Joseph A Goguen
Dept. Computer Science & Engineering
University of California, San Diego
It is difficult to design systems that satisfy users; failure is common, and even successful designs often overrun time and cost. This motivates user-centered design methods. But users often don't know what they need or else cannot articulate it (due to tacit knowledge), and also are often not aware of key impacts of organizational context on how they work. This motivates using ethnographic methods. However these can be slower and more expensive, and can still fail. We argue that values are key to the promise of socially sensitive design. Algebraic semiotics provides a rigorous notation and calculus for representation that is explicitly value sensitive, while compassion supports both better analysis and better ethics in design. Together with discourse-based value discovery methods and iterative design, these enable a method that we call value-centered design; some case studies are discussed.
Few would deny that design today lacks mature principles and methods, is more an art than a science, and often fails to deliver satisfying results. The following are among the challenges that a mature design discipline would have to overcome:
Much of the literature in design and management seeks theory-based, replicable methods for solving problems, in the style of mathematics, physics, or (at least) engineering. However, the rapid evolution of fads and buzzwords, and the ubiquity of spectacular failures (e.g., Enron and Windows1, to take just one example from each area) attest to the lack of significant progress. Two often cited obstacles are: giving precise formulations of realistic problems; and giving realistic metrics for the adequacy of solutions. I suggest that in general, these obstacles cannot be overcome, and in fact, that they are not even genuine problems, but rather are artifacts of a misguided reductionist program that attempts to apply successful methods from the hard sciences to domains having completely different characteristics.
Instead of longing for the stable, grounded world that seems promised by reductionist science, and seeking reductionist solutions, managers and designers should learn to live in the groundless semiotic world of social reality. Although certain specific problems can be reduced to predictable routine methods, management and design operate in open social environments, which implies that their most important problems are not reducible. Philosophers including Heidegger and Nishitani have developed deep insights into the groundlessness of the human condition, and how to live with it, as discussed here later. I believe that extreme reductionist tendencies are harmful, because they raise expectations that cannot be fulfilled, thus leading to disappointment, and fueling further cycles of hope and fear. For example, (Heidegger 1977) gives a powerful and very influential discussion of the dark side of technology, and (Burstall 1991) describes
some of the ways in which our involvement with computing may bias our overall point of view, leading to additional confusion and pain in our lives, both our working and our personal lives.This article suggests that, although extreme reductionism remains sterile, semi-formal approaches that take account of social processes can be valuable. It argues that values are the key to unlocking the mysteries surrounding the enormous opportunities and enormous dangers of contemporary technologies. Claims are often made that better engineering will solve the problems, or better management, or further progress in basic technical areas such as distributed algorithms, user interface design and ontologies, and no doubt all this can help, but until we understand not only what users want (as in requirements analysis, defined in (Goguen 1994) as the reconciliation of what is desirable with what is possible, so that a useful system can actually be built), but much more fundamentally, why they want, i.e., their fundamental underlying motivations, progress will be heavily interleaved with failure, and will continue be very expensive when it does occur, since users are notoriously unreliable at saying what they want, and traditional requirements engineering is very error-prone, as shown by the shockingly common failures of large software systems. A design method called algebraic semiotics is sketched, combining ideas from sociology and computer science.
We also reject extreme relativism, which claims that all social phenomena and human values are equally valid. However, it is not claimed that merely denying both absolutism and relativism solves any hard problems in design or management. Instead, our view that there are no definite foundations for such disciplines leads us to explore groundlessness, the lack of any definite foundation, and to discover that groundlessness can spark compassion, ethics, and even better design.
The research described in this paper draws on insights and methods from ethnomethodology, activity theory, discourse analysis, symbolic interactionism, etc., but it is focused on practical results, rather than ideological purity. CSCW (Ackerman 2000, Dourish 2001, Robinson and Bannon 1991) and related work in sociology of technology (Agre 1995, Bowker 1994, Bowker and Star 1999, Star 1989, Star 1989a) suggest relating activity to institutions of practice within particular communities, while ethnomethodology2 suggests viewing context as situated interaction, rather than attempting to reify it with precise (allegedly context independent) descriptions (Sacks 1992). Such insights naturally motivate the idea that values are inherent in all situations, and indeed, are what give them the coherence that allows us (whether as participants or as observers) to see them as situations. The formal side of the research also uses algebraic abstract data type theory as a basis for semiotic theories. However, this is not the place for a detailed exposition of the mathematics involved; for this, readers may wish to consult (Goguen 1999).
