We Have Always Been Action Theorists: Toward a Critical Theory of Language for the Era of “Large Language Models”
Scholars of literature and culture understandably place themselves among the world’s premiere experts on matters of language. But they also know that fields like linguistics and communication have their own ways of studying how people express themselves through speech and written media. A key difference concerns the theories and methodologies that inform how language is studied: thus, the language denoted by literary scholars, which often explores language in its textual forms, differs considerably from that denoted by many linguists, for whom language use—as codified by the psycholinguist Herbert H. Clark—may be the principal framework. These distinctions bear on cross-disciplinary dialogues that are beginning to take place as literary scholars increasingly turn their sights on large language models (LLMs); on the generative pretrained transformer (GPT) architectures through which LLMs are now developed; on the chatbot implementations through which they are packaged; and on the larger sociotechnical, politico-economic, and geopolitical projects of commercialized generative AI to which these technologies give rise.Few communities of practice are as skilled in the vital work of theorization, contextualization, and formal analysis as are literary and cultural critics, whose methods set aside shallow and positivistic truth claims in favor of situated inquiries from diverse standpoints. As I have argued elsewhere, the field of critical AI studies urgently needs to put the “humanities in the loop” of conversations surrounding “AI.”1 New collaborative projects will likely find literary critics studying the topic in partnership both with scholars they already know (for example, historians and anthropologists) and those they may know less well (e.g., computational linguists and specialists in Natural Language Processing [NLP]). While the theories and methods particular to these AI-adjacent disciplines may compel conversation and debate, such engagements could sharpen long-held assumptions about language in light of new and, indeed, urgent topics.2In this essay, I frame the challenges ahead in relation to recent critical interventions into generative AI (gen AI) while proposing that one preparation for cross-disciplinary encounters could entail critics thinking more about their own teaching and practice of writing. In “Against Output” (2023), writing studies and digital humanities (DH) scholar Annette Vee points out that the theories that underpin literary criticism only “rarely” discuss “the process of writing,” even though that very process undergirds both the form and practice of the scholarship that ensues.3 Reflection on the writing process could, I believe, help critics to recognize that they already know something important about what Clark identifies as the action tradition in language study, and what Vee—with the writing process in mind—describes as “the real action in authorship.” In his influential study Using Language (1996), Clark contested what had been a predominant “product tradition” in language theory, which “grew out of the study of sentences, words, and speech sounds” as “linguistic types”—that is, as the “products” of language use. In contrast, the “action tradition” he proposed explores using language—in speech, writing, and gesture—as a process involving the study of the specific “speakers, times, places, and circumstances” through which the language in question materializes.4Most literary scholars will not know Clark’s book: but his distinction is comparable to that made by educators in writing-based disciplines when they encourage students to distinguish the process of composing essays from the product handed in for a grade—not least since it is the process that supports active learning.5 Moreover, Clark’s distinction between “action” and “product” traditions (though explicitly addressed to “product” theorists such as Noam Chomsky) can be productively compared to a source many literary critics know well: Ferdinand de Saussure’s structuralist distinction between langue and parole as articulated in the Course in General Linguistics (1916). As I will argue, Saussure’s distinction points to long-disputed but surprisingly unresolved tensions within poststructuralist theories that are remarkably relevant to the new conjuncture. Indeed, it is possible that the “language” at stake for cultural critics has all along been conceived—if sometimes implicitly—through a theory of action harnessed to strong practices of textual interpretation whose implied author is typically the critic. If that proposition holds, then “we” have always been action theorists—by which pronoun I hail literary scholars and their colleagues in other writing-intensive humanities disciplines. From this standpoint, the “language” at stake for scholars influenced by the legacies of Jacques Derrida, Michel Foucault, and the Russian formalists was and remains responsive to an