Matthew Fuller
Know Your Sorts
Matthew Fuller & Andrew Goffey, Evil Media, MIT Press, 2012
Sorting takes a sequence of entities and permutates that sequence in order to arrive at a result which renders it more useful.
As a stratagematic force in its own right, therefore, sorting should
be understood both as something that yields results, in the form
of a ranking, and as something that generates its own terms of
composition, shifting relations between things that are sorted in
ways that imply multiple kinds of use and attention.
Amongst other qualities, permutation, the process of the shifting
and sifting of the order of things, has an aesthetics of its own
which renders it distinct from the conceptual lock-down of nominally Platonic essentialism favoured in certain kinds of mathematically grounded accounts of software. Such an aesthetics
establishes a vivid dynamic of interplay between algorithms, the
machinic context of hardware and software resources and the
data which is being handled, all of which makes demands on
the other and combine to render each permutational process
individual. Further iterations and enfoldings of sorting in other
media - such as social processes - make it particularly interesting. In such a context, knowing your sorts, gaining a sense of the
aesthetic dimensions of ordering is crucial. But aside from the
way in which it engages the sensorial aspect of being, sorting
has a profound and intricate relationship to systems of ordering.
Amongst these, sorting is something distinct from categorization,
to which it is naturally affiliated. Categorisation may be the result
of a sort, and categories may also be sorted, but it is the permutational moment and the kinds of power it produces and invokes
that we are concerned with here.
As a distinct field of thought, computer science usefully maintains intellectual and technical reserve in relation to its application, its wider place in the world. As such it maintains relations of
pretended universality, in that everything finds its place in computation, but also equally establishes its separateness. In such a
context of technical neutrality, sorts are evaluated in terms of the
optimal use of resources both for processing code and for han/ 119
dling data during runtime, and in terms of the speed of execution
in relation to different sorting problems. In the field, questions
of optimality or optimization of this kind may be complicated by
those of the efficiencies of management, but since most sorting algorithms are readily available within standard libraries such
forms of interference tend not to coincide.
Within the computational articulation of sorts, material is typically handled via an alphanumeric key, a pointer, that maintains
a relation to the records, numbers or other single or clustered
entities that it in turn is able to treat as satellites. What is sorted
then is not at first the ‘primary’ data, such as a record or file, or
what it may refer to, such as an event or a person. What sorting
first acts upon is the numeric values by which they are handled.
Once these are organised, sorting can concatenate out. As such,
a general literacy of sorting is to be recommended.
Arguments against instrumental reason1, averring that it is one
more form of knowledge which subordinates means to ends, is
usefully transformed by other forms of sorting in which not the
numeric handler but rather the data itself are understood to have
an intrinsic and indexical relation to things in the world. “Social
sorting” as it is termed by scholars in critical surveillance studies2
adopts a mode of sorting in which mechanisms for the management of entitlement, control and protection are deployed to maximise efficiency, convenience and speed. Opponents of sorting
tend not to concern themselves with the underlying logic of such
rules, but with those moments at which they become inefficient,
inconvenient and slow. They may also attend to the way in which
vague social classifiers such as race are mobilized to provide
surety and the opportunity for the randomized, unjust or unaccountable exercise of power. The likelihood of racial category
providing insight into someone’s level of criminality is roughly
equivalent to such analysis being made on the basis of shared
name. That this is so does not preclude either association being
made. Imagine the quandary of an eight year old boy who has
the same name as a person who is for some reason on a ‘terror
watch-list’, and who is therefore subject to body searches before
boarding any plane.3
His problem arises because he is being sorted on the basis of
a referencing system that specifies an alphanumeric identifier–
his name–which, due to social convention and a resulting limited name space, is most likely held by more than one person.
It is perhaps better to name children with strong passwords that
are unlikely to be duplicated or easily memorized. Alternately,
it is possible to identify citizens on the basis of unique identi120 / GLI.TC/H READER[ROR] 20111
fiers, such as passport numbers and identification cards, which
provide an easily exploitable sense of security. These may be
triangulated with probabilistically unique identifiers such as biometrics, nominally unchanging and accessible features that can
be turned into a record which in turn can be assigned an alphanumeric code.
The primary method of sorting in computing is sorting by comparison. In such cases, data is sorted in relation to other data,
for instance whether it has a lesser or greater numeric value.4
Sorting by comparison implies that the range of data to be sorted
is generally not known in advance, or does not need to be. In
cases where it is known, algorithms such as bucket sort, radix
sort and pigeonhole sort, amongst others, work with effective addressing schemas to allocate results. Differences between these
can be accounted for at the level of speed, for instance when
using a search engine to query for a common search term whose
results are pre-ranked, compared to those that are rarer or unprecedented, and thus need to be generated on the fly.
