We begin by describing a very different technolinguistic situation than our own to take up the questions posed by Matthew Kirschenbaum and Rita Raley throughout their essay about the crisis AI provokes, signifying the end of one set of approaches to language and potentially opening up another. Kirschenbaum and Raley acknowledge that AI opens a rupture that challenges the humanist foundations of our profession: “the leading sciences and technologies are no longer operated and mediated by language as such,” and that large language models (LLMs) betoken the advent of a new “general condition of language and life.” Kirschenbaum and Raley suggest a range of potential models for thinking about the affective impact of AI, models that they warn will quickly become “iterated, localized, and branded” before being sold back to us as tools ready to be integrated into the structures of corporate campus life. If we are to grapple with the effect of AI on the profession, the time is now. By that same token, the disruption to “language and life” arriving with these developments assumes an all-too-familiar form: to wit, this very demand for timeliness, for an adaptation to a constricted temporal horizon characteristic of the “organizational strategy” of the corporate university and of capitalism itself.

It may seem counterintuitive, then, to turn to late-sixteenth-century scenes of reading to think through the affective valences of AI; however, if AI represents a rupture to humanist models of learning, it is worth returning briefly to their origins, including how the skills of reading and writing became linked in a different moment of textual crisis, sparked by the technology of print and its radical transformation of scenes of writing. In their influential work on the history of reading, Lisa Jardine and Anthony Grafton argue that the mechanization of print engendered a deluge of text that provoked a crisis of attention: How were readers to manage an exponentially larger scale of information and transform it into “purposeful reading” (32)? The sheer volume of printed books made such efficiency challenging, requiring new ways of engaging with text and new techniques for managing print, including the use of heavy machinery like a book wheel (figs. 1 and 2). Book wheels were revolving desks that allowed readers to consult or handle multiple books simultaneously and quickly; though most were simple designs that held one or two volumes and rotated, more elaborate configurations imagine readers engaging with upwards of sixty to seventy large books, or more if they were smaller (Jardine and Grafton 46). Agostino Ramelli's version was designed for readers with limited mobility (Ramelli 316).

Fig. 1. Agostino Ramelli, Le diverse et artificiose machine, Paris, 1588, p. 317 [sig. Ee]. Image courtesy of the Folger Shakespeare Library.

Fig. 2. Bodleian Libraries, University of Oxford, MS. Canon. Class. Lat. 257, fol. 3r (CC-BY-NC 4.0).

Jardine and Grafton point out that such methods for reading are both “flexible” and programmatic (73). Readers were to actively reinterpret the books they read—annotating, transcribing, imitating, and transforming text into something new (Chartier 95). Reading books in this way was not necessarily designed for comprehension; it was designed for productivity and future-oriented usefulness (Jardine and Grafton 73). Book wheels reflect a desire to speed up what was already imagined to be an action-oriented approach to text: reading was designed “to give rise to something else” (Jardine and Grafton 30).

It is their description of humanist reading practices as giving rise to “something else” that we wish to consider in dialogue with Kirschenbaum and Raley's configuration of our current situation. AI represents a culmination of this desire to be ever more efficient in managing information. For those of us who are lucky enough to (still) have jobs working in higher education, especially in humanities departments, AI has transformed that sense of purpose-driven activity: What is the value of our disciplinary skills, particularly the skills of reading, writing, and interpreting, in this particular moment when the “symbolic ground, made up of human-constructed sign systems” seems to be eroding beneath us? This ground's “a priori” presumption, as adduced by Kirschenbaum and Raley, has, of course, a history, and this history leads back to the “something else” that Jardine and Grafton describe: an emphasis on writing as a humanist metric of learning. As many of us who teach premodern materials emphasize to our students (Parker and Silva), reading in the past involved writing; the multitude of commonplace books in archives document that history, as readers in the past recorded salient passages from some books in other books and reorganized them to prepare for future-oriented engagement.

Of course, things have changed: “The advent of large language models has radically transformed this technolinguistic situation, full stop.” Kirschenbaum and Raley invoke our own technolinguistic scene of writing to create a space for processing this new relationship with text, drawing attention not just to the model of word-processing most of us rely on to create text for human readers but also to what LLMs and natural language processing (NLP) do in terms of generating text designed to be read by machines. It is those “stops”—including the spacing between characters in words, between words in sentences, and between sentences in paragraphs—that encapsulate the computational and material dynamics of language both in early modern techniques of print and in AI systems.

The contributors to this forum have provided a range of models for how we might begin to connect the computational insights of language that AI provides to our professional models of interpretation. And, as Katherine Elkins emphasizes, we will need to theorize and adopt these models in real time; AI is already prompting a plethora of tools marketed to faculty and staff members and to students to manage the deluge of text that is shaping this new “episteme” (as Aarthi Vadde calls it), all of which come at a very high cost to the environment. Like the book wheel, whose solution to the mechanization of reading appears (from our vantage point) too literal, these tools will likely be inadequate, harnessing a similar desire to remain efficient when confronted with an exponentially increasing volume of text. It is likely that very few of these book wheels existed (Cambers 78; Considine 390). Rather, book wheels represent a humanist fantasy: “to master the whole world of learning and make it readily usable in political action” (Jardine and Grafton 75).