incipient theory of language use.To be clear: in urging reflection on language use, I do not mean that critics should abandon textual analysis in favor of writing studies, linguistics, or communication. I am arguing, rather, that because literary and cultural critics are invariably immersed in the teaching and practice of writing, they are already prepared to appreciate theories that conceive language first and foremost as enacted through social, embodied, and relational practices of meaning-making. Whether their pedagogy centers on the undergraduate classroom, the graduate seminar, or both, these scholars guide and evaluate their students’ writing processes in diverse ways—for example, by helping students to formulate critical reflections on existing bodies of knowledge and/or to explore new hypotheses, evidentiary pathways, theories, or methods. They are thus well positioned to recognize that today’s LLMs do not understand or enact language in a human-like way. That is because (as I will explain) these tools model the signifiers of human-generated language but not the signs of that language as typically conceived. It follows that text-centered approaches to the study of literature, culture, or history—however compelling in their own right—are insufficient to answer the most pressing questions about language in relation to text-generating language models.Fortunately, critics have begun to push back against the aggressive commercialization of and hype over generative AI. Just a few years after the November 2022 release of OpenAI’s ChatGPT, cultural critics have published cogent interventions, including Michele Elam’s “Poetry Will Not Optimize,” Vee’s “Against Output,” Kyle Booten’s “Build Word Gyms, Not Word Factories,” Eduardo Ledesma’s “Critical AI Studies and the Foreign Language Disciplines,” and Elizabeth Losh’s “AI Literacy and the Politics of Academic Labor.” These articles derive from four essay collections published, respectively, in American Literature; the blog of Critical Inquiry; Critical AI’s two-part special issue on generative AI; and a recent PMLA “Theories and Methodologies” cluster on AI.6 For the most part, these contributions to a now flourishing field of critical AI studies productively complement one another. In their introduction in PMLA, Matthew Kirschenbaum and Rita Raley argue that the spread of text-generating technologies calls for renewed theoretical attention to language, including questions of provenance (where language comes from), affordances (“what it can do, what can be done with it”), and contingency (“how it might be changing”).7 I agree. Nonetheless, as cultural critics answer such calls, they join ongoing interdisciplinary conversations about GPT-based products and practices—a technology still less than ten years old—while drawing on theories of language that coalesced decades before. Meeting the challenge of this new sociotechnical conjuncture thus requires evaluating theoretical propositions about the language in “language” models from new vantage points.In what follows I begin by describing the political economy from which gen AI research and development spring. After establishing that LLMs model signifiers, I turn to a theory of active language that, I hope, will sharpen humanist analyses of generative technologies while facilitating interdisciplinary dialogue. The latter goal speaks partly to my experiences editing Critical AI, a journal that resists the siloed disciplinary configurations that have dominated AI research since its Cold War-era inception—a research ecosystem that has enabled sociotechnical agendas that profess to simulate or augment “the human” with minimal input from the humanities or, indeed, from humanity at large.8When the DH scholar Katherine Bode and I co-introduced Critical AI’s inaugural issue in 2023, we proposed collaborations that could enlarge spaces of “dissensus” and create pathways to new knowledge.9 In doing so, we envisioned creative solidarities, the enlistment of design justice principles, and the blending of the onto-epistemic insights of the “new materialisms” with historiographic traditions—dialectics that we saw productively at work in Arturo Escobar’s Designs for the Pluriverse.10 Citing media theorist Sasha Costanza-Chock’s affirmative vision of a just digital commons, we hoped that even readers who disagreed could find “ways of living differently” in the journal’s pages.11 The unconventional partnerships we anticipated would not only cross the much-discussed divide between the liberal arts and the sciences but also, we hoped, permeate the myriad enclaves within the arts and humanities that shape disciplines, periods, and methodologies. For critical AI studies to create solidaristic but open-ended spaces of dissensus, we recognized, the field would need to eschew naïve or overweening dreams of freedom, consensus, techno-utopianism, or ideological purity. That is the aspiration too of the present call for shared reflection on language as it bears on the futures of writing, the teach