Comparison essentially involves the allocation of a position on
the basis of a greater than or less than calculation. Whilst it is
tempting to make the assumption that simply because something is sorted by comparison it is reduced to a place within a
schema of greater or lesser rank, this would be to over-estimate
its effects and possibly to misrecognise the importance of the
process of being sorted as significant in itself before a place
in such an array is determined. Ranking can be an extremely
useful effect in combination with a queuing system or resource
allocation process as a way of entraining what is ranked. Ranking regimes that are active through the differential ordering of
interacting entities of different scales are inherently interesting.
As an example, the ranking of academics by numerous interacting rank-based mechanisms (such as those of scholars, departments, institutions, articles, citations, journals) confirm the
benefit of such approaches in terms of the simplification of the
evaluation of research into a quantifiable metric. The ease with
which such a system can be interpreted and summarized allows
for all positions within it to adapt to and canalize the required behaviour. Fine-tuning of results can be achieved by more obscure
means of handling such as those evinced by social networks.
All forms of sorting require the use of resources. In resourceconstrained environments, choosing which sort may be adopted,
testing which sort may be being applied or to which one is subject, or whether to estimate the employment of a sort of any kind
as useful, it is advisable to evaluate its implications in terms of
Matthew Fuller / 121
calculation and processing. Because sorts imply such costs they
are often identified as implying a deliberate sacrifice of resources, especially time, at the altar of rather obscure gods. As an
important dimension of the experience of sorting this is something itself to take into account. Here, the deployment of sorts
can act usefully as a form of immobilisation, an occlusion of the
identification of the beneficiaries of the sorting process, or for the
generation of support for new resource requirements. Whilst in
some cases, the least amount of comparisons should be aimed
at for the sake of efficiency, each opportunity for ordering is one
that should also be taken as a test of the worth of ordering in
itself and should therefore be evaluated carefully.
An example of the unpredictable results of the introduction of
sorting practices is to be found in the work of postal delivery. By
its very nature the work requires numerous stages of sorting.
The number and range of address to be delivered to is generally known and fixed into the “frame” used to position the run of
things to be delivered during their preparation in the sorting office. One exemplary factor that complicates the work is often the
uneven physical distribution of the addresses. A street may be
laid out in a higgledy- piggledy fashion, various plots of land having perhaps been developed at different times, being of different
sizes, or arrayed non-uniformly due to natural features. Working out the optimal route to take may be further complicated by
many factors (such as the slope of the ground or the presence of
parcels in the delivery load). Thus, every person delivering post
experiences their own daily version of the Travelling Salesman’s
Problem.5 This problem is usually understood to be resolvable
only in exponentially calculable time, but is solved here by the
tacit knowledge and the labour of the postal worker who knows
and sorts the route. Attempts to automate the process of sorting
and route planning in ways that marginalize or contradict this
local and habitual knowledge – on the undisclosed assumption
that such knowledge lacks the quantifiable virtue of the explicit,
and so can easily be confused with stupidity - raise a number of
problems that are exploitable as stratagems, but that mitigate
against an optimal postal service.6 The case of the trickiness of
the postal sort remind us that the virtue of a stratagem must not
be mistaken for an illusory efficiency. It also makes evident however that certain problems of sorting can be offloaded by such
means. The efficient circulation of an illusion is something to be
appreciated.
122 / GLI.TC/H READER[ROR] 20111
.............................................................................................................................................................................
1
Max Horkheimer, Critique of Instrumental Reason, Continuum,
London 1983
2
See i.e. David Lyon, ed., Surveillance and Social Sorting,
Routledge, London, 2002
3
Lizette Alavarez, “Meet Mikey, 8: U.S. Has Him on Watch
List”, New York Times, January 13th 2010, http://www.nytimes.
com/2010/01/14/nyregion/14watchlist.html/
4
Algorithms exemplifying such an approach include heapsort,
quicksort, insertion sort, merge sort, cocktail sort, and others.
5
A ‘classic’ problem in computer science, in which the shortest
path between several different points (cities to be visited by a
salesman) is to be calculated.
6
Roy Mayall, Dear Granny Smith, Short Books, London, 2009
.............................................................................................................................................................................
Matthew Fuller / 123