Kafkaesque Algorithms: Kafka’s Writing in the Age of Artificial Intelligence

Abstract

This article uses the surge of recent AI-generated simulations of Kafka’s writing as an opportunity to reflect upon both what AI can teach us about Kafka’s writing and what Kafka’s writing can teach us about the age of artificial intelligence. Under the heading “Kafkaesque Algorithms”, this article explores three distinct but related questions that emerge at the intersection of stylistics, poetics, and media theory. First, do AI simulations of Kafka’s writing adequately capture Kafka’s style, and if not, why not? Second, is there perhaps something inherently algorithmic about Kafka’s poetics, in ways that might both tempt and resist simulation by AI? And third, is the notion of outsourcing low-level repetitive labor to machines—a common promotional strategy for contemporary AI writing aids—truly novel, or would it instead have already been familiar to Kafka himself?

1. Introduction

In the spring of 2021, The New Yorker published to relatively little notice two short, never-before-seen fragments from Kafka’s Metamorphosis. One relates Gregor Samsa’s childhood memories concerning “the bugs his father used to bring to him when he was still a young boy”; the other presents an extended ending to the novella, in which “Mr. Samsa” muses at length about the family trio’s upcoming move (quoted in Marche 2021). These are recognizable, almost overdetermined elements of Kafka’s imaginary repertoire; why then the muted response? For Kafka scholars, the appearance of previously unknown drafts of a familiar work is, if not surprising, nonetheless almost always welcome. The well-known particularities of Kafka’s compositional process (exacting but unstructured) and publication history (largely posthumous) mean that each of the many unpublished drafts and fragments that have trickled out over the past fifty years has offered new insights into a notoriously hermetic author and oeuvre.1 Yet the fragments from The Metamorphosis that appeared in The New Yorker had not been written by Kafka himself—at least, not directly. Instead, as the journalist and experimental novelist Stephen Marche (2021) confessed, they had been produced by an “artificial-intelligence application called Sudowrite”. Running at that time atop a modified version of OpenAI’s large language model GPT-3, Sudowrite boasted that it could produce a faithful continuation of any text that was fed into it, even—or particularly—literature, by using artificial intelligence to reverse-engineer the unconscious linguistic “algorithms” underlying an author’s personal style. After experimenting with algorithmically generated extensions to a number of shorter and longer excerpts from The Metamorphosis, Marche, at least, was convinced: “Brute computational power is why Sudowrite can write like Kafka”.

Of course, such claims can—and should—be subjected to critical scrutiny. Can a large language model really be re-trained in any meaningful way through such small text samples? (That said, it is striking that in the transition from “his father” to “Mr. Samsa”, the two fragments capture the subtle shift in narrative focalization within The Metamorphosis that follows Gregor’s death). But what is of greater interest than the accuracy of an assertion like Marche’s are its terms. Why test a new AI application against Kafka’s writing? What is it about Kafka in particular that makes him such an attractive subject for computer simulation? In the last few years, the connection has become a commonplace. In 2023, the website for the “Kafka AI Project” announced that a team had used GPT-4 to produce “missing chapters” of The Trial, while 2024 saw the premiere of Robert Schoen’s radio play Kaf*KI [Eng.: Kafk*AI], in which an author collaborates with an Odradek-like chatbot to produce a new version of “The Cares of a Family Man”, as well as the appearance of Viktoria Richtschaidt’s Kollwitz Trifft Kafka [Kollwitz Meets Kafka], in which an image-generating AI trained on Käthe Kollwitz’s art was used to produce illustrations for Kafka’s The Trial.2 Such a confluence cannot be explained on merely formal grounds. Although Marche (2021) gestures half-heartedly toward the alleged “plainness” of Kafka’s style as the justification for such an experiment, the real attraction of Kafka’s oeuvre is indicated in his article’s concluding lines. There, we read that in The Metamorphosis, “Gregor is subject to a miracle that is at once a revelation and a catastrophe. The human entity changes once again, in a way that is both magical and degrading”. Marche’s implication is clear: the advent of algorithms that can write like Kafka is itself a “revelation and a catastrophe”, subjecting “the human entity” to the kind of nightmare scenario and atmosphere of existential anxiety already familiar from Kafka’s writings.

Evidently, there is something Kafkaesque about the effects of the latest advances in machine learning and artificial intelligence. Since Gilles Deleuze’s (1992, p. 5) influential invocation of Kafka’s The Trial as an early index of the transition from analog “disciplinary societies” to digital and data-driven “societies of control”, much has been written on the “Kafkaesque” aspects of life in the age of digital algorithms (see Franklin 2015, pp. 107–11). The reasons are not hard to understand: the opaque decision-making procedures that once persecuted Josef K. from attic courts have migrated into software packages, where they continue to assert their power over individuals. From criminal sentencing to health insurance to consumer credit scores, “authority is increasingly expressed algorithmically” and thus asymmetrically, in so far as these algorithms are “black-boxed” or concealed from those subject to their perverse logic in ways familiar from Kafka’s writings (Pasquale 2015, pp. 8, 24). As a recent introduction to The Political Philosophy of AI put it, “in the 21st-century United States, Josef K. is black and is falsely accused by an algorithm, without explanation” (Coeckelbergh 2022, p. 2).

Yet the advent of algorithms which purport to write like Kafka signals a new phase in the encroachment of artificial intelligence upon all domains of human life: Where predictive policing algorithms make Kafka’s writing timely, applications like Sudowrite threaten to make it obsolete. (And not just Kafka’s writing—asking the latest versions of any contemporary large language model AI to expound upon the “Kafkaesque” dimensions of digital algorithms produces critical analyses that should make Kafka scholars uneasy.) This, perhaps, explains the attraction of using artificial intelligence to update, in a simulacrum of Kafka’s own voice, early-twentieth-century works like The Trial or “The Cares of a Family Man”: it evokes a vertiginous (because recursively Kafkaesque) scenario in which computer algorithms will soon supplant humans in the imagination of Kafkaesque scenarios.

To be certain, one should not take at face value the claims of Silicon Valley entrepreneurs or contemporary journalists when they assert that the style of great authors can now be flawlessly reproduced in silico. But literary scholars should also be wary, in response to the growing indistinguishability of works of mind and works of matter, of resurrecting Romantic myths of the inimitable because immaterial author-genius. Indeed, when explaining Sudowrite’s implications for literary culture, Marche (2021) often sounds less like an AI evangelist and more like a hard-bitten antihumanist critic, debunking our comforting illusions about individuality and expression: “Already, what GPT-3 shows is that literary style is an algorithm, a complex series of instructions. The reason a passage from Kafka […] doesn’t look like math is because the mind isn’t capable of holding a hundred and seventy-five billion parameters at the same time”. If this is correct, then Kafka’s signature style, far from being, in Buffon’s words, l’homme même, is now one algorithm among others—a “Kafkaesque algorithm” in more than one sense of the term.

