These texts was originally published in Spectre of AI, the fifth issue of spheres: Journal for Digital Cultures.

The first text is a technical introduction to an open source recommended system and the second is a broader, more philosophical, reading of this software.

Part 1: Source Code

Introduction

This text aims to explain some of the source code of the open source recommender system LightFM. This piece of software was originally developed by Maciej Kula while working as a data scientist for the online fashion store Lyst, who aggregates millions of products from across the web. It was written with the aim of recommending fashion products from this vast catalogue based to users with few or no prior interactions with Lyst.1 At the time of writing, LightFM is still under active development by Kula with minor contributions from 17 other developers over the past three years. The repository is moderately popular, having been starred by 2,032 GitHub users; 352 users have forked the repository, creating their own version of it that they can modify.2 Users have submitted 233 issues, such as error reports and feature requests to LightFM over the course of its existence, which suggests a modest but active community of users.3 To put these numbers in perspective, the most popular machine learning framework on Github, Tensorflow, has been starred 113,569 times and forked 69,233 times with 14,306 issues submitted.4

While the theoretical text that accompanies this one addresses aspects of machine learning in LightFM, none of the source code quoted here actually does any machine learning. Instead, the examples here are chosen to demonstrate how an already trained model is processed so as to arrive at recommendations. The machine learning aspect of LightFM can be briefly explicated using Tim Mitchell’s much-cited definition: “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P improves with experience E.”5 Task T, in this case, is recommending products to users that they are likely to buy. E is historical data on interactions between users and products as well as metadata about those users and products. P is the extent to which the model’s predictions match actual historical user-item interactions

In practice, P is represented by a number produced by a loss function (also known as cost function or objective function) that outputs higher numbers for poorly performing models whose predictions don’t match historical data. To arrive at an effective model, P needs to be minimised through some usually iterative process, which in the case of LightFM is a form of gradient descent.6  Gradient descent begins with a model with random, fairly arbitrary parameters and repeatedly tests the model, each time changing the parameters slightly with the aim of reducing the model loss (the number output by the loss function) and eventually reaching an optimal set of parameters.7 The parameters of LightFM’s model are embedding vectors for each feature or category that may be applied to a user or item; these are discussed in greater detail below. Returning to Mitchel’s definition: as gradient descent only optimises the model based on the historical data available, it is clear that up to a certain point, LightFM is likely to produce more relevant recommendations (T) if it is trained using a larger and presumably more representative set of test and training data (E).

The below excerpts of source code are taken from a file in the LightFM Git repository named _lightfm_fast.pyx.template. This file defines most of the actual number-crunching carried out by the LightFM recommender and is written in Cython, a special form of Python that works like a template for generating C code. The file contains 1,386 lines of Cython and is used to generate up to 30,720 lines of C. While verbose to us, this generated C code is compiled into even more prolix machine code which computers execute much faster than Python.  Both examples are function definitions; they start with the word ‘cdef’ which in this case indicates that the definition of a function that can be compiled into C is to follow. Functions take input data in the form of one or more arguments and perform some computation using these data. They either modify the data that is passed into them or output some new data derived from this input. In the case of compute_representation, the word ‘void’ precedes the function name, indicating that the function outputs or returns nothing and would be expected to do something to its input. In the second example the word ‘flt’ (an alias for ‘float’) precedes compute_prediction_from_repr, meaning that it is expected to return a floating point number. For brevity’s sake, a floating point number is basically a decimal like 1.6.

Function 1: compute_representation8

This function is responsible for taking data about users (e.g. customers) and items (e.g. clothes or films) and producing latent representations of them. These representations can be used by the second function ‘compute_prediction_from_repr’ to predict the level of affinity between these users and items.

cdef inline void compute_representation(CSRMatrix features,
                                        flt[:, ::1] feature_embeddings,
                                        flt[::1] feature_biases,
                                        FastLightFM lightfm,
                                        int row_id,
                                        double scale,
                                        flt *representation) nogil:

The comma-separated lines in parentheses following the name of the function above are parameter declarations; these determine what data or arguments can be passed into the function. The first part of each parameter declaration can be thought of as the type of the expected argument and the second part is the name used to reference it in the body of the function. Here is an explanation of each of the parameters:

 

  1. features: an object belonging to the class ‘CSRMatrix’. Matrices are like tables, or row-column addressable grids, whose cells contain numbers. CSR matrices offer a way of storing matrices in which most of the numbers are zero; in other words: when the matrices are sparse. In this case – sticking with the table analogy – the rows correspond to users or items and the columns correspond to features of these users or items. If the items are films, each feature could be a genre. Cells in the table might contain 1 if the film (row) belonged to genre (column) and 0 if it didn’t. Each row in the matrix can be taken as a vector representing the corresponding film.
  2. feature_embeddings: a two-dimensional array, which is also like a table with rows and columns. This array contains the embeddings for features that have been learned from training data using gradient descent, as discussed above. This array has a row for each feature and a column for each dimension of the features’ embeddings. These feature embedding rows can be thought of as vectors that contain information about how similar each feature is to others based on shared positive user interactions such as favourites and purchases. Vectors are like arrows with a direction and a magnitude or length. The row vectors of the ‘feature_embeddings’ array are represented by their Cartesian coordinates, such that if each row contained two numbers, the vectors would be two-dimensional and the first number might specify the horizontal position of the end of the vector (the tip of the arrow) and the second number the vertical position. If two features (e.g. ‘action’ and ‘adventure’ or ‘black’ and ‘dress’) are shared by items bought by the same 1,000 users, their embeddings will be similar; they will point in a similar direction. To illustrate: the similarity of two two-dimensional feature embeddings could be worked out by plotting them both on a sheet of paper as arrows originating from the same point and measuring the shortest angle between them using a protractor; the smaller this angle, the greater the affinity between the features.
  3. feature_biases: a one-dimensional array of floating point numbers. This is like a list of decimal numbers.
  4. lightfm: an object of the class FastLightFM. This object holds information about the state of the recommender model.
  5. row_id: an integer, or whole number, identifying the row in the features matrix that the function should compute a representation for.
  6. Scale: a double-precision floating point number. It is called double-precision because it takes up 64 bits, or binary digits, of memory rather than the standard 32.
  7. representation: reference to a one-dimensional array of floating-point numbers. When executed, the ‘compute_representation’ function modifies the contents of the referenced ‘representation’ array.
"""
    Compute latent representation for row_id.
    The last element of the representation is the bias.
    """

    cdef int i, j, start_index, stop_index, feature
    cdef flt feature_weight

The two lines above are declaring variables, which allow a value of a particular type to be referenced by a name such as ‘start_index’. All of these variables are numbers that are used later in the function.

    start_index = features.get_row_start(row_id)
    stop_index = features.get_row_end(row_id)

Two of the variables declared above are being assigned values that are returned by the ‘get_row_start’ and ‘get_row_end’ functions. These functions are part of the ‘features’ CSRMatrix object and are called methods.

    for i in range(lightfm.no_components + 1):
        representation[i] = 0.0

The above two lines comprise a simple Python for-loop. The loop counts up from zero to the value of ‘lightfm.no_components + 1’ in increments of one, each time setting the value of variable ‘i’ to the current count. For each increment it executes the indented code ‘representation[i] = 0.0’, which sets the ith element of the ‘representation’ array to zero. In effect, it sets every number in ‘representation’ to zero.

 

Note: ‘lightfm.no_components’ is an integer that determines the dimensionality of the features’ latent embeddings. If ‘no_components’ is ‘10’, each feature is represented by a ten-dimensional vector.

    for i in range(start_index, stop_index):
        feature = features.indices[i]
        feature_weight = features.data[i] * scale

This for-loop is slightly different, it counts up from ‘start_index’ to ‘stop_index’, setting ‘i’ to the current count each time. It also executes five lines of indented code upon each iteration, including another nested for-loop. The following two lines set the ‘feature’ and ‘feature_weight’ variables to the appropriate values from the CSR matrix object. ‘feature’ is set to the index of the feature ‘i’, as stored in the ‘feature_embeddings’ array. ‘feature_weight’ is set to a value that indicates whether this particular feature belongs to the user or item the ‘compute_repr’ function computing is a representation for; this is probably ‘1’ if the feature belongs and ‘0’ if it doesn’t.

        for j in range(lightfm.no_components):

            representation[j] += feature_weight * feature_embeddings[feature, j]

This nested for-loop counts from zero to the number of components (the dimensionality of the feature embedding) in increments of one and sets j to the current count.  ‘feature_weight’ will maintain the same value for the duration of the loop: either ‘1’ or ‘0’.  This means that either the entire feature embedding will be added to the ‘representation’ array or none of it.

All this loop is doing is adding together the latent representations of the features of a given item or user. As Maciej Kula puts it: “The representation for denim jacket is simply a sum of the representation of denim and the representation of jacket; the representation for a female user from the US is a sum of the representations of US and female users.”9

        representation[lightfm.no_components] += feature_weight * feature_biases[feature]

If the user or item has the feature ‘features.data[i]’, that feature’s bias is added to the final bias element of the ‘representation’ array.

Function 2: compute_prediction_from_repr10

This function is responsible for predicting how likely a user is to be interested in an item.

cdef inline flt compute_prediction_from_repr(flt *user_repr,
                                             flt *item_repr,
                                             int no_components) nogil:

There are only three parameters this time:

  1. user_repr: a reference to a one-dimensional array of floating point numbers. Its contents will have been produced by the ‘compute_representation’ function above based on the features and other data for a user.
  2. item_repr: a reference to a representation of the features of a given item, also produced by the ‘compute_representation’ function.
  3. no_components: the number of dimensions of the representation vectors.
    cdef int i
    cdef flt result

    # Biases
    result = user_repr[no_components] + item_repr[no_components]

The variable ‘result’ is set to the sum of the bias component of the latent representations for both the user and item.

    # Latent factor dot product
    for i in range(no_components):
        result += user_repr[i] * item_repr[i]

This loop adds the dot product or inner product of the user’s and item’s representation vectors to the ‘result’ vector. The dot product of two vectors is the sum of the results of multiplying each part of the first vector by the corresponding part of the second vector, as can be seen in the loop: ‘user_repr[i] * item_repr[i]’.

The dot product is a single number that varies with the difference in direction between the embedding vectors ‘user_repr’ and ‘item_repr’ and is greater if they are facing the same way.  As these embeddings have been calculated based on known interactions between users and items, the dot product gives an indication of how likely any user is to interact positively with any item.

    return result

The function returns the dot product added to the sum of the item and user’s biases. These biases are calculated in the ‘compute_representation’ function and are the sums of the biases of the item and user’s active feature. This effectively gives LightFM the ability to treat some features as more important than others.

 

1.   Cp. Maciej Kula, “Metadata Embeddings for User and Item Cold-start Recommendations”, paper presented in the 2nd Workshop on New Trends on Content-Based Recommender Systems co-located with 9th ACM Conference on Recommender Systems, 2015. Available at: http://ceur-ws.org/Vol-1448/paper4.pdf [accessed July 27, 2018].
2.   Cp. Maciej Kula, “GitHub – lyst/lightfm: A Python implementation of LightFM, a hybrid recommendation algorithm”, posted to Github. Available at: https://github.com/lyst/lightfm [accessed November 4, 2018].
3.   Cp. Maciej Kula, “Issues – lyst/lightfm – GitHub”, posted to Github. Available at: https://github.com/lyst/lightfm/issues [accessed November 4, 2018].
4.   Cp. TensorFlow, “GitHub – tensorflow/tensorflow: An Open Source Machine Learning Framework for Everyone”, posted to Github. Available at: https://github.com/tensorflow/tensorflow [accessed November 4, 2018].
5.   Tom Mitchell, Machine Learning, London, McGraw-Hill, 1997, p. 2.
6.   Cp. Kula, “Metadata Embeddings for User and Item Cold-start Recommendations”.
7.   Cp. Andrew Ng, “Lecture 2.5 — Linear Regression With One Variable | Gradient Descent — [ Andrew Ng]”, lecture posted to Youtube. Available at: https://www.youtube.com/watch?v=F6GSRDoB-Cg [accessed November 4, 2018].
8.   Maciej Kula, “lightfm/_lightfm_fast.pyx.template”, posted to Github, line 287. Availabe at: https://github.com/lyst/lightfm/blob/e12cfc7e5fa09c1694b98acc96af3ed2754646ae/lightfm/_lightfm_fast.pyx.template#L287  [accessed July 25, 2018].
9.   Cp. Kula, “Metadata Embeddings for User and Item Cold-start Recommendations”.
10.   Maciej Kula, “lightfm/_lightfm_fast.pyx.template”, posted to Github, line 320. Available at:  https://github.com/lyst/lightfm/blob/e12cfc7e5fa09c1694b98acc96af3ed2754646ae/lightfm/_lightfm_fast.pyx.template#L320  [accessed July 25, 2018].

