Generative AI has made fluent language cheap. It can produce summaries, explanations, drafts, and stylistic imitations at a speed that unsettles long-standing assumptions about writing, originality, and academic work. Most institutional conversations respond at the level of output: detection, policy, efficiency, and acceptable use. Those concerns matter. Ann E. Berthoff (1933–2019) offers a more clarifying starting point. Although she wrote before contemporary generative AI, her account of composing as meaning-making is newly urgent now that language production can appear to substitute for interpretation.
Berthoff’s account of composing treats writing as epistemic activity. Writers do not simply report what they already know. They use language to discover, test, and revise what they think. The work of composing involves selection, relation, naming, re-seeing, and re-making. In this view, a draft is not merely a preliminary version of a finished product. It is evidence of a mind at work. Writing externalizes thinking so that it can be examined, challenged, and changed.
In an age where generative AI excels at producing what looks like finishedness, Berthoff helps us name the risk. In terms of education, the risk is confusing surface fluency for understanding. AI can supply “naming” at scale, but naming is not knowing. Knowing requires a relationship to a problem, evidence, context, and consequences. Knowing includes accountability for meaning.
Berthoff’s emphasis on interpretation sharpens this distinction. Interpretation, for her, is not a specialized technique reserved for literary study. It is a fundamental human activity, the ongoing work of making sense of experience and information through language. Writing instruction, in this framing, is not chiefly about correctness or formula. It is about cultivating interpretive judgment: learning to generate meanings, to weigh meanings, and to take responsibility for meanings. That is why she resists pedagogies that reduce writing to the reproduction of content or the filling of predetermined slots. When students write only to comply with a format, they practice conformity rather than interpretation. They can learn to sound right without learning to think.
AI intensifies that danger while disguising it. A student can now produce plausible academic discourse with minimal engagement in the interpretive labor that gives discourse its value. The result can look like competence while bypassing the cognitive work that makes competence real. This is not simply an integrity problem. It is a learning design problem. If a course positions writing as deliverable rather than meaning-making, AI will naturally become the shortest path to the deliverable. If a course positions writing as inquiry, judgment, and accountable interpretation, AI becomes less substitutive and more obviously limited.
A Berthoff-informed stance toward AI does not require blanket prohibition, nor does it require uncritical embrace. It asks for alignment between our purposes and our designs. When AI is used as a scaffold that supports a writer’s interpretive work, it may reduce barriers to entry, especially for writers who struggle with initiation, organization, or anxiety. When AI is used to replace interpretive work, it produces the appearance of learning without its substance. The pedagogical question becomes: does the assignment and the assessment structure require the student to make meaning in ways that cannot be outsourced without loss?
This lens suggests several durable design implications.
First, design for interpretation rather than generic production. Tasks that depend on situated context, discipline-specific evidence, local constraints, or lived experience make meaning-making more visible and harder to counterfeit. A prompt that asks students to apply concepts to a specific case with real tradeoffs is not simply harder for AI. It is more educationally honest. It requires the student to decide what matters and to justify those decisions.
Second, design for visible judgment. If writing is a method of knowing, we should assess not only the final text but the interpretive moves that produced it. Brief process artifacts can do this without turning a course into a paperwork factory: revision memos that explain changes and rationales, annotated bibliographies that justify source selection, comparison drafts that show alternatives considered, short reflections that connect claims to evidence and consequences. These practices do not merely deter misuse. They teach students that writing includes choosing, not just producing.
Third, protect voice as agency rather than voice as style. AI can imitate tone. It cannot inhabit responsibility. A student’s voice becomes educationally meaningful when the writing demonstrates ownership of meaning: specificity, warranted claims, an identifiable stance toward evidence, and an ability to explain why the writing says what it says. In Berthoff’s terms, this is authorship as interpretive accountability. It is less about sounding unique and more about being answerable.
These implications extend beyond composition courses. Berthoff’s contribution is not confined to English studies. In any discipline where students must explain, argue, analyze, interpret data, design solutions, or communicate with public audiences, writing functions as a medium for disciplinary thinking. Students learn what counts as evidence and what counts as a good question through the act of composing. If AI supplies the discourse of a discipline without requiring students to internalize its reasoning, we risk producing students who can perform membership without developing membership. Berthoff helps us see why that matters: disciplinary language is not merely a code to be mimicked. It is a way of making meaning in a community of practice.
A compact way to state the challenge is this: AI changes the economics of language, but it does not change the educational purpose of writing. Berthoff reminds us that writing is a way of knowing. The task for faculty is not to defend writing as tradition or to chase every new tool. It is to design learning environments in which interpretation remains central, judgment remains visible, and students remain responsible for meaning.
If generative AI makes fluent language abundant, then the scarce resource in higher education becomes interpretive agency. Berthoff gives us a vocabulary for protecting that resource. We can treat writing as a site where students learn to make meaning, not merely to produce text, and we can design accordingly.