7. Some Case Studies
The first four sections below apply some of the concepts and theories that we have been studying, especially ANT, while the last gives some practical motivation for studying them, and outlines a method for applying them.
In Database Metatheory: Asking the Big Queries, Christos Papadimitriou (a former UCSD CSE faculty member who moved to Berkeley) applies falsifiability and notions from Kuhn to computer science, and in particular, to database theory. His main claim is that theoretical CS is now in a crisis. Although this is certainly true in some sense, due to the phenomenal growth and fragmentation of computer science today, I do not think it is true in Kuhn's more technical sense; part of the problem is that Papadimitriou seems to be using the term "paradigm" for phenomena that are much more fine-grained than intended by Kuhn. The example that Papadimitriou gives of a paradigm shift in the database community, namely the introduction of the relational model, seems very good, and I also like his Figure 1, as a graphical summary of Kuhn (though I doubt that Kuhn would have approved).
The two diagrams in Figure 2 can be considered actor-network diagrams, although Papadimitriou doesn't use (and doesn't know) this theory; his use of properties of random graphs in this connection is very clever. One could extrapolate a fascinating conjecture from his graph-based discussion of the theory/practice split in the database community, namely that normal science correlates with having a single well-connected "random" graph, while crisis correlates with having several major well-connected subgraphs, between which there are only weak links. However, Papadimitriou's discussion would have been better if the graphs in Figure 2 had some genuine empirical content, instead of being "unspecified," i.e., made up by the author.
Papadimitriou also poses the interesting question of what constitutes falsifiability in computer science, and more generally, in a "science of the artificial," but fails to reach a conclusion. Personally, I have doubts about both the notion of falsifiability and that of a "science of the artificial," and so am not very surprised at the inconclusive result here.
7.2 Four Case Studies for Actor Network Theory
In Traduction/Trahison - Notes on ANT, John Law (Department of Sociology, Univeristy of Lancaster, UK) discusses four case studies that apply ANT to various situations, in various ways. The paper is well written, and his comparative discussion is at least as interesting as the case studies themselves, revealing much about the evolution and somewhat elusive nature of ANT.
The first case study discussed by Law was done by Madeleine Akrich, and it concerns technology transfer, a subject on which much has been written. The use of ANT allows an insightful formulation of this study, showing that much more gets "translated" during the so called "transfer" than we might expect. This constitutes a counterexample to naive technology diffusion theories, which claim that technologies transfer simply by spreading out through some milieu, an assertion which is really a form of technological determinism. The work of Akrich shows that instead of diffusion, there can be complex interactions between the source network (in this case, in Sweden), the target network (in this case, in Nicaragua), and the technology itself (a wood waste compacting machine), during which all three may be changed in significant ways. In this case, it is the technology itself that undergoes the most significant alteration, a finding that is somewhat against the spirit of the "canonical" (i.e., original) actor-network theory of Paris in the 1980s. The alterations are the result of the compromises needed to adapt to the values of the new environment; moreover, the network also undergoes significant alterations, even replacing some major actants by others - e.g., instead of compacting wood waste, the new machine in the new network compacts cotton plants after they have been harvested; this alteration is due to the difficulty of recruiting waste wood in the Nicaraguan situation, and the availability of cotton plant waste.
The second study discussed by Law, done by Charis Cussins (while she was at UCSD), concerns a medical procedure called "in vitro fertilization." As Law points out, Cussins pays close attention to inconsistency, and in particular, she questions whether objectification (being treated as an object, e.g., in a medical procedure) is necessarily a bad thing, as is often assumed without much thought. Law also points out that Cussins' approach differs in some significant respects from canonical ANT, which tended towards making things appear consistent, by finding a place for them in a network, other actants, and events. Cussins' study also shows that time can play a dialectical role (involving a "thesis," "antithesis," and (perhaps) a final "synthesis" or resolution), with network relations changing not only over time and actor, but sometimes reversing in motivated ways, as actors look back, re-evaluate, and change their values. Even what is said to exist (the ontology) can change over time. Cussins emphasizes the work involved in making all this happen. The work needed to make things happen is a common theme in ANT, but some of the "things" involved here are unusual; in particular, studying the work done to change, even reverse, some key values, challenges ANT orthodoxy.
