Artificial Intelligence and the Critical Distance of Literature

The poets are getting restless. The novelists are biting their nails. The essayists can be found pacing the floors, sick with worry. They are talking to themselves, mumbling under their breath. If we take a step closer, we can hear it, the question on their lips, the one that keeps them up at night: Can artificial intelligence write literature? Are we witnessing the end of writing?

The funeral bells are ringing. Open your window and you will hear the alarms being sounded. The scribes are busy preparing the death certificates. The town criers, in the public square, are going hoarse, proclaiming the end of this, the end of that.

In the past few decades, we have seen it all, so many false endings. The end of history, ideology, grand narratives. The death of the public intellectual. Yet, somehow, before the tears have a chance to dry, the defunct get summarily resurrected. They appear once again on the world stage, rearing their heads, baring their teeth, seemingly strengthened by the attempts to repress them.

Francis Fukuyama watched the Soviet Union collapse and heralded the advent of a new golden age, the definitive triumph of Western liberal democracy and market capitalism, as the Crown Jewels of human civilization, the final forms of political and economic organization. The final resolution of history, he argued in The End of History and the Last Man (1992), had arrived. A few years later, we would all watch as The Cold War gave way to hotter forms of warfare, as the Washington Consensus inaugurated a predatory form of neoliberal economics, and as liberalism started backsliding yet again towards fascism. By the time the twin towers fell, history had been vigorously reinstated.

Likewise, Jean-François Lyotard, in The Postmodern Condition, proclaimed the death of the grand narratives of modernity, totalizing interpretative frameworks capable of explaining our place in history and society, inciting us to do the craziest of things, like working towards emancipation. In their place, we were to expect a post-political landscape marked by hearty skepticism and an incredulity towards metanarratives. The pernicious political passions of yesteryear had subsided. We had learned our lesson. The twentieth century saw to that. Having returned to our senses, we could expect smooth sailing from here on out. It didn’t take long, however, for the waters to get choppy again. As the twentieth century drew to a close, religious fundamentalism spread like wildfire throughout the West and East. It was coupled with a new expression of extremism, a particularly postmodern form of irrationality: market fundamentalism. No, the game had not come to an end. Increasingly, it seemed as if we were back to square one.

In 1987, Russell Jacoby published his influential book, The Last Intellectuals, an elegy to the passing of the once-revered figure of the public intellectual. In this classic of “declinist” discourse, Jacoby mourns the replacement of the well-rounded generalist, in possession of an encyclopedic knowledge by the hyper-specialized academic hack. As the humanities were dethroned from their seat at the center of university education, replaced by STEM fields, the scholar gave way to the technician. This shift coincides with the profound transformation of the university from a Modern institution founded on the principles of the Enlightenment to a technocratic institution, in which knowledge production is valued in terms of its utility to industry and commerce. A commitment to intellectual freedom, the disinterested pursuit of knowledge, education as a means upon itself, the cultivation of human potential, autonomy from other spheres of society –these founding principles of universities fell to the wayside as teaching and research were increasingly subjected to utilitarian criterion. The public intellectual of yesteryear, not having a place in the emerging world of hyper-specialization and techno-idolatry, was ushered off stage, displaced by the hacks of academia and the buffoons of the entertainment industry.

To be sure, many of the transformations described by Jacoby did indeed take place. In fact, they have run their course and reached the next stage in their development. We have witnessed the university change from a technocratic institution to a thoroughly neoliberal one. If the technocratic university was obligated to prove its utility to business, the neoliberal university became a business unto itself, ruled by the profit motive, a corporation among others. In its present iteration, each department and indeed each individual professor (and each graduate student for that matter) is expected to attract funds to the university, to generate profit. If you do not comply with this injunction, the university turns its back on you, revokes your funds, and allows your project to sink beneath the waves. Department after department in the humanities have been shut down for this very reason. We are witnessing the onset of Academic Darwinism, where fitness is measured in terms of profitability.