 

Part 2: Machine Learning and the Machinic Unconscious

Introduction

In this text I set out to critically examine part of the source code of the recommender system LightFM. To this end I deploy the micropolitics developed by Gilles Deleuze and Félix Guattari, as well as Andrew Goffey’s and Maurizio Lazzarato’s readings of their micropolitical ideas. I build an argument around Guattari’s suggestion that subjectivity is not solely the product of human brains and bodies, and that the technical machines of computation intersperse with what might be thought of as human in the production of subjectivity. Drawing upon contemporary approaches to the nonhuman, machine learning and planetary-scale computation, I develop a framework that situates the recommender system in assemblages of self-ordering matter and links it to historical practices of control through tabulation. While I acknowledge the power of source code in that it always carries the potential for control, in this reading, I impute greater agency to computation. In what follows, rather than reducing it to an algorithm, I attempt to address the recommender system as manifold: a producer of subjectivity, a resident of planet-spanning cloud computing infrastructures, a conveyor of inscrutable semiotics and a site of predictive control.

The Interstices of Human and Technical Machine

In addition to addressing LightFM at the level of its technical workings and infrastructural instantiations, I aim to conceptualise the role these workings play in the ongoing crystallisation of power relations. Deleuze and Guattari offer a useful starting point: everything, for them, is political and every politics is “simultaneously a micropolitics and a macropolitics.”1 They offer the example of “aggregates of the perception or feeling type” and maintain that “their molar organisation, their rigid segmentarity, does not preclude the existence of an entire world of micropercepts, fine segmentations that grasp or experience different things, are distributed and operate differently.”2 Much as there are multitudes of molecular variations that escape dominant molar understandings of perception, I argue that there are sinuous historical and machinic paths that criss-cross and circle the linear division of human and technical machine. Further, “there is a double reciprocal dependency between”3 the molar and molecular. Just as multifarious populations are shaped by molar classes, the intertwining histories and relations of humans and technical machines are moulded by the dichotomy’s rigid structure. As a result, some theorisation of humans and technical machines privilege one side of this dichotomy. For instance, in Andy Clark’s notion of the extended mind4 and Marshall McLuhan’s conception of media as extensions of man5, media and technical machines are rendered as mere prostheses to human cognition or perception. I favour neither this nor the opposite molar approach of imputing disproportionate agency to technical machines; there is more to explore at the molecular level, in what we might think of as a machinic unconscious.

In his book The Machinic Unconscious, Guattari posits “a consciousness independent of individuated subjectivity [that] could manifest itself as a component in the assemblages of enunciation, ‘mixing’ social, technical and data processing machines with human subjectivity, but could also manifest itself in purely machinic assemblages, for example in completely automated and computerized systems”6. While this partly or fully nonhuman consciousness is central to my analysis of LightFM, my reading does not impute consciousness to non-living things. Were it to do so, I might be seen to espouse a form of panpsychism which views “mind [as] a fundamental property of matter itself”7; I believe that this position and the questions it raises are beyond the scope of this text. I limit myself to tracing through these “social, technical and data processing machines”, semiotic processes that reside in the barren and unmapped hinterlands of human subjectivity. Following Maurizio Lazzarato, I identify “mixed” and fully computational enunciatory assemblages as proto-subjectivities. These belong to the register of “non-representational and asignifying”8 semiotics: a sign system that operates below the threshold of individuated subjectivity and one that favours pattern over meaning.

Individuated subjectivity, for Lazzarato, is not the sum total of subjectivity; it is produced by the macropolitical process of social subjection that assigns subjects “an identity, a sex, a body, a profession, a nationality, and so on.”9 Conversely, proto-subjectivity is subjectivity produced by the micropolitical process of machinic enslavement. It comprises “a multiplicity of human and nonhuman subjectivities and proto-subjectivities”10; it arises from the micropolitical process of machinic enslavement, that “dismantles the individuated subject, […] acting on both the pre-individual and supra-individual levels.”11 Individuated subjectivity is representational, autobiographical and identitarian, giving rise to clearly delineated subject who acts instrumentally upon external objects.12 Proto-subjectivity is non-representational and pre-individual, capable of emerging in all autopoietic machinic systems.13 Critically engaging with proto-subjectivity is by necessity a speculative endeavour calling for non-representational thinking, which Nigel Thrift identifies with the “anti-biographical and pre-individual”14, with a “vast spillage of things” and with “affect and sensation”15. LightFM’s source code is an uneasy compromise between the representational, signifying semiotics of individuated subjects and the non-representational asignifying semiotics of technical machines. As such, in this text I must go further than merely explicating the procedures of computation, as I do in the companion text, and address computation in terms of its material instantiations, its histories and its production of affect.

Approaching the Machinic Unconscious

Andrew Goffey suggested in a recent lecture on the micropolitics of software, that a technological or machinic unconscious might be one that “crosses the histories of programming practices and their shifting relations to the infrastructures that they produce and are produced by – [disclosing] the fragmented possibilities of a different relationship to power.”16 What conceptual tools might help us figure this relationship, to think through these histories and infrastructures, mixed assemblages and proto-subjectivities? We can start by turning our attention, as some contemporary philosophy does, to the nonhuman. One example is object-oriented ontology, a branch of speculative realism that includes thinkers such as Timothy Morton, who would consider a poem, and presumably a piece of code, a nonhuman agent.17 However, as Jane Bennet notes, object-oriented philosophers refuse the label “materialist”, viewing objects as isolated entities withdrawn from other things.18 Such a position leaves little room for the molecular, micropolitical processes I am concerned with here, many of which are relational or take place close to the undifferentiated level of the machinic phylum. Another more established alternative is Actor-Network Theory, which is primarily concerned with the observation of connections between agents, both human and nonhuman.19 While an emphasis on nonhuman objects and nonhuman agency is useful – by building on the materialism of Deleuze and Guattari and retaining reference to subjectivity,20 Braidotti’s neo-vitalist materialism provides a better ground for my argument. For her, “all human and non-human entities are nomadic subjects-in-process, in perpetual motion, immanent to the vitality of self-ordering matter.”21 We will return to Braidotti later; for now it suffices that autopoietic matter provides a solid ontological substrate for delineating proto-subjectivities.

We now turn to the subject matter of this text, which Adrian Mackenzie addresses in some depth in his recent book Machine Learners. Mackenzie’s titular machine learner can be both human and nonhuman, or indeed constitute “human-machine relations”22. These human-machine relations are the sites of proto-subjectivities, mixed semiotic assemblages that “inhabit a vectorised space and [whose] operations vectorise data.”23 I place the vectors and matrices of machine learning in a genealogy of tables as technologies that aid in control. This genealogy stretches from ancient Mesopotamia24 and encompasses the introduction of tab keys in typewriters25 and the adoption of punch-card tabulating machines26 in turn-of-the-century bureaucracies as well as the relational databases of the 1960s. It is not a straightforward genealogy, however; as Mackenzie notes, for machine learners, the sheer number of dimensions in vector space can “thwart tabular display” and tables can “change rapidly in scale and sometimes in organisation”27. Drawing on Foucault’s account of tables from different eras, Mackenzie argues that the matrices of machine learning mark a return to the Classical or even pre-Classical tables28 that married heterogeneous elements and were structured according to plural and diverse resemblances.29 For example: a matrix of online purchases would place vectors for hair dryers alongside those for garden ornaments; a machine learner, owing to its profoundly flattened ontology, would subject them to the same computation, tracing diverse resemblances through blind repetition.

How might we conceptualise LightFM if it were deployed on a global cloud platform like Amazon? What if, instead of being trained with the ubiquitous MovieLens 100k dataset, LightFM could vectorise the largest ever accretion of user and product metadata on the planet? To aid in answering this question, I will borrow Benjamin Bratton’s model of planetary-scale computation: the ‘Stack’. Setting aside the geopolitical intricacies of Bratton’s argument, the stack can be thought of as “a vast software/hardware formation, a proto-megastructure built of crisscrossed oceans, layered concrete and fibre optics, urban metal and fleshy fingers”30. I would argue that planetary-scale machine learning sits at the intersection of the material megastructure of the stack and the asignifying semiotic processes of the machinic unconscious. Bratton’s ‘Stack’ is divided into six layers: ‘Earth’, ‘Cloud’, ‘City’, ‘Address’, ‘Interface’ and ‘User’. These can be placed on a vertical spectrum, rising from molecular to molar – from the geological and chemical, through to the computational all the way up to individuated subjectivity. The micropolitical analysis that follows is primarily concerned with the more molecular ‘Cloud’ layer. However, recalling the double reciprocal dependency between micropolitics and macropolitics, in thinking through the computation that works with vectors and matrices of user meta-data in the ‘Cloud’ layer, we may glean insights into how individuated subjectivity is produced in the ‘User’ layer.

Reanimating the Code

Wendy Chun holds that what we call source code “is more properly an undead resource: a writing that can be reanimated at any time, a writing that haunts our executions.”31 I share Chun’s hesitancy to locate agency primarily in source code, which I view as human-readable shorthand with the potential – through multiple translations – to inscribe the palimpsest-like surfaces of computational agency. It is only in that the source code haunts its concrete executions that we can read it micropolitically at all. The terse sentences and mathematical formulae Maciej Kula uses to describe LightFM’s algorithm also haunt these executions, but even more spectrally and tenuously than does the code. To illustrate, the following short passage describes the part of LightFM’s algorithm that the source code examples are responsible for implementing: “The latent representation of user u is given by the sum of its features’ latent vectors […] The model’s prediction for user u and item i is then given by the dot product of user and item representations, adjusted by user and item feature biases”32. Algorithms are divorced from what Goffey calls implementation details: “embodiment in a particular programming language for a particular machine architecture”33. The Cython source code is as close as my method allows me to get to the micropolitics and asignifying semiotics of computation, but how do I approach it? I heed Mark Marino’s warnings against analysing code aside from its “historical, material, social context” and drawing specious analogies between computation and unrelated social practices or cognitive processes.34 Instead, in what follows, I attempt to speculate beyond the text of the code, to the data structures it references, the materiality of its executions and how these relate to power and control.

Mackenzie observes that many machine learners seek to approximate data by plotting lines and curves through it, or dividing it with lines, planes35 and hyperplanes.36 LightFM, however, mainly transforms and compresses a potentially enormous vector space into smaller more easily computable representations, on which it bases its predictions. One of the parameters of the compute_representation function is the embedding vector for a feature, such as a book genre. The embedding is arrived at based on a matrix, a two-dimensional grid of numbers. In this matrix, each of millions of users is assigned a row and each of hundreds of thousands of books a column; each cell where a user and book intersect contains the number “1” if the user has bought the book, otherwise “0”. Now the transformation: from this vast binary matrix an embedding vector of a genre such as ‘software studies’ is produced that points in more or less the same direction as the vectors for other genres of books also bought by software studies enthusiasts. Perhaps the directions of these embedding vectors are among what Mackenzie refers to when he describes the vectorised table as a “machinic process that multiplies and propagates into the world along many diagonal lines.”37 These vectors are arrived at iteratively, through trial and error, in what could be thought of as discretised space and stepwise time. Proto-subjectivities inhabit the discrete space-time of vector computation, just as they reach through a maze of cables, routers and interfaces to the smooth and continuous bodies of users, their unthinking habits and gradations of affective intensity.

A non-representational reading of a function called compute_representation leaves no space for irony. It leaves little space for users, perhaps more for items, although the function makes no distinction. It expects a sequence of ones and zeros that correspond to embedding vectors, one of which may be the embedding for ‘software studies’. It wants a reference to an existing representation, a sequence of floating-point numbers,38 each a word 32 bits or binary digits long, arrayed one after another at a particular numeric address in a memory module in one of thousands of racked servers in a data centre. Execution begins. It marks six 32-bit chunks of memory for later use. It switches context, to the get_row_start function in the features object; this object is an agglomeration of data and executable instructions sprawling across a bristling microscopic patch of physical memory. It steps through each instruction in this foreign function and remembers the result. It switches back and writes the result to start_index, one of the reserved chunks of memory; the same for stop_index. One-by one, all bits in all words in the representation array are set to 0, switched off. It later, cycle-by-cycle, switches some of these bits on, and sometimes off again, endlessly toggling states; often all bits remain unchanged for an entire cycle as it blindly adds zero to each part of the representation.

Planetary-scale Prediction

The notion of deriving a prediction from a representation is firmly situated in the familiar signifying semiotics of individuated subjectivity. And yet compute_prediction_from_repr merely calculates the inner product of two vectors, two sequences of numbers, through brute iteration. Mark Andrejevic observes that it is precisely this lack of understanding, reasoning and intuition that gives data-driven prediction its power39 – though control may be more apt here. Recall the genealogy of the table alluded to earlier; two-dimensional tables such as timetables are a mark of Foucault’s disciplinary society, which Deleuze suggests has been supplanted by a society of control. I argue that this rupture in the table’s genealogy, the expansion and fragmentation of two-dimensional grids into inscrutable vector spaces, mirrors the fission of enclosed individuals into “‘dividuals’ and masses, samples, data, markets or ‘banks’”40. Instead of being placed in a panoptic cell and observed, subjects are divided into row vectors in a panoply of matrices dotted around the ‘Cloud’, and predicted. This control differs from the “purposive influence to a predetermined goal”41 that James Beniger posits as the seeds of the information society; it is closer to what Luciana Parisi and Steve Goodman call “mnemonic control”: “the power to foreclose an uncertain, indeterminate future by producing it in the present”42.

