied science work ever attempted. The model explicitly integrated many components of skilled performance – perception, attention, short-term memory operations, planning, and motor behavior – to produce predictions about expert performance in real tasks.

The MHP/GOMS model, taken as an engineering model, was an advance over prior human factors approaches in that it explicitly described the cognitive structures underlying manifest behavior. In other words, it was an engineering model directly and explicitly grounded in scientific understanding of information processing psychology. But as a scientific account, this model was a huge step forward also: Cognitive science models and theories up to that point had not attempted such a level of integration. The comprehensiveness of these early models vis-à-vis cognitive science can be seen as directly caused by their purpose vis-à-vis cognitive engineering. In order to generate detailed quantitative predictions about user performance in realistic contexts, HCI models must make explicit assumptions about a wide range of human characteristics.

Comment 52

 

Not all engineering models in HCI are as focused and narrow as MHP/GOMS. For example, the various developments of usability engineering could be considered a collection of engineering models. The usability engineering textbook I wrote with Rosson (Rosson and Carroll, 2002) quite explicitly presents a system development lifecycle engineering model, a wide range of systematic methods that usability engineers can follow to better assure that their designs are useful and usable to people. Our conception of engineering is heavily influenced by Brooks; we emphasize many approaches to prototyping, and emphasize throughout the book that one of the primary mistakes to avoid is premature commitment: thinking that the first passably-acceptable solution generated is the stopping point for design. If I were to write that book today I would significantly broaden its treatment of nuances of quality in the user experience. We emphasized satisfaction, but qualities like fun and engagement deserve more consideration. Of course this would make usability engineering even more dependent on applied science, and even more empirical.

Comment 53

Just recently, the ACM SIGCHI Symposium on Engineering Interactive Computing Systems (http://eics-conference.org/2009/) was launched, but again the notion of engineering in this conference series is far broader than that of mechanical derivation of solutions with predetermined properties.

Comment 54

3.3. The science foundation for HCI is incredibly rich and fragmented

The science foundation of HCI in the early 1980s was cognitive science, chiefly cognitive psychology. But this quickly changed. Even by the end of the 1980s, many other approaches with roots in social psychology, sociology and anthropology were moving to the center of HCI. A technological reason for this was that collaborative systems were emerging and raised many questions that could not be articulated in a cognitive paradigm. More importantly, HCI was reaching out to understanding technology in use, and actually contexts of use nearly always involved multiple people, work practices, organizational structures and myriad factors beyond the purview of individual cognition.

Suchman’s (1987) study of photocopier use was iconic. She described a variety of usability problems with advanced photocopier user interfaces. The problems she identified were fairly typical of HCI studies of the time (Carroll, 2003). But her approach and analysis were distinct in important ways. First, she studied people doing real work in a workplace context, not in a laboratory setting, as was typical at the time. Indeed, many of her participants were scientists at Xerox PARC going about their research work, and occasionally struggling to make copies. Second, she analyzed the interaction between the person and the machine as a sort of conversation that can fail when the actions of the participants are not intelligible to one another. This raised the level of analysis to that of human agency, as opposed to operating characteristics of the mind-as-a-computer (memory limitation, incorrect rules, etc.). Third, she directed her analysis to very fundamental issues in cognitive science. Thus, based on the amount of creative improvisation she observed in people trying to fathom and use photocopiers, she concluded that the concept of plans as causal accounts of human action was fundamentally flawed. Plans might be resources for action, but could not determine action in circumstances of any significant complexity.

Comment 55

By the end of the 1980s, HCI had become an international research community. Threads of work that had been going on in the United States and in Europe were increasingly brought together through Europeans visiting the mostly-American Association for Computing Machinery (ACM) CHI Conference and Americans visiting the mostly-European International Federation of Information Processing (IFIP) INTERACT Conference. For example, Bjerknes et al. (1987) published their collection on participatory design. Although they do argue that involving users directly in the inner sanctums of design deliberation is technically effective, the primary theme and argument in their book links participatory design to democracy, and asserts that sharing power with users in the design process is a better moral choice. Issues of self-determination in the workplace and power sharing and participation in software design cannot be articulated in a purely cognitive HCI.

Comment 56

These developments contributed to a scientific foundation far more rich, far more diverse than the starting points of the early 1980s. By the end of the 1990s, HCI looked quite different. Social psychology concepts like production blocking, conformity, social loafing, and social pressure had become as commonplace as memory capacity, consistency, and index of difficulty in the 1980s.

Activity theory and distributed cognition were established paradigms in theory, in many ways eclipsing information processing psychology as the establishment in theory. Ethnographical fieldwork had become a methodological touchstone for understanding usability.

