The Wager of the Algorithm: Towards a Performatic Gesture
0
The fantasy of the algorithm and, by extension, artificial intelligence imagines that the digital machine performs by executing an operational task. The task’s most basic logic parses the operation into input, output, and a set of instructions for generating the second from the first. The digital process dreams of entering the highest orders of instrumentality in which inputs run unimpeded through increasingly complex algorithms to yield desired outputs. The driving ideal of artificial intelligence ostensibly advances the phantasmagoria already at work in algorithmic thinking: absolute efficiency, productivity, and laborless labor. This phantasmagoria crystalizes the central problematic, mechanism, and potentiality that the following exposition aims to articulate.
1
The paranoiac optimism of predictive analytics poignantly exemplifies these instrumentalist aspirations. The proleptic impulse in the machine strives to identify and correct in advance any wayward elements – in short, any glitches – that risk interrupting the operation and the interests it serves. Such a preemptive strike, programmed using evidential probability, is ideally achieved before the glitch is even registered in the system. This corrective hankering is what led Siegfried Zielinski to remark somewhere that the euphoric age-old phrase "digital revolution" was already a contradiction in terms. Today’s utopian vision longs for the ultimate desiring machine: a human liberated from toil whose neurological system has become so seamlessly integrated with the digital network that the ensemble can vow continuous satisfaction. The power, speed, and efficacy of the network’s anticipatory mechanism consistently will know or can learn how to realign any deviation within lived experience before the human detects it.
The legacy of Filippo Brunelleschi includes two great ruses. The first strategem consists of the "vanishing point" in Renaissance linear perspective. Despite its name, this point remains clearly visible or traceable on the horizon line. On the picture plane in the opposite direction on the virtual Z axis, however, another point stands in absentia. This corollary point corresponds to the implied vantage point inscribed imperceptibly in the picture. This vertex, unlike the so-called vanishing point indeed vanishes from the visible surface. (100 years later, da Vinci would envision the "perspective machine.") The second yet infamous ruse of Brunelleschi is recounted in Antonio Manetti’s story "The Fat Woodworker." In an act of brilliant pettiness, the fifteenth-century master of perspective convinces an unsuspecting woodworker named Manetto, through an intricate web of social contrivances, that Manetto is in fact not himself but rather a well-known Florentine named Matteo.
Peter Freund, Lost Grids, or How to Explain Digital Media to a Dead Socialist, 2018-Present, Code Interventions (Multiple Mediums)
Left:
<000_Marx's_Beard>, 2019, Digital Print (1/5)
Magnified glitch resulting from entering semantic text into the underlying digital code-bed of an image.
Source materials:
Text:
"Revolutionary theory is now the enemy of all revolutionary ideology, and it knows it."
(Guy Debord, The Society of the Spectacle, 1967)
Image:
A Portrait of Karl Marx
(John Mayall, 1875, detail: beard)
(Further project description below: "Exhibit A.")
< LOST GRIDS, OR HOW TO EXPLAIN DIGITAL MEDIA TO A DEAD SOCIALIST
Left + below:
Project manifestation as an artist book: six loose signatures, five of which include a distinct theoretical statement, introduction, network of image-text source materials, grid print, and sample of hexadecimal code underlying the given grid; the sixth, containing the five grid prints in A3 format.
(See Exhibit A annotation below for a conceptual introduction to the project and a list of many of its various manifestations.)
Exhibit A
Lost Grids, or How to Explain Digital Media to a Dead Socialist comprises a generative engine made from a calculated misuse of digital code. The various manifestations of this project spring from an initial two-step: Images and texts are appropriated from the cultural archive, individually coupled, and combined with other image-text pairs to form loosely associated thematic networks. These appropriated elements are drawn from philosophical, artistic, or quotidian materials. The relationship between each paired image and text presents a conceptual leap of sorts. The image-text networks provide the material sets (miniature archives) from which the following operation is performed to produce varied results (listed below).
The operation conducts a torsion of the already-detourned image-text montages. A single image-text pair is selected from each network. The text portion is entered directly into the underlying code-bed of the image with which it has been coupled. A rupture results in the digital image that visually distorts its recognizability in the form of misplaced, recombined, reassigned, or obliterated pixels. From this interruptive glitch, a portion of the pixelated distortion is isolated, magnified, extracted, and then scaled up to form a grid image: 273 variegated squares (21 x 13). The relation between grid as surface and network as depth forms the project's conceptual core.
The essence of surface and depth: incommensurability;
that is to say,
radical ornamentalism and radical obscurantism.
The Lost Grids project to date has yielded the following forms (a few samples shown above):
- an artist book enfolding six loose signatures or pamphlets, each comprising a different grid print, theoretical statement, specific image-text network, and sample from the resulting hexadecimal code of the specific grid.