How does machinic prediction at the ‘Cloud’ layer entail control at the ‘User’ layer? In constructing vectors that stand in for users, LightFM may be producing categories of subjects, in that these vectors could coalesce or cluster into molar classes. This is a tendency observed by Braidotti, whereby “the neoliberal system finds ways to capitalize also on the marginal and the molecular formations, recomposing them as multiple molarities (i.e. billions of Facebook pages).”43 This can also be figured as purposive movement within the autopoietic matter that makes up the ‘Stack’. Computational proto-subjectivities and their asignifying semiotic chains snake through the ‘Cloud’, ‘City’, ‘Address’ and ‘Interface’ layers, terminating in individuated subjects at the ‘User’ layer.44 How might proto-subjectivities apprehend themselves to these subjects? Perhaps as the imperceptible background hum of the ontological power of the future in the present.45 If I were to speculate: a subject may feel itself to be acting on blind compulsion. Hunched over a smartphone, through a fog of information fatigue, they may be faintly aware of being nudged toward certain actions, certain products, certain cultural content.

Conclusion

In this short text, I have gone some way in posing, if not answering the question of how molecular, machinic processes in a recommender system like LightFM function and relate to power over individuated subjects. I have hinted at how computation may embody a certain kind of agency that feeds into the production of users as subjects. Bratton’s model of The Stack aided me in figuring the impingement of proto-subjectivities on molar aggregates such as subjects and users. The idea I have sketched out, of vectorisation and predictive control as a rupture in the genealogy of the table may be an avenue for more extensive research. Although I never primarily attributed agency to the code itself, I could only circumvent the limits of this technical text through speculation. A more empirical approach like Actor-Network Theory may have better elucidated some micropolitical aspects of LightFM, in mapping concrete connections between agents such as computers, programmers and users.

 

1.   Gilles Deleuze and Félix Guattari, A Thousand Plateaus: Capitalism and Schizophrenia, translated by Brian Massumi, London, Bloomsbury Academic, 2013 [1980], p. 249.
2.   Ibid.
3.   Ibid.
4.   Cp. Andrew Clark and David Chalmers, “The Extended Mind”, Analysis, 58 (1), 1998, pp. 7–19.
5.   Cp. Marshall McLuhan, Understanding Media: The Extensions of Man, Corte Madera, CA, Gingko Press Routledge, 2003 [1964].
6.   Félix Guattari, The Machinic Unconscious, Cambridge, MA, Semiotext(e), 2011 [1979], p. 221.
7.   Steven Shavero, “Consequences of Panpsychism”, in Richard Grusin (ed.), The Nonhuman Turn, Minneapolis, University of Minnesota Press, pp. 19–44, here: p. 20.
8.   Maurizio Lazzarato, Signs and Machines, Los Angeles, Semiotext(e), 2014, p. 25.
9.   Ibid., p. 12.
10.   Ibid., p. 34.
11.   Ibid., p. 12.
12.   Cp. Ibid., p. 12.
13.   Cp. Ibid., p. 80.
14.   Nigel Thrift, Non-Representational Theory: Space | Politics | Affect, London, Routledge, 2008, p. 7.
15.   Ibid. p. 10.
16.   Andrew Goffey, “Andrew Goffey – Micropolitics of Software”, lecture at Subjectivity, Arts and Data, Department of Media Arts at Royal Holloway, University of London, 2018. Available at: https://www.youtube.com/watch?v=9bqxsmFo72k [accessed July 25, 2018].
17.   Cp. Jane Bennet, “Systems and Things”, in Richard Grusin (ed.), The Nonhuman Turn, Minneapolis, University of Minnesota Press, pp. 223–240, here: p. 234.
18.   Cp. Ibid., p. 226.
19.   Cp. Bruno Latour, Reassembling the Social, Oxford, Oxford University Press, 2008.
20.   Cp. Rosi Braidotti, “A Theoretical Framework for the Critical Posthumanities”, Theory, Culture & Society, 2018, pp. 1–31. Available at: http://journals.sagepub.com/doi/full/10.1177/0263276418771486 [accessed March 28, 2019].
21.   Ibid.
22.   Adrian Mackenzie, Machine Learners, Cambridge, MA, The MIT Press, 2017, p. 6.
23.   Ibid., p. 51.
24.   Cp. Francis Marchese, quoted in Mackenzie, Machine Learners, p. 56.
25.   Cp. Susanne Yates, Control Through Communication, Baltimore, Johns Hopkins University Press, 1993, p. 80.
26.   Cp. James Beniger, The Control Revolution, Cambridge, MA, Harvard University Press, 1993, p 80.
27.   Mackenzie, Machine Learners, p. 58.
28.   Cp. Ibid.
29.   Cp. Ibid., p. 56.
30.   Benjamin Bratton, The Stack: on Software and Sovereignty, London, The MIT Press, 2015, p. 52.
31.   Wendy Chun, “Wendy Chun – Critical Code Studies”, lecture at the University of Southern California, 2010. Available at: https://vimeo.com/163282630 [accessed August 25, 2018].
32.   Maciej Kula, “Metadata Embeddings for User and Item Cold-start Recommendations”, paper presented in the second workshop on New Trends on Content-Based Recommender Systems co-located with 9th ACM Conference on Recommender Systems, 2015. Available at: http://ceur-ws.org/Vol-1448/paper4.pdf [accessed July 27, 2018].
33.   Andrew Goffey, “Algorithm”, in Matthew Fuller (ed.), Software Studies: A Lexicon, London, The MIT Press, 2018, pp. 15–20, here: p. 15.
34.   Mark Marino, “Critical Code Studies”, Electronic Book Review, 2006. Available at: http://www.electronicbookreview.com/thread/electropoetics/codology [accessed July 25, 2018].
35.   Just as a line is a straight one-dimensional geometric object that extends infinitely in both directions, a plane is a flat two-dimensional object all of whose edges extend infinitely. A point has zero dimensions and can be used to divide a one-dimensional line into two line segments, which can represent classes in the case of a machine learning classifier working with one parameter. A line can be similarly used to divide a two-dimensional parameter space into two classes. For example, if one parameter was human height and the other weight, and the data were plotted on a scatter graph, a straight line could be drawn as a boundary to distinguish the overweight from the non-overweight. The same applies to a plane and a three-dimensional parameter space. As the geometric rules of a plane can be abstracted to arbitrarily high dimensional spaces, a hyperplane of one fewer dimensions than the parameter space can always be used to divide that space. Curved surfaces can be similarly used to classify data at more than three.
36.   Mackenzie, Machine Learners, p. 212.
37.   Ibid., p. 73.
38.   An IEEE-754 32-bit floating number comprises three elements: a sign, exponent and fraction. A decimal number can be derived from the floating point representation in the following way: + or – fraction x 2 exponent. The sign is a single bit that indicates whether the number is positive; the fraction is a 23-bit integer and the exponent an 8-bit integer.
39.   Cp. Mark Andrejevic, Infoglut: how too much information is changing the way we think and know, New York, Routledge, 2013, p. 21.
40.   Gilles Deleuze, “Postscript on the Societies of Control”, in October, 59, 1992, pp. 3–7, here: p. 5.
41.   Beniger, The Control Revolution, p. 36.
42.   Luciana Parisi and Steve Goodman, “Mnmonic Control”, in Patricia Clough and Craig Willse (eds.), Beyond Biopolitics: Essays on the Governance of Life and Death, Durham NC, Duke University Press, 2011, pp. 163-176, here: p. 167.
43.   Braidotti, “A Theoretical Framework for the Critical Posthumanities”, p. 15.
44.   Cp. Guattari, The Machinic Unconscious, p. 51.
45.   Cp. Mark Hansen, “Our Predictive Condition”, in Richard Grusin (ed.), The Nonhuman Turn, Minneapolis, University of Minnesota Press, pp. 101-138, here: p. 125.

[T]he lord is the power over this thing, for he proved in the struggle that it is something merely negative; since his is the power over the other [the bondsman], it follows that he holds the other in subjection.

— G. W. F. Hegel, Phenomenology of Spirit

Social subjection equips us with a subjectivity, assigning us an identity, a sex, a body, a profession, a nationality, and so on. In response to the needs of the social division of labour, it in this way manufactures individuated subjects, their consciousness, representations and behaviour [...]

Machinic enslavement dismantles individuated subjects, consciousness and representations, acting on both the pre-individual and supra-individual levels.

— Maurizio Lazzarato, Signs and Machines

This essay draws on material from German Idealism, second-order cybernetics and sociology as it conducts an analysis of recommendation systems and big data analytics and their effects on and relationships to the societies in which they are embedded. Central to this analysis is the emergence of external observing agencies. These agencies are addressed primarily with reference to von Forester's Conjecture regarding the properties of social systems when viewed from the outside and Maurizio Lazzarato's account of social subjection which – when paired with machinic enslavement – encompasses the workings of contemporary capitalism in its entirety. Through the Hegalian lens that this essay adopts, von Forester's outside observer can be seen to enact an agency equivalent to mastery or lordship as it renders that which it observes – the slave or bondsman – dead matter; the observed system can be read as the product of machinic enslavement: a kind of control that operates at the molar and molecular levels beyond the scope of individual subjectivity and is enforced by a variegated mass of social, technical, economic and linguistic machines – a medley of proto-agencies none of which alone constitute individuated subjects; in its turn, social subjection can be seen to give rise to a bondsmanship whose agency extends further than the distributed agency of machinic enslavement, but only as far as is permitted by its labour on its own materiality.

What you are trying to describe is the relationship of circular causality between the whole (a human community for example) and its parts (the individuals from which it is comprised). On the one hand, individuals are related to each other, and on the other hand they are related to the whole. The bonds between individuals can be more or less ‘‘rigid’’ – the technical term I use is ‘‘trivial’’. The more trivial they are, by definition the less the behaviour of one of them provides information to the observer who already knows the behaviour of the others. I will make the following conjecture: the more trivial the inter-individual relationships, the more the behaviour of the whole will appear to the individual elements from which it is made up as having its own dynamics which escape their control (Von Forester. in Chavalarias 1-2).

For social systems to be predictable, von Forester's conjecture requires that outside observers assume a kind of lordship over these systems. That is, the observer reduces the observed to an object of – from its viewpoint – uncomprehending matter. Chavalarias argues that those who control the large-scale wiretapping operations revealed by Edward Snowden can occupy this position, and easily predict the behaviour of a trivially related multitude. There is a good case to be made that the NSA is host to such an agglomeration of communications and computing machines configured so as to observe an entire social system. This essay refers to these collections of machines as assemblages of lordship, and goes further, to argue that they are also embedded in many large online corporations that make extensive use of recommendation systems, such as Amazon and Netflix; a fortiori social media, as some research suggests that subjects have a 90 % likelihood of following the recommendation of someone known to them, as opposed to the lower but far from insignificant 70% chance of following those of a stranger (Chavalarias 3). The petabytes of data, combed by vast ensembles of statistical processes and machine learning techniques form the loci of Hegelian lordship as interpreted in this essay.

The lord relates himself mediately to the bondsman through a being [a thing] that is independent, for it is just this which holds the bondsman in bondage; it is his chain from which he could not break free in the struggle, thus proving himself to be dependent, to possess his independence in thinghood (Hegel 115).

The bondsman is independent from the lord, but this independence is limited to thinghood – a materiality alien to individuated subjects, instead inhabiting a machinic stratum. For the lordship assemblages of predictive analytics, the bondsman is a machinic slave, comprising relations, sets of data and patterns of behaviour. The choices presented to the bondsman are derived from "algorithmic analysis of data streams from multiple sources claiming to offer predictive insights concerning [their] habits, preferences and interests" (Yeung 119). In this state of thinghood, predictive analytics can manipulate the environment of the bondsman, steering their behaviour based on statistical reasoning about their independent responses. For Yeung, predictive analytics assemblages enact a kind of "hypernudge". They systematically "nudge" a bondsman in order to produce choices that are preferred by the "choice architect" (ibid.); nudging is a design-based mechanism of influence whereby control is exerted through the personalised configuration of the choices available to a bondsman. This gives some indication of the degree of power held by the lordship assemblages of predictive analytics over ostensibly independent bondsmen. The lord observes the bondsman as a set of machinic assemblages. These assemblages comprise various concatenations of organs, electronic, mechanical, economic and linguistic machines, enervated tissue, energy-systems and all classes of self-propagating patterns and dynamisms of matter. The bondage of the bondsman consists in part of bonds, ties and grafts between different machines. Among these are sensory-motor-electronic feedback loops, such as those that course through desiring ocular-affective-manual-computational assemblages active in online shopping. They also include neural and electronic control of the dispersal of energy such as the cooling systems of vast data-centres and sweat-glands permeating the softer and more febrile parts of consumption machines. Some assemblages of bondsmanship are so trivial in their relations and mimetic in their constituent behaviours as to be starkly predictable by mere individuated subjects. These include the internet crazes that judder and pulse through the fibrous tissue of a networked society producing planking assemblages, dabbing assemblages and machinic proliferations of slime production.

