This section will locate Heidegger’s philosophy historically in order to highlight its contribution. It will show how his analysis addresses the key issues underlying a theory of computer support:

1. Heidegger’s emphasis on the priority of the tacit over rationalist philosophers’ stress on the explicit should be understood through a recognition of his place in the history of philosophy. The view that the world is interpreted—socially constructed—and not simply known is a result of modern philosophy.

2. Tacit preunderstanding is intimately related to the issue of intentionality. This issue, in turn, is critical for a theory of computer support, providing the ultimate argument for a people-centered approach to computerization.

3. Given that computer support relies upon the appropriate application of representations to innovative situations, the problem of application arises as a central issue. Gadamer has addressed this issue as integral to Heidegger’s philosophy of interpretation.

4. The problem of application leads to the related and more general problem of relevance, which pervades attempts at computer support for design.

1. The social construction of reality. The interpretation of Heidegger’s philosophy in this dissertation bears directly upon the problem of tacit and explicit understanding in computer support of cooperative design. His philosophy makes particularly clear the basic ways in which (fundamentally tacit) human interpretation differs from (necessarily explicit) computer representation. This is in contrast to the rationalist philosophy of functionalism[1], which is usually assumed to provide a basis for AI. Functionalism proposes that human cognition and computer computations share a common functional structure, i.e., that mind is adequately modeled as software running on the brain’s hardware. If one reviews the history of philosophy leading up to Heidegger, one can see clearly the roots of AI’s belief that computer representations could correspond to the structure of human understanding. One can also see that this belief is misleading and based on antiquated philosophical positions.

In philosophical terms, the problem with the traditional AI approach is that it assumes that a single interpretive framework can, at least in theory, be formulated that will be adequate for all representations within a given domain. That is, most influential AI systems define a representation for the domain of knowledge they are dealing with and then proceed to compute solutions to problems in the domain by manipulating elements of the representation. In contrast, this dissertation argues that problem solving is typically situated in ways which require the representation of the problem to be interpreted based upon the interpreter’s unique situation, perspective, and language.

The assumption that problem-solving intelligence is based on mental representations that can be known a priori can be traced back to Kant. In his Critique of Pure Reason (1787), Kant argued that the human mind imposes a set of elements or categories on sense data in order to understand the external world. These elements or categories of space, time, quantity, quality, etc. that Kant derived were claimed to be universal a priori. The idea in AI is to capture such objective categories in a representation scheme that could be determined in advance to be valid of necessity, in analogy with the example from mathematics or physics that within certain geometric domains all objects can be represented with Cartesian spatial coordinates. Kant’s approach was revolutionary in that he located the source of the objective representations or categories that we use to make sense of our world in the human mind, rather than in some divine or natural order. The objectivity of these categories derived from the view that all minds necessarily used the same categories.

However, Kant’s claim for the universality of our interpretive framework was soon criticized by Hegel, who argued that reason evolved through history. In the Phenomenology of Mind (1807), for instance, Hegel laid out the logical stages of reason's development in terms of a review of human history. So, for Hegel, our interpretation of reality depends upon the developmental stage reached by reason in our times. While there is a logic to the unfolding of reason, it happens historically (contingently). Therefore, the appropriate representations for understanding things change with socio-historical conditions.

Marx, in turn, tied this idealist history to the social development of production relations in Capital (1867). The basic categories for representing social phenomena within capitalist society—private property, exchange value, labor time, etc.—were themselves products of the historical development of capitalism and had to be interpreted through a hermeneutic process by people living within that society in order to avoid ideological conceptualizations. (See Stahl, 1975a, for a detailed discussion of the hermeneutic character of Marx’ method.)

Subsequent writers in the human and social sciences have shown many other aspects of how our representations and conceptualizations of reality are necessarily determined by our situation. Freud (1917), for instance, related an individual’s understanding to the person’s formative history of inter-personal relationships. Anthropologists and other theorists show how interpretation is necessarily embedded in rich traditions of social, cultural, and personal histories.