In design, it is often important to view some signs as representing other signs. This motivates the systematic study of representation, including what makes some representations better than others. Although transformations are fundamental in many areas of mathematics and its applications (e.g., linear transformations, i.e., matrices), transformations of signs seem not to have been previously studied in semiotics; in algebraic semiotics, semiotic morphisms are mappings between such spaces which preserve various significant properties (Goguen 1999). Just as semiotic systems are theories rather than models, so their morphisms translate from the language of one semiotic system to the language of another, instead of just translating the concrete signs. This may seem indirect, but it has important advantages over more common approaches based on set theoretic models, in that it is open, in allowing multiple models, as well as in permitting new structure to be added at later times.
Algebraic semiotics also provides precise ways to compare the quality of representations, and to combine representations, such that conceptual blending (in the sense of cognitive linguistics (Turner 1997, Fauconnier and Turner 1998, 2000)) is a special case. A number of algebraic laws have been proved about operations for combining representations, constituting the beginnings of a calculus of representations. Case studies for this theory include web-based displays for mathematical proofs that integrate motivation, background and explanation with formal details (Goguen 1999a, Goguen and Lin 2001), and information visualization (Goguen and Harrell 2003).
In many real world examples, not everything can or should be preserved, so that semiotic morphisms must be partial. For example, the table of contents of a book preserves structure and the names of major parts, but completely fails to preserve content (which is what makes it useful). The extent of preservation gives a way to compare the quality of semiotic morphisms (Goguen 1999). It is notable that semiotic spaces and semiotic morphisms are qualitative rather than quantitative, in that they concern structure, and their quality measures are partial orderings, rather than linear numerical scales. Design is the problem of massaging a source space, a target space, and a morphism, to achieve suitable quality, subject to constraints. This formulation applies just as well to managing an organization as it does to designing a website. In addition, various design principles can be stated and justified justified, including the following:
The situated abstract data type (SADT) notion arose from noticing many situations in which users recognize quite different complex signs as ``representing the same thing'' (Goguen 1994). For example, sports events often involve elaborate data, presented in different ways in different contexts, e.g., on TV screens, in newspapers, and on real-time scoreboards at the event. Moreover, specific events, e.g., the legality of a particular play, may be negotiated by various combinations of players, referees, coaches, rule bodies, etc. Thus, both producing and interpreting these displays are social achievements. Three other examples discussed in (Goguen 1994) are the so called ``waterfall model'' (see Figure 1), a normative process model of how software engineering should be done, a taxonomy of requirements engineering methods (see Figure 2), and the value hierarchy of a small corporate recuirtment firm (see Figure 3).
The lack of any definite ground for phenomena follows from coemergence. The groundlessness of the human condition is discussed in depth by (Nishitani 1982), who points out (following his teacher Heidegger) that much of the recent history of Western thought can be seen as a progressively refined questioning of absolutes. Among the responses to this questioning, two extremes are identified: nihilism, which is absolute relativity, the denial of any meaning; and absolutism, which is the denial of the questioning. Such absolutism may take the form of dogmatism, fundamentalism, or extreme reductionism. Moreover, there tends to be an unstable oscillation between these two extremes4. Thus groundlessness is not a stable, fixed state; indeed, it makes even less sense to reify groundlessness than other things. Nor is it passive. All living systems are dynamic, constantly rebalancing their state in order to achieve equilibrium within their environment.
Nishitani says there is a middle way which avoids the extremes of both nihilism and absolutism, as well as the unstable oscillation between them, by accepting groundlessness as a basis for being. The experience of groundlessness, and a path based upon it, have been described in many traditions with phrases such as ``dark night of the soul'' and ``cloud of unknowing.'' Results of practicing this middle way are said to include openness, compassion, and harmony with nature; joy, strength, and peace are also said to result. This is advocated in (Varela, Thompson and Rosch 1991) as a fruitful approach to cognitive science. Here, I suggest it also makes sense as an approach to design, dwelling in neither relativism nor reductionism, and drawing energy and inspiration from silence.