The third study discussed here is by Vicky Singleton, and concerns another medical procedure, the Cervical Screening Procedure (CSP), in the context of British socialized medicine. Singleton points out a large number of ambiguities and ambivalences in the attitudes of both doctors and patients about this procedure, and then argues that these are not a problem for the CSP, but on the contrary, serve to strengthen it. For example, a woman may feel more comfortable with a doctor who is not highly authoritative and authoritarian. This version of ANT differs significantly from canonical ANT in not speaking of enrolling actors and locking them into continuous chains of translation. Instead, it speaks of the utility of actors changing roles and values, and of the utilities of the inconsistency of values, and even of exclusion and conflict, in maintaining the network.
One conclusion that Law draws from this and the Cussins study is that in many cases, it is not possible to tell any grand narrative, or single coherent (possibly heroic) story, of what happened to/in/with a given actor-network, even when the network is successful; instead, it is necessary to tell many little stories, which in general are not consistent with each other, and which in particular may embody different values. Law then "goes meta" (as computer scientists like to say), and claims that ANT now has a similar status, in that it can no longer be told as a single story, but instead has evolved into a set of no longer mutually consistent stories. (See also the discussion of master narratives in How things (actor-net)work: Classification, magic and the ubiquity of standards, by Geoffrey Bowker and Susan Leigh Star, where the point is made that they work by excluding what does not fit.)
This discussion is reminiscent of postmodernism in the version of Lyotard, in his book listed in the recommended book section of the class homepage (there are many different mutually inconsistent versions of postmodernism, some of which seem to me a bit crazy, but the general idea of the more moderate notions is that many different world views can and should coexist). Lyotard speaks of local language games (in roughly the sense of late Wittgenstein), i.e., of narrative systems that are not in general mutually consistent; these of course embody value systems that also in general are not mutually consistent, and that sometimes may not even be internally consistent.
Law's fourth case study, by Annemarie Mol, concerns the diagnosis and tratment of arteroschlerosis in the Netherlands (though there is little doubt that the situation is similar in most developed countries). This study perhaps departs the furthest from canonical ANT, because it suggests that no single coherent pattern may actually exist in the data of the study. Law goes so far as to say
We are in the business of making the objects of our study. Of making realities, and the connections between those realities. Of making the realities that we describe.This is exactly the kind of extreme position that Sokal wants to discredit: that the social scientists are creating the reality they study, not nature itself, nor even the scientists (or in this case, doctors) that they are studying. However, in my view, Mol's study only warrants the weaker conclusion that in some cases there may not be any single coherent story, not even a coherent pattern of inconsistency, like that found by Singleton, but rather there may be just an on-going process of forging links and enduring clashes; and given Law's somewhat loose writing style in this paper, it is not clear that he would himself subscribe to the extreme position of the above quote - in fact, my own view is that he is probably just presenting it as one more possible evolutionary outgrowth of ANT.
Law uses variants of the term ontology to describe the several variants of ANT that arise here, including ontological choreography, a term introduced by Cussins for the work of creating coherence, and ontological patchwork for situations like those described by Mol. He concludes that ANT is diasporic (i.e., spreading out), being "translated" ("traduction" in French), and in the process, being "betrayed" ("trahison" in French), i.e., changing in ways of which the originators of ANT might not approve. His final conclusion is that this is a good thing. I agree, for the reason that growth and change demonstrate vitality. As Kuhn's theory says, any good theory will continue to adapt through periods of normal science, and will inspire revolutionary revisions after a period of crisis (see Section Section 5 of the class notes). The paper of Law would seem to suggest that ANT is now in a period of crisis; personally, I consider such a judgement premature.
Social scientists are often interested in the sociology of sociology, and in particular, in applying whatever theory they are working on to itself; this is of course a reflection of their own values. The word reflexive is commonly used for this phenomenon, though computer scientists would probably rather say "recursive" or "self-referential." The paper by Law, and to a lesser extent the paper by Bowker and Star discussed in Section Section 7.3 below, demonstrate this tendency, and in particular, show an interest in the sociology of actor network theory, and in applying actor network theory to itself. The results are interesting, though at times perhaps a bit difficult to untangle. Reflectivity in various guises is an important theme in the 20th century arts and humanities, as well as science, and can be seen, for example, in the famous drawings of Escher and short stories of Borges; the play Six Characters in Search of an Author by Luigi Pirandello is also well known in the theatre community, and there are many other examples. In computer science, I have seen language design guidelines recommening that every feature possible should be made recursive (which seems to me very foolish advice). In fact, excessive reflexivity seems to be poor form in all the arts, sciences, and humanities, although moderate amounts can be very healthy.