This shift has profound consequences not only for education but for the whole of society. It means that the very function of universities has changed. It is no longer the university, as an educational institution, that ensures the conditions for the production of knowledge in this or that field but the individual fields that create the conditions for the profit of the university as an economic enterprise. As if by sleight of hand, the pursuit of knowledge has been switched for the pursuit of profit. The social function of university, historically, has been tied to open-ended inquiry, the untethered pursuit of knowledge, the formation of informed citizenry, and the creation of a populace educated enough to participate responsibly in a democracy. These ideals, consecrated in the charters of nearly all universities, no longer characterized the overall social function of universities nor inform how they are currently organized and administered. And, if universities conduct themselves as businesses, as they increasingly do, it is unclear, then, what makes them different from other institutions. The university, it would seem, has lost its specificity.

The fact of the matter remains, however, that the extinction of the Modern University, the enlightenment principles of its founding long since thrown out the window, did not spell the death of the intellectual. If the hack stayed aboard, the public intellectual did not go down with the ship. As today’s universities become centers of mediocrity, intellectually sterile, as they increasingly are, intellectual life is migrating elsewhere. These are the conditions of the birth of the influencer, of independent researcher, the DIY scholar, who are nothing other than a 21^st century version of the public intellectual of the 18th and 19th century. It is not that history came to an end but, rather, that we have come full circle.

So, after so many false alarms, we might ask ourselves: is it really the end? Is literature over? How are we to know?

What we do know is that AI is a generative tool that, on the level of syntax, can generate a potentially infinite number of enunciations through the combination of discrete linguistic units and the application of a set of rules. On the level of semantics, the content of these enunciations is determined by input, that is, by the processing of what has previously been enunciated on a given topic. Both the rules of combination and the content of the new enunciations produced is determined by what has already been said. In the search of a mean, a form that is representative of the sum of previous enunciations, exceptional uses of language are necessarily factored out. In this sense, AI can be said to flatten language. Nonetheless, the fact remains that a considerable portion of the previously enunciations used to train AI have already been flattened.

A frequent critique levelled at writing produced by artificial intelligence is that it sounds generic, bland, stiff, something you would be more likely to hear in a corporation board room than an arthouse café. Yet, the same could be said of much of the human writing that is used to train AI. Are we right, then, to blame AI when the form and content of its output is determined by vast quantities of our own texts? As such, there is another question lurking behind the issue of whether machines can write literature: Can we ourselves still write literature? Can we write in a way that doesn’t sound like a corporate HR training session? When, how, and why did our writing become so bland, so automatized?

The market is, of course, one force of the homogenization of literature. What is sellable in the market is determined by what has already sold well. In an effort of minimize risks and maximize profits, publishing houses opt for the safe bet, which is to say for that which has already been proved profitable. This feature of the publishing industry encourages a seemingly endless repetition of the same, a certain zeal for the pre-determined that makes it difficult to see over the wall, much less likely to climb over it and break into new territory.

Take, for example, current conventions concerning the extension of novels. According to the guidelines for the submission of proposals, there is general agreement among publishers that somewhere between 70,000 and 80,000 is the “sweet spot.” While shorter novels are often accepted, it is much more difficult to get a longer novel published. Write more than 100,000 words and you can expect to have the door shut in your face. The rationale? No one, we are told, wants to read a long novel. Who has the time? Think of your readers.

The same goes for syntax. Whatever you do, don’t write sentences with more than fifteen words, and don’t use more than two subordinate clauses per sentence. The cognitive demand is too great. It’s simply too much to ask of readers. Clarity above all else! Keep description to a minimum, shorten scenes, speed the plot along, make the roles of the characters clear, and, above all, create the desirable effect, that is, narrative closure. If writers are to stay in the game, if they are to keep their heads above water, they are obliged to adhere to these implicit rules. The only other option is to self-publish... and start looking for a day job.  

There are historical reasons why this came to be the case, why this particular set of narrative conventions has come to be preferred over others.