Comment 57

HCI theory during this period was highly successful, in the sense of producing explanations and principled descriptions of human– computer interaction contexts (themselves largely produced through craft innovations) that have had great impact on cognitive science. Just as the MHP/GOMS model had led cognitive science in the 1980s, Suchman’s analysis of situated actions, distributed cognition and activity theoretic models, and studies of computer-mediated collaboration had substantial influence throughout cognitive science. For example, in 1993 a special issue of Cognitive Science, the field’s flagship journal, was addressed to reconsideration of Suchman’s 1987 book on the field of cognitive science. Similarly, a special issue of the Journal of the Learning Sciences, the flagship cognitive science journal in learning, was directed to reflection on Suchman’s contribution in 2006.

Comment 58

 

3.4. Design rationale as theory: the task-artifact framework

My own contribution to the Nottingham conference was more similar to Long’s than I realized at the time. Like Long, I also sought to formulate a programme for HCI.

Comment 59

Like Long, I had concluded that HCI could not comprehensively be constructed as applied cognitive science. In my paper (Carroll, 1989), I suggested that the most effective role for science in HCI design might be to interpret designs-in-use, to codify the knowledge implicit in designs so that it could be used more explicitly in future designs. I suggested that bringing this interpretative work into the design process itself might be the closest we could get to theory-based design. I saw this as augmenting the paradigm of craft practice to demystify the how and why, quite analogous to what George Sturt was trying to with the craft of wheelwrights.

Comment 60

Much like Sturt, I suggested that designed artifacts ought to ”read” as theories, that system design and development outcomes should be directly leveraged as knowledge outcomes (Carroll and Campbell, 1989). Implemented designs have nice properties, considered as knowledge outcomes: they are precise and complete; that is, they are complete enough to run and do whatever it is that they do, and they cannot leave things vague or make unrealistic simplifying assumptions the way a discursive theory can.

Comment 61

Designs must take a stand on every issue they encounter. Designs seamlessly integrate ideas from many sources and from many levels of analysis; that is, a design may embody ideas from GOMS with respect to keystroke-level interactions, ideas from Activity Theory about leveraging cultural practices in new work designs, and ideas from social psychology about how collaborators can quickly come to trust one another. Discursive theories are notoriously bad at spanning levels of analysis; indeed, many take it as given that levels of analysis can never be spanned. Finally, designs are also unavoidably testable, that is, when people use them, their use generates consequences, vivid, concrete and often poignant.

The main thing that makes designs poor as theories is that they are difficult to read. The propositions of an artifact’s theory are implicit, after all; they must be constructed by an analyst. And there’s the rub. How can we identify the theory that is implicit in an implemented design? My answer was design rationale, the documentation traditionally generated in the design process describing the issues, decisions, choices, and consequences that were considered, pursued, abandoned, and/or implemented. Of course, because design rationale is documentation, it is often viewed as tedious, boring, and bureaucratic. Moreover, because most designs ultimately fail in one way or another, creating an explicit and thorough design rationale is analogous to carefully leaving your fingerprints all over a spot you know will most likely be a crime scene.

Comment 62

In Longian terms, I wanted to imagine ways to better integrate HCI craft practices with applied social and cognitive science so as to do the least violence to the manifest effectiveness of HCI craft practices, but at the same time help those practices to be more deliberative, more auditable, and more manageable.

Comment 63

The key criterion for me was intelligibility to the practices and values of HCI designers. Thus, I eventually built my own methodological prescriptions out of scenario-based design, emphasizing the practical utility of narrative representations as well as their analytic utility in suggesting and contextualizing rationales (e.g., Carroll, 2000).

During the years after Nottingham, and to some extent caused by prodding from Long, I kept at this line of thinking, eventually reaching the programmatic claim that in HCI design rationale is the theory (Carroll and Rosson, 2003). During the early 1990s, I had many enjoyable interactions with Long. My recollection is that Long accepted that my approach integrated craft and applied science, but, perhaps not surprisingly, he felt our ”design rationale as theory” programme was underspecified as a disciplinary model, and that, in particular, it needed to be more explicit about defining effectiveness. Our discussion never came to an ending.

Comment 64

3.5. Making sense of HCI

In 1989, many in the field felt that HCI needed a better-defined paradigm, or as Long and Dowell termed it, a better-defined disciplinary model. For even in 1989, HCI was plainly a thriving and growing socio-technical endeavor that was diversifying much more than it was converging. Perhaps framing disciplinary models is just the human impulse to closure, to create figure from ground. Perhaps it is just researchers doing what they do. With another 20 years of hindsight, we can see more plainly now that HCI continues to diversify more than to narrow. Not only do Long and Dowell’s original contenders live on, but we have many variants of each.

Long and Dowell’s technical analysis wound up being more nihilistic than they most likely intended. I think this was because they faithfully and energetically applied overly rigid and a priori models of all three of their disciplinary models for HCI. To me this is the tragic pattern of positivism, which I take here as a paradigm– defining concern with how propositions are generated (discovery procedures, predictive models), warranted (usually in observable empirical phenomena), and logically related (e.g., by derivation or generalization). Positivism, as far as I can tell, is motivated by good and noble impulses: Escape subjectivism and capture universal truths, provide empirical foundation for knowledge statements, produce systematic, cumulative, and integrative knowledge-generating practices and knowledge descriptions and explanations (e.g., science) and so forth. The problem for positivism, and reason I see it as tragic, is that subjectivism is inescapable. Knowledge depends on context, on point of view, on history, on meaning making practices that are partially ineffable, and on levels of analysis that are incommensurable. Indeed, as emphasized by every philosopher since Kuhn (1962) science is a social institution, and what is regarded as sound, even what is regarded as true, is socially constructed.

Comment 65

My personal construction of this is that positivism belongs to that interesting category of wrong ideas we need to value. Positivism so overworries methodological issues that it produces results that are unproblematic but of little consequence. Still we are well advised to orient to positivist objectives. We should do so knowing that to take these objectives too seriously, too rigidly, will lead to paradigmatic dead ends. Throwing positivist cautions away can lead to empirical programmes for which we cannot know what methods were actually employed, what data were gathered, or what the results are really about.

Comment 66

Long and Dowell, I think, became snarled in an intellectual trap of their own design in characterizing the three disciplinary models for HCI. They characterized models that no one followed, and that no one ever has followed. They laid down positivistic criteria for disciplinary models that I suspect cannot be satisfied, and that, in any case, do not accord with what anyone actually does or has done in the craft, science, and engineering of HCI.

Comment 67

But they were in good company. On my side of the Atlantic, Allen Newell articulated an interestingly comparable disciplinary programme in his 1985 opening plenary address at the ACM CHI Conference (Newell and Card, 1985). Newell’s talk presented a vision of extending the early GOMS work into a much more comprehensive paradigm for science in HCI. He memorably said that psychology might be ”driven out” of HCI in the future if it were not pursued in a quantitative modeling paradigm (aka, ”hard science”). The talk provoked much controversy, discussion, and new research. It led to alternate proposals, modified proposals, replies and rejoinders (Carroll and Campbell, 1986; Newell and Card, 1986). Again, through the benefit of hindsight, we can see that Newell overstated the ”hard science” threat. His programme was ignored, yet psychology continues to thrive in HCI. As I look back, I think that what Newell really wanted was a science of HCI in which psychology (and other cognitive sciences) could play a central role. Newell’s real worry, I think, was not that the psychology of HCI might take a qualitative turn. He was worried about the interdisciplinary power balance between computer science and its human science partners. I have elaborated this historical reflection in Carroll (2006).

In my view, Newell and Long’s contributions must be taken in historical context. They were addressing the still-current threat of methodological fragmentation. They had helped to found HCI in the 1970s, and were trying to ensure that HCI could continue to prosper as it had in the early 1980s. Both proposed to achieve this through normative disciplinary frameworks. And both went a bit too far in this regard.

Comment 68

The conclusion I take away is that we should regard HCI as a sort of meta-discipline. I call it a community formed around the ever-expanding concept of usability (Carroll, 2009), because I think it is really just this shared pre-theoretic interest and commitment that causes HCI to cohere at all. HCI has no single disciplinary problem or specified set of practices, and certainly no single conception of effectiveness. Instead, the boundaries of HCI have expanded as the notion of usability became richer.

Comment 69

Usability was originally articulated naively in the slogan ”easy to learn, easy to use”. The blunt simplicity of this conceptualization gave HCI an edgy and prominent identity in computing. It served to hold the field together, and to help it influence computer science and technology development more broadly and effectively. However, inside HCI the concept of usability has been reconstructed continually, and has become increasingly rich and intriguingly problematic. Usability now often subsumes qualities like fun, well-being, collective efficacy, aesthetic tension, enhanced creativity, support for human development, and many others. The trajectory of its core concept explains how and why the HCI community has continued to grow and diversify. It also explains why a priori frameworks, such as those articulated by Long and by Newell, tend to look dated almost before they are formulated. More importantly, a dynamic view of usability suggests that what we have seen for the past three decades may just continue. Perhaps usability will always develop as our ability to reach further toward it improves.

