David Weinberger came to my briar patch last week to give his stump speech, entitled “The Realism of the Web”. There’s a bit of acknowledged cheekiness to that title given that Weinberger considers himself to be a “bunker” (as opposed to a “debunker”). As he puts it, “the Web isn’t hyped enough.” Ah.

For those unfamiliar with Weinberger’s recent work, here’s a quick run-down. He and some pals wrote The Cluetrain Manifesto (2000), a polemic warning business executives that their customers know a lot about their products and are using the Web to spread the word (“markets are conversations”) and organizational hierarchies are being similarly subverted as the Web lets employees talk to customers (organizations are becoming “hyperlinked”). The recommendation: drop the fortress mentality and marketing speak, then find new business models that make productive use of all the candid conversation. So far so good. Small Pieces Loosely Joined (2003) is a less popular stream-of-consciousness in which Weinberger tries to come up with a grand unifying-theory about how the Web changes everything. I’m not so sure that the book lives up to that lofty goal. The theory seems to be: it’s all decentralized, revolutionary, loosely coupled, and messy; which, as far as grand unifying-theories go, falls a tad short. Weinberger’s most recent book, Everything is Miscellaneous, is a slightly more scholarly discussion about how best to classify information to make it easier to find and use.

I’m interested in talking about that last book. It allows me to continue where I left off after reviewing Peter Morville’s Ambient Findability (2005). (I encourage you to read that review as background to this one.) It also gives me a chance to review a book I read recently on the same topic, Bowker & Star’s Sorting Things Out. And, as it happens, all three books have an anti-authority/anti-rationalist bent that’s worth scrutinizing.

So let me start by talking publicly about Weinberger’s book given that I’m one of his customers with access to the Web.

Everything is Miscellaneous: The Power of the New Digital Disorder by David Weinberger (Times Books, 2007), pp. 278.

Weinberger starts from the premise that there are three different orders of classification. First-order classification involves putting items into discrete bunches, as you would with cutlery in a cutlery drawer or books on a book shelf (like with like, each bunch in its own place). Second-order classification labels the items using some external classification scheme (metadata), as libraries do with their book catalogues. The number of labels is limited by the constraints of the cataloguing media. But as importantly, there is no guarantee that the labels used will be the same ones considered by users given their experience, context, way of thinking, or current task; nor will the set of labels for each item be exhaustive. The Web makes possible a third order classification that overcomes these limitations, according to Weinberger. Creators and users can collectively affix any number of labels (a.k.a. tags, descriptors, annotations) to an item according to their own opinions, uses, circumstances, or wants. Unlike other ways of classifying, the third order does not presume there is a neutral set of descriptors that will suit everyone or every use. Classification is a political choice that can bestow advantages. Third order classification places the power in the hands of everyone with a stake in the way items are sorted and labeled, not just in the hands of an overriding authority. The classification scheme emerges and evolves over time. And people find it easier to find things with such schemes when the number of items gets very large.

A good example from Weinberger’s book is the way the online retailer Amazon categorizes its books. There are second-order descriptors. But Amazon is more useful because of all the third-order ways of classifying books: book reviews by customers; reviews of customers who make the book reviews; collaborative filters, such as recommending books that other customers with similar tastes have bought; using computer algorithms to make on-the-fly connections between books based on their content; and book lists created by enthusiasts of a particular type of book. Blogs and blog aggregators (such as Digg.com and del.icio.us) also affix labels and links to information in a way that creates emergent categories over time. The main point for Weinberger is that these methods allow people to come across items through many different paths, not just a single way.

The reason why Weinberger thinks this is crucial is the inherent weaknesses of traditional attempts at classification. He talks about the Dewey decimal system for cataloguing library books, which has nine categories for religion, eight of which are about Christianity. There is the U.S. Food & Drug Administration’s system for classifying fish that places a diverse array of 17 fish species into the same category (including one of the most commercially lucrative species). And there are poorly thought-out schemes that place too many items under a “miscellaneous” category. There is also the case of conventional bar-codes, which are inherently limited in the amount of information they can associate with a consumer product. Weinberger even takes swipes at chronological ordering and alphabetization. He has problems with the arbitrariness of it all. There is often no one-best-way to classify items. And he argues that most of these attempts at classification make misguided rationalist judgments about what are the essential features that ought to differentiate one thing from another.

So what advice does Weinberger have for those with responsibility for classification systems (as many businesses are)? He lists four “laws of the jungle”. First, don’t create artificial scarcity by filtering what can be accessed by people. Instead, provide tools that allow individual users to filter out what is not relevant to them. Second, allow for as many descriptors to be placed on an item as possible. Third, in the case of documented information, everything in the document can be used as a label. Finally, Winberger suggests that gatekeepers give up control to the users of information. He also has lots of praise for the clever ways that technology and social networks are used to point people in the right direction.

