CHAPTER 1. OVERVIEW

The following chapters present a theory of computer support for innovative (non-routine), cooperative design based on an analysis of interpretation in design. They will argue that the central impediment to computer support of innovative design is that designers make extensive use of situated tacit understanding while computers can only store and display explicit representations of information.

The process by which designers transform their tacit preunderstanding into explicit knowledge is termed interpretation. (See Part I for an analysis of interpretation in design.) Interpretation is central to the process of solving design problems and is part of the process of collaborating with other designers; the explicit knowledge that is generated by this interpretation is therefore a natural by-product of innovative, cooperative design. (See Part II for a theory of computer support based on this generated knowledge.) Representations of this knowledge defined using computer-based design support systems can be captured by these systems for the support of subsequent design work, including the maintenance and modification of the designed artifacts. (See Part III for details of a computer system for supporting interpretation in design.)

Chapter 1 provides a chapter-by-chapter overview of the dissertation. It discusses the claims, arguments, and themes that arise in each of the subsequent chapters, without going into the detail necessary to defend the claims, support the arguments, or work out the themes. Its purpose is to provide a readers’ guide to the flow of the dissertation, motivating how one discussion leads into or provides the background for another. Section 1.1 offers a preliminary presentation of the central concept of interpretation, anticipating the analysis of this concept from various approaches in the dissertation. Each of the other sections provides an overview of a specific chapter.

1.1.      Understanding Interpretation

To say that interpretation is central to innovative design is to stress that in order to design the designer must to some degree understand and be able to articulate the significance of the artifact being designed. This may include, for instance, understanding what is desired in a task specification, how possible composite parts of the artifact will function and interact, or how people can use the designed artifact. According to the analysis presented below, such understanding is possible for people but not for computers. People understand things because they are actively involved with them in the world. The significance of artifacts for a person is determined by the artifacts’ relationships to other artifacts, activities, and people whose significance is already understood as part of the person’s situation. Understanding combines personal and socially shared perspectives on the world. All of this takes place primarily in tacit ways, i.e., unverbalized. However, one’s tacit understanding of something can be partially articulated or expressed explicitly in spoken, written, or graphical language—either to deepen one’s own understanding or to communicate with others.

Two aspects of the process of interpretation can be distinguished.

(1) There is a tacit preunderstanding based on previous background knowledge; items from this preunderstanding can be articulated explicitly.

(2) There is the possibility of revising that preunderstanding based on discoveries that are opened up by it.

That is, one can interpret something as something that one already knows about, or as a variation that differs from that in ways that are discovered as a result of the breaking of one’s tacit expectations. Accordingly, interpretation in innovative design involves both human understanding of extensive background and a creative ability to revise one’s understandings iteratively.

The analysis of interpretation developed below distinguishes three characteristics of interpretation: being situated, having a perspective, and using language.[1]

(a) Being situated means that what is to be interpreted is tacitly understood by virtue of its associations within an open-ended network of related artifacts, people, human purposes, and other concerns. All of these associations are themselves understood as part of one’s background understanding of one’s involvements.

(b) Having a perspective means that there is a focus on a certain aspect or that a specific approach is taken in interpreting something.

(c) Using language means that a particular vocabulary is available as part of a tradition that provides a conceptual framework for the interpretive task.

Each of these characteristics of interpretation is grounded in a form of preunderstanding that can be transformed through a corresponding phase of discovery. This two-dimensional structure is presented in Table 1-1.

Table 1-1. The structure of human interpretation.

In articulating tacit understanding, interpretation both discloses inherent implications and discovers unanticipated consequences in the situation. Through interpretation, designers might (a) try to externalize their expectations about a design situation by drawing a sketch and then discover surprises when they explore the sketch. Similarly, they might need to revise their understanding as a result of (b) shifting their focus on a problem and deliberating from alternative perspectives or (c) changing the way they conceptualize an issue and refining the definitions of terms in their language.

The structure of human interpretation carries over to design. The design process is a cycle or spiral of interpretation: (1) some item of the initial preunderstanding—the grasp of the design situation, the perspective for viewing, the language for conceptualizing—is made explicit, reflected upon, and further articulated in new design decisions. (2) This leads to the discovery of unanticipated consequences or contingencies and a new understanding that requires revisions to the understanding of the design problem, its viewpoint, or terminology. (1´) The new understanding then becomes re-submerged into a modified tacit understanding that forms the starting-point for the next iteration of interpretation and design.

The analysis of interpretation in design motivates a theory of computer support. According to this theory, computer support for interpretation in innovative design differs from autonomous software systems for routine design by focusing on supporting or augmenting human activities rather than automating them, because only people have the understanding and creativity required for interpretation. This computer support takes two general forms in order to support the two phases of interpretation:

1.   It provides access to a wealth of information that might be useful as a basis for interpreting new design tasks. This information for reuse is culled primarily from previous design experience, and includes (a) partial representations[2] of design situations, (b) alternative ways of considering tasks, and (c) terminology helpful for conceptualizing problems.

2.   It facilitates the revision of stored information so designers can tailor existing representations to novel problems and can capture innovative designs to extend the computerized knowledge-base and to communicate ideas to collaborators. This plasticity of representation—the ability to mold, form, adapt, alter, or modify the representations—applies to all design knowledge, including (a), (b), and (c) of point (1).

The proposed theory of computer support suggests an approach to building software systems that has been prototyped in a system named Hermes. Hermes is a substrate for building design environments to support interpretation in innovative design. Motivated by the analysis of interpretation, Hermes provides the following features to support reuse and plasticity of representations of each of the three characteristics of interpretation, being situated, having perspectives, and using language (see Table 1-2):

a-1. A persistent hypermedia network for storing partial representations of design situations and for browsing among them.

a-2. Efficient mechanisms for revising the representations (multimedia nodes) and modifying their associations (links).

b-1. A perspectives mechanism that organizes specialized or personal ways of filtering out information of interest

b-2. Procedures for switching perspectives or for creating new ones by merging existing perspectives and modifying their inherited contents in the new one.

c-1. An end-user language that provides useful domain terms, rules for critiquing designs, and queries for displaying stored information.

c-2. The ability to modify or generate new terms, critic rules, and queries or to use the language for defining computations.

Table 1-2. Computer-based mechanisms to support interpreta­tion in design.

Although computers cannot understand things the way people do, they can serve as a computational medium to support people’s interpretive processes. The computer support mechanisms listed in Table 1-2 can augment cooperative design in a number of ways, including:

a-1 As a long-term memory or repository for information that was created in past designing and is now available to be shared by designers using the repository.

a-2 As an external memory for representing and revising designs to see how alternative variations appear.

b-1 As a retrieval mechanism for organizing and managing design knowledge and filtering through just what is relevant.

b-2 As a display mechanism to define new personal and shared views of designs.

c-1 As a linguistic medium for expressing knowledge in a canonical form that can be used for computations by the software.

c-2 As a communication medium to generate new knowledge to be shared with others.

A comparison of Table 1-1 and Table 1-2 shows that the mechanisms of computer support are based on the structure of unaided human interpretation. The computer support is intended to extend the power of designers to operate under conditions of “information overload,” in which it is becoming increasingly difficult to work effectively without the use of computers.

Computer support will inevitably change the practices of collaborative design. This need not be considered harmful—particularly in cases where traditional procedures have become inadequate—if important factors like the characteristics of interpretation are preserved and adequately supported. Computational media have the potential for changing the activities of professionals even more than the media of written language did in the past, because of significant opportunities for the computer to play a computationally active role in organizing, analyzing, displaying, and communicating the information. The ways in which design tasks are accomplished will change dramatically as the computer augments and supports designers to do many of the same tasks they have done unaided in the past, like designing and modifying artifacts.