Finally, Heidegger (1927) generalized these historical perspectives by saying that we always understand from within the situation in which we find ourselves already thrown as a result of our past. The social, cultural, and personal traditions are part of the background that we bring to interpretation as part of our preunderstanding of the world. But Heidegger also added a second important dimension to this critique of Kant. Our interpretive perspective, he argued, is not simply a matter of categories that can be made explicit and stated in propositions. More fundamentally, it is a matter of understanding what it means to be a person and what it means for other things to be encountered in the world. This background knowledge is fundamentally tacit. It is not only tacit in fact (most background knowledge has never been expressed explicitly), but also in principle (tacit knowledge is the necessary foundation for having explicit knowledge at all). Dreyfus (1985) claims that Heidegger was the first in the history of philosophy to point out the tacit nature of pre-understanding.

It is only in terms of our ontological pre-understanding—which can be seen in the intentionality of our actions, in our grasp of linguistic meaning, in bodily adeptness, and in our interpersonal skills—that we can in the first place make things explicit and formulate propositional knowledge. Our understanding of our world, of artifacts in it, of ourselves, of other people, and of problems we have meeting our goals are structured by skills, preconceptions, and traditions that make up a social construction of reality. From the historical nature of understanding and its basis in tacit pre-understanding it follows that understanding develops through the hermeneutic circle of interpretation, in which the categories of understanding cannot be taken as pre-given but must evolve out of preconceptions, the situational context of meaning, and the process of iteratively interpreting the artifacts of interest.

The notion that our perception of reality is a social construction that fundamentally involves acts of interpretation that are essentially structured by our socio-historical context has had a profound impact upon contemporary thought and has driven the critique of traditional, rationalist outlooks.[2] As Resnick (1991) points out, both Mead (1934) and Vygotsky (1978)—two of the most important analysts of the social basis of human understanding—proposed that mechanisms of individual thought are best conceived as internalizations of ways of interacting socially with other people. Extending the ideas of Hegel and Marx, Mead and Vygotsky claimed that to understand the psychological development of an individual one must understand the social relations in which the individual has developed and operated. Resnick (1991, p.2) concludes, “as Vygotsky (1978) and Mead (1934) have independently suggested, social experience can shape the kinds of interpretive processes available to individuals.”

2. The problem of intentionality. Given the complexity and subtlety of the social situatedness of human categories of understanding, the representations proposed for AI systems look primitive and rigid indeed. Even if large amounts of commonsense background knowledge could in principle be represented in a computer system, as proposed by the CYC project (Smith, 1991), there are three major limitations to computer systems carrying out the interpretive tasks autonomously:

* The “background knowledge” for interpretation consists largely of procedural skills and ontological understandings that cannot effectively be made explicit. For instance, people know how to interact with broad ranges of artifacts (e.g., specialized tools) and how to behave in cultural settings (which involve recognizing the intentions of other people). They can identify different kinds of beings and are able to interact with them appropriately.

* Interpretation is not an algorithmic process. Although we know that interpretations are conditioned by various factors in the situation, we cannot say that a certain interpretation will arise given certain inputs. Interpretation seems to be an emergent phenomenon from a holistic context. Heidegger’s analysis argues that interpretation is a response to an open-ended set of preconditions and situational factors, but it gives no suggestion of causal effects that could be programmed into an autonomous computer system.

* The problem of intentionality probably presents the greatest barrier to defining an autonomous computer system for interpretation. Searle (1980) convincingly argues that computer software does not (and never can) understand the semantics of what is represented symbolically. Even a thorough cognitivist (functionalist) like Fodor (1981) must concede that the symbol systems of programs must be interpreted by people.

One useful way of stating the problem of intentionality is as the “symbol grounding problem” (Harnad, 1993). This refers to the fundamental principle of model theory, that regardless of the formal syntactic relations among symbols in a model, their truth or meaning depends upon a mapping to things in the real world. This mapping is not part of the model itself, but is a matter of the human interpretation of the model. Even if one takes a functionalist view of human thought and hypothesizes that thought takes place by the manipulation of formal or formalizable symbols, one must in addition assume that the thinking person has grounded the symbols of thought in some kind of understanding of their meaning.