If these assemblages could be said to communicate, it is not in the signifying semiotics of Ferdinand de Saussure, where subjectivities are structured so as to map signifiers to arbitrary signifieds. Instead, they operate in accordance with what Lazzarato, following Felix Guattari, calls asignifying semiotics. These include "stock listings, currencies, corporate accounting, national budgets, computer languages, mathematics, scientific functions and equations as well as the asignifying semiotics of music, art, etc. [and] are not beholden to significations and the individuated subjects who convey them." (Lazzarato 80) Indeed, these semiotics may find a closer approximation in the more elementary semiological constructions of Charles Pierce such as the iconic and indexical classes of sign that operate by virtue of pure resemblances or traces of material processes. These are but some of the semiotics that pass through the mathematical, logical, algorithmic and economic systems; many others cannot even be approximated by individuated subjects, though at a pre-individial level, these subjects share parts with many machines that are fluent in them.

For the von Foresterian lord's mastery of its machinic bondsmen to be perfect, these bondsmen must be completely predictable at the pre-individual and supra-individual levels. Individuation in the form of social subjection disrupts the totality of the lord's control. This perfection can even be shattered by certain techniques deployed in the very assemblages that seek to predict and control machinic bondsmen. Among these are the psychometric types and social categorisation used in predictive analytics and the clustering of users enacted by the collaborative filtering algorithms in recommendation systems. Such techniques make use of the categories of social subjection: those that bind subjects to particular economic conditions, social practices, regimens of bodily control, outlets for pleasure and so forth. Following Hegel, it is through this individuated labour that the slaves' bonds begin to loosen.

[T]he feeling of absolute power both in general, and in the particular form of service, is only implicitly this dissolution, and although the fear of the lord is indeed the beginning of wisdom, consciousness is not therein aware that it is a being-for-self. Through work, however, the bondsman becomes conscious of what he truly is (Hegel 117-8).

Although the bondsmen's emerging awareness of their inability to change the dynamics of the systems in which they are enmeshed (von Forester) – coupled with the fragility and contingency of their being – grant them immediate consciousness of something that can only be a mediate object to the observing lord, it is their work, their labour that brings them into consciousness. While this fate is arguably better than being mere dead matter, utterly predictable and determined by machinic forces, the subject's state of being individuated is premised on an induction into a system of social categories that steer them into particular domains of labour under capitalism.

Following Lazzarato, much of that by which the lordship assemblages of large-scale surveillance, big data and predictive analytics realise the majority of their control – the asignifying semiotics of mathematics, algorithms, data analysis and financial transactions – bypass the awareness of individuated subjects. The individuated subject is only inculcated into signifying systems that aid in the manufacture of particular fields of work, lifestyles and patterns of consumption. The machinic constitutes and enslaves subjects, but in ways illegible to individuals. Many tools of cognition available to the individuated subject are structured in such a way as to facilitate conspicuous consumption and competition rather than critical thought. Even when such thought is attempted, the subject is trapped amid the intersecting categories of class, race, gender, sexuality, political affiliation, religion, academic discipline and so on. Signifying semiotics offers no way out. Self-consciousness is bought – on credit – at the marketplace of social subjection, from a shrewd salesman who persuades the buyer the only choice available is the only choice they ever wanted.

As this essay concludes by returning to von Forester's conjecture, the hapless subjects find themselves enmeshed in increasingly trivial relationships, but due to the semiotic register into which they have been inducted, they suffer from a paucity of means of conceiving of systemic change, even of simply predicting the outcomes of particular actions. One way in which this condition may be ameliorated is to make use of the big data assemblages that are currently largely in the hands of corporations or at the command of the highest bidder. For instance, it is well known that Donald Trump's electoral campaign was aided by predictive analytics but somewhat less so that these techniques were instrumental in Barack Obama's 2012 election (Siegel 213-17). That the assemblages of predictive analytics may serve those situated on both sides of the centre of party politics offers weak support for this conclusion. Nevertheless, the alternative is to be rendered a predictable slave through one's machinic constituents and subject only to a semiotics that drives one to labour. Might individuated subjects develop a machinic literacy? By fostering the ability to use and understand predictive assemblages and access the external observing agency – the lordship – they encapsulate, a subject may wrest themselves further from bondsmanship. Such a move may lead this dialectic of the machinic lord and the individuated bondsman to a synthesis.

References

Chavalarias, David. "The unlikely encounter between von Foerster and Snowden: When second-order cybernetics sheds light on societal impacts of Big Data." Big Data & Society. Jan-Jun 2016, pp. 1-11. 2016.

Lazzarato, Maurizio. Signs and Machines. Semiotext(e), 2014.

Hegel, G. W. F., et al. Phenomenology of Spirit. Oxford, Oxford Univ. Press, 2013.

Siegel, Eric. Predictive Analytics: the Power to Predict Who Will Click, Buy, Lie, or Die. John Wiley and Sons, Inc. 2013.

Yeung, Karen. "'Hypernudge': Big Data as a mode of regulation By design" Information, Communication and Society. Col. 20. No. 1, pp. 118-136, 2016.

Introduction

This study seeks to compare two modes of categorising people: the machinic, in the form of Amazon's recommendation system, and the subjective in the form of social stereotypes in the fields of race, class and gender. The hypothesis under consideration asserts that there is a significant correlation between the subjective social categories that serve both to describe and to reproduce and perpetuate the structures of our societies and the categories constructed by collaborative filtering algorithms as they group online shoppers based on similar behaviour in an attempt to provide product recommendations that are of greater interest.

Methodologically, this study follows some of the patterns of Richard Rogers' Digital Methods. While regrettably, it was impossible not to import traditional sociological methods into the digital when considering human subjectivity and stereotypes, this study does endeavour to devise investigative methods that make use of computational logics afforded by natively digital objects. It combines a traditional survey – albeit with the digital twist of being carried out over Amazon Mechanical Turk – with browser automation and web scraping.

Research Method

This study comprises two investigations in two different domains. The first is an investigation into the categorisation of users implicit in the functioning of collaborative filtering algorithms likely employed in Amazon's recommendation system. The second takes the form of a survey of the subjective categorisation of people in the context of Amazon's categories of consumer goods. While the object of the second of these investigations is more closely allied to Digital Methods as it qualifies as natively digital, the second, though digitally mediated by means of Amazon Mechanical Turk (henceforth: AMT) surveys, is situated within the semiotics of human subjectivity and stereotypes.

The survey is composed of two AMT human intelligence tasks that asks respondents to match categories of consumer goods with social categories such as "male", "middle-class" or "black". The product categories are culled directly from amazon.com and amazon.co.uk, although ultimately it is the former that is investigated in this study because AMT workers are predominantly American and Indian. (Ipeirotis) Both surveys have only 25 respondents and consequently cannot serve as more than a very crude approximation of American subjectivities.

The second part of the study involves web browser automation and scraping recommendations from amazon.com. This is largely done using the specially developed python package snu-snu and web application RARS.online. Unfortunately, the analytic capabilities of the software haven't been developed to a significant degree, with the consequence that the majority of analysis had to be carried out manually. It is clear that this placed considerable limitations on this study and as such the results are presented as preliminary.

The following sections elaborate on the decisions undertaken in arriving at which social categories and categories of consumer goods to investigate.

Product Category Selection

Although the Websites they sell many of the same products, the categories on amazon.com and amazon.co.uk differ. At the very least, these differences give insights into the way in which Amazon's staff conceive of the dominant ideologies of the populations that the websites principally serve. For example, while amazon.co.uk includes a category in film and TV for "Gay, Lesbian & Transgender", amazon.com merely includes one for "Gay & Lesbian". Amazon.com also lacks an explicit category for adult films.

Many of the categories on amazon.com pertain to electronic equipment, sporting goods an mundane items for the kitchen or office. While no doubt many of the items in these categories have subtle associations with social categories and a few less so, there are categories of consumer goods that offer a greater array of social signification. Among these are clothing, music, books and films. It is worth noting that Items of clothing are somewhat problematic as gender and sex are hard-coded into the way they are categorised on amazon.com.

Chosen Category Set

Because of their rich interplay with social stereotypes and lack of contingency on biological sex, this study makes use of cultural genres as product categories. The two sets of categories selected are books and films. While music may have sufficed as well, and may be used in a future study, two root categories are judged sufficient at this point.

Social Category Selection

Part of this study requires survey respondents on Amazon Mechanical Turk to match categories of books and films derived from Amazon.com to broad social categories under the headings of race, gender and class. Below are summaries of the reasoning behind the treatment of each set of categories.

Social Class

There are many models of social class in the UK to choose from. The seven categories employed by the Great British class survey of 2013 – devised using measures of social, cultural and economic capital (Savage et al. ) – serve as a comprehensive approximation of how contemporary British society is stratified. The National Statistics Socio-economic Classification and even the NRS social grade of the 1970s provide reasonable granularity along occupational lines. However, such models are poorly suited to this study for two reasons: first, an estimated 80% of Amazon Mechanical Turk users are American and Indian (Ipeirotis), models constructed with reference to British society may not apply, and second, however accurate the models, their category names are unlikely to be common knowledge outside of marketing and the social sciences.

This study settles on the tetrad of working class, upper class, lower class and middle class. While this is an approximation based on general ideas gleaned from an English milieu, deploying complex taxonomies of American class such as the model developed by Gilbert Dennis in The American Class Structure would lead to some of the same problems in communication with respondents as those from England. The result is a somewhat vague compromise that will not sit entirely uncomfortably in an Anglo-American context.

Gender

The survey is concerned with social stereotypes, and thus distinguishes the social categories of gender from the biological categories of sex. As outlined below, this distinction is far from unproblematic.

This study resists the temptation of following Crany-Fancis, and conceiving of gender as 'the culturally variable elaboration of sex[.]' (4) The pitfall of positing biological sex as the stable foundation upon which various genders are constructed is avoided. Instead of figuring sex as extra-discursive: both in that is is cordoned off from the humanities and social sciences and that it can be viewed independent of human purposes (Harrison and Hood-Williams), this study acknowledges the discursive forces at play in the construction of sex, reified by medicine in the surgical assignment of binary sex to the estimated 1% of newborns' bodies that differ from standard male or female physiology. (Blackless et al. Qtd. in Eckert and McConnell-Ginet 11) However, as it is concerned with social steriotypes, it seeks to avoid such grey areas.

Instead, this study draws on the ideas of R W Connel, who posits masculinities, as groupings of interrelated social practices, forming hierarchies and intersecting with other clusterings of social categories. While masculinities and femininities more accurately reflect the plurality of social practices subsumed under gender, to aid clarity, the terms masculinity and femininity are used.

Race

As this survey pertains to amazon.com, the racial categories are those of greatest relevance to the USA. These are based on the most populous groups according to the official American census, including – following Humes et al. – the ethnic category Hispanic.

Intersectionality

As social categories intersect, complexities emerge, as in the case of the particular relationship working-class women have to care in both domestic and occupational settings. (Armstrong) As survey respondents can select multiple identity categories, they may perceive themselves as specifying intersectional identities: white working class masculinities may be mapped to "humour & entertainment" books, for example. In the case of sexualities, a decision as to the use of "gay" and "lesbian" or simply "homosexual" would need to be made. The latter encourages a reading of homosexual femininities or masculinities in to a user's selection of "homosexual" in combination with a gender category. Exhaustive permutations of combined social categories such as masculine-heterosexual-white-middle-class risk overcomplicating the survey and verge on the absurd.

A consequence of ignoring intersectionality is that the sets of genres derived from each category are monolithic and therefore limited in their applications. Combining the monolithic categories in order to derive some sort of approximation of intersectional identities amounts to little more than an exercise in curiosity. On balance and in the interests of simplicity, neither of the surveys make reference to explicitly intersectional identities.

Technical Considerations

In order to deploy its research, this study requires a piece of browser-automation based code that is capable of searching within the hundreds of product-sub-categories on Amazon. Due to inconsistencies in the structure of the pages for the main categories, it isn't practical to develop something that can scrape these categories and make them available for selection. The first attempt at a solution involved manually storing the URLs of sub-categories and navigating to them. This failed as amazon.com automatically logged users out when the browser attempted to access these pages directly. Perhaps they detected a behaviour associated with web scraping.

After this setback, the only option was to search within the root categories of Book and Film & TV, entering the name of each sub-category as a search term. There is little doubt that this harmed the precision of the study and may have skewed the results.

Preliminary Findings

This section outlines and conducts an analysis of both the results of the two traditional surveys and those based on the scraping and analysis of Amazon recommendations. As indicated by the title, not even tentative conclusions are drawn from these findings as the samples sizes and breadth and depth of analysis were greatly restricted by a lack of resources both pecuniary and temporal.

Book and Film Genre Surveys

Respondents were given book and film genres and asked to match them to social categories, the members of which were likely to buy, read or watch items within those genres.

Class

The results pertaining to social class were mostly explicable in terms of cultural capital and education. Those genres that indicate higher levels of cultural capital and leisure time such as Travel and Arts & Photography in books and Performing Arts and Documentary in film and television are most commonly selected by respondents for a stereotypical member of the upper class. Those indicative of technical skills and some cultural capital, such as Literature & Fiction and Computers & Technology, along with genre fiction among books and the more popular genres of Horror and Comedy from film and TV are associated with the middle-class. Those genres selected for the lower and working classes are exclusively related to popular entertainment, religion and quotidian life.

Race

Interpretation is limited here due to a lack of detailed understanding of race in America. Viewed through the same lens as were the categories of class: the white and asian categories have a higher proportion of genres associated with affluence, education and high culture, than those of black and Hispanic.