- a set of artist books each made of more than 500 unbound pages of hexadecimal code for every grid.
- a set of artist books organized according to the sequence of colors in each grid and designed as a series of swatches.
- video installation loops animating each color swatch book in a page-by-page series.
- video installations (13–15 mins) of scrolling hexadecimal code underlying each grid.
- a performance-based work presented as oral recitations of the hexadecimal code. (See "rehearsal sketch" at top of exposition.)
- a performance-based work translating the hexadecimal code into choreographical prompts.
- the textuality of the hexadecimal code sets presented in their pure linearity without line breaks, each in a single visible but illegible line.
We are concerned here with two technical equivalents of a parapraxis. On the one hand, the interruptive glitch is found in the "noise" (excess pixels, buzzing, or analog snow) that distorts or blocks the transmission of an audiovisual signal. On the other hand, the glitch of relentlessness can be exemplified by the out-of-control feeding machine that aims to streamline the factory assembly production in Chaplin's Modern Times. A real-word illustration presents itself in the code-breaking machine devised by Alan Turing during the Second World War; the device runs on, unclear if its persistence will ever find a solution and stop.
If the Fibonacci sequence, an avowed guarantor of aesthetic proportion, indeed comprises a series of ordinal numbers, then it is uncountable. That is, it is impossible to begin with the single initial integer from which it claims to launch. Simply put, if one attempts to abide by its arithmetical rule, the "recurrence relation," there is no way to get from zero to one. One is forced to conclude that the binary 0-1 itself begins the sequence. Such a foundational impossibility must lie beyond the ordinal system and aesthetic it initiates.
2
The inherent computational structure of the digital performance de facto fails to live up to its instrumental promise. Even where it might avert a specific interruptive glitch, the system then faces the prospect of a glitch of relentlessness: the "halting problem." In computation theory, the halting problem reveals an incompleteness, hole, or blind spot at the structural center of its apparently self-enclosed instruction set. This quandary began with a logical paradox in mathematical set theory. The paradox asserts that no set can be a member of itself. The set that constitutes the internal coherence of its membership must itself be excluded from the set. Put more paradoxically, the set determines what it contains but cannot determine itself. By homology, when given an algorithm and a set of inputs, no machine can determine whether its operation will ever arrive at the desired output and thus halt. Like the set, the digital system is fundamentally split at the undecidable nexus between the core algorithm and the input/output at its limits.
3
The parlance of contemporary generative AI and prompt engineering, like our everyday language, blurs an ambiguity brought out by early computation theory. In the latter, the "prompt" can be understood in two radically different senses. In one sense, the term indicates the presence and content of an input that expresses a desired output. (This input is then run through an algorithm that endeavors to generate that output.) Here, the prompt belongs exclusively to the internal composition of a distinguishable system in process. A second sense of the term "prompt" signals a placeholder and empty field – the flashing cursor – at the nexus of the system where the inputs are delivered to the algorithm or the outputs are displayed on the interface. This nexus signifies a structuring absence essential to the system's functionality. As generative AI and AI more generally approach greater autonomy and even possible sentience, these technologies have aroused an urgent outcry against their purported threat to human creativity and responsibility alike. Its merits notwithstanding, this cautionary alarm may miscalculate the conceptual scope of creativity and responsibility, not to mention the very definition of "sentience" and "intelligence." Two relevant points, however, emerge here: This unease minimizes the internal vulnerability and essentially compensatory nature of the digital performance. The disquiet, in turn, presents an occasion to ask, precisely in relation to this structural fragility, what practices might now count, within and without the digital domain, as art.
Today’s ever-worsening auto-correct technology presents an outstanding exemplar of the unconscious. A lapse originates from outside the subject rather than emanating from some hidden depth within the subject (e.g., "slip of the tongue"). It signals the unconscious the instant that the subject feels compelled to resist it. One grabs on to the dreaded auto-correction as a mistake modeled on one’s own singular resistances. Such anomalies could apply to more dramatic "AI hallucinations" as well.
Exhibit B
Floating Point presents a work of algorithmic cinema sprung from the tension within the moving image between its linear syntax and the nonlinear archive or database that it simultaneously embodies. Borrowing from Stanley Kubrick’s 2001: A Space Odyssey the scene in which the HAL 9000 computer is deactivated, the piece re-presents while reframing the twenty-five (25) shots of the original sequence. The frame of the unfolding re-presentation widens to reveal immediately adjacent shots in a grid system that tabulates the sequence’s twenty-five elements.