Let us now be a bit reflexive ourselves, and consider how ANT and Kuhn's ideas might illuminate each other, using Law's paper as a basis. The technology in Law's first case study, on technology transfer from Sweden to Nicaragua, might at first seem like a typical example of "normal science," involving an adaptation to new conditions. But a closer examination reveals phenomena that are far from those considered by Kuhn, such as the pest Amphiserus Cornutu, and the seasonal schedules of Nicaraguan farm workers. We can claim an analogy between the adaptation of scientific theories to new data, and the transfer of technology to new environments, but more than that would be difficult to justify.
Law's fourth case study, of the diagnosis of arteriosclerosis, seems a good illustration of pre-paradigmatic science, because it does not form a coherent whole, and medical diagnosis can reasonably be considered a kind of science, even though the actants involved are far more heterogeneous than would be entertained by Kuhn, and the scope of this activity is less than in Kuhn's exemplars. The second and third case studies in Law's paper seem to involve in a central way concepts that are too far from Kuhn's way of thinking for there to be any deep similarities.
Now let's try to look at ANT itself in Kuhnian terms, which though it requires some broadening of Kuhn's notion of science, seems quite reasonable. Is the diversity revealed by Laws' case studies a crisis, or normal science, or pre-paradigmatic science? Or is perhaps a revolution happening? Despite pronouncements by Latour that ANT is dead, I am inclined to say that this is normal science, with concepts and theories adapting and evolving in the light of new data, despite the fact that these concepts and theories are far more vague than in physics, and that close examination reveals many actants that Kuhn would not have considered, such as professional societies, manifestos, and death pronouncements.
By way of summary and synthesis, although I do feel that there is much of value in Kuhn's work, comparing it carefully with ANT produces a sense of its being overgeneralized, of lacking attention to the details of how science is achieved, including the actual work of science, its infrastructure, and its institutions. Moreover, we have seen that there is considerably more ambiguity, conflict, instability, diversity, confusion and even chaos than Kuhn's framework seems to admit, and that this occurs even in normal science.
In fact, ANT provides more refined, more nuanced reflections of Kuhn's ideas. Instead of a crude classification into pre-paradigmatic, normal, crisis, and revolution, ANT would consider in detail which actants are enrolled, and to what extent; it would also consider the degree to which various translations are successful, at various times. Instead of a dramatic revolution, we can look for many different kinds of change, some of which may be very slow, yet still yield a very different final state, others of which may affect only small parts of a large network. Similarly, Kuhn's notion of retrospective reinterpretation is too absolute and total; in ANT, reinterpretation is merely retranslation, which may occur to a certain extent in some, but not all parts of a network, and may happen in different ways, at different rates, at different times, or even not at all.
So we conclude that the phenomena described by Kuhn do occur, but are relatively rare, extreme forms of much more common phenomena that are often much more complex and subtle. Kuhn does not provide a language for describing such phenomena, but they are a basic aspect of ANT. On the other hand, ANT does not adequately address processes of change in networks, and in particular does not provide any classification of kinds of change, let alone a classification that is more refined than Kuhn's. Therefore Kuhn's ideas and terminology remain useful, though they must be applied with caution.
7.3 Classification, Standards and Databases
In How things (actor-net)work: Classification, magic and the ubiquity of standards, Geoffrey Bowker and Susan Leigh Star (technical report, from the University of Illinois at Champaign-Urbana in 1998 - though both are now at UCSD) give a most interesting discussion of classifications and standards, which shows that treating them with ANT allows addressing political and ethical issues, in part by making infrastructure non-transparent. This paper considers the Nursing Intervention Classification (NIC), an attempt by nurses to develop a new classification system for their work that better represents their values, such as comforting patients. The NIC is a move in the struggle by nurses to fit better into a healthcare environment dominated by HMOs with their heavy emphasis on financial accountancy. Two sample categories from the NIC are "hope installation" and "humor," where the latter is defined as "facilitating the patient to perceive, appreciate and express what is funny, amusing, or ludicrous in order to establish relationships." This clearly involves a very different value system from that of the HMOs, or the doctors, or even the patients.
From a computer science perspective, I would highlight the ways in which this paper raises important issues about the role of databases in healthcare (and other areas). Databases are significant for this study, because they mediate so much of modern healthcare. Classifications and standards are important, because standard representations of classified items are needed in order to enter information into a database. For this reason, the classification of work activities became a key area for debate between nurses and hospital administrators: even if some administrators believed that something not represented in any database was important, they would still have difficulty in justifying this belief unless relevant figures could be included in the reports that are read by administrators higher up the "chain of translations." However these reports are generated by application programs that take their data from the hospital databases, so that they cannot include details about the work of nurses if they are not in these databases.