Comment 70

This picture of HCI as a diverse community orienting to a concept whose meaning changes through time is unsettling. It implies that the methodological fragmentation Long addressed in his 1989 Nottingham keynote is endemic to HCI, not so much a problem to be remedied, but a characteristic to be accepted and leveraged, or at least coped with.

Comment 71 In either case, I think it remains useful to try to articulate disciplinary models – models that are evidenced in current practices, models from areas neighboring HCI that arguably could address current challenges in HCI, if they were to be adopted, and perhaps even models that we just invent. Some of these models could be the pure types that Long and Dowell worked with, perhaps too rigid to be implemented, but useful as analytic tools to characterize the more hybrid forms that can actually be observed.

Comment 72

Acknowledgements

This paper draws on several my own previous meditations on the history and foundations of HCI: Carroll, 1997, 2002, 2006, 2009. I thank the editors for organizing this project. I especially thank Peter Wright and two anonymous reviewers for their cheerfully insightful deconstructions of my essay. I believe that moments of reflection on the contributions of those who have helped to lead us in the recent past are not only appropriate celebrations, but also directly useful for us toward making sense of what we have done, and what we are doing now. Finally, I want to acknowledge and thank John Long, with whom I had a very stimulating debate at various conferences and workshops during 1988– 1993. These interactions were very helpful to me in motivating and focusing my own thinking and writing. Frankly, I think I benefited more, though John always seemed ready for another round. As far as I can remember and reconstruct, no minds were changed in these debates, but I recognize more clearly now that that is not necessarily the most important outcome of such a debate. The writing of this essay was supported in part by the Edward M. Frymoyer Chair Endowment.

References

Bentley, R., Hughes, J.A., Randall, D., Rodden, T., Sawyer, P., Shapiro, D., Sommerville, I. 1992. Ethnographically-informed systems design for air traffic control. In: Proceedings of the 1992 ACM Conference on Computer-Supported Cooperative Work, Toronto, Ontario, Canada, November 01–04, 1992. CSCW ’92. ACM, New York, NY, pp. 123–129. DOI: http://doi.acm.org/10.1145/143457.143470.

Bjerknes, G., Ehn, P., Kyng, M. (Eds.), 1987. Computers and Democracy — a Scandinavian Challenge. Aldershot, Avebury, VT.

Blackwell, A.F., 2006. The reification of metaphor as a design tool. ACM Transactions on Human–Computer Interaction 13 (4), 490–530.

Bornat, R., Thimbleby, H., 1989. The life and times of Ded, text display. In: Long, J.B., Whitefield, A.D. (Eds.), Cognitive Ergonomics and Human–Computer Interaction. Cambridge University Press, Cambridge.

Brooks, F., 1975. The Mythical Man–Month. MIT Press, Cambridge.

Buckley, P., 1989. Expressing research findings to have a practical influence on design. In: Long, J.B., Whitefield, A.D. (Eds.), Cognitive Ergonomics and Human–
Computer Interaction. Cambridge University Press, Cambridge.

Buxton, W., Baecker, R., Clark, W., Richardson, F., Sutherland, I., Sutherland, W., Henderson, A., 2005. Interaction at lincoln laboratory in the 1960’s: looking forward – looking back. In: CHI ’05 Extended Abstracts on Human Factors in Computing Systems, Portland, OR, USA, April 02–07, 2005. CHI ’05. ACM, New
York, NY, pp. 1162–1167.

Card, S.K., Moran, T.P., Newell, A., 1983. The Psychology of Human–Computer
Interaction. Erlbaum, Hillsdale, NJ. Carroll, J.M., 1989. Feeding the interface eaters. In: Sutcliffe, A.G., Macaulay, L.A. (Eds.), People and Computers V. Cambridge University Press, Cambridge, pp. 35–48.

Carroll, J.M., 1997. Human–computer interaction: psychology as a science of design.
Invited Chapter for the Annual Review of Psychology 48, 61–83.

Carroll, J.M., 2000. Making Use: Scenario-Based Design of Human–Computer Interactions. MIT Press, Cambridge, MA.

Carroll, J.M., 2002. Human–Computer Interaction. Encyclopedia of Cognitive Science. Macmillan/Nature Publishing Group, London.

Carroll, J.M., 2003. Situated action in the zeitgeist of human–computer interaction.
The Journal of the Learning Sciences 12 (2), 273–278.