The proposed theory of computer support for interpretation in design goes to the root of the problem of tacit and explicit understanding. Designers approach their task with a background of skills, know-how, and experience that they are generally not aware of as they design but that is a necessary precondition of their work as trained professionals. For instance, architects have the ability to understand the situations people might face in the buildings they design, they know how to sketch and visualize relationships from the perspectives of different concerns, and they move freely between various frameworks or traditions that provide meaningful languages or metaphors for expressing their insights. Computers have no such tacit preunderstanding; they can only retrieve and manipulate what people have already formulated in explicit propositions or drawings. People and computers are not analogous processors of information. If computers are to support human cognition effectively, then these differences must be understood and taken into account.

By describing the transformation of tacit to explicit human understanding, the analysis of interpretation not only clarifies how human cognition differs from computer information processing, but also suggests how computers can support the way people think. Philosophically, the analysis of interpretation provides the key to a theory of people-centered computer support. Technically, the analysis enumerates the functionality needed for computer support of interpretation in design. Practically, it points out that the process of innovative design and the requirements of collaboration generate both the need for computer support and the sources of explicit knowledge that make it possible. For instance, large, multi-person design projects often confront the problem of information overload, where computers are required to manage volumes of technical knowledge. At the same time, these cooperative design processes naturally articulate much explicit knowledge that could prove useful in subsequent computer-supported design work.

The theory of computer support for interpretation in design is presented in three Parts: in Part I, Chapters 2, 3, and 4 develop the analysis of interpretation in design. In Part II, Chapters 5, 6, and 7 draw the consequences of the problem of tacit and explicit understanding for computer support. In Part III, Chapters 8, 9, and 10 describe how the technical features of the Hermes substrate support interpretation in collaborative design.

The analysis of interpretation is developed by reviewing insightful descriptions of design by design methodologists Alexander, Rittel, and Schön (Chapter 2). Characteristics of design enumerated in that review are then used to guide a study of transcripts of a design session involving a task of lunar habitat design (Chapter 3). The design process—as characterized by design methodology and as illustrated with lunar habitat design—is then conceptualized as a process of interpretation by using Heidegger’s philosophy of interpretation (Chapter 4).

The consequences for computer-based design systems are drawn by further developing the analysis of tacit and explicit understanding in design (Chapter 5), and extending it to include a theory of the computer support of interpretation (Chapter 6). This theory is applied to evaluate traditions of software design environments and design rationale systems; useful techniques in these previous systems are explored and their limitations noted (Chapter 7).

The technical description of computer support for cooperative design describes the central functionality of Hermes. It has a hypermedia knowledge-base to support (a) the representation of design situations (Chapter 8). A virtual copying mechanism provides (b) perspectives on design knowledge (Chapter 9). An end-user language is used for (c) articulating formerly tacit understandings in explicit language (Chapter 10)

The order of presentation in the dissertation corresponds to the process of interpretation. First, in the Introduction and Part I a preunderstanding is sketched to provide a starting point for interpreting the problem of computer support for innovative design. A review of design methodology provides a perspective from which to understand design, formed by merging the perspectives of the three design methodologists. A lunar habitat design project provides a concrete design situation for grounding the developing understanding of design. Heidegger’s philosophy provides a language and conceptual framework for talking about interpretation in design. Second, in Part II this preunderstanding is used to explore possibilities for computer support that are opened up by the preunderstanding. This is accomplished partially by drawing out the theoretical consequences in order to extend the analysis of interpretation in design to include a theory of its computer support. It is further accomplished by discovering the achievements and the limitations of previous software systems in providing the kind of support for design that is called for. Third, in Part III the arrived at understanding allows for a discussion of the Hermes system as an explicit illustration of possible responses to the problem of supporting interpretation in design .

Predecessor systems to Hermes (principally Janus and Phidias) were already headed in the direction that Hermes adopts. Discussions of these earlier design environments made frequent reference to Alexander, Rittel, and Schön, for instance, and insisted on supporting rather than automating design. The theory of computer support for interpretation in design presented in this dissertation extends this approach theoretically and practically. Its focus on interpretation situates its people-centered approach unambiguously in an analysis of human understanding. By providing a coherent perspective for viewing systems to support design, the theory suggests principled extensions to the functionality of design environments, such as those incorporated in the Hermes substrate. It provides an explicit language as a basis for a coherent conceptual framework.

Each section in the remainder of Chapter 1 provides an overview of a chapter of the dissertation. The first three sections each provide an argument for interpreting design as a process of interpretation. The other sections draw the implications of this argument for the computer support of design. The three characteristics of interpretation run through all the chapters. Table 1-3 shows the correspondences between the central themes in the different chapters. These correspondences link the theoretical analysis of interpretation to the operational mechanisms that provide computer support for these characteristics. For the sake of simplicity, the table does not indicate that each of the entries involves both reuse of past information and creative modification, however this is true both for the three characteristics of interpretation and for their corresponding software mechanisms, as already shown in Tables 1-1 and 1-2.

Table 1-3. Correspondences among the chapters.

Note that the three mechanisms of Hermes in Chapters 8, 9, and 10 correspond to the three characteristics of interpretation that permeate and structure the dissertation.

1.2.      The Methodology of Design

A central claim of this dissertation is that design can be viewed as fundamentally a process of interpretation. In this interpretive process, elements of the designer’s tacit background preunderstanding are made explicit. The first evidence in support of this claim is a review of the writings of three influential design methodologists. It is argued that their diverse but complementary descriptions of the design process highlight characteristics of what is here called interpretation. They recognize the importance of both tacit understanding and explicit representations, as well as the iterative movement between them. Among the three writers, the dimensions of (a) the situation, (b) perspectives, and (c) language are all stressed. Furthermore, each of these dimensions is recognized to entail both (1) traditions of past knowledge to start from and (2) an ability to revise that knowledge to promote and grasp innovation.

Alexander (1964) pioneered the use of computers for designing. He used them to compute diagrams or patterns that decomposed the structural dependencies of a given problem into relatively independent substructures. In this way, he developed understandings of the design situation for solving a task based on an analysis of the unique design situation.

Later, Alexander (1977) assembled 253 patterns that he considered useful for architectural design, based on an extensive study of successful past designs. These patterns were to be reused and modified to form personal pattern languages for expressing the individual perspectives of different designers. They were schematic enough to be adapted to a broad range of specific design situations.

Alexander felt that the design profession necessarily made explicit the understanding that was “unselfconscious” in traditional cultures in which everyone designed their own artifacts. His structures and patterns were meant to be tools for explicitly representing design situations for “self-conscious” design. However, he always also recognized the need for tacit or intuitive understanding as a basis for good design.

For Rittel (1973), the heart of design was the deliberation of issues from multiple perspectives. In a collaborative effort, each participant may bring different personal interests, value systems, and political commitments to the task. Also, people with different technical specialties or professional skills may contribute to a design. These are actually different kinds of “perspectives.” The theory of computer support in Chapter 6 distinguishes three classes of perspectives that need to be supported:

*    personal or group perspectives

*    technical specialties (e.g., plumbing)

*    domain traditions (e.g., residential kitchens)

However, they all provide the same function of determining what issues will be addressed, what alternatives will be considered, and what criteria will be applied. Because they all determine the organization or relevance of information in a similar way, they can be discussed as one kind of determinant of interpretation and can be operationalized and supported with one software mechanism (a perspectives mechanism).

The important thing for Rittel was not the subjective character of interpretation deriving from its basis in personal perspectives, but the way in which deliberation among perspectives can lead to innovative solutions that would not have arisen without such interaction. Deliberation is an interpretive process in which understanding of the problem situation and of the design solution emerges gradually as a product of iterative revisions subject to critical argument from the various perspectives. This can take place for an individual designer as well if the designer consecutively adopts different perspectives on the issues. Rittel foresaw computer support for this. His idea of using computers to keep track of the various issues at stake and alternative positions on those issues led to the creation of issue-based information systems.