In what follows, I take up the subject of “Kafkaesque algorithms” announced in this article’s title from a slightly different perspective than the power–political dimensions normally connoted by the term “Kafkaesque”. Here, the term “Kafkaesque algorithms” is meant to call attention to issues of the author’s proper name and personal style, where “Kafkaesque” means quite literally: Kafka-like. As the examples from Marche et al. suggest, the advent of generative artificial intelligence that can ostensibly “write like Kafka” poses new and urgent questions about the status of (digital) literature and (digital) authorship. Most immediately, however, the production of allegedly “Kafkaesque” writing via algorithm raises issues of empirical judgment and taste: Do these applications adequately simulate Kafka’s literary style? If so, how, and if not, why not? Further, Marche’s contention that generative AIs like GPT-3 reveal that literary style was an algorithm all along, albeit one that we were incapable of recognizing as such, touches upon questions of poetics and stylistics: Was there something formulaic, algorithmic even, about Kafka’s writing? Finally, the existence of software like Sudowrite, which promises to liberate would-be authors from a kind of manual drudgery—what the application’s founder refers, brazenly, to as “the low-level work of pumping out word by [sic] word by word” (quoted in Marche 2021)—raises broader literary–historical and literary–theoretical issues about the significance of writing in modern media environments. The implication is that, in the age of text-generating AI, there is a kind of stubborn asceticism inherent to any contemporary writer’s decision to keep placing one word after another, in the manner prescribed by tradition. But how new is such a mindset, really? Would such an attitude have been foreign, or even inimicable, to Kafka? These are the questions and issues pursued in the following.

2. Kafkaesque Algorithms (I): Computer-Generated Kafka

The philosopher of technology and early Kafka critic Günther Anders once spoke of a kind of “Promethean shame”, which he defined as the mild or acute sense of embarrassment modern human beings felt when “confronted by the ‘humiliatingly’ high quality of fabricated things” (Anders 2016, p. 30). In a remarkable recent essay, Hannes Bajohr (2022, p. 30) has extended this notion to contemporary discourses surrounding computer-generated art, asking whether one should not more properly speak of a “Promethean anxiety”—a fear that “humans might cede their status as creators to machines”. As is now well known, Kafka was an avid, if not entirely enthusiastic, follower of the technological developments of his era, and traces of such a “Promethean” attitude can be found in a number of his writings. One thinks, for instance, of the many letters to Felice in which Kafka confesses his anxieties and embarrassments in the face of her company’s automated speaking- and writing-machines (“Parlographs”), or the letter to Milena concerning the media wars between innovative transportation technologies, which facilitate co-presence and keep humans in the loop, and novel communication devices, which progressively replace humans with ghostly doubles and will soon render us irrelevant. Examples of this anxiety abound in Kafka’s fiction as well, ranging from the Traveler’s ambivalent attitude toward the inscription apparatus of “In the Penal Colony” to the concern of the eponymous “Family Man” that the artifact named Odradek “is likely to outlive him” (Kafka 1971, p. 429).3 The last text seems particularly suggestive in the present context: Is it possible that works derived algorithmically from Kafka’s own texts would, like Odradek, supplant their father in turn?

When it comes to computer-generated Kafka, such “Promethean anxieties” prove premature, at least for the time being. As the fragments produced by Sudowrite from The Metamorphosis are relatively short, let us consider instead a representative passage from the Kafka AI Project mentioned above, which used one of the latest large language models, OpenAI’s GPT-4, to “fill in the gaps” in the fragmentary manuscript of The Trial. The excerpt constitutes the opening of the third and final AI-generated “missing chapter”:

After his discourse with the priest, Josef K. vacated the cathedral and roamed the city streets, his mind heavy with ruminations of the enigmatic court. The world around him seemed to twist and shift subtly, the familiar urban terrain taking on an uncanny, dreamlike quality. It was as if the very essence of the court had bled into the fabric of the world, transmuting it into an embodiment of the labyrinthine bureaucracy that had ensnared him.

(KAIP 2023)

For Kafka scholars, the narrative origin of the passage is obvious: it must follow immediately after Josef K.’s infamous discussion with the priest, in which the parable “Before the Law” is told and then interpreted. No less apparent, however, is that the origin of the passage is not Kafka: the adjectives are didactic, the metaphors predictable, the sentence structure repetitive—not to mention the fact that it is written in (imperfect) English. The progression of the “chapter” is similarly belabored: Josef K.’s growing despair at the realization that “the labyrinth of bureaucracy was not just a physical space, but a pervasive force that consumed everything it touched” gives way, at the chapter’s conclusion, to the buoying literary–critical or existential insight that the trial had been “a journey through the recesses of his own mind, a trial held within the chambers of his psyche” (KAIP 2023). In recognizing that he was the unconscious instigator of his own trial, he is at least partially relieved of its burden. Unfortunately, this concluding twist founders upon the fact that the end of the computer-generated chapter is not the end of the novel as a whole. Thus, the AI’s allegorical auto-interpretation of K.’s “trial”—an ending somewhere between “it was all a dream” and “it was all a metaphor”—is followed, incongruously, by Kafka’s own ending, in which the same K. dies “like a dog” in a rock quarry.

It seems unlikely, then, that Kafka’s digital offspring will replace their analog progenitors in bookstores any time soon. (This not least for commercial reasons, as US and EU courts have ruled that AI-generated works of art cannot be copyrighted). Before congratulating ourselves on a victory over our tools, though, it is worth taking a second look at the passage. For its source, while not Kafka, is nonetheless easily recognizable as something else—namely, the pool of clichés and caricatures of Kafka’s work that circulate in the mass media under the heading of the “Kafkaesque”. From the invocation of “enigmatic courts” to the “uncanny, dreamlike quality” of once-familiar “urban terrain” to a “labyrinthine bureaucracy” in which Josef K. has been “ensnared”, the algorithmically generated text resembles not so much Kafka’s writings as tropes that circulate about Kafka. The effect is uncanny, as if someone who had heard much of Kafka but had never read him had been asked to imagine one of his novels. As Oliver Jahraus (2008, p. 232) points out, this is a general challenge of intermedial adaptations of Kafka’s work: “every Kafka film adaptation is farthest removed from Kafka where it attempts to depict [bebildern] not Kafka, but the Kafkaesque”.