Sorting Things Out: Classification and Its Consequences by Geoffrey C. Bowker and Susan Leigh Star (MIT Press, 2000), pp. 389.

Bowker & Star’s Sorting Things Out isn’t a sweeping survey of how to classify, label and find things, unlike the books by Morville and Weinberger. Bowker & Star instead choose to focus on attempts, since the late nineteenth century, to create a comprehensive and discrete system of classifying medical conditions. This is a formal taxonomy which attempts to organize medical conditions based on consistent principles into mutually exclusive categories in a comprehensive way. (There is a chapter on race classification in Apartheid-era South Africa but the authors don’t offer much of an explanation for the tangent other than to say that they want to show the social impact of nasty classification schemes.)

Bowker & Star show how formal taxonomies evolve over time and build up inertia, as the product of both expert judgment and bureaucratic negotiations. Bowker & Star also expand on a theme mentioned by Morville and Weinberger in passing: how political power, aesthetic preferences, and the prevailing morality distort classification schemes and, given the inertial force, entrench political advantage (often in ways that are scarcely noticed). And to the extent that classification schemes are ubiquitous, they become part of the infrastructure that shapes the way we perceive and act.

The book itself is a series of historical episodes and philosophical commentaries. There’s mention of how homosexuality was once classified as a medical malady. They tell of how one classification becomes superseded by another (e.g., Gay-Related Immune Disorder becomes Acquired Immune Deficiency Syndrome) and how the name of a category can carry a political sting. They tell of how suicide poses dilemmas for a classification system based on medical science when moralistic and cultural forces intrude. Bowker & Star also describe how practitioners often have a hard time using the taxonomy because it seems geared more towards bureaucratic reporting than medical diagnosis. Reliability of reporting may be undermined as practitioners go through the motions just to fill out the paper work. There’s usually no data about how observations were conducted (a.k.a. paradata), making it difficult to determine overall accuracy and reliability of diagnoses. The taxonomy may be woefully behind the state of medical science because it tends to evolve slowly. Restrictions in reporting procedures may encourage practitioners to point to a singular cause of poor health when multiple causes may be involved. Medical problems differ across the world (e.g., tropical diseases) and universal taxonomies may not reflect local circumstances, which sometimes creates international conflict. All told, these are problems caused by the essentially arbitrary nature of a taxonomy and the contortions practitioners need to make to conform to (or game) the system. I would add that complexity of the subject area has something to do with it.

These are all excellent insights. But there is much to quarrel with in Bowker & Star’s book given the way they conduct their study. In particular, the authors claim that these are highly representative problems (or in some cases universal problems) inherent to formal taxonomies. Yet their analysis of a single case (notwithstanding the tangent involving the extreme Apartheid case) is not suited to drawing such sweeping conclusions (especially given the emphasis on older episodes in the case).

It’s not difficult to figure out how this leap in logic takes place. Bowker & Star use Michel Foucault’s brand of analysis, which tends to boil down to: if you want to know how tyrannical a system can really get, look through its dirtiest laundry (my term, not his). The choice of a medical taxonomy is not entirely coincidental, for Foucault focused on medical institutions in order to study how society treated lepers or the mentally ill (prisons are the other main source of dirty laundry for him). The authors’ reliance on Bruno Latour’s cynical post-modernist take on science reinforces the leap in logic. I’m just as interested as Latour in identifying political, bureaucratic, and ideological intrusions into scientific inquiry. But Latour, like many post-modernists, denies the possibility of separating them out in practice. Bowker & Star commit a similar conflation of (a.) the imperfect state of scientific knowledge at any one time and (b.) the corruption caused by moralizing (as distinct from ethical) bureaucratic/political interference. Any self-respecting scientist or medical practitioner acknowledges the former and resists that latter, contrary to the stereotype being perpetuated here.

I’ll leave the jousting over the epistemological assumptions and ideological predisposition of post-modernists to another day. I’ll just note that this type of thinking also influences Weinberger. Weinberger acknowledges that “social constructivists” have a point when it comes to the way classification schemes can unduly influence practitioners (the Foucauldian author Ian Hacking is cited here). I agree. Indeed, I agree above that Bowker & Star make several insightful points about that. But there is still the outstanding issue of how prevalent these problems are and whether they indict the entire enterprise of formal taxonomy. I confidently keep my doubts. Plus, I couldn’t help but notice that Weinberger uses two of the extreme examples Bowker & Star go out of their way to emphasize (the treatment of homosexuality as a medical disorder and the Apartheid racial classification scheme). Yes, there is some cherry-picking going on here.

Review by Peter Stoyko

REFERENCES