Schön (1983) argued that designers constantly shift perspectives on a problem by bringing various professionally trained tacit skills to bear, such as visual perception, graphical sketching, and vicarious simulation. The designer’s intuitive appreciations shape the problem by forming a subsidiary background awareness of the design task’s patterns, materials, and relationships. By then experimenting with tentative design moves within this tacitly understood situation, the designer discovers consequences and makes aspects of the problem explicit. As this is done, certain features of the situation come into focus and can be named or characterized in a language. When focus subsequently shifts, what has been made explicit may slip back into an understanding that is again tacit, but is now more developed.

Schön (1992) provided empirical evidence for the roles of the situation, perspectives for viewing, and conceptual frameworks in the iterative process of interpretation in design. His experiments showed how the designer uses tacit skills and preunderstandings to uncover unanticipated discoveries, to reflect upon them, and to develop new understandings, new perspectives, and new articulations of the evolving design situation.

1.3.      The Example of Lunar Habitat Design

A second argument for understanding design as a process of interpretation is presented in Chapter 3. Here, a protocol analysis of designers collaborating shows that most of what went on was interpretation.

As part of the research for this dissertation, a study was undertaken of lunar habitat design. Lunar habitat design is a task that is not well understood compared to many other, more mundane design tasks. It is not a routine matter that can be done according to well-formulated rules or by applying available template solutions. Furthermore, it is representative of a broad range of high-tech design tasks. Such tasks typically involve extensive technical knowledge. They seem to call for computer support.

The volume of information available to people is increasing rapidly. For many professionals this “information overload” means that the execution of their jobs requires taking into account far more information than they can possibly keep in mind. The lunar habitat designers here provide a prime illustration of such professionals. In working on their high-tech design tasks, they must take into account architectural knowledge, ergonomics, space science, NASA regulations, and lessons learned in past lunar missions. These designers turn to computers for help with their complex, technical problems. That is why a group of lunar habitat designers initiated the software development effort that led to this dissertation.

Providing the computer support needed by lunar habitat designers is not straight-forward. Designers need to be able to consider wide varieties of experience, professional know-how, technical concerns, and previous solutions that are relevant to their current tasks. However, the problem is not so much one of storing large amounts of information as one of deciding what information to retain that might be relevant to novel future tasks and how to present it to designers in formats that support their mode of work. It is a problem of how to manage the information and present it so that it can usefully serve the design process. The necessary decisions must be made by the designers who are involved with these tasks. Computer techniques for capture and display of information must be under the control of people engaged in the interpretation of the information.

As part of the effort at developing computer support for lunar habitat designers, thirty hours of design sessions were videotaped and analyzed. The designers were asked to design a 23 foot long by 14 foot wide cylindrical habitat to accommodate four astronauts for 45 days on the moon. A protocol analysis of segments of the video recording was conducted.

The analysis of the videotape of the designers’ activities shows that design time is dominated by processes of interpretation, i.e., the explication of previously tacit knowledge in response to discoveries of surprises. As part of the interpreting by the designers, graphical representations were developed for describing pivotal features of the design situation that had not been included in the original specification; perspectives were created for looking at the task in different ways; and language terminology was defined for explicitly naming, describing, and communicating formerly tacit understandings. The definitions of the situated understanding, perspectives, and language continually evolved as part of the design process in an effort to achieve an adequate understanding of the design task and the evolving artifact.

The nature of interpretation and the three dimensions of preunderstanding are illustrated in Chapter 3 with an example from the lunar habitat design sessions. This designing primarily consisted of sketching and discussion that explicated visual and conceptual expressions used for understanding, explaining, and guiding the emerging design. The example analyzed has to do with the tacit notion of privacy and a default perspective on bathroom design related to this notion. The following paragraph briefly summarizes the example.

The designers felt that a careful balance of public and private space would be essential given the long-term isolation in the habitat. This is an important concern that receives limited treatment in official NASA design guidelines. An early design decision proposed that there be private crew compartments for each astronaut. An initial sketch revealed problems with adjacencies of public and private areas, leading to an interpretation of privacy as determining a “gradient” along the habitat from quiet sleep quarters to a public activity area. In the process, the conventional American idea of a bathroom was subjected to critical reflection when it was realized that the placement of the toilet and that of the shower were subject to different sets of constraints based on life in the habitat. The tacit American assumption of the location of the toilet and shower together was made explicit by comparing it to alternative European perspectives. The revised conception permitting a separation of the toilet from the shower facilitated a major design reorganization.

In this way, a traditional conception of “private space” as a place for one person to get away was made explicit and explored within graphical representations of the design situation. As part of the designing process, this concept was revised into a notion of “degrees of privacy”, which facilitated the design process. The failure of the NASA guidelines to provide significant guidance despite a clear recognition by NASA of the importance of habitability and privacy considerations raises the problem of how to represent effectively notions like privacy that are ordinarily tacit. This problem provides the central test case for this dissertation. In Chapter 9, a scenario shows how designers using Hermes can define interpretive critics to evaluate the distribution of public and private spaces in a lunar habitat. A detailed analysis of how these critics are defined in the Hermes language is then presented in Chapter 10.

In this and other examples, the designers needed to revise their representations to enhance their understanding of the problem situation. They went from looking at privacy as a matter of individual space to reconceptualizing the whole interior space as a continuum of private to public areas. The conventional American notion of a bathroom was compared with other cultural models and broken down into separable functions that related differently to habitat usage patterns. The new views resulted from argumentative discussions motivated by design constraints—primarily spatial limitations and psychological factors of confinement. In these discussions, various perspectives were applied to the problem, suggesting new possibilities and considerations. Through discussion, the individual perspectives merged and novel solutions emerged. In the process, previously tacit features of the design became explicit by being named and described in the language that developed. For instance, the fact that quiet activities were being grouped toward one end of the habitat design and interactive ones at the other became a topic of conversation at one point and the terminology of a “privacy gradient” was proposed to clarify this emergent pattern.

1.4.      The Analysis of Situated Interpretation

Chapter 4 presents a third argument for focusing on interpretation in design: computer support of innovative design should be based primarily on an analysis of human understanding. As Norman (1993) puts it, “Without someone to interpret them, cognitive artifacts [like computer support systems] have no function. That means that if they are to work properly, they must be designed with consideration of the workings of human cognition.” The philosophy of interpretation provides just such a consideration.

This contrasts with many previous approaches to computerization of design and to artificial intelligence, which lean toward theories on the natural science model (e.g., mathematical physics), like information theory and predicate logic formalisms. Human sciences (e.g., cultural anthropology or non-behaviorist psychology), however, necessarily center on human interpretation because their subject matter is defined by what people consider to be important and by how people construe things. As one moves from routine design to highly innovative tasks, the distribution of roles in the human-computer relationship shifts more onto the people involved, and it becomes increasingly important to take into account their cognitive functioning.

An initial framework for clarifying the respective roles for computers and people in tasks like lunar habitat design is suggested by theories of situated cognition. Several influential recent books[3] argue that human cognition is very different from computer manipulations of formal symbol systems. The differences imply that people need to retain control of the processes of non-routine design because these processes rely heavily upon what might be called situated interpretation. Computers can provide valuable computational, visualization, and external memory aids for the designers by supporting such interpretation in design.

Situated interpretation, as used here, refers to a view of human understanding as taking place within tacit contexts of background skills, human concerns, and linguistic traditions that provide its grounding. Interpretation is not just a function of a disinterested rational mind, but relies on the interpreting person or people being actively involved with the situation, which includes the artifact being interpreted and supplies the basis for that artifact’s significance. (See Heidegger’s fuller definition of situation below and in Chapter 4.)