The problem of intermedial adaptation can serve as a reminder that, although these AI-generated works are presented to the reader in the same text form as Kafka’s originals, they are the product of specific and historically datable media technologies. Indeed, the eerie, washed-out quality of these simulations—not so much Kafkan as Kafkaesque—is due to the same innovations in artificial intelligence that make such human-like text generation possible in the first place. One of the more significant drivers of advances in AI in the last two decades has been the resurgence, due to larger data sets, more computing power, and innovations in machine learning, of what is termed “sub-symbolic” or “connectionist” AI over earlier “symbolic AI” approaches, now sometimes referred to as GOFAI or “Good Old-Fashioned AI”.4 At an abstract level, the contrast can be likened to two ways of teaching someone to speak Kafka’s German. Symbolic AI provides in advance a set of symbols and formal rules for their manipulation, such as “First, place a temporal adverb from Table 1 or a subject noun from Table 2 in the first position; next, take a finite verb from Table 3, conjugate it in agreement with the subject noun, and place it in the second position; now, place …”. This, in fact, is more or less the algorithm used to produce the first digital simulation of Kafka’s prose. In 1959, the German computer scientist and part-time literary theorist Theo Lutz used a Zuse Z22 computer belonging to the Technical University of Stuttgart and an edition of Kafka’s The Castle to produce what was long held to be the world’s first piece of computer-generated literature (Beals 2018, p. 158), entitled “Stochastic Texts: A Selection”. The idea seems to have been inspired by Lutz’s affiliation with the philosopher, experimental author, and information theorist Max Bense, whose seminar Lutz attended. By means of a relatively simple program, Lutz (2004) was able to have a computer randomly select nouns from a pool of 16 terms chosen from Kafka’s text, modify them with a logical operator, and then link them via a copula to a randomly selected adjective. An additional loop allowed for the first clause to be completed with a period, or, with significantly lower probability, to be linked to a second clause via a conjunction. The resulting text consisted of well-formed, vaguely ominous sentences, a Kafkaesque vocabulary in both the strict and the loose senses of the term: “NOT EVERY GAZE IS NEAR. NO VILLAGE IS LATE. A CASTLE IS FREE AND EVERY COMMONER IS DISTANT”, it begins, before—as in Kafka’s Castle—“THE STRANGER” [DER FREMDE] appears. The limitations of the program, however, quickly became evident. In contrast to GPT-4, whose rendition of The Trial attempts to produce a narrative arc, Lutz’s program, linked to a teleprinter, would simply proceed to generate simple predicative sentences until it was shut off.

The linguistic sensibility of symbolic AI, then, can be likened to a diligent student of German armed with a grammar and a dictionary (or perhaps a tourist with a phrasebook); its predictability and logical rigor comes at the cost of fluency and spontaneity. Sub-symbolic AI, by contrast, operates in a manner akin to the “immersion method” employed on infants and contemporary foreign language learners. Through prolonged exposure to a vast number of examples, as well as some gentle correction, organic and artificial neural nets alike will eventually learn to produce new outputs that resemble, at a statistical level, the inputs they were trained on. This training operates both at the levels of syntax and vocabulary—in Jakobsonian terms, on the syntagmatic and paradigmatic axes—through techniques like “(self-)attention”, which allows neural nets to better process relations among sequential data like text, and “word embeddings”, which use co-occurrence in texts to map relations between words onto many-dimensional (300+) semantic spaces. This comes with two challenges, however: first, in contrast to human language learners, such neural nets cannot be adequately trained on the outputs of one native speaker—or even one hundred. Vast linguistic corpora are required, comprising not just the collected works of an author, but those of their readers (and their readers’ readers) as well. The result is, as we have seen in the cases of computer-generated versions of The Metamorphosis and The Trial, not models whose texts reads like Kafka’s works, but instead ones whose productions read like how people tend to write about Kafka.

Of course, it is conceivable that this problem of pale imitation of Kafka’s German could be resolved through further improvements in machine learning, or through clever workarounds. In a recent piece, Kurt Beals (2024) moots as a thought experiment the possibility of an AI trained via machine learning on the Prague German-Jewish language milieu in which Franz Kafka grew up, then fine-tuned to emphasize the styles of Kafka’s self-declared literary “blood relatives” (Kleist, Dostoevsky, Flaubert, Grillparzer, etc.), such that it could perfectly simulate Kafka’s prose style. Here, however, we encounter the second challenge: even if such a flawless “KafkAI” could somehow be produced, it would only be of marginal use to Kafka scholars. Recall our neural net trained to speak a foreign language via immersion: while it can produce sentences indistinguishable from those of a native speaker, it cannot explain to you its own rules for doing so, at least, not in a way that you could consciously implement or imitate. The same would be true of a neural net trained to write like Kafka. As the computer scientist Melanie Mitchell (2019, p. 108) puts it, “‘showing their work’ is something that deep neural networks—the bedrock of modern AI systems—cannot easily do”. This is due to the basic nature of sub-symbolic AI, which as its name implies operates at a level below the formal rules and symbolic manipulations of conscious thought and natural or artificial languages. In contrast to something like Lutz’s program for generating Kafkaesque sentences, there would be no code to inspect, merely countless weighted nodes in a many-layered network, linked in extremely convoluted ways. This could be reformulated as a kind of paradox or aporia of “Kafkaesque algorithms” in the digital age: when the algorithms subtending computer-generated Kafka texts can be explained, their outputs will not resemble Kafka, and when the texts truly are Kafka-like (or even just Kafkaesque), it becomes difficult to explain why.

Put slightly differently, the challenges facing a “KafkAI” may be representative of the challenges facing contemporary sub-symbolic AIs more generally: they do not really understand what they are talking about, despite or because of the fact that they are able to “parrot” those who do. This, at least, is the charge advanced in an oft-cited paper, in which contemporary large language models are referred to as “stochastic parrots” (Bender et al. 2021). The deflationary term is meant to emphasize that such models operate not on the basis of a reconstructable symbolic logic or adequate understanding of their subject matter, but instead according to sheer linguistic probability. Of course, it is important to highlight and understand the distinctions between organic and artificial neural nets, particularly when it comes to language production. Nonetheless, the authors’ reference to “stochastic parrots” not only draws unwittingly upon a long history of opposing authentic human speech to mere animal mimicry—one which, in its preference for using birds and simians as unthinking foils (Siegert 2015), Kafka’s Red Peter might object to. It also presupposes that the human speech such models were trained on was itself immune to the avian dynamics of chatter.

Perhaps just this blind spot is what the Kafka AI Project sought to expose. Read carefully, the activities of the “team” responsible—whose members are never named—seem not so much secretive as subversive. Nearly all of the text promoting the Kafka AI Project is generic and bland, obviously itself AI-generated, with one notable exception: in response to a user’s comment on a Reddit forum for digital literature which asks if, with “AI Kafka”, the “overuse of the term ‘Kafkaesque’ is finally coming full circle?”, the account associated with the Kafka AI Project momentarily breaks character to respond that this is “precisely why” the work of Kafka was selected over that of any other author for simulation. This suggests that the aim of the GPT-generated edition of The Trial may not have been homage or displacement, but instead social satire. For in the age of machine learning on large, uncurated data sets, the “overuse” of a term like “Kafkaesque” in discussions of AI results not in linguistic aversion but in overrepresentation in the training data. Given that Reddit—an online social news aggregator and discussion forum that is one of the most-visited websites in the world—was and is a major source of training data for OpenAI’s GPT language models, the subsequent promotion of the Kafka AI Project on Reddit represents the discourse of the Kafkaesque coming full circle. What users could find in these GPT-4 generated “chapters” of The Trial were not Kafka’s speech patterns but their own, Kafka seen through the funhouse mirror of the contemporary Kafkaesque. What the Kafka AI Project ultimately sought to reveal about “Kafkaesque algorithms”, then, were not the historical algorithms underpinning Kafka’s writing, but instead the automatisms, tropes, and clichés that inform the ways we now talk about Kafka—and about AI.