The term intentionality has the same implication as the term grounding. They both indicate that when a person uses a word, sentence, or symbol that refers to something, then that person “intends” the thing referred to. In other words, the person’s understanding of the word is “grounded” in the thing. Very few philosophers have much idea about how this grounding takes place. Searle makes vague references to biology. Marx would locate the grounding in social practice. Wittgenstein speaks of a “form of life.” For Heidegger (1927 and 1975; see also Dreyfus, 1991), the structure of being-in-the-world provides the solution to the problem of intentionality. The fact that people are in-the-world in Heidegger’s sense means precisely that they have direct, meaningful, semantic access to—are grounded in—things in the world. The situation is the network of the understood things with which one is more or less involved. The disclosure of the world as preunderstood is what makes interpretation possible. This is very different from a simplistic argument that knowledge is often “in the world” rather than “in the head.” Whether we are understanding an artifact in our physical environment or one represented mentally, we rely on preunderstandings that are grounded in our interpretive situation.

Computers lack being-in-the-world. They merely manipulate ungrounded symbols. As Searle (1980) argued, even if computers are placed in robots that move among and interact with things in physical space, they lack intentionality of those things. This means that when computers are used in tasks like innovative design that involve interpretation, they cannot accomplish the entire task autonomously, but can at best support people in the required interpretations.

Computers lack intentionality. They can only manipulate explicit, ungrounded symbols; they have no tacitly-based sense of the semantics of the formal symbols. This has been identified in the present dissertation as the fundamental problem of tacit and explicit understanding that must be addressed by a theory of computer support for interpretation in design. Suchman (1993) has formulated this problem as a lack of access by computers to “semantic resources” and has agreed on its centrality. She summarized her book (Suchman, 1987) as an attempt to locate the “sense-making ability for machines in the limits of their access to relevant social and material resources, and identify the resulting asymmetry as the central problem for human-machine communication” (Suchman, 1993, p.73).

The problem of intentionality or symbol grounding underlies the problem of tacit and explicit understanding. The asymmetry in the relationship of people to computers—the fact that people have intentional understanding but computers do not—means that computers can only support the interpretive processes of people. This means that (at least within application domains like innovative design) a theory of human-computer interaction should be framed as a theory of computer support for (human) interpretation. People’s intentional grounding is, according to Heidegger’s analysis, primarily a matter of tacit situated understanding. Computers, on the other hand, can only operate with explicit symbolic representations. This poses the core problem for a theory of computer support: how the computer’s manipulation of explicit symbols can support people’s fundamentally tacit understanding.

3. The problem of application. Decontextualization of knowledge presents problems for the subsequent application of that knowledge in new contexts of interest. Because hermeneutic theory claims that all interpretation is situated in concrete circumstances, the problem of application of knowledge is an important issue. In particular, the theory of computer support of interpretation must address the question of how the explicit, formalized, and decontextualized information that can be provided by computer systems can be applicable to the human tasks of tacit interpretation that this information is supposed to support. For instance, in Section 3.2 this problem arose in the context of how to apply specific patterns from Alexander’s pattern language to particular decisions in the design of a lunar habitat.

Gadamer (1960) addresses the problem of application as a central issue for his hermeneutic theory of interpretation. Although Gadamer is primarily interested in the human sciences and bases his discussion of application on examples of ethics, law, and theology, his characterization of the role of application in interpretation has broad generality. Schön (1983) makes similar arguments concerning the application of scientific principles in design and engineering.

Generalizing from his analysis of Aristotelian ethics, Gadamer (1960) concludes that application is not a secondary phenomenon of understanding, but an essential determinant of understanding as a whole from the start. That is, a textual statement in ethics or some other subject matter of interpretation cannot be interpreted in the abstract first and then subsequently applied to the situation of the interpreter:

The interpreter dealing with a traditional text seeks to apply it to himself. But this does not mean that the text is given for him as something universal, that he understands it as such and only afterwards uses it for particular applications. Rather, the interpreter seeks no more than to understand this universal thing, the text; i.e., to understand what this piece of tradition says, what constitutes the meaning and importance of the text. In order to understand that, he must not seek to disregard himself and his particular hermeneutical situation. He must relate the text to this situation, if he wants to understand at all. ( p.289 / S.307)

Granted, historical texts arose within situations that are different from the situation of the current interpreter. This is particularly clear in stories from the Bible or legal case law. Here the moral or precedent of the story was originally situated in a context that could be removed by thousands of years and vast cultural distances from the person who tries to understand it now. But for Gadamer, a religious proclamation is not to be understood strictly as an historic document, but is to be taken in a way that exercises its religious effect upon the interpreter. Similarly, a legal case is not simply an historic fact, but needs to be made concretely valid as a precedent through being interpreted in a contemporary context. Gadamer claims, “The text, whether law or gospel, if it is to be understood properly, i.e., according to the claim it makes, must be understood at every moment, in every particular situation, in a new and different way. Understanding here is always application” (p.275 / S.292).