Situated cognition theory disputes the prevalent view based on the natural sciences model that all human cognition is based on explicit mental representations such as goals and plans. Winograd and Flores (1986) hold that “experts do not need to have formalized representations in order to act” (p.99). Although manipulation of such representations is often useful, there is a background of preunderstanding that cannot be fully formalized as explicit symbolic representations subject to rule-governed manipulation. This tacit preunderstanding even underlies people’s ability to understand representations when they do make use of them. Suchman (1987) concurs that goals and plans are secondary phenomena in human behavior, usually arising only after action has been initiated: “When situated action becomes in some way problematic, rules and procedures are explicated for purposes of deliberation and the action, which is otherwise neither rule-based nor procedural, is then made accountable to them” (p.54).

This is not to denigrate conceptual reasoning and rational planning. Rather, it is to point out that the manipulation of formal representations alone cannot provide a complete model of human understanding. Rational thought is an advanced form of cognition that distinguishes humans from other life forms. Accordingly, an evolutionary theorist of consciousness like Donald (1991) traces the development of symbolic thought from earlier developmental stages of tacit knowing (e.g., episodic and mimetic memory-based cognition). He shows how these earlier levels persist in rational human thought as the necessary foundation for advanced developments, including language, writing, and computer usage.

Philosophers like Wittgenstein (1953), Polanyi (1962), Searle (1980), and Dreyfus (1991) suggest a variety of reasons why tacit preunderstanding cannot be fully formalized as data for computation. It is too vast: background knowledge includes bodily skills and social practices that result from immense histories of life experience. We are unaware of much of it: these skills and practices are generally transparent to us. It must be tacit to function: the examples of biking, swimming or playing a musical instrument suggest that procedural knowledge at least gets in the way of skilled action if it is explicit. More generally, tacit knowledge is a precondition for explicit knowing: we cannot formulate, understand, or use explicit knowledge except on the basis of necessarily tacit preunderstandings.

The philosophical foundations of situated cognition theory were laid out by Heidegger (1927), the first to point out the role of tacit preunderstanding and to elaborate its implications. For Heidegger, we are always knowledgeably embedded in our world; things of concern in our situations are already meaningful before we engage in explicit cognitive activity. We know how to behave without having to think about it. For instance, an architect designing a lunar habitat knows how to lift a pencil and sketch a line or how to look at a drawing and see the rough relationships of various spaces pictured there. The architect understands what it is to be a designer, to critique a drawing, to imagine being a person walking through the spaces of a floor plan. Such tacit, background skills or preunderstandings of the design situation are necessary prerequisites for being able to design an artifact.

Heidegger defines the situation as a person’s interpretive context—including the physical surround­ings, the available tools, the circumstances surrounding the task at hand, the person’s own personal or professional aims, and social or cultural relations. The situation constitutes a network of signifi­cance in terms of which each part of the situation is already meaningful. That is, the person has tacit knowledge of the situation as a whole; if something becomes a focus, it is perceived as already understood and its meaning is defined by its relations within the situation. Everything is tacitly understood in its relations to other things and to the whole.

According to situated cognition in contrast to rationalist views, to an architect a rectangular arrangement of lines on a piece of paper is not first perceived as meaningless lines that need defining attributes (to be determined by subsequent rational thought). Rather, given the design situation, it is already understood as (say) a sleep compartment for astronauts. The sleep compartment is implicitly defined as such by the design task, the shared intentions of the design team, the other elements of the design, the availability of tools for revising the drawing, the sense of space conveyed by the design, the prevailing NASA terminology. This network of significance is background knowledge that allows the architect to think about features of the design, to make plans for changes, and to discover problems or opportunities in the evolving design. At any given moment, the background is already tacitly understood and does not need to be an object of rational thought manipulating symbolic representations.

At some point the architect might realize that the sleep compartment is too close to some source of potential noise, like the flushing of the toilet. This physical adjacency would come into focus as an explicit concern against the background of relationships of the preunderstood situation. Whereas a common sense view might claim that the sleep compartment and toilet were already immediately and objectively present, and that therefore their adjacency was always there by logical implication, Heidegger proposes a more complex reality in which things are ordinarily hidden from explicit concern. In various ways, they can become uncovered and discovered, only to re-submerge soon into the background as our focus moves on.

In this way, our knowledge of the world primarily consists neither in mental models that represent reality nor in an unmediated and objective access to objects. Rather, our understanding of things presupposes a tacit preunderstanding of our situation. This is analogous to the view of Kuhn (1962), who argues that scientists’ experimental observations presuppose their tacit ability to use their experimental equipment and to apply their frameworks of hypotheses and theory. Only by being already situated in our world can we discover things and construct meaningful representations of them by building upon, explicating, and exploring our tacit preunderstanding. Situated cognition is not a simplistic theory that claims our knowledge lies in our physical environment like words on a sign post: it is a sophisticated philosophy of interpretation.

According to the philosophy of situated interpretation, human understanding develops through interpretive explication involving both (1) preunderstanding and (2) explorative discovery of the situation. In Heidegger’s analysis, interpretation provides the path from tacit, uncritical preunderstandings to reflection, refinement, and creativity. The structure of this process of interpretation reflects the inextricable coupling of the interpreter with the situation, i.e., of people with their worlds. One’s situation is not reducible to one’s preunderstanding of it; it offers untold surprises, which may call for reflection, but which can only be discovered and comprehended thanks to one’s preunderstanding. Often, these surprise occasions signal breakdowns in a person’s skillful, transparent behavior, although one can also make unexpected discoveries in the situation through conversation, exploration, or external events.

A discovery breaks out of the preunderstood situation because it violates or goes beyond the network of tacit meanings that make up the preunderstanding of the situation. To understand what one has discovered, one must explicitly interpret it as something, as having a certain significance, as somehow fitting into an understood background. Then it can merge into one’s comprehension of the meaningful situation and become part of the new background. Interpretation of “something as something” requires a reinterpretation of the situated context if the discovery does not fit into the previously understood situation.

For instance, the lunar habitat designers discovered problems in their early sketches (their representations of the design situation) that they interpreted as issues of privacy. Although they had created the sketches themselves, they were completely surprised to discover certain conflicts among the functions of adjacent components, like the sleep compartments and the toilet. The discoveries could only occur because of their situated understanding represented in the drawings. The designers paused in their sketching to discuss the new issues. First they debated the matter from various perspectives: experiences of previous space missions, cultural variations in bathroom designs, technical acoustical considerations. Then they considered alternative conceptions of privacy, gradually developing a shared vocabulary that guided their revisions and became part of their interpretation of their task. They reinterpreted their understanding of privacy and articulated their new view using the terminology of a privacy gradient.

These themes of being situated, having perspectives, and using explicit language correspond to the three-fold structure of preunderstanding in Heidegger’s philosophy. He articulates the pre-conditions of interpretation as: (a) pre-possession of the situation as a network of preunderstood significance; (b) pre-view or expectations that things in the world are structured in certain ways; and (c) pre-conception, a preliminary language for expressing and communicating. In other words, interpretation never starts from scratch or from an arbitrary assignment of representations, but is an evolving of tentative prejudices and anticipations. (1) One necessarily starts with a preunderstanding that has been handed down from one’s past experiences and inherited traditions. (2) The interpretive process allows one to reflect upon this preunderstanding methodically and to refine new meanings, viewpoints, and terminologies for understanding things more appropriately.

The analysis of interpretation based on Heidegger’s philosophy stresses the role of tacit preunderstanding as the basis for all understanding. Preunderstanding consists primarily of the characteristics of prepossession, preview, and preconception. It also implicitly incorporates the structure of “something as something.” Through interpretation, this preunderstanding is articulated. The resultant explicit understanding can be externalized in discourse. This can be taken further through the methodologies of science to codify knowledge. Each stage in this process preserves the original structure of the preunderstanding. It is because of this structure that metaphors, speech acts, and scientific propositions have the structure they do of something as something, something is some predicate, or something has some attribute.