Admittedly, Ramelli's book wheel seems quaint, especially when viewed in terms of the amount of text AI is capable of generating. But it documents how readers of the past grappled with their own “textpocalypse” (Kirschenbaum), managing “the flood of information that the presses poured over them” (Jardine and Grafton 77). Its circular fantasy—of a machine created to manage a crisis of text produced by other machines—isolates the problem we face now. The humanist fantasy of reading, in its long passage from the Cartesian insistence on the “clear” and “distinct” to the Weberian-positivist emphasis on rationality as a set of rules made formal and explicit, is—and has been—entangled with modes of production and consumption that provide readers with more and more text. Read in this way, the image of a Renaissance reader (white, male, privileged, and rich) alone in his private study with his absurd book wheel prefigures a much longer history of industrialization: the reader is not unlike the spinners at mechanical spindles or the weavers at water- and later steam-powered looms, embodying a long history of labor driven ever deeper into domination by the efficiency of machines. But it also speaks to the radical shift that AI has created and that Kirschenbaum and Raley are asking us to consider: What does it mean that “language itself” now forms, in a sense, not only the interface to but also the inner workings of the machine?

At the very least, word processing now entails a very different scene of writing, including the rapid generation of text based on probabilities of linguistic patterns, derived “from training data composed of sequences of tokens that have been converted into numerical representations.” It is not yet clear what this shift will entail. On the one hand, AI initiates a set of desires about the role of text as a tool of efficiency, including (potentially) a renewed investment in the power of text. Marketing about the power of NLP to create “autosummary on dialogic demand” promises to help us navigate the new textual wastelands, offering “productivity schema laid across the full spectrum of the postindustrial knowledge economy.” But on the other hand, AI also hints at speculative possibilities—expanded consciousness (Watts), communicating with animals (Wong), and, perhaps more banally but also most relevant to readers of this journal, the death of the college essay (Marche), at least in terms of its requirement for “robotic” rehearsals of knowledge (Gray).

Whether or not the college essay is doomed, its seemingly imminent and ignominious demise at the hands of a chatbot inspires fears that knowledge itself, at least as practiced in the academy, can “no longer [be] operated and mediated by language as such” (according to Kirschenbaum and Raley). There is no doubt that the ground of our profession has shifted. To navigate its new contours requires that we let go of what a humanist-driven focus on language has bequeathed to us and embrace instead a reconfigured and vastly expanded understanding of what language entails. It is tempting to lean into concepts of affect or aesthetics, defending (once more) “the thing itself, language” (as Elkins writes). But these models generate thin responses that sound good at first but upon scrutiny sound like a sort of fustian drawn from the archives of previous crises. The question of the value of our professional training in the face of AI is not immaterial: tuition is quantifiable (and expensive). Our collective commitment as a profession to the new contours of language, knowledge, and text must grapple with the fact that even as we seek to include a wider array of interpretive and performative tactics in our discipline, our tools are, by the rubrics dominant in the neoliberal academy, inefficient.

For it seems beyond dispute that we are on the verge of new forms of enclosure: “an enclosure of the language commons,” as Kirschenbaum and Raley note. Like previous iterations, this enclosure functions through its propensity to increase the efficiency of labor toward the ends of profit by curtailing our collective resort to any resources not under the domination of capital. Reading, listening, thinking, and writing take time—mostly unprofitable time. But the fantasy embedded in the promotional video for Blackbox.AI cited by Kirschenbaum and Raley is not unlike the fantasy embedded in the image of the book wheel: the solution to the crisis of AI is more AI, just as the solution to the crisis of print was more print. Both foster a fantasy of production linked to increased consumption of text. And no doubt this form of enclosure, like its predecessors, will prove capable of absorbing externalities and pressing them into profits.

Kirschenbaum and Raley's insistence that AI is not the prompt but the punctuation is an important call to action, in terms of both wider posthuman engagements with language in the future and, just as important, a wider posthumanist engagement with the textual record of the past. Like Kirschenbaum and Raley, we are not yet sure what the outcomes of AI on “literary and domain-specific historical research” may be. But the conversation represented in these essays reinforced our shared belief that it was worthwhile to explore possibilities; we partnered in teaching a graduate English class this spring, developing a module on machine reading for a course on the history of reading. Instead of reinforcing either early modern or postmodern fantasies about productivity, we challenged ourselves to explore inefficiencies as an integral part of previous histories of literacy; this allowed us to connect a computational approach to language with these other models. The goal was to create space for ourselves and our students to explore what insights we might develop through a range of approaches to language itself—that is, without a demand to package these insights according to the rubrics of efficient and marketable use. Framed in this way, reading is less a programmable set of actions designed to master ever-increasing amounts of information through extraction and more a kind of game or experiment (as Junting Huang notes). But this kind of framing, too, takes time—both the hours of faculty instruction in the classroom and the hours of student engagement outside it.