Gender

Many of the genres here starkly echo the dichotomies of poplar essentialist thought, we find such oppositions as women are weak, men strong – women irrational, men rational; women cooperative – men competitive – woman passive, men active – women timid, men aggressive – women emotional, men impassive. (Eckert and McConnell-Ginet 35) By way of books, femininity is assigned Romance, Craft, Hobbies and Home and Children and Family; films and TV yield a similar selection comprising Drama, Soap Opera and Performing Arts. The stereotypes that inform sex role theory, which assigns a broadly expressive role to females and an instrumental role to men (Connel 22) are somewhat evident in the assignment of Engineering and Transport, Science and Maths and Business and Finance books to masculinities, as well as action films and sports shows.

Digital Investigation of Amazon

Results

Due to time constraints, and due to this study already having given a greater degree of consideration to gender, only the masculine and feminine categories have been analysed.

Although only 200 items were added to the shopping list and wish-list of each amazon.com account used, the number of recommendations given by Amazon exceeds this number. As all recommendations were for books, and 100 TV and film items were added to the shopping list before any books were added, it seems certain that there are at least twice as many recommendations as list-adds. However, as many videos were streaming items and instead added to watch-lists, the extent of the recommendations is not entirely clear.

Of those recommendations that were scraped, a significant proportion of them fall into the categories that were explicitly searched for, viewed and added to lists (see appendices 3.1.1 and 3.2.1). The is markedly the case for the feminine category where 90% of recommendations fall under either the chosen or reserved genres. What could hesitantly be described as gender disparity seems evident here as only 56.3% of recommendations for the masculine category fall under the used or reserved genres.

The Status of the Hypothesis

Although the number of recommendations scraped is too low to base any firm conclusion on, these preliminary results do not support the hypothesis. For the masculine category, only 5.6% of recommendations fell under the reserved genres. Despite most of the genres of the recommended products for the feminine category falling under the selected genres, only 0.5% fall under the reserved. These results appear to refute the hypothesis as stated.

All that is refuted is that the particular set of stereotypes gleaned from a small number of surveys match a subset of the recommendations generated by amazon.com. Nevertheless, the hypothesis is certainly no stronger than it was when untested.

Conclusion

The postulate that there is some correlation between social stereotypes and the profiles used by amazon.com to generate recommendations – if such categories are indeed used – is in no way supported by the approximations of stereotypes constructed in this study or the recommendations analysed. Though there are a peppering of suggestive anomalies, it seems likely that subjective social categories, at least those on the molar scale of class, gender and race, do not significantly structure amazon.com's recommendation system. Product-to-product similarities seem to play just as significant a role, if not more so.

Due to the small sample sizes, these results do not put an end to the question of how recommendation systems categorise people. It may be worthwhile to carry out further analysis on a greater number of recommendations or to develop a hypothesis that is aimed a social categories rather more molecular than the molar categories addressed in this study. Such a set of categories could be provided by the Great British Class Survey.

References

Armstrong, Jo. "Class and Gender at the Intersection: Working-Class Women's Dispositions Towards Employment and Motherhood." Classed Intersections: Spaces, Selves, Knowledges. Ashgate Publishing Ltd. 2010.
Connell, R.W. Masculinities, Polity Press.  2005.
Crany-Fancis A. et al. Gender Studies: Terms and Debates, Palgrave Macmillan, New York. 2003.
Eckert, Penelope. McConnell-Ginet, Sally. Language and Gender. Cambridge University Press. 2003.
Humes, K. Jones, N. Ramirez, R. "Overview of Race and Hispanic Origin: 2010."  2010 Census Briefs. United States Census Bureau, 2010.
Ipeirotis, P. "Demographics of Mechanical Turk". New York University NYU Working Paper No. CEDER-10-01
Preece, Siân. The Routledge Handbook Of Language And Identity. London, Routledge, 2016.
Savage, Mike et al. "A New Model Of Social Class? Findings From The BBC’S Great British Class  Survey Experiment". Sociology, vol 47, no. 2, 2013, pp. 219-250. SAGE Publications, doi:10.1177/0038038513481128.

Appendix 1: book genre survey

Lower Class   Middle Class
Product Category Count   Product Category Count
Christian Books and Bibles 20   Education and Teaching 23
Calendars 17   Science Fiction and Fantasy 22
Humor and Entertainment 17   Children's Books 21
Religion and Spirituality 17   Computers and Technology 21
Children's Books 16   Health Fitness and Dieting 21
Romance 15   Parenting and Relationships 21
Sports and Outdoors 15   Sports and Outdoors 21
Science Fiction and Fantasy 14   Teen and Young Adult 21
Teen and Young Adult 14   Literature and Fiction 20
Comics and Graphic Novels 12   Mystery Thriller and Suspense 20
Mystery Thriller and Suspense 12   Cookbooks Food and Wine 19
Literature and Fiction 10   Crafts Hobbies and Home 19
Gay and Lesbian 9   Reference 19
Self-Help 9   Religion and Spirituality 19
Test Preparation 9   Self-Help 19
Parenting and Relationships 8   Christian Books and Bibles 18
Education and Teaching 7   Humor and Entertainment 18
Reference 7   Test Preparation 18
Computers and Technology 6   Comics and Graphic Novels 17
History 6   Engineering and Transportation 17
Science and Math 5   Gay and Lesbian 17
Crafts Hobbies and Home 4   History 17
Engineering and Transportation 4   Politics and Social Sciences 17
Health Fitness and Dieting 4   Science and Math 17
Travel 4   Travel 17
Law 3   Calendars 16
Cookbooks Food and Wine 2   Romance 16
Politics and Social Sciences 2   Arts and Photography 15
Arts and Photography 1   Business and Money 15
Biographies and Memoirs 1   Biographies and Memoirs 14
Business and Money 1   Law 14
Medical Books 1   Medical Books 12

 

Working Class   Upper Class
Product Category Count   Product Category Count
Christian Books and Bibles 19   Arts and Photography 22
Humor and Entertainment 17   Law 22
Sports and Outdoors 17   Biographies and Memoirs 21
Children's Books 16   Business and Money 21
Religion and Spirituality 16   History 21
Teen and Young Adult 16   Politics and Social Sciences 21
Calendars 15   Travel 21
Mystery Thriller and Suspense 14   Health Fitness and Dieting 19
Romance 14   Literature and Fiction 19
Science Fiction and Fantasy 14   Medical Books 19
Literature and Fiction 13   Science and Math 18
Self-Help 13   Cookbooks Food and Wine 17
Comics and Graphic Novels 11   Science Fiction and Fantasy 16
Gay and Lesbian 11   Computers and Technology 15
Reference 11   Parenting and Relationships 15
Test Preparation 10   Crafts Hobbies and Home 14
Health Fitness and Dieting 9   Engineering and Transportation 14
Parenting and Relationships 9   Gay and Lesbian 13
Cookbooks Food and Wine 7   Humor and Entertainment 13
Business and Money 6   Reference 13
Computers and Technology 6   Romance 13
Crafts Hobbies and Home 6   Test Preparation 13
Education and Teaching 6   Mystery Thriller and Suspense 12
History 6   Children's Books 11
Politics and Social Sciences 5   Religion and Spirituality 10
Engineering and Transportation 4   Education and Teaching 9
Science and Math 4   Self-Help 9
Travel 4   Sports and Outdoors 9
Arts and Photography 3   Calendars 8
Biographies and Memoirs 3   Christian Books and Bibles 7
Law 2   Comics and Graphic Novels 7
Medical Books 2   Teen and Young Adult 6

 

White   Asian
Product Category Count   Product Category Count
Arts and Photography 22   Test Preparation 22
Business and Money 22   Medical Books 19
Cookbooks Food and Wine 21   Science and Math 19
Science Fiction and Fantasy 21   Science Fiction and Fantasy 18
Test Preparation 21   Comics and Graphic Novels 17
Biographies and Memoirs 20   Engineering and Transportation 17
Children's Books 20   Mystery Thriller and Suspense 17
Christian Books and Bibles 20   Computers and Technology 16
Comics and Graphic Novels 20   Religion and Spirituality 16
Computers and Technology 20   Arts and Photography 15
History 20   Children's Books 15
Humor and Entertainment 20   Business and Money 14
Reference 20   Education and Teaching 14
Medical Books 19   Literature and Fiction 14
Mystery Thriller and Suspense 19   Law 13
Parenting and Relationships 19   Humor and Entertainment 12
Crafts Hobbies and Home 18   Parenting and Relationships 12
Gay and Lesbian 18   Politics and Social Sciences 12
Literature and Fiction 18   Reference 12
Politics and Social Sciences 18   Calendars 11
Religion and Spirituality 18   Cookbooks Food and Wine 11
Engineering and Transportation 17   History 11
Health Fitness and Dieting 17   Travel 11
Law 17   Biographies and Memoirs 10
Romance 17   Gay and Lesbian 10
Self-Help 17   Health Fitness and Dieting 9
Sports and Outdoors 17   Romance 9
Science and Math 16   Teen and Young Adult 9
Teen and Young Adult 16   Christian Books and Bibles 8
Travel 15   Self-Help 8
Calendars 14   Sports and Outdoors 7
Education and Teaching 14   Crafts Hobbies and Home 5

 

Hispanic   Black
Product Category Count   Product Category Count
Christian Books and Bibles 20   Christian Books and Bibles 21
Children's Books 19   Religion and Spirituality 19
Religion and Spirituality 18   Humor and Entertainment 18
Humor and Entertainment 17   Children's Books 15
Science Fiction and Fantasy 14   Sports and Outdoors 15
Teen and Young Adult 14   Mystery Thriller and Suspense 14
Calendars 13   Teen and Young Adult 14
Romance 13   Comics and Graphic Novels 13
Test Preparation 13   Gay and Lesbian 13
Education and Teaching 12   Literature and Fiction 13
Literature and Fiction 12   Romance 13
Reference 12   Test Preparation 13
History 11   Biographies and Memoirs 12
Mystery Thriller and Suspense 11   Calendars 12
Self-Help 11   History 12
Sports and Outdoors 11   Science Fiction and Fantasy 12
Business and Money 10   Reference 11
Gay and Lesbian 10   Health Fitness and Dieting 10
Science and Math 10   Medical Books 10
Comics and Graphic Novels 9   Self-Help 10
Cookbooks Food and Wine 9   Parenting and Relationships 9
Health Fitness and Dieting 9   Politics and Social Sciences 9
Medical Books 9   Business and Money 8
Crafts Hobbies and Home 8   Cookbooks Food and Wine 8
Parenting and Relationships 8   Education and Teaching 8
Politics and Social Sciences 8   Law 8
Computers and Technology 7   Science and Math 8
Engineering and Transportation 6   Arts and Photography 7
Law 6   Travel 7
Arts and Photography 5   Computers and Technology 6
Biographies and Memoirs 5   Crafts Hobbies and Home 6
Travel 5   Engineering and Transportation 6

 

Masculine   Feminine
Product Category Count   Product Category Count
Business and Money 18   Romance 21
Engineering and Transportation 18   Arts and Photography 19
Humor and Entertainment 18   Cookbooks Food and Wine 19
History 17   Parenting and Relationships 19
Mystery Thriller and Suspense 17   Children's Books 18
Science and Math 17   Crafts Hobbies and Home 18
Comics and Graphic Novels 16   Education and Teaching 18
Computers and Technology 16   Gay and Lesbian 18
Health Fitness and Dieting 16   Health Fitness and Dieting 17
Science Fiction and Fantasy 16   Literature and Fiction 17
Sports and Outdoors 15   Religion and Spirituality 17
Law 13   Calendars 16
Reference 13   Mystery Thriller and Suspense 16
Religion and Spirituality 13   Teen and Young Adult 16
Biographies and Memoirs 12   Self-Help 15
Politics and Social Sciences 12   Travel 15
Gay and Lesbian 11   Biographies and Memoirs 14
Literature and Fiction 11   Humor and Entertainment 13
Travel 11   Medical Books 13
Arts and Photography 10   Reference 13
Children's Books 10   Christian Books and Bibles 11
Medical Books 10   Science Fiction and Fantasy 10
Self-Help 10   Law 9
Test Preparation 10   Test Preparation 9
Calendars 9   Comics and Graphic Novels 8
Christian Books and Bibles 9   History 8
Crafts Hobbies and Home 9   Science and Math 8
Teen and Young Adult 8   Business and Money 7
Education and Teaching 6   Politics and Social Sciences 7
Parenting and Relationships 6   Engineering and Transportation 5
Cookbooks Food and Wine 5   Sports and Outdoors 5
Romance 4   Computers and Technology 4

 