The accompanying grid diagram (above) shows the conceptual geometry that structures the piece. The central white portion of the numbered grid lays out the sequence path from the first shot to the twenty-fifth (indicated by the narrow green line progressing in a rectangular spiral), which Floating Point follows and replicates from the original film. The gray rectangular overlays indicate samples of the expanded frame composition that runs for the duration of the entire piece. The outer green region schematizes in a flat, two-dimensional view a three-dimensional topology in which the inner white grid has been folded back onto itself such that its edges and corners meet. To the extent that these points of contact cannot be formed simultaneously in physical space, this topology stipulates an "impossible geometry" – that is, a geometry that cannot be visually configured or imagined in three-dimensional space but only conceptualized through an ordinal system. Floating Point thus folds back on itself in the manner of a conceptual origami. At this level, the piece offers up a moving image for the mathematical or computational eye rather than the human or retinal eye.
The resulting work is produced through computer programming (Marc Anglès + Arnau Giralt), and when screened in its intended form runs live as a software performance that in real time configures simultaneously the grid and the progression of shots within it. To screen the work as such is to submit the performative unfolding of machinic operations to the semi-precarious time frame of a contingent production, which the expedient logic of the algorithmic is normally meant to streamline. If the film hints at any political ambition, it seeks to assert in a modest and oblique way the principle of non-identity operating in the self-evident by hinting at the actual potential already at work in the identity of the moving image. The performance of Floating Point withdraws from the utopian gesture of recombinatory intervention typically associated with generative art. Each presentation instead generates here the same input/output relation. The exact same outcome appears to result from the same algorithm, unless one fundamental condition comes to the fore: The hardware fails to support the fluidity of the generative software performance. This fragility, even when visually undetectable, nonetheless endures. (Names for this structural entropy – which spans aesthetics, technology, political economy, or all three – include "dead or defective links," "bit rot," "designed obsolescence," and "host subscription lapses.") It is in the apparent lack of difference produced by the reproducible, positive, successful operation of the system that one is asked to see already at work its very opposite. Herein lies the "performatic gesture," which will be elaborated below.
The term "performatic" originates from a linguistic discrepancy between the English and Spanish languages. Historically, Spanish has substituted the word "action" ("actuación," "acción") for what in English is called "performance." Thus one routinely finds in Spanish-speaking countries the translation "action art" in lieu of "performance art." The word "performático" ("performatic") was later introduced into the Spanish language in the context of discussing work such as performance art and theories of performativity.
The instrumental dream does not simply encounter a gap that it aspires to overcome; rather, it begins with a gap without which there would be no dream in the first place. The slumberer who is otherwise strangely reassured by the prohibition implied by every procedural operation awakens promptly. The awakening is not one of pure enlightenment but rather one of disorienting obscurity. In the lapse between the algorithmic prompt and the contingent, inopportune auto-corrections that intervene lies the force of necessity. The particular resistance to the ostensibly external auto-correction offers a coup d'œil into some truth about the dreamer's innermost being.
4
Now consider three ways to produce, present, or enact a work of art. First, the work can be performed: A set of instructions is executed with inputs in order to produce desired outputs. Imagine a script or a score enacted on stage. The instrumentalized act draws a creative interest by offering variant live versions. Second, the work can be performative: The act of performing makes a reality so. Imagine a police team saying, "You are under arrest." God declares, "Let there be light" and there is light. Or an artist sends a telegram for a gallery show: "This is a portrait of Iris Clert if I say so." Finally, the work can be performatic: An instruction set is initiated but meets with a glitch of interruption or relentlessness. To be performatic, the work functions antithetically to the formalist, merely transgressive fetish of the "postdigital" glitch. Imagine, by contrast, holding up the instruction set itself as the artistic marrow while leaving the outcome as an independent, variable, and compelling curiosity or even detritus. Take a page from the Oulipeans, or think Sol LeWitt, Josep Maria Mestres Quadreny, or Sophie Calle. The three performing modes outlined above each turn out to be less a different work than a distinguishable orientation toward performance. In contrast to "performance" and "performativity," the "performatic" gesture breaks off the operational movement precisely at the interface between input and algorithm or between algorithm and output. This gesture reveals already, as immanent to the system, a partial transcendence of it: The input and the output always introduce an element of uncertainty, incompleteness, and ambivalence. The gap in the operational nexus – expressed by the proverbial flashing cursor – presents a structural break that indicates the non-instrumental, unknowable kernel of the machine. The instruction set for producing a visual artwork or act of performance, for example, can suddenly resemble a poem; a musical score can resemble a beautiful diagrammatic framed drawing; a lost personal telephone directory found on the street can resemble a pen for plotting the negative space that outlines a portrait of its unknown owner.