The process of explication through interpretation forms the basis for computer support by transforming tacit understanding into increasingly explicit forms that can eventually be captured in computer-based systems.

1.5.      Tacit and Explicit Knowledge in Design

Heidegger’s analysis of interpretation must be applied to the realm of design before it can be used as the basis for a theory of computer support of design. Three general problems must be considered:

*    First, although his philosophy is presented in a very general way, Heidegger’s examples come primarily from people’s relations to physical things in the world, rather than to imagined artifacts that they are designing.

*    Second, he stresses that things are always understood on the basis of preunderstandings we already have, which makes it hard to say how innovative design ideas are understood.

*    Third, of course, Heidegger (writing in the mid-1920’s) did not address the issue of computer representations as a form of explicit knowledge.

Chapter 5 accomplishes the application of Heidegger’s analysis to design in three steps.

*    First, it shows that Heidegger’s philosophy can be extended naturally to design.

*    Second, it discusses the problem of application, which addresses the issue of how previously captured knowledge can be adapted to innovative new designs.

*    Third, it spells out a taxonomy of transformations of tacit understanding to explicit knowledge adequate for providing a basis for computer representations of normally tacit design knowledge.

Heidegger’s concept of the situation transfers well to design. As the network of relationships in the understood world, the situation corresponds closely to the set of constraints and adjacencies that are of concern in design and that are sometimes even represented explicitly in design documents. Heidegger’s definition of interpretation as the explication of tacit understanding, involving discoveries, is also applicable to the process of design, in which relationships are explored and discoveries made. Consideration of interpretation in the design context clarifies how breakdowns in action require repair to the tacit underlying understanding of the situation. Although Heidegger’s examples focus on the individual, his recognition of the social dimension and the importance of shared understanding allows his analysis to be extended to design, which is largely collaborative.

Heidegger’s philosophy occupies an important position in the twentieth century recognition that reality is socially constructed. People have access to their world (intentionality) because the world is in many ways a human, social creation. Of course, reality also has an immanence which can contradict our expectations and present surprises, just as we can make discoveries in designs of our own creating. The point is that an understanding of the world or of innovative designs requires the situated interpretation of a person: it cannot be reduced to a set of rules or a computer algorithm. The same goes for knowledge, which encapsulates understanding. To apply knowledge from past cases to a new design, one must apply it within a situated, perspectival, linguistic understanding. That means that computer software for designing should be people-centered and should support the situated, perspectival, linguistic character of human understanding.

Chapter 5 defines tacit as being expressed without words or speech, and explicit as being fully revealed or verbally expressed. It defines a taxonomy of forms of information along the continuum between these extremes and describes the transformations from one form to the next based on Heidegger’s analysis. These transformations are summarized in Figure 1-1. Each transformation involves a reinterpretation of the informational content in a new medium. With that comes a gain in precision balanced by a loss of grounding. As a result of the increased clarity and the change of form, new discoveries are made about the content of the information.

Figure 1-1. Transformations of tacit to explicit information.

The left-hand column lists consecutive forms of information. The right-hand column names the transformation processes from one form to another.

Heidegger uses the term discourse for the fundamental shift to putting one’s understanding into words, even if the words are not yet asserted in speech to be shared with someone. After tacit preunderstanding is articulated in discourse as explicit understanding, this understanding can then be asserted and externalized in spoken or written language (such as documented design rationale). Such knowledge can be further codified in accordance with formal procedures (e.g., scientific methods). These are important transformations for creating widely shared knowledge. The movement from externalized to codified information can go from informal to formal (i.e., capable of being processed by computer). Shipman (1993) discusses this stage of formalization and methods for supporting it within computer-based design environments. This is relevant to the further stages of articulation, which involve computers: capture of the information in computer representations and modification of these representations to adapt them to new requirements. The theory of computer support for design proposed in Chapter 6 suggests that all stages of information articulation can take advantage of computer support. If designing takes place within a computer-based design environment, then designers can use and modify computer representations to support the design process from the start. As Reeves (1993) recommends, the design environment can serve as a medium of communication to support collaboration. In the process, design information can be captured automatically without becoming a burdensome task to be done in retrospect.

1.6.      Consequences for a Theory of Computer Support

The ideas of situated cognition and Heidegger’s philosophy of interpretation stress how different human understanding is from computer manipulations of symbols. These analyses suggest a people-centered approach of augmenting, rather than automating, human intelligence. According to this view, software can at best provide computer representations for people to interpret based on their tacit understanding of what is being represented. Representations used in computer programs must be carefully structured by people who understand the task being handled thoroughly, because the computer itself simply follows the rules it has been given for manipulating symbols, with no notion of what they represent. People (e.g., software designers or software users) who understand the domain must codify their knowledge into software rules sufficiently to make the computer algorithms generate results that, when interpreted by people, will be the desired results. Only if a domain can be strictly delimited and its associated knowledge exhaustively reduced to rules, can it be completely represented in advance (by the software designers) so that tasks in the domain can be automated.

Many tasks like lunar habitat design that call for computer support do not belong to well-defined domains with fully catalogued and formalized knowledge bases. These tasks may require (a) exploration of possibilities never before considered, (b) assumption of creative viewpoints, or (c) formulation of innovative concepts. Software to support designers in such tasks should provide facilities for the designers themselves (as the software users) to create new representations and to flexibly modify old representations. As the discussion of Alexander emphasizes, the ability to develop appropriate understandings of the situation dynamically is critical to innovative design. Because they capture understandings that evolve through processes of interpretation, representations need to be modifiable during the design process itself and cannot adequately be anticipated in advance or provided once and for all. Lunar habitat design is an example of an exploratory domain in two senses: (1) it is a new domain with relatively little in the way of accepted conventional knowledge, and (2) it involves continual innovation to meet novel, over-constrained, politically sensitive mission specifications.

The assumption of the existence (even in principle) of an objective, coherent body of domain knowledge that can be used without being reinterpreted in new situations and from different perspectives is misleading. As Rittel says, non-routine design is an argumentative process involving the interplay of unlimited perspectives, reflecting differing and potentially conflicting technical concerns, personal idiosyncrasies, and political interests. Rather than trying to supply all knowledge in advance, software to support this type of design should capture alternative deliberations on important issues as they arise and document specific solutions. Then, these can be available to support interpretive deliberations. Furthermore, because all design knowledge is relative to perspectives, the computer should be used to define a network of over-lapping perspectives with which to organize issues, rationale, sketches, component parts, and terminology to reflect the different viewpoints designers adopt. That will facilitate the retrieval of information relevant to a particular interpretive stance.

As Schön emphasizes, design relies on moving from tacit skills to explicit conceptualizations, and on the ability to reformulate the implications in linguistic expressions. Additionally, design work is inherently communicative and increasingly collaborative, with high-tech designs requiring successive teams of designers, implementors, and maintainers. Software to support collaborative design should provide a language facility for designers to develop a sharable vocabulary for expressing their ideas, for communicating them to future collaborators, and for formally representing them within computer-executable software. An end-user language that provides an extensible domain vocabulary, is usable by non-programmers, and encourages reuse and modification could help provide support for designers trying to express their interpretations..

Heidegger’s analysis of interpretation says that new interpretations are based on preunderstandings developed in the past or handed down by tradition. In this sense, it is likely that the information designers need most when they reflect on problems may have previously been made explicit at some moment of interpretation during past designing. Accordingly, one promising strategy for accumulating a useful knowledge base is to have the software capture knowledge that becomes explicit while the software is being used. As successive lunar habitats are designed on a system, issues and alternative deliberations can accumulate in its repository of design rationale; new perspectives can be defined with their own modified representations, terminology, and critic rules; and the language can be expanded to include more domain vocabulary, conditional expressions, and query formulations.