3. Kafkaesque Algorithms (II): Word-Processing in Kafka’s Prose

Nonetheless, this should not be taken to mean that there is nothing algorithmic about Kafka’s style. Indeed, it may be that contemporary large language models struggle to imitate Kafka’s writing not because of the ways in which it is not machine-like, but because of the ways in which it is. Perhaps Kafka simply employed different—and thus more difficult to assimilate—algorithms. While Kafka was not among those early-twentieth-century authors who made explicit use of algorithms in the generation of text, his work does reflect what David Wellbery (2010, p. 276) has referred to as the emergence of a “constructivist concept of form” in the early twentieth century, characterized by its combination of inorganic, logical (or pseudo-logical) rigor and iterative repetition. Since Max Brod, critics have tried to name the peculiar “spiral” tendency characteristic of Kafka’s involuted prose, in which each successive sentence seems to build upon—but also feed off of—the sentence that preceded it. Why not refer to this combination of dynamism and proceduralism as “algorithmic”, a term which has famously been defined as “logic + control” (Kowalski 1979)? In keeping with Borges’ well-known proclamation that Kafka “created” his own precursors, by enabling them to be recognized after the fact as anticipations of Kafka’s style or themes, it is conceivable that even unsuccessful attempts to implement “Kafkaesque algorithms” via digital literature can sensitize us to their “precursors”, to what was always-already “algorithmic” in Kafka’s prose. How else is one to make sense of Kafka’s maddeningly recursive narratives, his endless iterations in the diaries on the same themes, if not as the expression of an almost machine-like drive at the core of his literary production? This approach allows us to reappropriate, for the purposes of Kafka criticism, both Marche’s anti- or post-humanist insight—i.e., the assertion that any strong literary style was always in some sense “algorithmic”—as well as the evident attraction exerted by Kafka’s writing upon those who seek to produce art through artificial intelligence.

The task of this section is thus to reverse-engineer, via analog methods, the algorithms underpinning Kafka’s writing. A first pointer is provided, ex negativo, by the failures of the Kafkaesque digital texts examined above. What is striking about the passages generated by contemporary sub-symbolic AI is that, although they contain many of Kafka’s favorite themes—the law, the court, the nightmare, the enigma, the labyrinth, and so forth—they are conspicuously devoid of any reference to what is perhaps the most Kafkan topic of all: writing. The computer-simulated Josef K. of the Kafka AI Project’s version of The Trial suffers through a recognizable series of interrogations, confrontations, and pursuits by shadowy forces, yet there is no equivalent to the oft-cited passage in The Trial in which K. considers writing and submitting to the court a comprehensive review of his entire life. Instead, the production of text drops out entirely, rendering these productions not truly Kafka-like but instead merely Kafkaesque. This discrepancy is in keeping with what one of the few critical commentaries on “the Kafkaesque” describes as this discourse’s blind spot: “The Kafkaesque is what most emphatically conceals Kafka from view, because it overlooks something which, in the case of Kafka, is decisive and indeed constitutive for the relationship between life and work: namely, writing” (Jahraus 2006, pp. 30f.). The Kafkaesque is thus not simply Kafka-lite, a sort of abridged or truncated version of Kafka, but is instead a kind of anti-Kafka, insofar as it fails to acknowledge the centrality of writing to everything that he wrote. (With any other author, the sentence “writing was central to everything that he wrote” would be a mere tautology; with Kafka, it is a fundamental truth.) For this reason, Jahraus concludes, “the Kafkaesque must be negated in order to understand the full significance of writing [for Kafka]” (p. 31).

This suggests that if there is something algorithmic about Kafka’s prose, it lies not in the invocation of familiar Kafkaesque themes, but instead in the continual relation of what is written to the act of writing itself, such that the generation of text becomes both theme and motor of a given work. The historical conditions for such autonomous, self-thematizing textuality can be found in what Friedrich Kittler (1999) has referred to as the end of the “alphabetic monopoly” around 1900. According to Kittler’s innovative media-historical periodization, the simultaneous introduction of three new “technical media” around 1900—gramophone, film, typewriter—put an end to the vivid, sensory-rich worlds of Romantic and Realist literature. During the eighteenth and nineteenth centuries, the word had been the privileged medium for encoding, symbolically, sights and sounds, which were subsequently decoded and reanimated in the mind through the act of reading. Yet the creation of analog media that, as the names phono-graph and cinemato-graph imply, could “write” audible sound and visible movement directly deprived language of its pre-eminence in the creation of audio–visual hallucinations (F. Kittler 1985). What remained of literature was the pure textuality of the letter itself, which in Kafka’s era, was increasingly filtered through the keys of the typewriter. The machine’s forceful inscription of standardized black letters upon the white page simultaneously foregrounded the material and machinic qualities of writing, while distancing the author’s subjectivity from what he or she wrote. As Kafka typed to Felice Bauer in 1912, in his very first letter to her, the typewriter would make him a better correspondent—even if his “mood doesn’t happen to feel equal to a letter, there are still my fingertips to do the writing” (Kafka 1973, p. 5; see W. Kittler 1984).5 With the aid of the typewriter, the writing subject could become what the device’s German name (Schreibmaschine) already implied: a machine for writing.6

Of particular significance for the subject of “Kafkaesque algorithms”, however, is a sentence that Kafka types only a few lines later, as he begins the second page of his letter to Bauer: “I notice while inserting a new sheet of paper that I may have described myself as far more difficult than I am. If I have made this mistake it would serve me right, for why do I choose to write this letter after six hours in the office, and on a typewriter I am not used to” (Kafka 1973, p. 5).7 What is important here is not so much the typewriter as the textual feedback loop that it enables, insofar as feeding a new sheet of paper into the machine allows Kafka the opportunity to feed his own text back into itself, generating further commentary. This initial scene of writing about writing—the recursive re-entry of letter-writing into the letter being written—would set the tone for the subsequent correspondence. In the months and years that followed this first missive, Kafka overwhelmed Bauer by compulsively authoring ever-new letters about the letters he had just written, each of which demanded a response in turn, resulting in what Friedrich Kittler terms “word-processing in an infinite loop” [Textverarbeitung in Endlosschleife] (F. Kittler 1999, p. 223). Kittler’s use of an information-theoretical vocabulary to describe Kafka’s writing is no coincidence: according to one of his bolder hypotheses, a computer resembles nothing so much as an electrified typewriter, capable of processing text autonomously. Elsewhere, Kittler (2013, p. 186f.) points out that Alan Turing’s conception of a “Universal Discrete Machine”, which operates by advancing an alternating read/write head step by step across an infinitely long strip of paper, not only resembled “a typewriter stripped down to its fundamental principle”, but was in fact preceded by Turing’s own youthful invention of primitive writing machines. In the humble “word-processing” of the typewriter, Kittler finds the seed of those algorithms that, as computer programs like Sudowrite or GPT-4, could read and write themselves.

