Placing Classification

I’ve been thinking about the places of classification, meaning both where it takes place and how different classification systems incorporate space and place. This is in part a reaction to collaborative work I’ve been doing (with a group from the Algorithms Studies Network), since algorithms are (in)famous for traveling across domains without attending to the specificities of particular contexts.

Two of the most well known sources on classification are Star and Bowker Sorting Things Out, who attend to the work that classifications do and how they influence the classifiers, and Foucault’s The Order of Things, where he develops the concept of the episteme. Foucault begins by quoting Jorge Luis Borges’s “Celestial Emporium of Benevolent Knowledge” where Borges satirically offers a system for classifying animals, supposedly taken from an ancient Chinese encyclopedia. The list includes categories of animals such as, among others:

  • Those that belong to the emperor
  • Mermaids (or sirens)
  • Those that are included in this classification
  • Those that have just broken the flower vase
  • Those that, at a distance, resemble flies

Borges is actually satirizing the work of a European scholar by comparing it to this fantastical list, which looks like a kind of categorization without reason. In the preface to the Order of Things, Foucault notes that Borges’s list caused him to break out in laughter, because it “shattered…all the familiar landmarks of thought.” (p. xvi, my emphasis, also discussed on Wikipedia)

But even as Borges uses the passage to comment on European classifications, it also has an Orientalist aspect, which it both mocks and reinforces in the sense that a vague ancient China is used as the repository for the unexpected and strange. In fact Borges could just as easily have used, for example, European heraldry, one ancestor of the flags of nation-states, where the crest of families and monarchies included elements like “three pairs of testicles”, and even a run of the mill coat of arms can be described as follows:

A crown of five arms, as follows: Azure a mullet of six points Or above a crescent argent; Azure two bars gules; Azure three leopard heads caboshed Or; Azure a goat statant Or unguled and armed gules….

There are also the overtones of eugenics in dog breeding, whose focus on idealized and segregated taxonomies places it in Foucault’s “classical episteme”:

A Bulldog’s ears:                                         The rear of a Chihuahua:

screenshot 2019-01-14 17.11.25

screenshot 2019-01-14 17.11.15








Yet these forms of classification do make some sense in their respective contexts, and indeed it’s the contexts that are often the subject of judgment rather than the classifications themselves. Similarly, the one given by Borges might in context as well—at least it might if it weren’t (I suspect) made up in order to appear strange.

One further reason why Borges’s classification might seem amusing is because there isn’t any clear organizing principle to the categories, which overlap and are nested within each other. Mythical creators like mermaids are interspersed with categories defined by actions in relative points in time (“those that have just broken the flower vase”) and space (“those that, at a distance, resemble flies”).

Although this can seem incredible, there are plenty of classifications organized through place and experiences in place. For example, the terms for cave features are unexpected when listed, as there’s not outside organizing principle visible, but they make sense when viewed as attempts by spelunkers from particular contexts and backgrounds to describe the unknown through categories that are relatable to their experience. This diagram notes features like “soda straws”, “clouds”, “pearls”, and “frost”, but the organizing principle isn’t legible without the cave, and particularly the experiences of particular cave divers in specific caves:


Similarly, the divisions in an orchestra don’t make much sense without understanding both the particular histories of instruments in particular places—including why some became common while others, like the saxotromba, saxhorn, and saxtuba did not—as well as the fact that the instruments are ordered through sound, and more specifically, the way instruments sound in particular places and situations, like classical music concerts in classical music halls.

The saxtuba:

saxtuba1867 In addition to these ways that places inform systems of classification, classifications also incorporate place and geography by, for example, including specific dogs—and histories/geographies of their origins—in a list of approved breeds particular nations in an international organization like the UN, which I deal with in a paper that’s currently under second review (fingers crossed).


What the classifications used and generated by algorithms might miss, depending on how they’re implemented and where, is a reflective take on their own emplacement. No algorithm is a panacea but rather is perhaps a type of the boundary infrastructure noted by Starr and Bowker, as well as the ineffable qualities of particular classifications, the distancing and commensuration work they do.

At times algorithms are implemented in specific places or, to consider things more broadly, contexts without extensively taking three factors into account:

  1. Those contexts of application
  2. The contexts where the algorithms originate, and
  3. The ways algorithms reproduce both of those contexts and their relationship with each other and the algorithm itself

Then, even though particular algorithms aren’t always adapted extensively to respond to the contexts where they’re implemented, they are sometimes used to justify all kinds of actions within those contexts.

The saying “when all you have is a hammer (algorithm), everything begins to look like a nail,” can then be modified: when you believe a hammer can solve every problem, you end up smashing everything to bits. Or even better: don’t turn everything into nails just because you think you have a hammer, when in reality your tool is much more heterogeneous than that.

The ways that algorithms are constructed and used tend to reduce everything to particular kinds of data and this makes one lose sight of the versatile and particular weirdness of both algorithms and the classifications that they both enact and replace.

screenshot 2019-01-14 17.11.30

The side of the Chihuahua