Carroll, J.M., 2006. Soft versus hard: the essential tension. In: Galletta, Dennis,Zhang, Ping (Eds.), Human–Computer Interaction in Management Information Systems. In: Zwass, Vladimir (Ed.), Advances in Management Information Systems Series. Armonk, NY, M.E. Sharpe, Inc., pp. 424–432.

Carroll, John M., 2009. Human computer interaction (HCI). Interaction-Design.org, <http://www.interaction-design.org/encyclopedia/human_computer_interaction_ hci.html>. (Retrieved 19.05.09).

Carroll, J.M., Campbell, R.L., 1986. Softening up hard science: reply to Newell and Card. Human–Computer Interaction 2, 227–249.

Carroll, J.M., Campbell, R.L., 1989. Artifacts as psychological theories: the case of human–computer interaction. Behaviour and Information Technology 8, 247– 256.

Carroll, J.M., Rosson, M.B., 2003. Design rationale as theory. In: Carroll, J.M. (Ed.), HCI Models, Theories, Models and Frameworks: Toward a Multidisciplinary Science. Morgan-Kaufmann, San Francisco, pp. 431–461.

Diaper, D., 2002. Scenarios and task analysis. Interacting with Computers.

Dix, A.J., Harrison, M.D., 1987. Formalizing models of interaction in the design of a display. In: Proceedings of Intract’87 Second IFIP Conference on Human–
Computer Interaction. Elsevier Scientific, Amsterdam, pp. 409–413.

Dowell, J., Long, J., 1989. Toward a conception for an engineering discipline of
human factors. Ergonomics 32, 1513–1535.

Grudin, J., 2005. Three faces of human–computer interaction. IEEE Annals of the
History of Computing 27 (4), 46–62.

Hammond, N.V., Allinson, L.J., 1988. Development and evaluation of a CAL system for non-formal domains: the hitch-hikers guide to cognition. Computers and Education 12, 215–220.

Heath, C., Luff, P., 2000. Technology in Action. Cambridge University Press, Cambridge. Hutchins, E., Hollan, J., Norman, D., 1985. Direct manipulation interfaces. Human–
Computer Interaction 1, 311–338. Johnson, J., Roberts, T.L., Verplank, W., Smith, D.C.,

Irby, C.H., Beard, M., Mackey, K., 1989. The xerox star: a retrospective. Computer 22 9 (11–26), 28–29. Kuhn, T., 1962. The Structure of Scientific Revolutions. University of Chicago Press, Chicago.

Latour, B., 1987. Science in Action: How to Follow Scientists and Engineers Through
Society. Harvard University Press, Cambridge, MA.

Latour, B., Woolgar, S., 1986. Laboratory Life: The Construction of Scientific Facts, second ed. Princeton University Press, Princeton, NJ.

Long, J.E., Dowell, J., 1989. Conceptions of the discipline of HCI: craft, applied science and engineering. In: Sutcliffe, A., Macaulay, L. (Eds.), People and Computers IV, Proceedings of the Fifth Conference of the BCS HCI SIG. Cambridge University Press, Cambridge, pp. 9–32.

Myers, B.A., 1998. A brief history of human computer interaction technology. ACM Interactions 5 (2), 44–54.

Newell, A., Card, S., 1985. The prospects for psychological science in human– computer interaction. Human–Computer Interaction 1, 209–242.

Newell, A., Card, S., 1986. Straightening out softening up: response to Carroll and Campbell. Human–Computer Interaction 2, 251–267.

Nielsen, J., 1994. Usability Engineering. Morgan Kaufmann, San Francisco. Norman, D.A., 1982. Steps toward a cognitive engineering: design rules based on analyses of human error. In: Proceedings of the 1982 Conference on Human Factors in Computing Systems, Gaithersburg, Maryland, United States, March 15–17.

Orr, J.E., 1996. Talking About Machines: An Ethnography of a Modern Job. Cornell University Press, Ithaca, NY.

Plouznikoff, N., Plouznikoff, A., Robert, J.-M., 2005. Object augmentation through ecological human–wearable computer interactions. In: IEEE International Conference on Wireless and Mobile Computing, Networking and Communications, 22–24 August, WiMob’2005, pp. 159–164.

Rosson, M.B., Carroll, J.M., 2002. Usability Engineering: Scenario-Based Development of Human–Computer Interaction. Morgan-Kaufmann, San Francisco.

Shneiderman, B., 1983. Direct manipulation: a step beyond programming languages. IEEE Computer 16 (8), 57–69.

Sturt, G., 1923. The Wheelwright’s Shop. Cambridge University Press, London.

Suchman, L.A., 1987. Plans and Situated Actions: The Problem of Human–Machine Communication. Cambridge University Press, New York.