The presence of such activity in Kafka’s correspondence suggests an unsuspected affinity with Turing, Kittler’s preferred twentieth-century source on symbol-processing machines. For Kittler, however, Kafka’s “word-processing” is primarily of interest as a symptom of the nascent “desexualization” of authorship around 1900, with the emergence of female typists and the new genre of women’s novels that describe, “in infinite loops”, how these typists become authors. As a consequence, he misses—or chooses to overlook—the fact that such “word-processing in an infinite loop” is a core element of Kafka’s poetics more broadly, that he wrote letters about the letters he had just written in more senses than one. Put slightly differently, the looping or spiraling dynamics of the passage from the first letter to Felice (“I notice while…”) are in fact typical of Kafka’s “word-processing” in fiction as well, insofar as it displays something like a self-aware textual system which reproduces itself, line by line, through a kind of feedback loop. The principles underlying such a system could be described as a Kafkan algorithm for text production, a set of rules for accomplishing a common or repeated task. (This could be described as a Kafkan “poetics” in the etymological sense of the word, where poiesis refers to a kind of generative activity.) While such an algorithm cannot be specified with the formal rigor needed to code it, the attempt which follows to spell out some of its principles may shed new light on Kafka’s works.

A particularly suggestive demonstration of Kafka’s word-processing is provided by one of the fictional texts that he had submitted for publication around the same time that he was beginning his correspondence with Felice. This is “The Excursion into the Mountains”, which appeared in Kafka’s first book Betrachtung [Contemplation], a collection of short prose pieces, in 1912. (It was while discussing the final arrangement of texts for Betrachtung that Kafka first met Bauer, the significance of which has been emphasized by Canetti (1974).) Among the merits of “Excursion” in this context are not only its brevity and formal elegance, but also the conspicuous absence of those thematic elements (the law, the labyrinth, the bureaucratic, the enigmatic) frequently associated with the Kafkaesque.

The Excursion into the Mountains

“I don’t know”, I cried without a sound, “I just don’t know. If nobody comes, then nobody comes. I’ve done nobody any harm, nobody’s done me any harm, but nobody is helping me. Nothing but nobodies. Yet that’s not how it is. It’s just that nobody is helping me—otherwise, nothing but nobodies would be rather fine. I’d love to go on an excursion—well why not?—with a company of Nobodies. Into the mountains, of course, where else? How these Nobodies jostle each other, all these extended and interlinked arms, these many feet separated by such tiny steps! Of course they are all in tail coats. We go so gaily, the wind blows through us and the gaps that we and our limbs leave open. Our throats swell and are free in the mountains! It’s a wonder that we don’t burst into song”.

(Kafka 1971, p. 3838)

Much has been written about this remarkable piece, both in the context of other early writings and against the horizon of Kafka’s oeuvre as a whole (see in particular Lehmann 1984; Endres 2013; Krings 2018). While “Excursion” is one of Kafka’s first works to appear in print, its origins lie even further back—in the Description of a Struggle, an early experimental prose narrative which Kafka subsequently pillaged for his first publications. Here, I would like to call attention to just three features of this text which I take to be representative of fundamental aspects of Kafka’s poetics. I term these processuality or open-endedness; recursivity or feedback; and reflexivity or self-thematization. With these categories, I make no particular claim to originality, only to a compressed but comprehensive account of Kafkan—and not Kafkaesque—“algorithms”.

Processuality/Open-Endedness. A basic feature of algorithms is that they are processual, i.e., that they govern a series of steps that are carried out diachronically. This coincides with one of the more striking features of “Excursion”, its air of a dynamic and open-ended unfolding in time. This is emphasized both by the abrupt beginning in medias res, in which it is never explained what the narrator “just doesn’t know” or in what situation his discourse appears, as well as by the repeated markers at the level of both form and content of a narration that is unfolding diachronically. The “Excursion” is neither the transformation into language of a static, pre-existing scene, as is underscored by spontaneous, dialogical interjections like “well why not?” [warum denn nicht] or “where else?” [wohin denn sonst?], nor does its comparatively threadbare narrative suggest an attempt to conjure up a compelling fictional world. Instead, we see the ragged outline of a plot emerge (the narrator, alone, is joined by a pack of “nobodies”; together, they mount an “excursion into the mountains”) in what, borrowing from Kafka’s “blood-relative” Kleist, could be described as die allmähliche Verfertigung einer Geschichte beim Schreiben, “the gradual formulation of a story while writing”. One remarkable demonstration of this tendency is the abrupt transition from the optative of the seventh sentence (“I’d love to go …”) to the present indicative of the ninth sentence (“How these nobodies jostle one another…”). What was lacking in the fictional scenario moments prior is suddenly present, the subject of a description, as if the necessary “company of nobodies” had simply been conjured into existence by the narrator in between sentences.

Such spontaneity could, of course, be feigned—nothing prevents Kafka from having conceived in advance a work that only appears improvised. Yet this processuality is in keeping with what is arguably Kafka’s fundamental statement on his own poetics, as reported by Max Brod (1957, p. 267): “One must write as if in a dark tunnel, straight into the darkness, without knowing how the figures will develop”. In the memorable phrasing of the philologist and Kafka editor Malcolm Pasley (1980, p. 14), one has the sense that Kafka’s works were not “ge-” but rather “er-schrieben”, written into existence step-by-step. Thanks to his work with Kafka’s manuscripts, Pasley is able to cite as evidence not just Kafka’s own assertions but also the “mute testimony” of the archive itself, which reveals that Kafka rarely wrote according to a pre-fashioned plan, but instead revised, modified, expanded his sentences as he put them to paper. This explains the thematic appeal of journeys and voyages to Kafka, as well as the fact that his protagonists tend to err, wander, double back, or become disoriented. However, this radical open-endedness also poses a problem: if one does not decide in advance what to write about, where does the material come from? Where does Kafka draw the next sentence from, if not an artistic vision that preceded the work’s transcription? One answer is: from the sentence that preceded it.

Recursivity/Feedback. With this, we encounter a second prominent feature of “Excursion”: its recursivity. From the Latin recursio, running backward, the term came to be increasingly associated with the domains of mathematics and information sciences in the early twentieth century and featured prominently in Turing’s early work on computers (Krajewski 1997; Chabert 1999, pp. 455–80). Although definitions vary by field, recursion generally refers to “self-calling algorithms”, or those systems which, in keeping with the term’s etymology, ‘run forward’ by ‘running backward,’ using their own past states in the present to produce their future states. Typically, such an algorithm has a defined initial state, to which the recursive function is then applied to generate the next term in the sequence—such as the 0 and 1 which begin the Fibonacci sequence, or the arbitrary explanandum which initiates the child’s recursive series of “Why?”s. As Geoffrey Winthrop-Young (2015, p. 448) puts it, “Recursion refers to an algorithmic procedure in the course of which the output of a particular stage in the operation of a system is fed back into the system as the input for the next iteration”.

This too is a hallmark of Kafka’s prose, which often seems to iterate obsessively upon an apparently unmotivated starting point. Admittedly, according to Winthrop-Young’s definition, all fiction has a certain degree of recursivity; characters, scenes, plot points are introduced at the outset in order to be drawn upon, revised, and expanded for the story’s progression (see Luhmann 2008). Yet the effect is particularly pronounced in Kafka’s “Excursion”, which progresses only by repeatedly drawing upon the previous sentence’s use of “nobody” and modifying it slightly. In contrast to other typewriter modernists, Kafka’s word-processing occurs not in simple loops (“A rose is a rose is a rose”) but in spirals: the initial invocation of ‘nobody’ as an actual absence of an interlocutor, an indefinite pronoun, acquires a kind of progressive substantiality, as ‘nobody’ takes on form, becomes a positive entity (“Nobody”), multiplies, and begins to sing. “Excursion” itself becomes a circuit or a positive feedback loop, amplifying the gain with each go-round until it threatens to blow out the text’s speaker.

Of course, accretion or gain is not the only possible outcome of recursivity. There are many ways in which a prior output can serve as an input for later stages of a process, and many different kinds of self-calling algorithms; not all spirals ascend upward, particularly in Kafka. One thinks, for example, of the rolling negation or negative feedback loop characteristic of another famous text from Betrachtung, the “Wish to be a Red Indian”. There, in a kind of inversion of “Excursion”, the starting condition—a fantasy of oneself as an “Indian” galloping on horseback—is systematically and recursively negated, clause by clause: “until one dropped the spurs, for there were no spurs, until one tossed the reins, for there were no reins, and could scarcely see the country before one as smooth shorn heath, already without the horse’s head and neck” (Kafka 1971, p. 390).9 This feature of Kafka’s poetics of word-processing—in “Excursion”, the word is “nobody”; here, it is “rider”10—has been described in a penetrating article by Hans-Thies Lehmann (1984, p. 214) as the “deconstruction” [Abbau] or “withdrawal [Entzug] of reference”; the phrase forms the basis of an interpretation of Kafka which he supports through discussion, among other texts, of “Excursion”. Yet what Lehmann’s term and reading miss is that the recursivity of Kafka’s prose is neither exclusively negative nor is it exclusively deconstructive. Instead, it is quite often productive, at the level of form and content. In “Excursion”, recursion progressively lends shape to the text and substance (or body, as it were) to the initially empty “nobody”, such that the word itself (“niemand”) rises by the text’s conclusion to the status of a proper noun (“Niemand”).

Reflexivity/Self-Thematization. This typographic dimension allows us to further specify Kafka’s poetics: Not all algorithmic processes are recursive, and few recursive processes are as relentlessly self-reflexive or self-thematizing as Kafka’s writing. A third fundamental characteristic of “Excursion” is that it is a text in which nearly every sentence can be read two ways: as a statement about a fictional world or as a statement about the creation of such worlds through writing. Consider the process by which the first sentence unfolds. The opening, enigmatic exclamation—“I don’t know”—is immediately qualified by the no less enigmatic “… I cried without a sound” [rief ich ohne Klang], the only words in all of “Excursion” not contained within quotation marks. What is one to make of a “silent cry”? While the logical contradiction can be read as the articulation of a Kafkaesque nightmare scenario, it is also an elegant description of Kafka’s writing itself, as mute “speech” captured in text. It is as if the second clause—“I cried without a sound”—was an immediate commentary upon the first. As before, an initial output of the narrative system is fed back into it as input and transformed, generating a new output—albeit one that reads as reflection upon the output that preceded and produced it.

This first moment of processual, self-reflexive word-processing is not an isolated incident, but instead something like the motor of the work as a whole. After the opening sentence, the text dithers briefly, turning in circles around various iterations on “nobody”, before the untranslatable locution “lauter niemand” appears. One can scarcely imagine Kafka’s joy upon hitting upon this richly ambiguous phrase: When “lauter” is read as the indeclinable adjective lauter, linked etymologically to notions of purity, “lauter niemand” means something like “nothing but nobodies”. However, when “lauter” is read as the form of the adjective laut (cognate to the English loud) declined for a masculine noun, “lauter niemand” means: “a loud Nobody”. The allusion to Odysseus’ use of the pseudonym “Nobody” [Outis] in Book 9 of The Odyssey is deliberate, and Kafka’s adaptation into German of the Homeric pun or “paronomasia” (Strowick 2004), through which the indefinite pronoun is treated like a proper noun and nobody becomes a Somebody, is subsequently concretized grammatically and typographically: “niemand” gives way to “Niemand” in the sentences that follow.

At the same time, the formulation “lauter niemand” allows Kafka to exceed the Classical intertext, concluding “Excursion” with what is undoubtedly one of the most compelling depictions of the strange combination of presence and absence, materiality and ideality, that is the letter or linguistic sign. As Lacan (Lacan 1992, p. 121) liked to joke, a signifier is like a macaroni, or a cannon: “a hole with something around it”. The reference to a nothing “in a tail coat” is an inspired depiction of the typed letter, set out in black against the white paper, while the “jostling” limbs and “feet separated by such tiny steps” evoke the letters as they “run” together across the page. The text thus proceeds by writing into being a scenario that is, at the same time, a description or metacommentary on the scenario being written. Even the tension between written and spoken language is realized in the final lines of “Excursion”, as an imagined breath or spirit (pneuma) passes through the “gaps” in the letters’ “limbs”, causing the narrator and his substantially insubstantial companions to break into a voiceless jubilation. With this, the text has returned to its starting point, albeit in altered form: what was a “cry without sound” at the text’s outset becomes, at its conclusion, the “miracle” of a silent song. However, this success comes at a substantial cost: Kafka’s narrator finds his “voice” only by receding into a choir of textual “Nobodies”, a typically Kafkan conflation of authorial victory and defeat.11

A final note on beginning and ending in Kafka’s word-processing: Since Turing, algorithmic activity has been associated with so-called “halting problems”, issues of whether the operations of an algorithm-driven, symbol-processing machine will eventually come to a stop. Similarly, while Kafka regularly lamented the problem of beginning his stories, this was often with an eye towards his difficulties concluding them; when they do not halt abruptly, they tend to spiral out of control. From this perspective, it is surely no coincidence that “Excursion” ends when it returns, in a spiral movement, to its beginning. Not all of Kafka’s word-processing ends as neatly as the pieces in Betrachtung, a fact which has often bolstered the claim that the style or themes of these early prose texts are somehow fundamentally distinct from those of his later work. Nonetheless, there is much to indicate that Kafka’s conception of the writer’s task, and the algorithm that he used to address it, remained the same throughout his life. Consider, as one example that must stand in for many, the late prose fragment “A Delicate Task” [Eine heikle Aufgabe], which runs as follows:

A delicate task, a tip-toeing across a crumbling board set down as a bridge, having nothing underfoot, having to first scrape together with one’s feet the ground upon which one will walk, walking on nothing but the reflection that one sees in the water below, holding the world together with one’s feet, hands cramping in the air above to survive this ordeal.

(Kafka 2020, p. 2312)

Evidently, the “task” in question is writing, and the fragment itself an attempt to carry it out. Whether it succeeds or fails is something of an open question; the short text never really manages to find its footing, but this also seems to be the point. Its “failures” generate a welter of successful images, each an iteration on the unspoken master-word or “hypogram” Schreiben [writing, to write].13 (This can be seen by mentally adding the word to the opening line, such that it reads “Writing: A delicate task, a tip-toeing …”, an act which also underscores that every apposition following the first is an extended infinitive clause, devoid of subject).

The mechanisms of Kafka’s word-processing at work here can be described with even greater precision, for despite the shift in tone and gap in years, the underlying poetic principles of “A Delicate Task” are clearly the same as those of “Excursion”. The piece is overtly processual and open-ended, advancing clause by clause like the “tip-toeing” that it represents; it is recursive, in that each successive verbal infinitive construction takes up, but modifies, the image immediately preceding it; and it is brilliantly self-thematizing, deploying a kaleidoscopic sequence of metaphors (walking on a “crumbling board”, walking on ground that one has created oneself, walking on a “reflection” or illusory referent, walking on nothing at all) for the simultaneous “movement” or “progression” of a text across the page and of narration across imaginary space and time. In a last, inspired twist, the final image of “hands cramping in the air” can be read as the sign of concentration on behalf of the literary tight-rope walker, or as the upstretched arms of “surrender”—a concluding pun on the noun “Aufgabe” with which the piece begins, which can mean to give something out (like a task) or to give something up (like writing). The tension between the two readings of the final image is in keeping with Kafka’s own internal contradictions: for him, the task of writing is synonymous with the author’s inevitable defeat, even as the “ordeal” produces ingenious and inimitable prose.

4. Kafkaesque Algorithms (III): Analog Authors, Modern Machines

A high-wire act, an attempt to walk on water, a suspension in mid-air: Kafka’s “A Delicate Task” offers a heroic but rather bleak view of the author’s work. The modern writer is depicted as an exacting wordsmith, agonizing—in the tradition initiated by Kafka’s nineteenth-century idol Flaubert—over every syllable of his prose. Would it not be easier, in the twenty-first century, to turn this “task” over to artificial intelligence, so as to be able to focus on the realization of one’s own artistic vision? This is the argument of Amit Gupta, founder of the “AI Writing Partner” Sudowrite discussed above, who asserts in conversation with Stephen Marche (2021) that “visual artists are using really advanced tools that automate a lot of the processes. But writing has been stuck in the past”. While their colleagues make use of 3-D modelers or image-editing software like Photoshop, authors—equipped with word-processors or, worse, “paper and pen”—remain comparatively speaking in the Stone Age. For those who use Gupta’s aptly named Sudowrite, however, “the writer’s job becomes as [sic] an editor almost. Your role starts to become deciding what’s good and executing on your taste, not as much the low-level work of pumping out word by word by word” (ibid.). Here, in the intersections of new media theory and new media entrepreneurs, silicon in theory and in the Valley, the antihumanist reframing of writing as “algorithmic” activity melds with late-capitalist hierarchies of labor. Why perform the “low-level work of pumping out words” when this manual labor can be automatized and delegated, elevating the scribbler to the upper echelons of the “curator class” (Kompatsiaris 2024)?

It is tempting, in response to such rhetoric, to juxtapose the banal statements of Sudowrite’s founder with Kafka’s declaration that “My job is unbearable to me because it conflicts with my only desire and my only calling, which is literature” (Kafka 1976b, p. 230).14 Tempting, too, to declare: “Thank God Kafka didn’t live to see this!” Yet this would be to consign Kafka’s writing to a mythical era of pre-history, one in which the now “low-level work” of writing was not subject to question. In fact, as I have argued above, Kafka’s poetics of word-processing emerged in response to an age of medial—and not just medial—upheaval, when the emergence of new technologies made literature itself seem laborious, dated, and disconnected from reality. Indeed, Kafka seemed determined to demystify and demythologize the work of the author in modernity. Between the years of 1917 and 1920, he authored a number of short pieces in which he revisited figures from classical mythology and literary history, offering ironic and deflationary accounts of the poet’s task. Conspicuously, Kafka modernizes these myths through recursive, self-reflexive retellings, which invariably thematize the untimeliness of literature in modernity. In Kafka’s “The Silence of the Sirens” [Das Schweigen der Sirenen], for example, the “modest means” by which Homer once guided his protagonist past oblivion’s call have become “childish” and “inadequate” (Kafka 1971, p. 430f.), while the temptations of the now irresistible sirens resemble those of the nascent culture industry’s audio–visual media.15

Most suggestive, however, for the topic of “Kafkaesque algorithms” is Kafka’s “Poseidon”, the last of his updates to figures from classical mythology. It opens with a depiction of the ruler of the seas as a conscientious but dyspeptic administrator, a bureaucrat chained to his desk. Poseidon, it seems, is drowning in paperwork. The “administration of all the waters” leaves the Olympian with little time to do anything other than “calculate” [rechnen]; although he does not particularly enjoy the work, Poseidon takes his “office” [Amt] seriously and is reluctant to delegate even the most menial tasks, insisting upon “calculating everything through himself a final time” [rechnete alles noch einmal durch].16 Occasionally, he considers taking up a new position, but none of the possibilities on offer would befit the God of the Seas, nor would they allow him to do “less calculating” [rechnerische Arbeit]—only “lesser calculations”. What particularly irks Poseidon are the rumors that circulate about his professional activities: namely, that he spends his time traveling the waves with trident and chariot, when in truth, “he sits on the ocean floor and calculates without rest” [rechnet ununterbrochen]. In fact, he has never really seen the seas that are his domain, although he likes to say that he is waiting for a quiet moment just before the end of the world, when—after he has looked over his final “calculation” [Rechnung]—he will have time to “make a quick little tour”.