Conditioned as we are to respond to demands for a timeliness that subordinates the thickness of the historical present to a vanishing margin of future value, it is tempting here to speak of “skills”: the skills that will help us grapple with algorithmically generated text, the skills that enable us to move within the limits of the historical record and capture a wider understanding of the past than what is recorded. But these multiple literacies are perhaps better regarded as forms of labor; to attend to them as such requires grappling with institutional demands for efficiency. And though we share Kirschenbaum and Raley's skepticism of calls for interdisciplinary partnership, we maintain that partnership is necessary to grapple with the scale of transformation AI brings to higher education, in terms of both new models for academic writing (see Martin's contribution here) and transparency about use (see Laquintano and Vee's). We emphasize that librarians are already organizing and responding to the material crises initiated by AI to meet the pedagogical and archival challenges that arrive in its wake. Partnerships like ours, beginning with an effort to enrich existing courses on the history of reading with computational approaches to language, provide a shared terrain for developing something new in the face of such challenges. Partnerships between faculty members and librarians can offer opportunities to improve pedagogical outcomes for our students while also creating space to organize a resistance to the demand for efficiency and the policies it imposes.

Libraries institute a resistance to the enclosure of the commons on multiple fronts. They contest the encroachment of aggressive copyright laws and licenses. They maintain the infrastructure for open-access publishing and open-source software and data, and they socialize the value of such venues and practices among the communities they serve. Libraries preserve, in open form, the cultural heritage materials that would otherwise be lost to institutional neglect or seized by corporate publishers, whose shoddy but costly “digital archives” reify the past as a luxury good. Finally, on the pedagogical front, libraries cultivate, through their spaces, their practices, and their very ambience, the experience of that serendipitous swerve among disciplines, genres, languages, and styles that rewards the verve of the desultory reader with the delight of discovery and the profound shock of recognition. Unlike the algorithmically manipulated bitstreams that shape social media engagement, libraries reveal institutional and epistemological investments, physically connecting the legacies of techniques of curation, ordering, and description rooted in outmoded models for efficiency with their newer iterations. For this reason, the library (as opposed to the increasingly surveilled classroom described by Kirschenbaum and Raley) may provide space for alternative imaginaries.

It's not hard to imagine a day when the text-generating tools seamlessly embedded into applications like Microsoft Word or Google Docs impose end-user licenses that limit what we can do with the text thus produced, or when publishers market texts authored “on demand”: individually customizable reading experiences in everything from news outlets to genre fiction. Libraries already find themselves elbowed into the role of contractual agents, unwilling mediators between corporate publishers and the publics that they are intent on converting into consumers. But at least the content thus licensed already exists, and it can be cataloged and, to some extent, repurposed as an occasion for shared experiences that the library hosts. If all content is generated on demand, “personalized” according to the rubrics of models themselves proprietary and shielded from public inspection, then reading ceases to be an act of multiplying publics; it promises to become, rather, the production of an endlessly splintered and ephemeral privacy—a “templated” privacy whose contours repeat endlessly throughout the population but without creating the kinds of relations that can exist among people on the basis of a shared, if contested, experience and a common point of reference.

Even elite humanist readers like Gabriel Harvey, alone in his study with his book wheel, wondered whether there might not be a better way of reading, one that did not involve endless annotation and the consumption of more and more information, but that instead would foster thought without a commodifiable outcome or purpose. It was a mode of reading without pen in hand, a sense of purposeless thinking (Jardine and Grafton 77). The rarefied book wheels that flood our fractured commons with content algorithmically selected, and now algorithmically generated, enclose and commodify attention itself. Or they seek to—attention is not a resource but a relation. As scholars in the humanities, we are trained to attend to what's missing, and we reach for a history of reading that documents something other than a totalizing program of mastery and optimization—something like a record of resistance. These insights are not new, especially as they build on scholarship defined by this history of resistance, but they are especially relevant now as the field of critical AI emerges (Raley and Rhee).

If the humanities have a place to defend amid various models of ruination (and if the word ruin should apply when, for the most part, our academic buildings are not reduced to rubble), it seems at once too much and woefully insufficient to suggest that the practices of humanistic pedagogy—of close reading, of compassionate listening, of impassioned but circumspect speaking and writing—are among those that lend themselves to one of the great tasks of this moment: to keep our collective intelligence from becoming an artifice to itself, and hence a thing to be dispensed with. But we'll need other insights as well, including critiques of those pedagogies developed in fields that are defined by this sense of resistance. In naming this task, we do not appeal to any binarist conception of human essence or human nature. Rather, we refer to the cultivation of the lag that this pedagogy introduces and the critique that it hopefully inspires in its wake: a hiatus or suspension, in which conversation becomes possible, and a moral imagination, which is only the flesh, in its thickness irreducible to any model, at work.