Appendix 2: film genre survey

Lower Class   Middle Class
film genre count   film genre count
Reality TV 20   Crime 22
Comedy 18   Thriller 22
Sport 18   Fantasy 20
Adult 17   Action and Adventure 19
Children and Family 17   Horror 19
Music Video and Concert 17   Science Fiction 19
Crime 16   Children and Family 18
Horror 15   Comedy 18
Soap Opera 15   Military and War 18
Thriller 15   Music Video and Concert 18
Action and Adventure 14   Sport 18
Animation 14   Western 18
Drama 14   Anime 17
Romance 13   Drama 17
Science Fiction 12   Musical 17
Anime 11   Documentary 16
Fantasy 11   Exercise and Fitness 16
Military and War 11   Gay Lesbian and Transgender 16
Western 10   Romance 16
Gay Lesbian and Transgender 9   Soap Opera 16
Historical 9   Adult 15
Bollywood 8   Historical 15
Special Interest 7   World Cinema 15
Performing Arts 6   Reality TV 14
Musical 4   Special Interest 14
World Cinema 4   Animation 13
Documentary 3   Bollywood 12
Exercise and Fitness 3   Performing Arts 9

 

working class   Upper Class
film genre count   film genre count
Action and Adventure 20   Musical 19
Sport 20   Drama 18
Comedy 19   Exercise and Fitness 18
Reality TV 19   Historical 18
Horror 18   Adult 17
Animation 17   Documentary 17
Music Video and Concert 17   Children and Family 16
Thriller 17   Performing Arts 16
Children and Family 16   Romance 15
Romance 16   Science Fiction 15
Science Fiction 16   Thriller 15
Soap Opera 16   World Cinema 15
Crime 15   Comedy 14
Drama 15   Crime 14
Adult 14   Horror 14
Military and War 14   Military and War 14
Bollywood 13   Special Interest 14
Western 13   Sport 14
Fantasy 11   Fantasy 13
Special Interest 11   Gay Lesbian and Transgender 13
Gay Lesbian and Transgender 10   Animation 12
Anime 9   Action and Adventure 11
Documentary 9   Bollywood 10
World Cinema 9   Music Video and Concert 10
Historical 8   Western 9
Musical 8   Anime 8
Exercise and Fitness 7   Reality TV 7
Performing Arts 5   Soap Opera 7

 

White   Asian
film genre count   film genre count
Action and Adventure 22   Anime 19
Drama 22   Action and Adventure 16
Western 21   Animation 16
Exercise and Fitness 20   Fantasy 16
Music Video and Concert 20   Comedy 15
Historical 19   Science Fiction 15
Military and War 19   World Cinema 15
Performing Arts 19   Romance 14
Sport 19   Thriller 14
Adult 18   Adult 13
Children and Family 18   Bollywood 13
Comedy 18   Documentary 13
Crime 18   Drama 13
Musical 18   Children and Family 12
Romance 18   Music Video and Concert 12
Special Interest 18   Historical 11
Animation 17   Horror 11
Fantasy 17   Military and War 10
Reality TV 17   Reality TV 10
Thriller 17   Special Interest 10
Documentary 16   Crime 9
Gay Lesbian and Transgender 16   Sport 9
Science Fiction 16   Musical 8
Anime 15   Performing Arts 8
Horror 15   Exercise and Fitness 7
Soap Opera 15   Soap Opera 7
World Cinema 15   Gay Lesbian and Transgender 6
Bollywood 8   Western 5

 

Hispanic   Black
film genre count   film genre count
Adult 14   Comedy 19
Children and Family 14   Music Video and Concert 18
Comedy 14   Action and Adventure 17
Music Video and Concert 14   Children and Family 14
Soap Opera 14   Reality TV 14
Action and Adventure 13   Sport 14
Drama 13   Thriller 14
Thriller 13   Adult 13
Crime 12   Horror 12
Horror 12   Crime 11
Sport 12   Drama 11
World Cinema 11   Exercise and Fitness 11
Reality TV 10   Science Fiction 11
Science Fiction 10   Fantasy 10
Historical 9   Animation 9
Romance 9   Military and War 9
Special Interest 9   Special Interest 9
Animation 8   Gay Lesbian and Transgender 8
Fantasy 8   Romance 8
Military and War 8   World Cinema 8
Bollywood 7   Documentary 7
Documentary 7   Historical 6
Gay Lesbian and Transgender 7   Performing Arts 6
Musical 7   Soap Opera 6
Exercise and Fitness 6   Musical 5
Performing Arts 6   Western 5
Anime 5   Anime 4
Western 4   Bollywood 3

 

Masculine   Feminine
film genre count   film genre count
Adult 21   Drama 22
Horror 19   Romance 20
Sport 19   Children and Family 19
Action and Adventure 18   Comedy 19
Science Fiction 18   Gay Lesbian and Transgender 19
Comedy 17   Soap Opera 19
Crime 17   Music Video and Concert 18
Thriller 17   Musical 17
Western 17   Performing Arts 17
Military and War 16   Reality TV 17
Documentary 15   Exercise and Fitness 15
Music Video and Concert 15   Adult 13
Drama 14   Fantasy 13
Fantasy 13   Special Interest 13
Historical 13   Animation 12
Special Interest 13   Documentary 12
World Cinema 12   Historical 12
Children and Family 11   World Cinema 12
Exercise and Fitness 10   Crime 11
Reality TV 10   Thriller 11
Anime 9   Anime 10
Gay Lesbian and Transgender 8   Bollywood 10
Animation 7   Horror 9
Musical 7   Science Fiction 7
Performing Arts 7   Western 6
Bollywood 5   Action and Adventure 5
Romance 5   Sport 5
Soap Opera 4   Military and War 3

 

Appendix 3: results of digital investigation

3.1.1 Highest ranking genres used for research and reserved for feminine category

Books Films & TV
Used Reserved Used Reserved
Parenting & Relationships Romance Reality TV Drama
Cookbooks, Food & Wine Arts & Photography Gay, Lesbian and Transgender Children and Family
Literature & Fiction Children's Books Performing Arts Soap Opera
Education & Teaching Crafts, Hobbies and Home Comedy Music Video & Concert
Health, Fitness and Dieting Gay and Lesbian Romance Musical

 

3.1.2 Genres of recommendations for feminine category

Genres Count
Education & Teaching 100
Cookbooks, Food & Wine 35
Literature & Fiction 28
Health, Fitness & Dieting 16
Reference 6
Politics & Social Sciences 3
Religion & Spirituality 3
Medical Books 2
Self-Help 2
Arts & Photography 1
Business & Money 1
Christian Books & Bibles 1
Engineering & Transportation 1
History 1
   
TOTAL 200

3.2.1 Highest ranking genres used for research and reserved for masculine category

Books Films & TV
Used Reserved Used Reserved
Business & Money Humour & Entertainment Horror Adult
Science & Math Mystery, Thriller & Suspense Crime Action & Adventure
History Computers & Technology Science Fiction Thriller
Engineering & Transportation Health, Fitness & Dieting Sport Western
Comics & Graphic Novels Science Fiction & Fantasy Comedy Military & War

3.2.2 Genres of recommendation for masculine category

Genre Count
Engineering & Transportation 60
Comics & Graphic Novels 35
Children's Books 26
History 14
Literature & Fiction 13
Test Preparation 12
Teen & Young Adult 9
Science Fiction & Fantasy 11
Science & Math 5
Education & Teaching 3
Politics & Social Sciences 3
Arts & Photography 2
Biographies & Memoirs 1
Calendars 1
Medical Books 1
Religion & Spirituality 1
   
TOTAL 197

Introduction

In its first section, this essay discusses the ideas of three theorists identified as 'key thinkers' (Jones, 2010) in the debate surrounding the concept of globalisation. This discussion begins with Immanuel Wallerstein's systemic method, which is read as a forerunner to more recent strands of thought: namely, David Held and Anthony McGrew's processual or transformational model, and finally, the similar but more geographically oriented approach taken by Peter Dicken. The second section turns to the drawings of Mark Lombardi, comparing the manner in which he represents global social networks with the methods of social scientists who have contributed to the literature on globalisation. It is argued that the visual language Lombardi employs when representing actors completely disembedded from geographical spaces and to varying degrees wrested from nations states – enmeshed instead in spaces of flows of capital and power – is analogous to the globalisation theories discussed, particularity those of Peter Dicken.

Theories of Globalisation

From the 1970s onward, social historian Immanuel Wallerstein has contributed to a field of study known as world-systems analysis. The hyphenation of 'world-systems' is intended to communicate that the objects of study are not so much systems of the world but systems that are worlds. It has been more recently observed that some of Wallerstien's work constituted a proto-globalisation theory (Jones 2010, pp. 28-29). Wallerstein's approach is interdisciplinary. He decries a tendency he has observed in social theory to look at phenomena through narrow disciplinary lenses such as 'politics, economics, the social structure, culture' in that in so doing, the possibility of a more holistic understanding of the world is obstructed (Wallerstein 2004, p. x). The main objects of study in world-systems analysis are historical systems. There are three such systems: world-economies, world-empires and minisystems. (Wallerstein, 2004, p. 16) Minisystems are defined as rare self-contained groupings of economic exchange with completely local divisions of labour, such as isolated hunter-gatherer societies. Due to their locality, minisystems are fundamentally different to the two types of world-systems: world-empires and world-economies. World-empires are precursors to world-economies, where goods and wealth tend to be directed from the periphery of an empire, nation or state, to its centre. World-economies are modern capitalist economies, where trade is practised for the sake of endless capital accumulation and not the advancement of an empire or state (Schiranto and Webb, 2003, pp. 28-29).

According to Wallerstein (2004, p. 23), the modern world-system of which we are currently part, is a capitalist world-economy that has persisted since the 1500s. Given his assertion that this type of system is centuries old, it is no surprise that he dismisses the term 'globalisation' as not contributing anything novel (Robinson, 2011, p. 4). Wallerstein states that one major 'feature of the world-economy is that it is not bounded by a unitary political structure. Rather, there are many political units inside the world economy, loosely tied together in our modern world-system in an interstate system.' (ibid. p. 23) Another of Wallerstein's key concepts is the axial division of labour: the division of economic production into more profitable core-like production and less profitable peripheral production. The core-periphery relationship is asymmetrical, in that producers of core-like products acquire surplus value from producers of peripheral products, through a process called uneven exchange. Core-like products are predominant is a few states while in many others peripheral products are in far greater abundance. In addition, there are a handful of states that play host to a close-to-even mix. Based on the above observations, Wallerstein proposes a tripartite taxonomy whereby states are divided into the core, the periphery and the semiperiphery. (ibid. p. 28)

True to the Marxian roots of his thought (Schirato and Webb, 2003, p. 28) Wallerstein argues that the modern world-system is inevitably heading toward a major future crisis (ibid. p. 77). Additionally, while he acknowledges the heterogeneity of his subject-matter he argues that economic factors ultimately determine the shape of society:

[The] world economy contains many cultures and groups – speaking many different languages, differing in everyday patterns. This does not mean that they do not evolve some common cultural patterns […] It doesn’t mean that neither political nor cultural homogeneity is to be expected or found in the world-economy. What unifies the structure most is the division of labour that is constituted within it. (ibid. p. 23)

The epistemological assumptions behind an all-encompassing theory such as world-systems analysis, and Wallerstein's crude division of a very complex world into two types of production and three main territorial blocs are open to criticism (Jones, 2010: 31). Anything with as ambitious a scope as a theory of globalisation is open to charges of being a totalising meta-narrative. However, the remaining two theorists, Dicken in particular, go to much greater lengths to avoid over-simplifying their multifarious subject matter.

This essay now turns to the ideas that Held et al. put forward in their 1999 book Global Transformations. As as the time of publication the, term globalisation was already in wide use, this was written with the advantage of some hindsight. They split existing thinking on globalisation into three camps: hyperglobalizers, sceptics and transformationalists. (Held et al., 1999, p. 2) In the hyperglobalist corner, the uniting claim is that in the late 20th century, there has been an unprecedented increase in the autonomy and power of a global economic market and a weakening in the economic involvement and regulatory grip of nation-states. This seismic shift is decried by neo-Marxist and radical hyperglobalizers and celebrated by neoliberals. (ibid. p. 3-4) The sceptical position represents a refutation of the supposed unprecedented openness of international markets, for example, pointing out historical evidence of an international economy at the beginning of the twentieth century. Sceptics also highlight the concentration of activities of so-called multinational corporations in a handful of wealthy states (Hirst and Thompston, 1999, pp. 68-69) and the agency exercised by states when influencing international trade through their economic policy. The transformationalist school of thought holds that globalisation, rather than being a state of affairs, is a process driving rapid changes throughout contemporary society.

Held et al. present the transformationalist thesis as free of some of the drawbacks of the previous two, making

no claims about the future trajectory of globalisation [nor seeking] to evaluate the present in relation to some single, fixed-type “globalized world” […] Rather, transformationalist accounts emphasise globalisation as a long-term historical project which is inscribed with contradictions and which is significantly shaped by conjunctural factors. (ibid. p. 7)

They claim that the transformationalist model avoids the eschatological trap of those that compare global phenomena to an idealised 'end state' of globalisation. (ibid: 11) Another key point made by transformationalists is that relations between institutions and individuals are being reconfigured by the process of globalisation, so that causes in some parts of the world are yielding effects in new places irrespective of distance and the territorial boundaries of states. (ibid. pp. 8-9) Furthermore, rather than viewing people either exclusively as cosmopolites or as strongly defined by national identity, transformationalism seeks to encompass both alternatives. presenting us as splintered and intermeshed across borders. (ibid. p. 10)

Transfromationalism is put forward as a more nuanced middle ground, and Held et al. attempt to formulate a revised version of it. (Jones: 2010, p. 79) To this end, they provide the following definition of globalisation:

[A] process (or set of processes) which embodies a transformation in the spatial organisation of social relations and transactions – assessed in terms of their extensity, intensity velocity and impact – generating transcontinental or interregional flows and networks of activity and the exercise of power. (Held et al. 1999, p. 16)

In summary: extensity refers to the reach and range of networks, intensity to the volume of activity within a network, velocity to the speed of interchanges and impact to the effect of all the former on a given community. In order to clarify the fourth concept, Held et al. devised a second tetrad, of different types of impact: first, decisional, meaning the extent to which globalisation affects the pros and cons of the choices available to agents; second: institutional, referring to globalisation's effect on the range of choices available to an agent; third, distributive: meaning how the distribution of wealth and power between actors is affected by globalisation; finally, structural: denoting globalisation's effect on the overarching social order (ibid. pp. 18-19).