Arguments against rule based approaches to ethics are well known, e.g. (Johnson 1993); they are similar to arguments against reductionist approaches to other areas, e.g., management and design. Fixed rules can never anticipate the complexities of the human condition, and in any case require interpretation, while second order rules (such as Kant's categorical imperative) require even more interpretation than first order rules (like ``Thou shallt not kill''). Although rules can certainly be very valuable as guidelines, as argued above with respect to design, the usual philosophical problems of reductionism arise when they are elevated to universal principles. A perhaps surprising result is that human nature is sufficient for ethical behavior, once it has been sufficiently refined. Groundlessness then becomes a ground for authentic behavior, including genuine ethics, as well as effective and creative design; indeed, from this perspective, effective behavior cannot be separated from genuine ethical behavior. Compassion is the central value here, and other values include an appreciation for groundlessness, and the avoidance of both nihilism and absolutism. Clearly, other, more specific, values arise in the many specific situations of life.
According to ethnomethodology, when events occur in a social context, members apply their concepts and methods to account for what happened: the technical term accountability refers to this process, which simultaneously produces new assertions, and expresses what the group values by highlighting some aspects while ignoring or downplaying others. Thus information and values do not exist as abstract ideal entities, but rather emerge interactively through accountability in actual situations; everything in social life attains meaning through the relations of accountability in which it participates, and therefore always has an inherent ethical component. Moreover, information always arises through the particular relations of accountability that tie it to a particular social group and the work done in a particular context to produce particular interpretations. A foundational approach to values based on these ideas is developed in (Goguen 1997), where the following definition is given:
An item of information is an interpretation of a configuration of signs for which members of some social group are accountable.To summarize, groups, values, and information are coemergent in the sense that each produces and sustains the others: groups exist because members share values and information with one another; values exist because they are shared and communicated within groups; and information arises as groups with shared values cope with a dynamic world. None of these three should be considered more basic than the others. Values are also a necessary presupposition of analysis, because it is members' accounting, based on their shared values, that renders their concepts and methods visible to analysts. (Jayyusi 1991) puts this point as follows:
What emerges from both Garfinkel's and Sacks' work is the understanding that all communicative praxis presupposes, and is founded in, a `natural' ethic - an ethic, that is, which is constitutive of, and reflexively constituted by, the natural attitude of everyday life.The sense of reflexivity here is the same as that of coemergence above.
But all this theory leaves open the question of how a working designer (or manager, or systems analyst) can actually discover values. For this, we can draw on practical ethnography (participant observation, field notes, audio and video recordings, etc.), work of (Labov 1972) on the embedding of evaluation in stories, and of (Sacks 1974) on interactions of speakers and audience during the telling of jokes. Case studies (Goguen 1996) with small groups show that value systems can be obtained by using (Labov 1972) and (Sacks 1974) plus discourse analysis, to extract value-laden discourse fragments, and the KJ method (Kawakita 1975) to classify them. For example, part of a ``value tree'' expressing the value system of a small corporate recruitment firm (Figure Figure 3 from section 3.6 of (Goguen 1996)). Later case studies have probed the values implicit in database interfaces (Goguen 2003) and in mathematical proofs (Goguen 2004).
It is interesting to look at the four examples in (Goguen 1994) in light of the definition of information above, because each is socially situated in a different way. Sports scores are constructed by players, referees, etc., while the value hierarchy consists of discourse fragments from firm members, as chosen and arranged by the analysts, the taxonomy of methods is purely an analysts' construction, and the waterfall model is a traditional diagram found (in varying forms) in many texts, and often imposed by contract on software engineers.
Actor network theory (Latour 1987, 1988) can contribute to system design through its emphasis on the whole network of relations that constitute, support, and use a system. For example, not only end users and their local environments should be included, but also equipment manufacturers, supplies of communications infrastructure, system maintainers, etc. And the notion of immutable mobile can shed light on how different SADTs (and even different parts of the same SADT) can be socially situated to different degrees, in different ways (Goguen 1994). Finally, the negotiations that occur along edges between actants crucially involve value translations, i.e., the values in the two nodes must somehow be resolved in a mutually satisfactory way, through negotiation and then exchange.
In brief, our proposed method calls for first extracting values, and using this information to determine key SADTs; these should be expressed as semiotic systems, which are then implemented; for user interfaces, defining semiotic morphisms is a useful intermediate step, and here it is also important to examine natural situations, rather than what users say they might do in imagined situations (Goguen and Linde 1993), though the latter can be useful for discovering values. Appropriate actors should be involved in all processes as much as feasible, and iterative development should be employed throughout. For example components could be procured in partnership with manufacturers, and interested end users could be supported in customizing applications, or even programming their own. It is hoped that exploring such ideas will lead not only to systems that better satisfy users and their managers, but that also better satisfy all the actors involved, are ethically produced and used, and enhance society as a whole.
Most papers by the author are available online and are easily found by search engines.