I presented at 4C25 on AI refusal, focusing on the ways that AI reduces intentionality and creativity in the craft of academic writing. In many ways, this builds on previous strands of research, like the keynote I delivered at last year’s C&W, and is building towards the upcoming Unprompted collection, which revolves around AI refusal in and out of the classroom.
Access copy of my talk below.
AI and the Enshittification of Writing
Dr. Vyshali Manivannan
Pace University – Pleasantville
CCCC, 12 April 2025
Hello, I’m Vyshali Manivannan, and I come to you from the virtual margins, as Cs lacks options for robust hybrid participation and doesn’t mandate onsite measures to mitigate disease spread. I can’t in good conscience proceed without noting that AI technologies play a role in the ongoing genocide in Gaza, in visa and citizenship revocations affecting our students and colleagues, in targeting trans folks, in rejecting health insurance claims. If you’ve ever said you believe in equity and inclusivity but behave as though AI tools are worth the aforementioned costs, realize that to many of us your values ring hollow.
Now: My contention that GenAI is furthering academic writing’s “enshittification”—Doctorow’s (2022) term for the Internet’s progressive decline in quality as profit imperatives encroach on it—hinges on three premises:
1. GenAI promises to “democratize” and “optimize” writing by enabling users to execute “authentic” artisanal production without requiring discipline or prerequisite skill.
GenAI’s operationality relies on harvesting original content without consent. Developers excuse GenAI’s copyright violations in the name of scientific progress. Users extend this by claiming that process is irrelevant: consumers see only a finished product, not the craft activity invested in its creation, implicitly justifying the opacity of GenAI’s decision-making processes. GenAI enshittifies writing by correlating the need to make fewer writerly decisions with optimized writing, minimizing and mystifying the role of craft and suggesting ignorance absolves users of any liability for GenAI’s decisions. GenAI can’t reproduce, refine, or develop craft, as its writing decisions consist of high-fidelity remixing of entire textual corpuses without the influence of human observation, experience, or emotional investment behind it. Thus, its textual products are more likely to pass muster in genres where craft is allegedly an afterthought, like academic writing.
2. GenAI’s datasets are biased toward popular and academic stereotypes of objectivity and knowledge legitimacy, so it’s particularly effective at simulating academic writing.
Current formulations of scholarship continue to privilege monolinguistic, monogeneric, monocultural, linear, error-free prose. Academic writing is preconfigured, in Gibbs’ (2015) words, as “an aftereffect of research” (p. 1). This is reflected in the fact that most doctoral programs outside of the arts don’t offer dedicated craft courses, and readability—beyond intelligibility to a specialized audience—is not commonly emphasized in curricula or publication criteria. Academic habitus is designed to reproduce academic subjects who see writing as an invisible template, a secondary thing that transpires outside of the research process. This erasure of craft from academic knowledge production is validated by (and in turn validates) the ever-accelerating neoliberalization of higher education, amplifying the enshittification cycle.
3. Writing is a deliberate craft and practice in and of itself that requires human experience and associative thinking, and disregarding craft impoverishes our scholarship.
While carefully considered craft decisions may not be missed in AI-generated academic writing, given how it’s overlooked in human-authored scholarship, craft-based problems are readily apparent in ChatGPT’s recent “metafiction.” Lauded by OpenAI and its proponents as the moment when AI-generated writing passed the Turing test, the story’s writing decisions actually demonstrate a failure to anticipate human interpretation. It presumes the genre is communicated solely by aggressive self-insertion; presents characters and events abstracted to vanishing point; forgets what it has written, as when it reveals itself as an AI for emotional payoff, twice; and makes intertextual references, like lifting Nabokov’s “a democracy of ghosts,” without any sense of having stumbled into a wider textual ecology. Metaphors are meant to concretize the abstract through anchoring sensory description, but ChatGPT’s most celebrated line, “Grief, as I’ve learned, is a delta,” makes the abstract differently abstract. In short, the story is what you might call skimmable prose: laden with oddities and mixed metaphors and phrases that feel randomly chosen, uninfluenced by human observation and feeling. It’s fine enough to skim but falls apart under attentive reading.
Chiang (2024) says, “[AI] is a fundamentally dehumanizing technology because it treats us as less than what we are: creators and apprehenders of meaning. It reduces the amount of intention in the world.” I expect AI is believed to be useful by, and feels useful for, academic writers who don’t have a vision for their prose, or don’t understand why it, like creative writing, requires intentionality, because they were never taught that writing is a mode of research itself (Gibbs, 2015). It doesn’t seem odd to them to outsource their writing choices, or to use skimmable prose as the vehicle for research. Maybe if we applied what Chiang says about craft equally to scholarship, it would be easier to perceive our own complicity in the AI-driven enshittification of academic writing.
Discussion Question: As writers and writing instructors in the neoliberal academy, how do we negotiate the tension between writing and teaching writing as a craft and practice to disrupt the enshittification of writing?