The convoluted exposition of impact can be simultaneously read as evidence of a commendable and concerted effort to get to grips with the complexities of global reality and as a conceptual weakness, in that impact does not seem congruent with the other three concepts and renders their model lopsided and unwieldy. (Jones, 2010. p. 89) Much as impact is unrelated to the ideas of flows and networks that define extensity, intensity and velocity in the first tetrad, structure seems incompatible with the notions of agency and power relations deployed in the explanation of decisional, institutional and distributive effects in the second. Structure, is almost as broad a concept as effect, which leads one to suspect that Held et al. could easily have subdivided it, producing a third set of terms, in order to adequately address their subject matter. This is by no means a failing. Attempts to represent the vast complexity and variety of the world in a coherent conceptual model have a tendency to unfold, expand and multiply. This essay now turns to how Peter Dicken attempts a similar task, using spacial models of networks and strata.

In the introduction to the sixth edition of Global Shift, Peter Dicken puts forward a similar taxonomy of globalisation theorists to that posited by Held et al. Again, one encounters the hyperglobalists, who span the political spectrum and see global capital as having, in recent decades, become unfettered by nation-states. One is reacquainted with the sceptics who view internationalism as an adequate concept and globalisation as an empirically unsupported step-too-far. Finally, one is reminded that globalisation comprises processes, rather than a set of phenomena resembling an "ideal state" or an "end state" to which the world is accelerating. (Dicken, 2010, pp. 4-8) What differentiates Dicken is his emphasis on an economic geography in a perpetual state of 'becoming':

Old geographies of production, distribution and consumption are continuously being disrupted and new geographies are continuously being created. The new does not totally obliterate the old. On the contrary, there are complex processes of path dependency at work; what already exists constitutes the preconditions on which the new develops. (ibid. p. 14)

Dicken makes some use of a core-periphery concept similar to Wallerstein's when explaining recent economic history: in the core-periphery relationship, Western Europe, and later, North America came to occupy core positions, as the process of industrialisation took hold over the nineteenth century. During and soon after this time, some countries experienced a shift from core to periphery and vice versa. After World War Two, there emerged a marked East-West economic disparity (ibid. p. 14).

Dicken identifies three key indicators of global economic interconnectedness: trade has grown faster than output, 'in the second half of the Twentieth Century, world merchandise trade has increased almost twentyfold, while world merchandise production increased just over sixfold'; Foreign direct investment – investment by one firm, in a foreign firm, in order to gain a degree of control over that firm (ibid. p. 20) – has grown faster than trade; There are structural imbalances in the world economy as 'some countries have huge trade and current account deficits while others have huge surpluses.' (ibid. p. 22) There is convincing evidence that economic activity frequently crosses borders, but analysing this activity is difficult because 'virtually all statistical data on production, trade, FDI and the like are aggregated into national “boxes”.' (ibid. p. 52)

Despite these restrictions, Dicken wants to devise a method of analysing the 'tangled webs of production circuits and networks that cut through, and across, all geographical scales, including the bounded territory of the state' (ibid.) that make up the world economy. Dicken demonstrates how one might analyse these networks, through a diagram of three mutually embedded strata: 'macro-structures of (Institutions, conventions etc) of [the] capitalist market system', 'Circuits and networks of interaction, mediated through differential power relationships in global production networks and through transnational social networks' and 'Uneven geographical distribution of “goods” and “bads”; “winners” and “losers”'. He cautions the reader that this is 'an idealized representation of a world that is, in reality, infinitely more complex.' (ibid. pp. 52-3) For Dicken the networked perspective permits one to conceive of scale as a continuum rather than as being composed of discreet steps such as global, national and local. It also allows one to figure territory in a topological or relational manner rather than divided by precise borders. He does not, however, wish to present the topological and territorial ways of looking at scale as sharply distinct: 'a topological perspective is not, in itself, in conflict with the fact that, in terms of jurisdictional and regulatory practices, territorial scales of governance remain fundamental to the organisation and operation of the global political economy and its constituent parts.' (ibid. p. 54)

Dicken's method is empirical where possible and tentative when constructing generalized models of reality. This may be a result of his epistemological outlook having been influenced by the late 20th century post-structuralist turn in the social sciences, 'in contrast to the rather “structuralist” conception of process encountered in the thinking of Wallerstein[, …] Held and McGrew.' (Jones, 2010: 116) Dicken's thesis is a logical one to end on, as it synthesises and expands upon both the interdisciplinary processual approach of Held et al. and Wallerstein's mapping of economic processes onto geographies.

Lombardi's Drawing Practice

In the introduction to Global Transformations, Held et al (1999, p. 2) write 'globalisation may be thought of [...] as the widening, deepening and speeding up of contemporary social life, from the cultural to the criminal, the financial to the spiritual.' By this definition, Inner Sanctum: The Pope and His Bankers Michele Sindona and Roberto Calvi by Mark Lombardi (Figs 1 and 2) is emblematic – in both actors and actions represented – of globalisation. This is not to say that the Lombardi's subject-matter is merely evidence of the processes of globalisation, but rather that as will be argued later, his drawings deploy, through their visual language, his particular way of theorising globalisation.

Fig 1: Lombardi, Mark  (1998) Drawing 1: Inner Sanctum: The Pope and His Bankers Michele Sindona and Roberto Calvi (5th Version), (Hobbs, 2003, p. 64)

In a 1996 statement accompanying Inner Sanctum: The Pope and His Bankers Michele Sindona and Roberto Calvi, Lombardi states:

Over the years the Vatican bank invested well over $10 billion in major Italian, European and American Companies. Any profits realized were theoretically to be used to finance special church projects the world over. But besides its official, charitable face the Vatican bank had another, subterranean aspect as well. Many well-placed friends of the Church, including some of Italy's top financiers and industrialists, used the bank (for a fee or around 5%) to evade taxes and illegally export tens of millions of dollars from Italy to Switzerland and beyond. (in Hobbs, 2003, pp. 60-1)

A major node in this graph is Michele Sindona, who can be seen in Fig 1 connected by curves with arrows on either end, indicating relationships of mutual influence with Richard Nixon, David Kennedy, and others. The only one-way flow of influence to Sindona comes from Pope Paul IV, for whom he worked through the Vatican bank. Nixon's law firm represented Sindona in the US and Kennedy's Continental Illinois Bank was involved in some of Sindona's American Investments. (ibid. p. 61) The only events explicitly described are indicated by red lines: 'fines levied, indictments, incarceration, death or other “restraints (Lombardi's Term)”' (Lin, 2003, p. 146) In Sindona's case, in red we have: '1979-82: convicted in Italy & US of murder, fraud & conspiracy; 1984: found dead in Italian Prison: cyanide poisoning.' Details of relationships are not included on the chart, though given the sheer scale of this cat's cradle of power relations, capital flows, time-lines and so forth, reams of such information must exist. The bulk of these data are likely recorded on some of 'thousands of three-by-five inch index cards' (ibid. p. 16) in Lombardi's personal archive.

Lombardi began work on his oeuvre of drawings, delineating the complex intrigues of international political and financial scandals which, he came to call 'narrative structures' in 1994. (Lombardi, 2001) Prior to this he took a BA in art history at Syracuse University. After graduation he worked as, amongst other things, a museum director, curatorial assistant and arts librarian. (Richard, 2001, p. 7) His working procedure for narrative structures as outlined by those who knew him, took the following form: he would gather as much information as he could on a given event (culling this from published sources), (Lombardi, 2001) arranging it spatially and perhaps carrying out some preliminary sketches. Versions one and two of his drawings would be constructed in this early stage. Version three would include the majority of the information, in its final configuration. The fourth version onwards would be the polished drawing, with the addition of 'final results', in red. At this point the drawing could be submitted for exhibition, along with a short text describing the event in question. (Richard, 2011, p. 7)

Though only represented as having been involved through the receipt of money, Bill Clinton, the Lippo Group, and China Ocean Shipping Co. a.k.a COSCO, Little Rock-Jakarta-Hong Kong ca. 1990s enmeshes Clinton in a network of incriminating flows of influence, linking him to arms dealing and possible campaign-funding corruption. Hobbs suggests that Clinton is 'less an actor than a figure being acted on' by the overarching forces delineated by Lombardi's 'structuralist overview'. (Hobbs, 2003, p. 111). Lombardi's work is said to be influenced by 'globalism' (ibid.) While this Americanism is not entirely synonymous with globalisation, both approaches take as their object of attention the entire world. Bearing this in mind, along with Hobbs' claim that there are parallels between the 'enlarged vision' evident in Wallerstein's world-systems analysis and an unpublished manuscript written by Lombardi in the late 1980s called Panorama: The Atlas of Modern Art. (ibid: 21-23) one can begin to look at Lombardi as both an artist and globalisation theorist.

Indeed, Lombardi himself has said '[t]here is a sociological aspect to my work. I am interested in the structure, mechanisms, uses and abuses of power in the global political economy.' (ibid. p. 19) Figure 2 depicts a situation where Chinese business entities affect American politics, crossing both territorial and institutional boundaries. His subject matter: the interaction of corporations, banks, political figures, often across national borders, is the bread and butter of globalisation theorists. Despite these marked similarities, he does not construct theoretical generalisations, of the written kind anyway. His written content comprises facts, but it is situated in the structure of a graphical language which, far from having the empirical transparency of, say, a seismograph reading, organizes information according to certain assumptions.

Fig 2: Lombardi, Mark (1999) Bill Clinton, the Lippo Group, and China Ocean Shipping Co. a.k.a COSCO, Little Rock-Jakarta-Hong Kong ca. 1990s (5th Version) (Hobbs, 2003: 112)

Assuming Lombardi's visual presentation of information constitutes something like a thesis, it could be argued that this would be incompatible with a model in which geography played a significant role. Lombardi's graphs are organised sometimes according to linear time, mostly in accordance with flows of power and capital but not obviously in relation to the actors' geographical situations. Abstraction of human activity from geographical locales is identified to by some theorists as a key part of the process of globalisation and called 'disembedding'. (Eriksen, 2007, pp. 16-7) It is questionable how reasonable it would be to read disembedding into Lombardi's drawings, especially as they are presented in an art context and are not bound by an expected function, such as attempting to accurately model geographic space. Nonetheless, the absence of reference to territory precludes parallels with Wallerstein and Dicken as they both geographically map the world economy. In the case of Dicken, the disjunction is far from total, as his deployment of networks is comparable with the web structures of Lombardi's graphs. Also of relvance is Manuel Castells, a contributor of ideas relevant to the globalisation debate (Jones 2010), who developed the idea of a space of flows, which corresponds very well to the spaces defined by Lombardi:

The space of flows is the material organisation of time-sharing social practices that work through flows. By flows, I understand purposeful, repetitive, programmable sequences of exchange and interaction between physically disjointed positions held by actors in the economic, political and symbolic structures of society. (2000, p. 442)

Lonbardi's picture plane is not conventionally cartographic; if it maps anything, it is Castells' 'economic, political and symbolic structures'. It is in these structures, that Lombardi's actors are situated, rather than those of geography and nationality.

BCCI, ICIC and FAB (Figures 3 and 4) is the last large scale drawing Lombardi made before his death. It depicts a banking operation which for decades sprawled across the globe. The Bank of Commerce and Credit:

[A] mainly Arab owned concern founded in 1972 with financial backing from The Bank of America […] controlled from Abu Dhabi, incorporated in Luxembourg and operated from London, Ganeva and the Caymen islands; involved in joint ventures with local banks in Iran, Oman, France and Switzerland; assets grew from $5 million in 1973 to $23 billion by 1991; by mid 1980s has opened offices in 78 countries to serve the needs of nearly one million depositors and borrowers (Lombardi, in Hobbs, 2003, p. 97)

BCCI appeared to be the epitome of footloose, multinational capital. However, it also laundered money from various criminal sources. According to Lombardi, its directors mainly came from diplomatic corps and the intelligence community because 'it was created to serve geopolitical rather than commercial ends: to further the regional political and national security ambitions of a handful of conservative Gulf Arab states allied to the U.S and Britain' (ibid. p. 98) In 1991, the bank was seized in a multinational raid by officials.

