Between De-familiarization and Re-familiarization: Rethinking the Transfer of ELA Skills in AI Literacy

There has been, of late, a minor flurry of well-meaning proclamations about the unexpected virtues of literary education in our age of algorithmic interlocutors. A recent example by Nick Potkalitsky offers what might be termed a pedagogical rapprochement between the disheveled textualities of fiction and the peculiar smoothness of generative AI. The claim, put plainly, is that those trained in the art of literary analysis—especially the kind that involves reading both with and against the grain—already possess the requisite cognitive flexibility to manage the duplicitous charm of language models. These readers, it is said, know how to feel with Gatsby and still suspect Fitzgerald, know how to weep for Anna Karenina while noticing the hand on the narrative tiller. This skill, so the argument goes, prepares them nicely for navigating interactions with ChatGPT, Claude, or whatever quasi-autonomous verbosity engine next emerges from the depths of venture-backed infrastructure.

But the analogy also breaks down in some important ways. And in those breaks, we find the contours of a deeper epistemic shift that “double reading” alone can’t capture.

In literary reading—especially within formalist or modernist traditions—”defamiliarization” plays a central role. Literature (and here I also acknowledge the instability of that term) doesn’t just teach us to keep belief and skepticism in tension. It stages aesthetic rupture—tonal disjunction, syntactic interruption, irony that cuts against affect.These ruptures elicit gaps in thought itself, a momentary awareness of the limits of our rational capacity to make quick sense of the world. In the process, these “literary” moments estrange language from its apparent meaning and invite reflection not how form shapes meaning. 

AI-generated language, by contrast, is optimized for the opposite effect. In fact, if literature can be understood as a project in defamiliarization, LLM’s are designed for refamiliarization. Its fluency masks the recursive, statistical, and infrastructural processes that produced it. There is rarely any rupture or  shift that signals, “something is off.” Instead, there is often the rhetorical equivalent of Polonius in Hamlet: contradiction that flows without resistance, irony drained of affect, coherence that feels like meaning but lacks ground. And even Polonious, must be read, but through a different lens. 

I agree that close reading matters. But the object of that reading must be refigured, if only because what we are reading in AI-generated text belongs to a different epistemological regime than even the most capacious definitions of the literary. One needn’t tether literature to individual intention to make this distinction. Even when texts are conceived as cultural artifacts—emergent from discursive formations, shaped by the unconscious grammars of ideology, sedimented through the non-sovereign agency of authorial subjectivity—they still presuppose a field of meaning produced through and for human interpretation. Their rhetorical textures, however diffuse, still implicate a horizon of recognition, a social embeddedness, a set of historically intelligible signifying practices. AI-generated language, by contrast, emerges from recursive optimization and stochastic training processes; it does not so much participate in culture as it assembles simulacra of its surfaces. The interpretive gesture here shifts. One is no longer deciphering intention or excavating subtext. One is engaging in a form of infrastructural tracing—reading for the procedural, the recursive, the computationally interpolated logics by which sense is simulated. The question is not what the text expresses, nor even what it conceals, but how it came to be legible in the first place—and to whom.

And that raises a pedagogical point: it may not be enough to say that literature instruction “transfers” cleanly into AI literacy. Because transfer itself is not a neutral pedagogical concept. It assumes a generalizable skill abstracted from context. But if meaning in AI is always situated—shaped by infrastructure, prompt history, system architecture—then we need to teach reading not as transferable content knowledge, but as situated epistemic awareness. (In fact, some might argue more generally against the goal of transfer on the grounds that all learning is situated and cannot be transferred beyond context–that is the premise of theoretical work in situated.)

This is not a call to abandon rhetorical analysis. Far from it. But it does require that we locate rhetoric itself within the logics of platform infrastructure. A well-crafted sentence must be read not simply for its connotation or cadence but as an index of training data, interface constraints, and prompt engineering. The poetics of machine-generated language lie as much in their latency as in their lyricism. Meaning, if it happens at all, happens at the intersection of recursive input history and statistical inference, a point that invites interpretation but also disperses agency in the process.

In other words, I agree with the need to cultivate literary sensitivity, but not to the flattening of difference between aesthetic rupture and infrastructural smoothness. This also invites a reflection on the evolving history of literary pedagogy itself. The kind of close reading often invoked in defenses of literary transfer is historically grounded in New Criticism, with its emphasis on ambiguity, irony, and formal complexity as signs of aesthetic depth. Think of William Empson’s Seven Types of Ambiguity, or Cleanth Brooks’s notion of the “well-wrought urn”—the literary text as a self-contained object whose meaning emerges from the tension between its internal parts. This was a model of reading that prized paradox, contradiction, and the interpretive challenge of unresolvable structure.

Later, this model gave way to cultural studies, deconstruction, and poststructuralist approaches that displaced the author as the source of meaning. Here, texts became nodes in larger cultural and ideological systems. Meaning was no longer housed in the unity of the work, but in its relational participation in discursive formations—in how it aligned with or subverted dominant norms. Reading became less about aesthetic complexity and more about historical legibility, institutional power, and representational politics.

But this shift also brought new challenges—especially in pedagogy. When ELA teachers began incorporating popular or “non-literary” texts (advertisements, memes, sitcoms, social media posts) as valid objects of analysis, they often struggled to model the same kind of nuanced interpretive moves that literature had trained them to perform. Many of these texts didn’t offer irony, ambiguity, or defamiliarization in the same way—or at all. They were not “well-wrought urns” but ephemeral artifacts of affect and circulation, texts that functioned differently and thus demanded a different kind of literacy. Even something as canonical as Shakespeare poses a similar challenge: how does one model a close reading of wordplay, wit, or rhetorical flourish, especially when those features resist direct paraphrase or moral interpretation?

This is where the analogy between literary reading and AI output again shows strain. AI-generated text, like much popular discourse, does not reliably perform literariness in the traditional sense. It doesn’t announce its irony, stage ambiguity, or rupture its own surface. Yet it still demands interpretation—just through different frames, ones attentive to infrastructure, recursion, and epistemic formation rather than to textual unity or symbolic resonance.

So if the goal is to teach students how to read AI outputs with the same care they bring to literature, we must also be honest about how literary pedagogy itself has always been in tension with its objects—and must continue to evolve not just in content, but in its conceptual grounding. The death of the author was one step. The de-centering of the text may be next.