We are still on the outer reaches of a long-standing reaction against literary modernism. Throughout much of the 19^th century, the realist novel was the hegemonic literature form. This genre, a compliment to positivism in philosophy, presupposes the transparency of language and its capacity to unproblematically apprehend social and historical realities. To borrow an analogy from José Ortega y Gasset, realism approaches language as a transparent windowpane to focus on what lies beyond, a garden perhaps, in all its myriad details, the tire swing hanging from the tree branch, a wicker chair in the grass, the gnarled roots of the tree pushing through the soil, right down to a single blade of grass. Modernism, then, came along to dirty the windowpane. It shifted attention away from the garden to the sheet of glass itself, to language as an artistic medium, to the problems of representation. This implied a metaliterary turn. The literary work folded in upon itself to inspect its own mechanisms. If realism flattened language, minimizing its nature as artifice, emphasizing the referent over the medium, modernism fattened it, slowing it down on its way into the body.

This shift had implications on the level of syntax. Liberated from the demand to realistically apprehend an extralinguistic referent, the literary work was free to explore the limits of language. The formal experimentation that ensued put a considerable cognitive demand on readers. The familiar signposts of plot development were often jettisoned, the chronological order of events was fragmented, the unity of perspective was shattered, narrative closure was frequently eschewed, and it was not uncommon for value judgments to be withheld, all of which left readers adrift amid vast swells of language and the free play of literary devices. Some authors, such as Jorge Luis Borges, made a point of antagonizing their readers, equivocating when not outright falsifying their intentions, placing demands on their reader that would have previously seemed unreasonable.

In the interwar period, the historical avant-garde took the exploration of the limits of language to new levels, to the point of dissolution, unintelligibility, the absurd. These movements mobilized the destructive side of the literary work to attack art as an institution and to forge a new relationship between aesthetics and everyday life. The epic confrontation between the avant-garde and the world of art did not end in cataclysm but fizzled out rather undramatically as critiques against the institutions of art were themselves institutionalized. The ready-mades did not bring down the walls of museum but were simply allotted their own special place within those same walls. Unfortunately, the story of the historical avant-garde is a truncated one: these movements of radical rupture did not survive the conflagration of the Second World War.

The postwar period saw the advent of Pop Art and a renegotiation of the relationship between high art and popular culture. The postmodern novel grew out of these developments. While it conserved certain features of literary modernism, such as the fragmentation of the chronological order of the plot and the multiplication of narrative points of view, it also borrowed elements from the culture industry, most notably a dramatically reduction in the cognitive demand placed on readers, now recast as consumers. Syntax was correspondingly simplified. Sentences were shortened. Clarity and accessibility became the new criteria of the acceptability and indeed marketability of the literary work.

Although literature today has new demands on its content, on the type of stories that are worthy of being told, we have inherited the fragmented structure of the modernist novel, and the criterion of clarity of the postmodern novel, as well as the low cognitive demand its places upon readers. These features are no longer perceived as polemic but simply the way that literature is, a new status quo. As if it were a natural inclination and not a cultural perspective, as if it were universal and not particular, something outside of history, we value texts that conform to this norm. Those that do not are ignored, overlooked, when not openly disparaged and rejected. 

The language of today’s novel mirrors that of other sectors of society. Whether news, entertainment, or education, we prefer a prosaic language, transactional, commonsensical, informal, and above all undemanding. Pick up any book from the New York Times best-sellers list, and you will find the same language used in articles from the Huffington Post or NPR. Turn on the news, scroll through social media posts, read your favorite blog, chances are that you will find it there too. Unsurprisingly, it is also the idiom of corporations and advertising, sectors whose profit margins depend on anticipating your linguistic needs. It’s in the tone in the safety instructions video of your next flight. It’s in my blog and most of the blogs I read. It’s all around us, this sameness, the irreflexive repetition of a discursive convention.

What, then, is the specificity of literary language? What distinguishes it from other types of texts? If literary discourse is conterminous with other neighboring uses of language, then what sets it apart? How do we even know when we are in the presence of literature? Does such a thing even exist?