5
Performance, performativity, and the performatic each mobilize a distinct conceptualization of causality harkening back to the Aristotelian typology and all its problematic clarifications. But only the third orientation, the performatic, indicates causality as operative insofar as the operation is interrupted or unrelenting. In this third orientation, cause results at a moment when the instrumental proceeds but simultaneously hesitates, stutters, or outright fails to realize its purported goal. In the perplexing lack or surplus emerges – without guarantee – an occasion for what psychoanalysis has called the "sublime": "an object raised to the dignity of das Ding." The instrumental is neither refused nor put in question as a fantasy. Rather, the performatic transfigures the instrumental into something properly and nobly inaccessible. The operation that exchanges inputs for outputs and uses the algorithm to do so is here barred from the expected libidinal economy of exchange and use value. As a specific encounter with an aesthetic obstacle, the sublime meets the prospect of a critical shift from contingency – a chance accident – to necessity – a compulsion to repeat the circuit as a concrete, self-differentiating concept. The sublime of the performatic gesture signals a jump from a prohibition, which demands correction, to an impossibility, which invokes a systemic non sequitur, a partial transcendence of the operative structure, and a leap of faith into the void that lies beyond and at the same time within the dream of "laborless labor."
The distinction between "prompt engineering" and "prompt art" hinges on a shift in function of the exact same gap at the operational nexus. Might one assemble a compendium and exhibition of prompts à la Fluxus for a set of acts or executions that are conceptually held, even momentarily, in abeyance?
The digital machine comprises an assembly of intricately coordinated and reputedly harmonious causalities. In the Aristotelian jargon, the machine functions via four causes: end (the aim or purpose of the system); agent (the mechanisms that drive the system’s work); form (the pattern that performs the work); and material (the substance mediated through labor that embodies the agent).
Exhibit C
1.618 Things I Know About Her presents an AI-generated replica of a short folkloric film sequence. The piece mobilizes the predictive logic of text-generation machine learning to regenerate scan-line by scan-line the 1967 film purportedly documenting a Sasquatch (or Yeti) walking urgently through a Northern California forest. The specific type of machine learning used for this project comprises a more general and complex version of the algorithm operating in auto-correction and other predictive text technologies. We encounter such algorithms when we begin to write a text message and the device completes and/or alters our incipient entry.
The act of training a neural network aims to prepare the foundation for a predictive algorithm. Undergirded by this foundation, such algorithms are designed to produce outputs that conform to an established recognition system. Whatever eludes the system of recognizability is anticipated and corrected so that it meets the system’s existing standards. The attempted replica in our project essentially fails at the instrumental level, narrowly conceived, but succeeds by leaving visual traces of the structural gap that enables the algorithm to run in the first place. The performatic gesture in this project emerges at the point where the algorithm generates and gives a visibly unfolding form to the very gap – a blind spot and misrecognition – at the heart of this and every other AI system.
The algorithm’s predictive aspect in 1.618 Things I Know About Her is set up by dividing each scan-line into a series of tiny two-part units each based on the mathematics of the aesthetic gold standard: the Fibonacci or phi proportion (1.618:1). The algorithm makes two passes through the entire film sequence in order to generate the replica. The first pass preserves the ratio’s larger portion in each unit (the 1.618 length) as a reference to the original film’s fragment while deleting the smaller portion from the original (the 1.0 length) to create an empty field that is then filled via a prediction based on the reference material and drawn from the trained neural network. The second pass reverses the process by deleting the larger portion in each unit (the 1.618 original reference material) and by then adopting the artificially produced smaller portion (the 1.0 length) as the reference to predict and generate new material to fill the now-empty 1.618 lengths. After preparing and running the AI system for well over a month without interruption, the process was arbitrarily halted, resulting in a new, entirely fabricated 3,175-frame film, named in an homage to a 1967 Jean-Luc Godard film, 2 or 3 Things I Know About Her. Our right sidebar and main text here in tandem hint at the meeting point of subject matter and procedural form: The alleged hoax of the folkloric film footage and the predictive fantasy of the machine learning that reconstructs the film overlap in a refusal to reckon with the explosive "plasticity" implied in each. Such a refusal is entangled with the quasi-reassuring aesthetic of what might be called flexible containment. 1.618 Things I Know About Her endeavors to glimpse a plasticity rattling inside the flexible container of machine learning.
She might have reminded you of Julia Pastrana, the "Non-Descript" – paraded under the name "the Bearded Lady" – who by her own account lived and died happy. But in fact she doesn’t resemble Pastrana at all. Like others of her kind, she is constitutionally elusive. There also persists a degree of uncertainty about whether she was happy at the time she was caught on film walking through the woods of Northern California in the autumn of 1967. The anomalous existence of the footage testifies to the fact that her species, sapient but neither pre- nor post-human, does not answer the call to be recognized in and by "our" human social order. This withdrawal, even indifference – in short, this "retraction" – offers no reason to romanticize or exoticize the mystery. It instead shows up a fundamental blind spot in the anthropological regime. The common rejection of the footage as a hoax presents the trivial, symptomatic side of this motivated blindness. The elusive character of her kind – even if a complete fiction – exposes what is more intractably at stake: not the transiency of our current era in deep time but the very coherence of the concept of the Anthropocene.