Many facets of an organization's life are reflected in its databases, including the non-representation of certain features. The reason the nurses wanted to get their concerns into the hospital databases is that there is a tendency to regard only what is explicitly represented in such databases as "real," with the rest being implicit, or in the terminology of Bowker and Star, relegated to an invisible infrastructure. That which is not explicitly represented will have trouble getting the attention that it may deserve. On the other hand, some groups of nurses oppose the NIC on the grounds that not being represented gives them more flexibility, also noting that classifications and standards provide a basis for more detailed and oppressive monitoring of their work, and that high status groups, like doctors, are in general subject to less detailed description and supervision.
There is also a tendency for database intensive organizations to resist change, and hence to freeze a status quo, because it can be very difficult to change the structure of a large database once it has been deployed; hospital databases are some of the best examples one can find of inflexible legacy systems. The difficulty of changing database structure also implies that any associated classification schemes and standards will be difficult to change. Moreover, mechanistic and reductionist views of organizational operation tend to be reinforced, because they are easier to implement in a database and its application programs for analysis and reporting, than would be the more "holistic," "humane," or "ecological" views that the nurses might tend to favor; for the same reason, quantification is reinforced.
7.4 Information Technology "on the Fringe"
In the late summer of 1999, I participated in a panel discussion at a conference in Buenos Aires, Argentina, on information technology in developing nations. The panel was entitled "Information Technology on the Fringe," and during my short presentation, I used ANT to clarify (I hoped) some terms that are often used in discussing this important issue, which is in many ways similar to the "digital divide" issue which is currently the subject of much discussion in the US. Clearly, the main motivation for addressing these issues is an ethical concern for disparities in the level of development of information technology and its social effects.
It turned out that the word "fringe" referred here to what are often called "developing" countries, and sometimes called "second world" countries, among which Argentina may be included. (It is interesting to notice that all these terms are euphemisms, used instead of something crude like "poor country," and that "fringe" is a new euphemism.) However, the term "fringe" is suggestive, and can be given a fairly precise meaning, by saying that it refers to geographical areas where the density of links in the relevant network is low-ish, but not very low (being very low would bring us into "undeveloped" or "third world" countries). So Silicon Valley, with its very high connectivity, is definitely not fringe, nor is the Pacific Ocean, with its very low connectivity. (Actually, the word "fringe" makes the most sense if there is a sudden drop in density, rather than a smooth gradual drop.)
By the way, the use of euphemisms is often an indicator of a key value in some network.
For two other examples, the term appropriate technology can be defined as a technology the required links of which will fit in well with links that are already in place; those existing links are then considered "infrastructure" for the new technology, again relative to the network that is appropriate. For example, PCs need reliable electrical power, good repair technicians, quickly available spare parts, and a reasonable telephone system. ("Appropriate technology" is often recommended as a powerful guideline for introducing technology into developing countries, following the work of E.F. Schumacher, author of the very nice book Small is Beautiful.)
It is important here that both the technology and its context are represented as networks, so that we can talk about the relationships between them; it is also important that these networks are heterogeneous, including both human and non-human actors. Of course, my definitions are no big deal, and in particular, they are not very precise by scientific standards. Moreover, following the discussions in Section 7.2, there will often be problems with applying them in practice. But they can bring a greater degree of precision than is usual to discussions of this important topic, by encouraging us not to ignore the infrastructure that is generally invisible in developed countries, to include both human and non-human actants in that infrastructure, and to inquire more deeply into the values of those actors.
It seems that certain errors are repeated again and again in information technology businesses. One of these is making an overly ambitious and overly precise business plan, and then trying to follow it to the bitter end. This is particularly common in startups, which by their nature are often committed to going all out after an ambitious goal. But what we have learned from actor network theory suggests that business plans should avoid being overly committed and precise, and instead should be more flexible, and in particular should include contingency planning: they should sketch and cost out all scenarios that at that time seem the most plausible, and should also explicitly budget for replanning at some point, where the most plausible scenarios will again be sought. We all know that IPOs are a gamble, and that this gamble usually fails; this empirical fact can be seen as a strong argument for the impossibility of making precise predictions about the social effects of technology, and in favor of flexible planning and re-evaluation.
Anyone who has worked in the computer industry, and especially in software development, will have seen many instances of the phenomena described by actor network theory, and will also have seen many instances of the kinds of myth and foolishness that it is capable of exposing, including naive optimism, hagiography, value conflicts, and technological determinism. In my opinion, a careful contemplation of actor network theory, including a number of good case studies, should be required preparation for high technology managers.
Here are some hints on how to use ideas from ANT in a "quick and dirty" style that may help in resolving business decisions: