This illimitable sentence-producing capacity

I had been wanting to read Fredric Jameson’s Fables of Aggression (1977) for a while, but as enticing and timely as the subtitle sounded to me — “the Modernist as Fascist” — I could never talk myself into plunging into a deep study of the works of misogynist Nazi sympathizer Wyndham Lewis, whose books I knew nothing about. But after reading several tributes to Jameson in the wake of his death that, when highlighting notable titles from his vast body of work, made a point of mentioning Fables of Aggression, I figured the time was as good as any.

Jameson describes the book as a “practical exploration” of Althusser’s definition of ideology as a “representation of the Imaginary relationship of individuals to their real conditions of existence.”

Two features of this definition need to be retained: first, that ideology must always be necessarily narrative in its structure, inasmuch as it not only involves a mapping of the real, but also the essentially narrative or fantasy attempt of the subject to invent a place for himself/herself in a collective and historical process which excludes him or her and which is itself basically nonrepresentable and nonnarrative. This is the sense of Althusser’s otherwise scandalous description of History as a “process without a subject or a telos,” and leads us to the second important implication of the definition in question. For the “Real” on this view is conceived, neither as an unknowable thing in itself, nor as a string of events or set of facts you can know directly in the form of some “true” or “adequate” representation for consciousness. It is rather an asymptotic phenomenon, an outer limit, which the subject approaches in the anxiety of the moment of truth —moments of personal crisis and of the agonizing political polarization of revolutionary situations; and from such an approach to the Real the subject then tends to retreat again, at best in possession of abstract or purely intellectual schemata when not of personally charged narrative representations.  

This seems like a useful way to understand people claiming that the government controls the weather and so on. It is not a question of “false consciousness” but of what happens to people when confronted with “history’s resistance to the fantasy-structures” they are trapped in and which circumscribe their understanding of meaningful action. (There is more on Jameson’s ideas about conspiracy theorizing here.)

Maybe I am totally ideologically poisoned in a different way, but I was surprised to discover that Fables of Aggression, in its concern with “dead language,” offers some interesting ways to think about contemporary generative models.  

Until I read his obituaries, I didn’t know Jameson studied under Erich Auerbach, but that is unmistakable in the Wyndham Lewis book. Devoting entire chapters to close readings of a few short passages of Lewis’s, Jameson applies an adapted version of Auerbach’s methodology from Mimesis (which traces the history of what was understood as representable in text and how) to investigate the political determinations and implications of the techniques for representing reality that were available to modernist writers (i.e. Jameson’s idea of the “political unconscious”).

With no familiarity whatsoever with Lewis, I figured I might be lost in the details of various close readings, but Jameson doesn’t focus on the content of Lewis’s books much at all; he instead dissects their style by taking apart a few characteristic sentences. The overall point is to “historicize the gap between style and narrative” — historical significance inheres at the level of style and can be traced there almost independently from the ideas expressed as content. “We will therefore read Lewis’s sentence-production as a symbolic act in its own right, an explosive and window-braking praxis on the level of the words themselves,” Jameson announces in the prologue. And in Jameson’s analysis, the significance of Lewis’s style is that it reveals with undiminished potency the urgency of modernism’s response to what was perceived as a crisis in representation, an exhaustion of the vitality of language at the culture-wide level.

That crisis stems, if you accept Lukács’s theory, from the forces of reification unleashed by capitalism, a crisis that has only continued to worsen somehow, despite the best efforts of artists and writers to forestall its progress. Jameson’s description of Nathalie Sarraute’s novels captures what reification means at the level of language:

It is as though the old common language of everyday life had ceased to be an adequate vehicle for individual expression, let alone communication. Brittle with cliché, great surfaces of it corroded by publicity and received ideas, that alienated and conventionalized language begins to break apart, leaving deserts of silence visible between the cracks. Here genuine human life continues to exist, but as it were underground, beneath the dead surface of social routine and convention. 

I think of generative models as automating capitalist reification at scale, eliminating the subjective component of experience and presenting as something straightforwardly quantifiable and calculable. Generative models are brute-force attempts to address crises of representation by purporting to solve them statistically. They dispense with any problems brought on by subjectivity or collectivity or intentionality by eliminating them — something that modernists like Lewis (in Jameson’s argument) assumed would be felt as a critique and not a confirmation of capitalist reification. But it turns out that decentering the subject and treating language as a self-perpetuating machine can be presented as mundane productivity hacks. As the world fills with automated text circulated in place of human interaction, the deserts of silence will become vaster and more arid.

If the diagnosis is reification, than “the role of the novelist,” Jameson points out, “becomes the recuperation of that more authentic reality” — the one behind the dead walls of automated words — “and the invention of a new and fresh, nonalienated originary language in which the latter’s preverbal or nonverbal events and incidents can somehow, beyond all fallen speech, be adequately restored.” In other words, novelists should make language work in ways that originate rather than predict or conform to probabilities. But that task is racing against a technology intent on making it seem impossible, that can assimilate and reify and alienate all forms of language before they have a chance to reinvigorate discourse and its ability to represent reality as something thought and lived.

(My least favorite Jamesonian tic is that opaque use of the word “latter” when a repetition of the referent would be perfectly fine — perhaps it is part of how his style deliberately frustrates a smooth and speedy reading pace, but it is also very annoying.)

Lewis’s approach to sentence making is offered as a flailing attempt at reinvigoration, or at least a

willful and desperate, impossible solution: the disintegration of the literary exchange as a socially guaranteed and institutionalized compact between writer and reader. the disappearance of a shared code, and the problematization of even those proairetic linguistic unities or naming systems which allow us to feel that we are discussing the same shared universe; the anomie and fragmentation of social life itself, which systematically dissolves the older classes and the homogeneous publics that developed out of them, at the same time that it isolates the now equivalent monads of the market system so effectively from one another that the writing monad can entertain no realistic hope of awakening analogous private experience or personal reference in the equally monadic reader.

But that is just what LLMs do. Many of Jameson’s descriptions of Lewis’s style could be taken for descriptions of ChatGPT. “To face the sentences of Wyndham Lewis is to find oneself confronted with a principle of immense mechanical energy.” It sometimes manifests as “mindless babble,” or as “‘tireless’ pages in frightening quantities,” making an “ambiguous monument” to an “illimitable sentence-producing capacity.” But a capacity that was once “a figure for human productive power in the industrial age” appears as a cancelation of that power in the late information age. Where “the sentence-producing mechanism” could once figure “the random fission of the metonymic impulse,” now it further testifies to the deep grooves laid in culture for the repetition of the same phrases and cliches, the same patterns that speak to the calcified “reality” of existing social relations and power distributions.

If once a writer like Lewis could make set phrases and dead words appear as “an amalgam of heterogeneous forces which must not be allowed to congeal,” generative AI demonstrates how they can be irresistibly forced to congeal — any sentence, no matter how dissonant and stylistically obtuse, can be summarized by an AI engine and rendered inert for the audience that accepts such summaries as part of their fantasy structure for dealing with the Real.

According to Jameson, Lewis’s writing works to compel us to “jettison the mirage-ideal of intersubjective communication altogether” to reawaken the reader’s capability to move in language at the level of language.

The text does not reproduce the process, or the data, of Lewis’s own inner vision and demand that we raise ourselves to its imaginative level: rather it merely proposes a certain number of mental operations, of which the final step is the breath of life, the infusion of private experience or personal knowledge which can have no possible correspondence or relationship of adequation with Lewis’s own, always assuming that he had such an original vision, or experience, or imaginative starting point, to begin with. The reader’s reconstruction of that “starting point” is quite properly unverifiable and undecidable and serves merely as a sign that the operations have been performed and the reading completed, rather than as any approximation to Lewis’s “original intent.’’ Yet the absence of this authorial referent is itself masked in turn by the immense energies released by our own demanding practice as readers: whence the exhilaration that unexpectedly takes the place of the more predictable pathos that such an experience of the breakdown of social language ought logically to inspire. 

When LLMs still made flarf, one could almost make this case for the machines themselves. They have no intention, so they exhilarate readers who get to project their own onto it. “For it is the reader who is ultimately called upon to reinvent the external form itself, to mint the vast, slow curve of which the sentences are broken traces and fragments, to restore to the vivid microlife of the novelist’s language that absent whole of which its moments are the parts.”

ChatGPT, “upgraded” with the rote reinforcement provided by the reactions of underpaid, checked-out workers, is designed to prevent that from being necessary. Instead of experiencing a breakdown of social language, the LLM serves it up in its most reified and alienated form, compelled from workers who could care less but have been paid a wage to pretend to make sense of a moment of attempted communication at its most basic level.