At first glance, “Poseidon” would seem to present a familiar critique of a debased, even Kafkaesque modernity. Fallen from the heights of Mount Olympus, the once-vital Greek gods have been reduced to petty bureaucrats; rather than rule over their domain directly, as in days of yore, they are now obliged to spend their time at the office, attending to endless paperwork. What is more, through his desk job, Poseidon himself has been reduced to little more than office equipment, a literal “computer” or “calculator” [Rechner]. While English translations of the text often refer to Poseidon’s task as “going over the accounts”, this conceals not just the consistency of Kafka’s use of rechnen and its derivatives throughout the text, but also its contemporaneity. In Kafka’s time, the model for such algorithmic “computation” would have been the electronic calculating machines of Hermann Hollerith, the German American immigrant whose company would later be renamed IBM (Austrian 1982). Although Hollerith’s programmable punch-card tabulator plays a prominent role in Kafka’s “In the Penal Colony” as a kind of inscriptive torture apparatus, the uses of calculation in “Poseidon” are more faithful to the machine’s true purpose—the administration of populations in the modern nation-state, or “bio-informatics” (Dotzler 2008). Hollerith machines had been employed to great effect in the US for the 1890 census, and quickly caught the attention of European commentators, including the Austrian national statistician Heinrich Rauchberg, later Kafka’s professor of “General and Austrian Statistics” at the Charles University in Prague (Wagner 2005). Even before Kafka’s “Poseidon”, the impact of such novel technologies upon the traditional prerogatives of the Olympians was clear; as the Electric Engineer reported in 1891, “The apparatus works as unerringly as the mills of the Gods, but beats them hollow as to speed” (quoted in Austrian 1982, p. 70). Kafka’s modern parable thus processes the end of rule from on high, and its replacement—or usurpation—by the authority of algorithms.

As before, however, what is excluded from the conventional Kafkaesque interpretation is the status of writing in and for the text. For another prominent feature of “Poseidon” is Kafka’s clear identification with the ruler of the seas, which suggests that Poseidon’s fate represents not just that of the modern ruler but also that of the modern writer. This presents an interpretive wrinkle, insofar as Kafka’s portrayal of the god, as an Olympian who spends his days doing elementary sums, bears an undeniable resemblance to Gupta’s portrayal of the contemporary author as a frustrated demigod, compelled to churn out expository dialog and scene transitions “word by word by word”.17 Is Kafka’s implication that it is only Poseidon’s stubborn insistence upon doing his calculations by hand, his refusal to delegate the algorithmic, “low level work” of computation to machines, which prevents him from re-attaining his true stature? If so, would the same not hold for Kafka himself, that he should delegate the more mechanical aspects of his vocation, in order to restore the writer’s former glory? This seems—even for a contrarian reading of Kafka—a bridge too far. A more conventional solution to the dilemma is that Kafka’s affinity with Poseidon rests upon their shared sense of workplace alienation. Did not Kafka famously lament the “horrible double-life” he was forced to lead, in which the “demands” of a tedious office job sapped his creative powers (Kafka 1976b, p. 38; but see Wolf 2006)?18 Like Poseidon, Kafka spent his days trapped behind a desk, going over paperwork. In contrast to the Olympian, however, there were rare occasions when Kafka was able to break free of an overly rationalized, overly bureaucratized modernity—such as the night in September which he wrote “The Judgment” in a single sitting, subsequently immortalized by its author as a dark, organic process in which the story “came out of me like a real birth, covered in filth and mucus” (Kafka 1976b, p. 214).19 “Only in this way can writing be done” (ibid., p. 213), Kafka exalted—through a sort of regression to a primal state, far from the world of automation, algorithms, and office equipment.20

This is a powerful myth about Kafka’s writing, indeed, a myth perpetuated by Kafka himself. But it remains, for all that, a mis- or myth-reading, one that cannot withstand close attention to the texts themselves. First, Poseidon is not, in fact, particularly alienated from his work; despite his grumbling, he refuses reassignment and finds that even the idea of a career change makes him “nauseous”. Instead, what “annoys him most” about his job is the false impression that others have of it: they imagine that his profession consists of exotic visions, journeys to distant lands, prominent displays of the trappings of symbolic authority. His objection is not that he sits, alone, at his desk, “calculating without rest”, but that others—perhaps even Zeus, who appears as a source of interruption and aggravation—are not aware that this is what his “office” entails. (The parallels with Kafka’s own relationship to his writing, here and in what follows, are apparent.) Second, upon first reading, it seems as if Poseidon’s desk job “administering all the waters” [Verwaltung aller Gewässer] is meant to thematize the tension between sovereign walten and bureaucratic verwalten, between Poseidon’s traditional “office” [Amt] as Ruler of the Seven Seas and the “office” [Büro] which prevents him from carrying it out. Read carefully, however, Kafka’s text suggests that Poseidon’s continued authority is something of a farce: the gods’ prolonged existence “in office”, it is implied, is not so much necessary as it is tolerated, a curiosity or historical relic of a bygone era. Like a Homeric bard in the age of cinema, or an unassisted author in the age of Sudowrite, a Greek god who remains at his post is a kind of stubborn anachronism, an object of pity and not of reverence.

Why does he remain? Kafka’s identification with Poseidon, and his commitment to a poetics of word-processing, is never stronger than when he relates that Poseidon’s refusal to delegate his computation to assistants is a result of the fact that “he takes his office [Amt] very seriously”. The implication is that what binds Poseidon to his position is not divine right or economic necessity, but instead what Kafka elsewhere refers to as an “oath of service” [Diensteid], a lapsed covenant that is now enforced only by the subject’s own conscience. In his and his protagonists’ stubborn adherence to such “oaths” of office, Kafka provides not so much a defense of the writer’s vocation as an image of those for whom it remains a subjective necessity—even (or particularly) in the age of its superannuation. The short parable from which the reference to the writer’s “oath” derives, written just as Kafka was beginning his engagement with classical myth, relates a world in which there are no “kings”, only “kings’ couriers”; as a result, the couriers “race through the world shouting to each other messages that, since there are no kings, have become meaningless”. Although they would gladly put an end to their miserable existence, they do not dare, due to their “oath of service” (Kafka 1954, p. 43).21 The affinities to Kafka’s “Poseidon” are apparent: both depict the modern author as a kind of “word-processor”, the holder of an office which has grown obsolescent through societal transformation. Similarly, Poseidon and couriers alike continue to faithfully perform their duties, even though these, seen objectively, seem to amount to little more than a kind of frenzied, algorithmic, even Kafkaesque activity. There is, however, a transformation that occurs between the two texts, one that suggests a kind of reconciliation with the writer’s fate through the embrace of process: In “Poseidon”, Kafka’s double is both monarch and messenger, meaning that unlike the couriers, his actions have an addressee—even if it is only himself. What is more, he seems to have found a kind of pleasure in his position, however muted; the end of the world appears in the later text not as a release from suffering, but instead as the opportunity for a brief vacation, after Poseidon has made his final revisions and laid down his pen. This, perhaps, is Kafka’s countermyth, his challenge to the image of the modern author as alienated laborer or savvy curator of “ideas”: Alone at his desk at the bottom of the sea, isolated from the outside world, Poseidon dutifully carries out his paperwork. One must imagine Poseidon happy.