Famously, Terry Eagleton addressed this problem in the conclusion to his Literary Theory: An Introduction (1983). Unable to identify any positive quality that distinguishes literature from other texts, he settled upon an institutional definition. Literature is merely what the guardians of literature decide it is. The inner sanctum, of course, is academia, and the role of guardianship is assigned to the priests of the temple, that is, the literary critics and theorists who haunt the lecture halls of the ivory towers.

Terry Eagleton, however, had been unable to foresee what his nemesis, Harold Bloom, had predicted with such clarity in The Western Canon (1994): the dethroning of the humanities, the radical reorganization of the university along different ideological lines, and the transformation of the university into an entirely different type of institution. Eagleton, in other words, overestimated academia and the power of its guardians, for the fact of the matter is that nobody much cares these days what the academic hacks of the neoliberal university say. The value of the work of literature is determined elsewhere. In the culture industry. The market. Online perhaps.

Aside from the fate of the institutions in question, Eagleton’s definition leaves a lot to be desired. Surely, the use of language by a minstrel reciting epic poems and lyrical ballads in the High Middles Ages is not the same as the inventory of a stockpile of weapons in the same period. The organizing principles of the text of my last speeding ticket were surely not the same as those of the hip-hop lyrics that I was listening to when I received the ticket. So, what is it then, the difference in these approaches to language? There must be something?

Well before Eagleton’s times, with the Russian Revolution as a background to their intellectual endeavors, the Russian Formalists devised a theory to address this question. They started with a distinction between poetic and prosaic or practical language. Our everyday lives are governed by prosaic uses of language. We presuppose linguistic devices in order to accomplish our pragmatic goals. This utilitarian approach to language seeks to communicate meaning in the most efficient way possible. The linguistic medium of representation should be as transparent as possible, taken as a given, in order to facilitate the unproblematic coding and decoding of messages. In our transactions, we necessarily forget about language as a code, as a structure, as a medium of representation, as artifice, and mobilize linguistic devices irreflexively. The Russian Formalists refer to this feature in terms of automatism. Our habitual uses of language and indeed our perceptions of the world are automatic. We do not generally reflect upon how language organizes our experiences and constructs our perceptions of the world any more than we calculate the precise angle and force necessary to lift a teacup from the tabletop to our lips.

Source: Henri's Walk To Paris (1962), written by Leonore Klein and illustrated by Saul Bass

It is against the backdrop of prosaic language and its corresponding automatism that the Russian Formalists understand poetic language. Its specificity, they contend, consists in what they call “ostranenie” or defamiliarization. Poetic language has the capacity of defamiliarizing our automatized perceptions. The literariness of a work of literature consists in its ability “to roughen the surfaces of language” and make us aware of language as language, as form, as artifice, as Viktor Schklovsky argues in “Art as Device” (1917). But that is not all. The defamiliarizing effect of poetic language helps us distance ourselves from our own presuppositions and thus comprehend the world in new ways.

This account of literariness has the virtue of being based on positive traits and not simply on what literature lacks, as is the case of Eagleton’s institutional definition. In addition, it is particularly relevant to the discussion of artificial intelligence. What could possibly characterize the machine writing better than the category of automatism? AI generates enunciations automatically by matching patterns, calculating mathematical probabilities, and combining discrete textual categories according to a pre-determined set of principles and parameters. It is not capable, however, of a poetic language in the way that the Russian Formalists understood it for the simple reason that it has no meta-literary awareness. Artificial Intelligence generates new combinations of language but has no self-awareness or true understanding of what it says. As such, defamiliarization as an awareness of language raised to a second degree is simply out of its scope. Machine writing is prosaic by design, trained to recognize and reproduce the most efficient uses of language. In other words, its job is to smooth the edges of language rather than roughening them up, as only poetic language knows how to do.