The term application may be misleading because of its rationalist innuendoes. Gadamer is not talking about taking a decontextualized meaning and applying it to some set of particular conditions by somehow adjusting this pre-given meaning the way one thinks of applying a scientific law to a practical problem by adjusting parameters or taking into account confounding factors like friction. As discussed in Section 4.3, according to Heidegger understanding always takes place within the preconditions of prepossession, preview, and preconception. Application of a text to an interpretive situation in Gadamer’s sense means that the text is necessarily interpreted within the preunderstanding of the current interpreter. This preunderstanding includes an anticipation of what the text is all about. For instance, if we are reading a text from the Bible, then our background knowledge and prejudices concerning the Bible come into play. These include the results of a long history of biblical interpretation and religious traditions through the ages, which has sedimented in our preunderstanding. So, for Gadamer, our “openness to the text” always includes placing its meaning in relation to the whole of our own understandings.

In this sense of application, the problem becomes not one of somehow adjusting a pre-given meaning to our circumstances, but of making sure that our preunderstanding provides access to the text as something that transcends (i.e., can surprise) our preunderstanding of it. This is the role of interpretation: to start from a preunderstanding and to go beyond it on the basis of it. This involves a process of critiquing the assumptions of the preunderstanding in terms of the text (as revealed by that preunderstanding): “Methodologically conscious understanding will be concerned not merely to form anticipatory ideas, but to make them conscious, so as to check them and thus acquire right understanding from the things themselves” (p.239 / S.253). This is why interpretation must be a critical reflection upon its presuppositions. The restructuring of the network of significance (the situation) that takes place in interpretation takes place on the basis of the anticipatory preunderstood situation but questions its adequacy in the face of discoveries made of the text as disclosed by that preunderstanding. This dialectical process of anticipation and discovery—and not some objective viewpoint—provides the foundation for the validity of interpretation. Thus, validity and rigor of interpretation are situated in the process of application.

4. The problem of relevance. The problem of application is related to the larger question of relevance. Given a task—whether a design task or a task of textual interpretation—the question arises as to what past experience is relevant to the accomplishing of that task. Once the relevant past experience has been selected, it can then be applied to the task at hand.

There are basically three ways a computer system can “know” what information is relevant to a given design situation: First, there are often useful heuristics that can be programmed into a system for use in strictly delimited domains. Second, people can be in control of crucial aspects of the system’s decision making and can use their human interpretive powers to determine what is relevant. In this case, the computer may be able to provide support for the person’s decisions and it can store representations of the decisions for future reuse. Third, the computer can present these stored past decisions for a person to approve reusing in the current case.

In general (excluding the narrowly confined domains where appropriateness can be algorithmically defined in advance), the judgment of what is relevant to a particular task at hand requires the tacitly-based judgmental skills that require the involvement of people. As suggested above, the decision of relevance involves carrying out to some extent the process of interpretation in which experiences recalled from the past are applied to (interpreted within) the current situation. Being based on tacit preunderstanding, this process cannot be carried out in explicit computer algorithms. Furthermore, as already discussed, the judgment of relevance relies upon an understanding that is intentionally grounded in being-in-the-world with the artifacts of the current task and of the past experience. Without human intentionality and interpretive powers, questions of relevance are intractable. The explicit nature of computerized knowledge means that computers may be able to support human judgments of relevance, but they cannot replace them. The following chapter explores how computers can support interpretation in domains of non-routine design such as lunar habitat design.

[1] This position was most prominently formulated by Putnam (1967), although he has more recently (1988) renounced functionalism and moved much closer to Heidegger in the sense that he recognizes the ultimate necessity of founding any formalism upon unformalizable human interpretation.

[2] The entrenched rationalism of AI is just starting to be subjected to such critique: see the collection of articles in Floyd, et al. (1992), based on a 1988 conference on Software Development and Reality Construction.

5.2. The Social and Human Grounding of Interpretation