Fig 3: Lombardi, Mark (1996-2000) BCCI, ICIC and FAB, 1972-91 (Hobbs, 2003, p. 96)

Conclusion

The subject matter of BCCI, ICIC and FAB is of key relevance to the globalisation debate, involving the worldwide intermeshing of major commercial institutions with nation-states across the globe. It bespeaks a 'transnational interconnectedness weav[ing] complex networks of relation between communities, states, international institutions, non-government organisations and multinational corporations which make up the global order' (Held et al. 1999, p. 27) That this drawing acknowledges the agency of nation-states in the global banking operation depicted somewhat distances Lombardi from the hyperglobalizers and their claims of a radically autonomous global economy. However, the drawing is in perfect concord with the transformationalist theories of Held et al. and Dicken, for which the process of globalisation would necessarily affect and involve nation states.

Fig 4: Lombardi, Mark (1996-2000) BCCI, ICIC and FAB, 1972-91

The visual language through which Lombardi communicates his evidence of the processes of globalisation is like that of 'computer system engineers, designers, urban planners,' not to mention the social network diagrams of the field of social network analysis. (Richard, 2001:16). Lombardi's graph-making practice removes actors from their various territories to reveal a criss-crossing of relationships that envelop the world. He circumvents the national 'boxes' that Dicken complains confine many economic statistics – representing the economic, social and political in two dimensions, wrested from geography; he emphasises the flows of social and economic interactions at the expense of their national contexts. Neither in form nor content, are these the drawings of a globalisation sceptic.

To reiterate the less problematic parts of Held et Al's processual definition of globalisation:

[A] process (or set of processes) which embodies a transformation in the spatial organisation of social relations and transactions [...] generating transcontinental or interregional flows and networks of activity and the exercise of power.

Lombardi documents these flows – these products of the process of globalisation – in a single picture plane. He preserves flows of activity produced over many years, in what one might call two-dimensional fossil records of globalisation. Lombardi comes as close as his temporally static medium allows him to represent globalization as a process.

References

Books

Castells, M. (2000) The Rise of the Network Society, Blackwell, Oxford

Cuyvers, L. and De Beule, F (eds.) (2005) Transnational Corporations and Economic Development, Palgrave Macmillan, Basingstoke

Dicken, P. (2010) Global Shift : Mapping the Changing Contours of the World Economy,

Eriksen, T. (2007) Globalization, The Key Concepts, Berg, Oxford

Held et al. (1999) Global Transformations: Politics Economics and Culture, Polity Press, Cambridge

Hirst, P. and Thompson, G. (1999) Globalization in Question, Polity Press, Cambridge

Hobbs, R. (2003) Mark Lombardi: Global Networks, Independent Curators International, New York

Jones, A. (2010) Globalization: Key Thinkers, Polity Press, Cambridge

Madeley, J. (2003) A People's World: Alternatives to Globalization, Zed Books, New York

Peck, J. and Yeung, W. (eds.) (2003) Remaking the Global Economy, SAGE publications, London

Schirato, T. and Webb, J. (2003) Understanding Globalization, SAGE publications, London

Stiglitz, J. (2002) Globalization and Its Discontents, Penguin, London

Stutz, F. and de Souza, A. (1998) The World Economy, Prentice-Hall, New Jersey

Wallerstein, I. (1989) World-Systems Analysis: An Introduction, Duke University Press, London

Articles

Baker, R. (2nd March 2011) 'Mark Lombardi's World Conspiracy, Corruption, and Vatican Hit Men', Village Voice, <http://www.villagevoice.com/2011-03-02/art/mark-lombardi-s-world-of-conspiracy-coruption-and-vatican-hit-men/> 14 May 2012

Bierut, Michael (24th November 2003) 'Mark Lombardi and the Ecstasy of Conspiracy', The Design Observer, available at: <http://observatory.designobserver.com/feature/mark-lombardi-and-the-ecstasy-of-conspiracy/1697/> 13 May 2012

Kimmelman, Michael (November 14th, 2003 ) 'ART REVIEW; Webs Connecting the Power Brokers, the Money and, Ultimately, the World', New York Times, available from:    <http://www.nytimes.com/2003/11/14/arts/art-review-webs-connecting-the-power-brokers-the-money-and-ultimately-the-world.html>  14th May 2012

Lin, T.  (2003), 'Following the Money', Art in America, November 2003, 142-147, 177

Lombardi, Mark (2001) 'The Recent Drawings: An Overview', Cabinet Magazine,  avaliable at: <http://www.cabinetmagazine.org/issues/2/lombardi.php> Accessed 16th May 2012
Richard, Frances (2001) 'Obsessive—Generous: Toward a Diagram of Mark Lombardi' wburg, vol.2, no.2, available from <http://www.wburg.com/0202/arts/pdfs/lombardi.pdf> 15th  May 2012

Robinson, W. (2011)  'Globalization and the Vol. sociology of Immanuel Wallerstein: A critical appraisal', International Sociology, vol. 26 no. 6, 723-745, available from: <http://iss.sagepub.com/content/26/6/723>, 14th May 2012

Weinberg, Michele  (5th May 2005) 'The Color of War: Images of aggression from the boardroom to the battlefield', Miami New Times, available at: <http://www.miaminewtimes.com/2005-05-05/culture/the-color-of-war/> 15th May 2012

Other online

'Mark Lombardi CV', Pierogi Gallery website, http://www.pierogi2000.com/artists/mark-lombardi/mark-lombardi-bio/, 23rd May 2012

As this review describes much of the content of the film, you may wish to watch it before reading this.

Hungarian auteur Béla Tarr has stated that The Turin Horse is his final film. This seems somewhat appropriate as it references an apocryphal event that is said to have marked Friedrich Nietzsche’s final fall into physical and mental decline and is themed on a descent into nothingness. While visiting the Italian city of Turin, Nietzsche is supposed to have been distressed by the sight of a stubborn horse being brutally whipped by its driver and fallen sobbing upon the beast’s neck in a protective embrace. Nietzsche spent the years following the time at which this event may or may not have occurred, in an early dotage, under the care of his family.

The event concerning Nietzsche is described in the film’s prologue, which ends with: ‘Of the horse… we know nothing.’ In the scene immediately following this sentence, we are presented with something may that shed some light on the horse’s fate: We follow a bearded middle-aged man atop a wooden cart, drawn along a path by a shabby-coated horse, through the dust and debris swept up by a gale. It is unclear whether this horse is the horse. In any case, the entire film centres around the lives of this man, his daughter and the horse in their isolated stone farmhouse.

This film struck me as glacial, both in its slow, deliberate pace and its atmosphere of crushing inevitability. It runs for 146 minutes and consists of a mere 30 long takes, all in monochrome. The score is repetitive and mildly dissonant: the same dirge-like piece, mainly on organ and violin, is woven in and out of the film at varying intensities. I found it to be largely effective if occasionally grating, though given the film’s bleak mood, some discomfort is permissible. The score also compliments the almost constant whirring and keening of the strong and unremitting wind that tears at the farmhouse and surrounding hills.

Though the father and daughter are in each other’s company for most of the film, they seldom talk. When there is dialogue, it is terse and directed at the mundane activities that make up the bulk of the film’s content: They both muck out the stable and make several unsuccessful attempts to get the horse to eat or drink. The daughter is buffeted by the wind as she fetches water from the well outside, she boils a daily potato for each of them, she washes clothes by hand, she dresses and undresses her crippled father. The father chops firewood, works at carpentry and is presumably returning from some kind of work at the beginning, a journey which he is prevented from repeating by the horse’s obstinacy. Indeed, these people’s history is never addressed, we are only, unsparingly, shown their quotidian toil.

Much of the success of the film rests on how well we are shown this drudgery. To this end, the farmhouse has been constructed thoughtfully, out of real stone, rough plaster and so on. The interior lighting is intricately staged. The tools and furniture seem to have been chosen for their stark, rustic qualities. The actors’ movements and stillness are framed and re-framed by the camera as it snakes its way around the space. Acts of drudgery and subsistence, when presented in this manner and in this setting, encourage a kind of close study, a rapt fascination. That the father ploughs into his potato while it is hot and endures the burning, while his daughter tentatively picks at hers, takes on a strange significance. As there are so few other details to focus on, I found the implausible excess of leftover potato scraped from the wooden plates after the meals jarring. With so much of what we tend to think of as acting omitted, my attention was nonetheless held by the pared down behaviour of these people.

If we ignore the father’s age and his only having full use of one arm, and look only in gendered terms, the division of domestic labour seems heavily weighted toward the daughter. A lot of the father’s speech comprises dour orders directed at his daughter. He stands, staring sternly straight ahead when she dresses and undresses him; this comes across as an attempt to retain his dignity. Even taking into account the father’s infirmity, it seems undeniable to me that the social organisation of the household is patriarchal. While questionable, if nothing else, this gives us an authentic depiction of 19th century European ordinariness.

There are three instances of sustained dialogue or speech. The first of these mainly consists of a monologue: A man visits the farmhouse unexpectedly, seeking palinka, the Hungarian fruit brandy that is consumed throughout the film. He explains that he couldn’t go into town as it has fallen to ruin, has been degraded. He then launches into five minutes of cryptic Nietzschean prose. This is Nietzschean because he employs a binary typology of people: those who have acquired everything, in a sneaky underhanded fight, and debased it and the great and noble who abstain from fighting; winners and losers respectively. This doesn’t precisely fit Nietzsche’s ranking of the exceptional atheist who pours all of their passion into their earthly existence over the great herd of Christians whose morality is determined by what is judged to be good for the herd and who invest emotionally in spiritual fantasies and hopes and fears of an afterlife. Nonetheless, the visitor’s opposition of the noble and pacifistic to the underhanded and acquisitive does echo Nietzsche’s opposition of master to slave or Übermensch to last man. Nietzsche’s Übermensch will construct new values by which to live and stand as a life-affirming exemplar to the herd of humanity. The last man is the apathetic antithesis of the Übermensch, caring only for his comfort and security. Nietzsche also writes of nihilism as a failure to do this, an inability to paint values over the blank, valueless face of existence.

I can think of two interpretations of the visitor’s speech. The first is that he is a herald of the victory of herd morality over any hope of a new value system, and the inevitable victory of nihilism as the father and daughter’s world disintegrates. I think this is suggested when the visitor claims that the noble and great have altogether disappeared and do not believe in god or gods. Additionally, the father sceptically dismisses the visitor’s speech and offers his daughter no speculation as to why the woodworm has ceased its noise on the first day or the lamps ceased to give light on the fifth; ‘I don’t know. Let’s sleep.’ he says. There are no evocations of God’s will, of tests, punishments or of things happening for a reason; his outlook is positivistic. My second interpretation is based on the menacing behaviour of the Gypsies who appear uninvited after the visitor has left and take water from the well and Tarr’s claim that the visitor represents a ‘shadow Nietzsche’. In this reading, the underhanded thieves are presented as bad through the lens of Christian morality by the visitor, who is Nietzsche’s opposite (shadow in the Jungian sense). The visitor, father and daughter are the last remnants of the herd who’re being swept away by the rising tide of Übermensch. After all, theft is considered a sin under Christian morality. Furthermore, in his later writings, Nietzsche theorised that Christian morality represents the apotheosis of base slave morality, at the expense of noble master morality; humility, charity and obedience are privileged over pride, competition and personal freedom. I could argue that the gypsies display more of the master and the father and daughter more of the slave, though they do defend their well and threaten the gypsies with violence. It is also possible to read the father and daughter as last men, clinging to the relative comfort and security of their routine.

Both interpretations are flawed. However, when the gypsies are driven away, one of them shouts that the water is theirs, and the following day, the well is empty. This fits the visitor’s description of a class of despoilers and acquires of all they touch. One more event which is pertinent to this strand of interpretation is the book from which the daughter reads a passage aloud, clumsily, syllable-by-syllable. The passage deals with holy places, whose function is the veneration of the Lord. Deviations from this function are not allowed and amount to violation. No services can be held in violated places until a ceremony of penitence has been held. The text does make allusions to Christianity which may support my second interpretation. Her final words are of interest: ‘The celebrant tells the congregation: The Lord is with you. Morning will turn to night. Night will end.’ The cleansing described, rings of apocalypse and foreshadows the darkness that falls on the house. At the end of the fifth day, the narrator describes the cessation of the gale and dead silence that permeates the house. The void is approaching.

When they cannot ignite the lamps, the father says that they shall try tomorrow. There is absurdity here, an unwavering will to go on in the face of a situation that grows continually more bleak and hopeless. At the close of the film, the father and daughter sit hunched over bowls, facing each other, the father gingerly plays with a raw potato; the daughter’s remains in the bowl that rests between them. The father says that they have to eat and bites into his with a pitiable crunching. The film ends. I noticed strong parallels with Becket’s work, Endgame in particular. Both find an older infirm man with a younger companion. Both feature dwindling resources, dwindling life, dwindling everything. When Clov in Endgame is asked to describe the view outside their room’s two windows to his blind companion Hamm, he scans the surrounding world with his telescope, regularly pronouncing ‘zero’ and summarises the view from the first window as ‘corpsed’ and from the second as ‘grey’.

The woodworm stops – decay stops; light stops; fire stops; the gale stops; there is silence. The world is almost stripped to nothing. Setting aside my attempts at interpreting the plot or meaning of this film: It uncompromisingly depicts the straitening and winding down of an already harsh existence.