There have been other attempts to identify the specificity of literary discourse that can help clarify the discussion of writing generated by artificial intelligence. In his work as an intellectual historian and cultural theorist, Raymond Williams, in the period from the 1950s to the 1980s, wrestled with one of the problems that Structuralism had been unable to answer, namely the question of how literature changes throughout time. If literature is conceived of as a closed system, composed of constitutive units, each with its own function, then theory can prove a static synchronic view of the various interactions between all the composite parts, but it cannot address how the balance is disrupted and how the system is reorganized into a subsequent iteration. To address these concerns, he proposed, in Marxism and Literature (1977), to analyze literature, and indeed cultural production in general, not so much as a static system than as a process of dynamic interrelations between residual, emergent, and pre-emergent elements. Literature, according to Williams, operates as a selective tradition, in which some residual aspects, formed in the past but are still active in the present, interact with emergent meanings and values to form the dominant or hegemonic cultural forms of a given epoch. Inherited traditions, interpreted and re-signified in light of the demands of the present, interact with emerging practices to form the sensibilities of dominant culture, a common sense of sorts. This is not, however, the whole picture.

For culture to develop throughout time, for one dominant cultural formation to be displaced by another, it is necessary to transgress existing norms and conventions, that is, to move beyond the dominant forms. This is possible because no dominant culture encompasses all culture forms, all modes of interpretation, and all frameworks of intelligibility. There will always be artistic and intellectual practices that fall outside its purview. These zones of unintelligibility, the blind spots of dominant culture, offer the possibility of seeing over the wall, so to speak, of forging new sensibilities, and thereby pushing culture in new directions. Literature, in particular, has the possibility of articulating what Williams denominates pre-emergent discourse. This means that some literary works have the ability of anticipating an emerging sensibility before it has been fully articulated. These works articulate an image as it is coming into focus, before it has been properly named. Some of these new meanings and values fall to the wayside, while other become emergent discourses that interact with residual elements to conform new dominant cultural forms.

The point is that culture is more than the eternal repetition of the same norms and conventions. There must be some act of creation that moves beyond existing forms and introduces sensibilities, meanings, and values that were not previously contemplated within the overall system. This account, to be sure, is indebted to the Romantic conception of the artist, who doesn’t merely follow pre-existing rules and forges new ones to meet the supernumerary demands of the creative process. The whole argument rests on a distinction between the generation of new enunciations, new content, through the repetition of pre-existing norms and conventions and the creation of new forms whose construction exceeds the possibilities of those same norms and conventions.

Artificial intelligence, to be sure, can accomplish the former, but it cannot achieve the latter. Machine writing has proven singularly efficient in producing familiar and highly readable texts through an automatized reproduction of pre-existing norms, fossilized tropes, time-tested devices, and lifeless formulas. It is capable of an endless repetition of the same, as we ourselves are. But can it situate itself outside existing frameworks of intelligibility? Can it anticipate sensibilities, values, and meanings that exceed the current cultural codes? Can it approach language in a defamiliarize manner? Literature, I would argue, is precisely such a meta-discursive awareness of language. As such, it resides in the critical distance, the gap between what we already know and what we have yet to discover, between a cultural world that is dying and one that is coming into being. It starts in the margins, the blind spots, the zones of unintelligibility and, from there, propels us towards a future different from the present and past. Artificial intelligence, to date, is not capable of writing this type of literature. It remains to be seen whether we ourselves are still capable of doing so and whether such an endeavor is even something that still interests us.

Disclaimer & Note to my South American readers:

My characterization of the current state of university education pertains primarily to North American and European models and does not contemplate important exceptions, most notably the case of Argentina, singular on the world stage, exemplar in so many ways, a feature that makes it both the target of current anti-intellectual neo-fascist political movements and a crucial site of resistance. The same holds true for my generalization about the state of literature. A hotbed of literary production and seemingly untameable by the market, Latin American literature still retains its specificity from other modes of discourse. It seems to follow its own course, with the market running behind, always eager to catch up, not the other way around.

Source: All in One Day (1955), written and illustrated by Lucy Ozone.