This is an evolutionary, bootstrap approach, where the software can not only support individual design projects, but simultaneously facilitate the accumulation of expertise and viewpoints in open-ended, exploratory domains. This means that the software should support designers in formalizing their knowledge when it becomes explicit. The software should reward its users for increasing the computer knowledge base by performing useful tasks with the new information, like providing documentation, communicating rationale, and facilitating reuse and modification of relevant knowledge.

The theory suggested by the analysis of interpretation in design is diagrammed in Figure 1-2. As the cycle of interpretation proceeds, driven by the needs of designing and collaboration, explicit knowledge that is generated can be captured by the computer support system. The computer system relies on a combination of stored representations (for representing situations, defining perspectives, and articulating language expressions) and plasticity (for tailoring the existing representations to the requirements of the specific design process). This combination makes support of interpretation in design possible and simultaneously drives an evolution of the stored knowledge base.

The theory proposed in Chapter 6 views the computer as a design medium. It is a multimedia device capable of representing the diverse forms of information used in design: text, graphics, pictures, pen sketches, numbers, voice, animation, and even video. It can use all these media to externalize design concepts and to store them for future use, serving as a medium of externalization and long-term memory. This means it can be used as a medium of communication among team members and a medium for embedding an artifact’s design history within the design of the artifact itself—Reeves (1993) argues for the role of such a medium in supporting collaborative work.

Figure 1-2. The theory of computer support for interpretation in design.

The cycle of human interpretation (illustrated on the top) is mirrored by a cycle of evolution of the computer knowledge base (below), that uses captured explicit knowledge to support future interpretation.

The uses of the computer as designing medium mentioned in the preceding paragraph are primarily passive uses. The impact of written language on civilization shows that even passive media can be powerful. However, the computational power of the computer suggests using it as an active medium as well. Certainly, numerical computations can be left to the computer: calculate square footage of designs or total their costs. But information can also be made dynamic, with representations modified on the basis of the state of other parts of a design. Furthermore, the information stored in the computer can be managed by it, perhaps organizing and displaying information based on a structure of defined perspectives. A language can make the system programmable by designers, so they can adjust displays to their changing needs. Part III will show how Hermes accomplishes this by means of a computationally active form of hypermedia, that integrates a perspectives mechanism and an end-user programming language.

One of the most powerful consequences of designing in a computational medium is the possibility of integrating all the relevant information. An example of this is the mechanism of interpretive critics (see Fischer, et al., 1993a). It is an extension of specific critics (Nakakoji, 1993). Specific critics are critiquing rules that analyze the representation of a design situation and optionally display a message depending on the results of the analysis. For instance, if two appliances are closer than they should be in the design of a habitat, a critic might display a warning, suggest looking at related design rationale issues, and show similar stored designs that avoid the problem. The specific critics are dynamically computed based on the design specification that has been entered into the design rationale. The critic thus integrates information about the graphical design, the textual rationale, the computational critic rules, and other designs. It does this in a way that supports the needs of the designer without providing overwhelming amounts of information. Interpretive critics are even further embedded in the contexts of design because they can be defined differently in different interpretive perspectives. Their active behavior depends on the current perspective and the way in which terms in the language are defined in that perspective. They use the language that is being used for the particular design, they are tied to the currently active perspective, and they analyze the represented situation.

The view of computer support systems as computationally active communication media is consonant with a liberatory view of the role of computers in society. Feenberg (1991) argues that the expert system approach based on technical rationality philosophy is profoundly anti-democratic and that an alternative approach to computers as communicative media is needed to give people control over their lives:

Systems designed for hierarchical control are congruent with rationalistic assumptions that treat the computer as an automaton intended to command or replace workers in decision-making roles. Democratically designed systems must instead respond to the communicative dimension of the computer through which it facilitates the self-organization of human communities, including those technical communities the control of which founds modern hegemonies. (p.108)

The theory of computer support presented in this dissertation pursues the democratic alternative, founding it on a respect for the irreducible nature of human interpretation.

The key is control. Computer systems are sophisticated tools for exerting control of information. As powerful as they are, computers have no understanding of the information they manipulate. Even in autonomous AI systems, all the interpretation is done by people—typically by the programmers who set up the system and the users who view the output. Innovative design is an arena in which the interpretation cannot be done in advance because this design requires understanding and interpretation at every step. Therefore, the role of computers in non-routine design must be to support designers. Human designers must retain control over (a) how things are represented, (b) which things are stored together, and (c) what terms are used to articulate ideas. Unless this control is vested in people who can use their interpretive skills, questions concerning what information might be relevant in a given context or in the future remain intractable for all but carefully delimited, well-understood, completely codified domains. The only heuristics proposed for the management of design knowledge are those tacitly followed by traditional design practice: (1) that knowledge represented, organized, and articulated in the past may be useful in the future, and (2) that designers will need to use their powers of interpretation to modify and apply reused knowledge in unique situations. (The problem of application addresses the fact that every situated, perspectival, linguistic understanding is unique and yet must be interpreted as similar to other cases; it is discussed in Chapter 5.)

The theory of computer support provides a principled basis for designing a computer system to support innovative design in such tasks as lunar habitat design. Before exploring the ideas suggested for such a system, the existing tradition of design environments is considered. This is a tradition of computer systems supporting the augmentation of human design efforts. It provides a basis upon which new ideas can be developed through extensions that are guided by the theory.

1.7.      Previous Software Systems for Design

For thirty years now, at least since Alexander (1964), efforts have been underway to use computers to support design. Much work in the area of computer support for design has concentrated on two approaches that will not be explored here:

*    Providing stand-alone tools for drafting and modeling, where the computer system has little or no representation of the semantics of what is being designed—e.g., so-called “computer aided design” (CAD).

*    Automating the design process, where the computer is given a specification of a problem and is expected to produce a design with minimal interaction with a human user—e.g., an expert system for design.

Although these approaches have proven useful for certain tasks or within restricted domains, in general they have been shown to be quite limited. Winograd & Flores (1986) and Dreyfus & Dreyfus (1986), for instance, have argued that expert systems are in principle essentially limited when it comes to tasks like creative design. They have based their arguments largely on Heidegger’s philosophy and other ideas that are discussed in this dissertation. Rather than duplicating their line of criticism, Chapter 7 will draw their positive implications for building software systems that can support innovative design.

There have always been some researchers who sought ways to use technology to augment human problem solving (e.g., Bush, 1945; Engelbart, 1963), rather than to model, simulate, or replace it. More specifically, there is a tradition in design methodology and design rationale capture efforts, going back to Rittel and his associates (Rittel & Webber, 1973; Kunz & Rittel, 1984) that advocates the use of computer-based design systems as cognitive aids for human designers.

Recent work in this tradition is reviewed in Chapter 7 and used as a starting point for designing a system to support interpretation in design. In particular, the design environments that will be reviewed (Janus, Modifier, Phidias) are domain-oriented in the sense that they try to embody generally accepted knowledge of their specialized design domains. In contrast, the domain-independent design rationale capture systems (Krl, Pie, Drl) focus on capturing and displaying potentially opposing perspectives on design issues. By synthesizing ideas from these different systems, the new approach will extend the notion of domain-orientation to support multiple interpretations of the domain as well.

The consequences of the theory of computer support for interpretation in design developed in Chapter 6 motivate and guide the survey of previous software systems. Established techniques implemented in the computer-based design assistants are reviewed and their limitations are critiqued on the basis of the theory. While mechanisms for representing situations, defining perspectives, and using language are found in some of these systems, the plasticity and integration of these mechanisms are quite limited. In many ways, these systems retain principles from expert system theory and are not oriented toward supporting interpretation in design even when they happen to provide some mechanisms that could be used for that.

Janus (Fischer, et al., 1989) is a design environment combining graphical and textual representations of the design situation. It introduces a multi-faceted architecture that includes a palette of design components for building graphical representations of kitchen layouts, a catalog of stored design cases, an issue-base of design rationale, and a daemon mechanism for active critics. This system provides an important model of a design environment. Its lack of support for users to create new representations is recognized and addressed by a successor system named Modifier.

Modifier (Girgensohn, 1992) defines all the knowledge representations with parameterized property sheets. Then it provides a user interface to these system internals. While it offers extensive support for the user to modify representations, this still involves the user in modifying Lisp expressions, altering hierarchical inheritance trees, and generally having to be concerned with system internals. Thus, it supports the user (with extensive help text, examples, checklists, and even critic rules concerning modifications) to engage in tasks of maintaining a sophisticated software system rather than supporting the user in interpretive tasks of design. Another problem with Modifier is that it provides no mechanism for organizing modifications into alternative versions to support personal and shared versions.

Several systems for knowledge representation and design rationale capture propose the use of multiple perspectives, a mechanism that this dissertation recommends. Krl (Boborow & Winograd, 1977) presents a sophisticated formal language for knowledge representation that incorporates a mechanism for perspectives. Pie (Boborow & Goldstein, 1980; Goldstein & Boborow, 1980a, 1980b) develops the ideas of Krl further as the basis for a design environment for software development. Drl (Lee, 1990; Lee & Lai, 1991) explores issues in design rationale capture using languages based on Rittel’s Ibis as well as Krl and Pie. These systems provide invaluable experience in designing languages for knowledge representation and design rationale, and in using perspectives mechanisms. However, their implementations lack the generality called for in certain ways. Furthermore, they are not particularly appropriate to the kind of hypermedia structure that seems useful for representing a broad diversity of design information. They provide important examples and recommend useful principles for the kinds of languages and perspectives mechanisms useful in supporting design. The lessons from these systems are combined in Chapter 8 with two design criteria: (1) the implementations should be appropriate to a hypermedia structure of knowledge representation and (2) end-users should be able to revise and extend the vocabulary of the language and the structure of the system of perspectives.

Phidias (McCall, et al., 1990) is another design environment like Janus. It does not include as many components or a critiquing mechanism, but it does demonstrate the utility of a query language for users to define displays of design rationale. The Phidias language has a number of important features: it is designed for navigation of hypertext and it is based on several syntactic characteristics of English. Vocabulary in the language can all be defined by users, so it supports adaptability. Phidias uses a form of hypertext that has a fine granularity; thus textual displays of design rationale, for instance, may be computed dynamically through the use of queries defined in the language. The Phidias language provides a good starting point for the design of a computationally powerful language that is appropriate to hypermedia and that can support interpretation.

In response to the shortcomings of previous systems, an integrated software prototype named Hermes is proposed. Hermes is a persistent hypermedia substrate for building design environments to support interpretation in design. Its mechanisms operationalize the positive design principles of the analysis of interpretation and the theory of computer support for interpretation in design.

1.8.      Hypermedia in the Hermes System

In Greek mythology, Hermes supported human interpretation by providing the gift of spoken and written language and by delivering the messages of the gods. As part of the research for this dissertation, a prototype software system named Hermes has been designed to support the preconditions of interpretation (a) by represent­ing the design construction situation to support prepossession, (b) by provid­ing alternative perspec­tives to support preview, and (c) by including a language to support preconception.

Hermes supports tacit knowing by encapsu­lat­ing mechanisms corresponding to each of the preconditions:

*    Interpretive critics (Fischer, et al., 1993b) are used for analyzing the design situation, which is represented in arbitrarily complex hypermedia data structures. These critics are expressions in the Hermes language that perform an analysis of the current state of some representations and then optionally display a message. The evaluation of the critic expressions or rules is dependent upon the currently active interpretive perspective, which determines the versions of the expression, of its constituent terms, and of the representations being analyzed.

*    Named perspectives (Stahl, 1993b) organize and manage alternative sets of information relevant to different purposes. By switching to a new perspective by selecting its name from a list, a designer can change how the representation of the situation appears, what interpretive critics are active, and in general what contents of the hypermedia network are “visible” from the viewpoint.

*    Language terms (Stahl, et al., 1992) define computations across the knowledge base. While these expressions can be arbitrarily complex if viewed in complete detail, they are typically constructed in a series of stages. At every stage, the components of the term’s definition can themselves be given names.

With each of these mechanisms, complexities are hidden from the user by being encapsulated in named objects. These complexities can gradually be made explicit upon demand so the designer can reflect upon the information and modify it. Together, these and other mechanisms make Hermes a computationally active medium in which designers can do their work.

Hermes is a knowledge-representation substrate for building computer-based design assistants like the Lunar Habitat Design Environment (Lhde) shown in Figure 1-3. It provides a hypermedia structure for designers to build representations of design knowledge.

Figure 1-3. Arranging sleep compartment bunks using Hermes.

Windows shown (left to right) include a dialogue box for browsing the hypermedia content, a selection from the design rationale issue-base, a critic rule’s message, a graphical sketching area, and a button for changing interpretive perspectives.

The network of knowledge corresponds to the design situation. Multi-media nodes of the knowledge representation can, for instance, be textual statements for the issue-base, CAD graphics for sketches, bitmap images to illustrate ideas, or language expressions for critics and queries. The inter-linked hypermedia structure facilitates browsing by designers. It can also be used to support associative memory (Hinton & Anderson, 1989) or case-based dynamic memory (Schank, 1982; Kolodner, 1984). All displays are defined by queries that dynamically assemble arbitrary collections of multimedia items. For instance, the Design Rationale window in Figure 1-3 shows the textual issues, answers, and arguments that resulted from a query that was executed by a user’s request to see the “discussion” of a previously viewed issue.

The hypermedia knowledge representation structure of Hermes is designed to facilitate the representation of design situations and to encourage their tailorability. Its generalized node and link structure models the network character of the situation as a network of inter-related, pre-understood significances and their associations. Its object-oriented implementation allows for the integration of information in different media—reflecting the need to bring together many forms of information in design. It provides graphics for sketching, text for issue-bases or design rationale, and other media for annotations to support the exploration of represented situations. All the media and mechanisms are designed to maximize plasticity of representation. The Hermes hypermedia structure incorporates a perspectives mechanism for managing and viewing all information and an end-user language for defining queries for displays, as discussed below.

Special emphasis is placed on the synergistic integration of the hypermedia, perspectives, and language mechanisms in the Hermes substrate. Definitions of perspective hierarchies and language expressions are stored in the hypermedia network so they can be browsed and modified like all other information. By using nodes of the hypermedia network to define the names of perspectives and links to determine the inheritance relationships among perspectives, the Hermes system can support annotation of these nodes to store information related to the purpose or origination of the perspectives. Similarly, the nodes that define terminology and expressions in the Hermes language can be linked like a semantic network (Quillian, 1967).

In turn, the definition of the hypermedia structure itself incorporates both perspectives and language expressions. Instead of having a fixed content in some medium, nodes can have their content defined by the evaluation of an expression in the language. Nodes and links can be conditional upon some computation defined in the language and involving other nodes and links. Furthermore, hypermedia information to be displayed is always dynamically computed in the currently active perspective—even language expressions can have different effects in different perspectives. In these ways, node contents can be dependent upon the state of other data in the hypermedia network. The interactions of the integrated hypermedia, perspectives, and language provide significant control and malleability for the designer. Design environments built on this substrate can have many features that support interpretation in design with consistent abilities to represent knowledge and to tailor the representations.

1.9.      Perspectives in Hermes

Hermes includes a perspectives mechanism for organizing all knowledge represented in the system. This mechanism is general and can be used to define a variety of different kinds of “perspectives” for categorizing information and for organizing inheritance of information among perspectives. For instance, hierarchies of perspectives can be defined for technical specialties (e.g., plumbing, ergonomics), knowledge domains (kitchen design, partial gravity design), worldviews (Bauhaus, austere missions), specific designs (i.e., cases), individual preferences, shared team decisions, and experimental “what-if” versions. New perspectives can merge information from multiple existing perspectives and then modify the information as seen through the new perspective without affecting it in the original perspectives. This can facilitate periodic, non-disruptive reorganizations of the knowledge base as it evolves.

The perspectives mechanism of Hermes helps to support the collaborative nature of design by multiple teams. Drawings, definitions of domain terms in the language, computations for critic rules, and annotations in the issue-base can be grouped together in a perspective for a project, a technical specialty, an individual, or a team. A new perspective can be defined to archive a version of a design for historical purposes so it will not change as team members continue to work on new versions. Every action in Hermes takes place within some defined perspective, which determines what versions of information are currently being accessed. Perspectives can collect knowledge according to various categories. For example, there can be perspectives for individual designers or design teams; for technical or professional specialties; for traditional or cultural domains; for specific projects; or for historical versions of projects.

Since information in Hermes is always viewed through a perspective, switching perspectives can support the deliberation of alternative approaches. By redefining in different perspectives the same graphic objects or linguistic terms used in conditionals, queries, and critics, one determines how things will be displayed (interpreted) differently in different perspectives. Thus, as shown by a scenario in Chapter 9, critics in a “privacy perspective” might analyze habitat layouts using a concept of privacy gradient defined in that perspective, whereas the same critics would in effect have different definitions in other perspectives and would therefore produce different results. The interpretive critics for privacy that are used in the scenario are analyzed and explained in detail in Chapter 10 as a case study in use of the language.

The approach of Hermes supports communication among designers. The representations of the design situation may include documentation of design rationale by specifying resolutions of issues in an issue-base. For lunar habitat design, such documentation is contractually required by NASA. Requirements traceability and clear communication of rationale are necessary for a design to move from the original design team to subsequent groups for approval, technical elaboration, mock-up, and eventual construction. Documentation is notoriously difficult to produce. Design rationale is most effectively captured when it is an explicit concern. Formulations developed in the Hermes language by designers in the midst of designing can supplement the situation representations, stating for the benefit of future designers looking at their work what aspects were originally considered important and what rules of thumb were developed then. Viewing the design from the original team’s perspective preserves their interpretation, while subsequent groups can define their own modified perspectives. Individuals in work teams can share ideas, viewpoints, and definitions by using group perspectives that inherit from and modify the contents of their different personal perspectives.

1.10.     The Hermes Language

Hermes features a language for designers to use. The language is defined as a series of subset languages to facilitate learning by new users. First it should be noted that previously defined terms and expressions are used most of the time. These are simply selected from lists of relevant terms. Then there is a beginner’s version of the language that is very similar to the Phidias language, which proved easy to use for non-programmer novice users. This level of the language suffices for defining or modifying most common terms and queries. An intermediate level provides access to virtually all features of the language except those related to graphics. Finally, an advanced level can be used for graphics-related tasks, like defining interpretive critics. Most system displays and component interfaces are defined in the language, so they can be modified through use of the language.

The Hermes language defines domain vocabulary for referring to represented objects and their associations (the nodes and links of the hypermedia). It also provides expressions for stating queries to define displays and for stating rules to critique designs. The expressions fall into three major syntax categories: (a) definitions of lists of nodes, (b) expressions for filtering out nodes not meeting stated criteria, and (c) operations to traverse various kinds of associations. These support the situated, perspectival, and linguistic character of interpretation by naming representations of things in the design situation, filtering out objects for display based on viewing criteria, and providing expressions for exploring associations. Objects in each of these categories can be either (1) reused or (2) refined by combining expressions in useful ways. This defines the six primary syntactic classes in the language; four other classes provide auxiliary terms and features. The syntactic classes are listed with brief descriptions in Table 1-4.

The language provides a tacit form of language usage for non-programmers. Most of the sequential processing is kept implicit, due partially to the declarative form of the language structure. Also, expressions that were originally figured out explicitly are given names in domain terminology. In Figure 1-3, for example, the user clicked on an issue about sleep compartment bunks and then chose the “Predicate” (Computed Association), discussion. This predicate was already defined to produce a hierarchy of issues with their answers and arguments. The user did not have to be concerned with the recursive structure of this hierarchy or its iterations through multiple links. All of those computational matters were implicit in the definition of the predicate. The user could simply select the predicate by name. This example of choosing “discussion” from a list of predicate names in Figure 1-3 is typical of how the language is used in Hermes. Even when one is creating a new expression, one selects syntax options in dialogue boxes and selects predefined terms from lists. This minimizes the need to remember syntax and terms, prevents many kinds of errors, and avoids the impression that one can simply use free-form English to define expressions.

Table 1-4. Syntactic classes of the Hermes language.

The Hermes language pervades the system, defining mechanisms for browsing, displaying, and critiquing all information. This means that designers can use the language to modify and refine the representations, views, and evaluations of all forms of domain knowledge in the system. All vocabulary in the language is modifiable by the designers. Every language expression (and every component of a larger expression) can be encapsulated by a name, so that many statements in the language can be defined with common terms from particular design domains. Considerable effort was put into the design of the language to make the appearance of expressions as easily interpretable as possible. Chapter 10 presents many examples and discusses the techniques used to achieve a readily interpretable appearance. This is just one way in which the language is designed to support tacit usage. Much of the knowledge that people must explicitly use in writing programs in conventional programming languages (assignment, variables, functions, quantification, etc.) has been hidden from the user in the Hermes language (see Chapter 10 for a detailed description of this). The power of these mechanisms is available through the language, but designers need not think in terms of the computational mechanisms. However, when it is necessary for a designer to explore the definition of a user-defined expression in the language in order to understand it more explicitly, this can be done.

Combined with the perspectives mechanism, the language permits designers to define and refine their own interpretations. This allows the Hermes substrate to extend systems beyond the domain-oriented approach of the knowledge-based design environments that Hermes grew out of, by supporting multiple situated interpretations of the domain. That is, the previous systems pre-defined most domain knowledge in a single, generic knowledge base. However, all representations are relative to human interpretation and interpretation is perspectival. Hermes lets designers reinterpret linguistic expressions of knowledge already in the system and store them in appropriate perspectives. This retains the relationship of design knowledge to interpretive perspectives. It also replaces the notion of a single body of domain knowledge (whether fixed or evolving) with a system of multiple perspectives on the domain. Furthermore, this extension encourages inter-related or relevant knowledge from diverse domains to be brought together in specific perspectives.

1.11.     Conclusion

The analysis of situated interpretation argues that only people’s tacit preunderstanding can make information meaningful in context. Neither people nor computers alone can take advantage of the huge stores of data required for many design tasks; such information is valueless unless designers can use it in their interpretations of design situations. The data handling capabilities of computers should be used to support the uniquely human ability to understand. The theory of computer support for interpretation in design suggests that several characteristics of human understanding and collaboration can be supported with mechanisms like those in Hermes for refining representations of the design situation, alternative perspectives, and linguistic expressions. The theory provides a coherent framework for a principled approach to computer support for designers’ situated interpretation in the age of information overload.

In elaborating the argument of the previous paragraph, this dissertation seeks to make three kinds of contributions: to a philosophy of interpretation, to a theory of computer support, and to a system for innovative design.

*    It makes a philosophic contribution by clarifying the foundations of situated cognition theory in Heidegger’s philosophy of interpretation and extending that philosophy through an analysis of interpretation in design and through a theory of computer support for interpretation in design.

*    It makes a contribution to computer science by arguing that systems to augment human skills in innovative design should be oriented toward providing support for the processes of interpretation.

*    It makes a practical contribution by prototyping three crucial mechanisms for design environments: a hypermedia substrate that integrates a perspectives mechanism and an end-user language.

These contributions reflect a belief that our age calls for alternatives to a technical rationality philosophy, an expert system approach to computerization, and a view of the designer as an isolated and unaided subject.