Academic Erasure: The Disappearance of Complexity Under AI-Supported Writing

Synthesis: Nwagboso & Atuba (2026) introduce academic erasure — the way generative AI produces fluent, structurally compliant writing while eroding the epistemic labor, critical thinking, and authentic voice that academic writing is meant to cultivate. Analyzing 49 Reddit posts from educators, they identify seven forms of erasure (epistemic labor, authentic voice, educator efficacy, trust, pedagogical method, the human–AI boundary, and institutional legibility) and frame the phenomenon as the joint effect of two mechanisms: epistemic delegation (offloading thinking to AI) and epistemic singularity (the homogenization of voice into an AI-optimized register). The study reframes AI's risk in higher education as less about plagiarism than about the loss of intellectual struggle, urging Assessment and pedagogy toward process, voice, and depth.

Key Findings

• Academic erasure is a distinct phenomenon from plagiarism. AI-generated writing preserves (even enhances) surface fluency — grammar, structure, coherence — while diminishing the epistemic, critical, and disciplinary dimensions of writing. It operates through compliance rather than deviation: students can meet every formal criterion yet produce epistemically vacuous work, a simulacrum that mimics scholarly form while bypassing the labor of inquiry.
• Two mechanisms drive it: epistemic delegation and epistemic singularity. Epistemic delegation is the displacement of cognitive labor onto AI — bypassing the iterative work of formulating arguments, working through tensions, and synthesizing sources. Epistemic singularity is the resulting homogenization: disciplinary, cultural, and individual variation collapsing into a dominant AI-optimized register, trained on standardized English and mainstream discourse.
• Seven forms of erasure emerged from 49 educator posts: (1) erasure of epistemic labor, (2) of authentic student voice, (3) of educator efficacy and Well-Being, (4) of trust and epistemic certainty, (5) of traditional pedagogical methods, (6) of the human–AI distinction in learning infrastructure, and (7) erosion of institutional legibility.
• Educators bear an unacknowledged affective cost. The corpus registers derealisation, grief, and burnout as educators police AI use — what the authors call affective labor at scale, worsened by inconsistent institutional policy and the normalization of AI as infrastructure. The authors link this to automation bias and the instrumentalisation of higher education.
• Assessment culture makes AI use rational, not deviant. Where rubrics reward fluency, polish, and citation over argumentation and intellectual risk, AI becomes the optimal response to the system's incentives. The authors therefore argue that treating AI use purely as misconduct individualizes a structural problem.

What this means for practice

• Instructors. Redesign assessment so the labor of thinking is visible — process logs, staged drafts, in-class composition, and oral defenses AI cannot meaningfully complete — because AI-generated writing satisfies every surface criterion while bypassing the inquiry writing is meant to develop.
• Instructors. Stop rewarding fluency and polish: where rubrics privilege grammatical compliance, structure, and citation over argumentation and intellectual risk, the paper argues AI use becomes the rational response to the system's incentives rather than a deviant act.
• Assessment designers. Grade epistemic depth — disciplinary positioning, engagement with contested ideas, and intellectual risk — so a well-formed but vacuous submission cannot earn credit for thinking it did not do.
• Administrators. Treat educators' grief, derealisation, and burnout while policing AI as a structural problem worsened by inconsistent institutional policy and the normalization of AI as infrastructure, not as individual teacher stress.
• Instructors. Move the classroom from surveillance to dialogue: the paper frames current uncertainty as a pedagogical opening and warns that framing AI use purely as misconduct individualizes a structural problem.

Limitations

• The dataset is 49 Reddit posts retained from roughly 380 candidates gathered between October 2023 and February 2025; the authors describe them as informal, unverified, subjective, and emotionally inflected, and note that poster anonymity rules out follow-up.
• The corpus captures educators only — self-identifying educators, plus three student posts — and excludes students, institutional leaders, and policymakers.
• The interpretive design supports conceptual insight rather than generalization: the authors state the findings are not statistically generalizable and rely on cross-sectional data, offering a snapshot rather than a longitudinal view of change.
• Both researchers are UK lecturers who had marked essays they suspected were AI-assisted and had redesigned assessments under uncertainty; the paper acknowledges this proximity as a limitation as well as a resource, and its planned comparison of AI-generated outputs with pre-AI student writing is reported as unused.

Citation

Nwagboso, C., & Atuba, S. (2026). Academic erasure: The disappearance of complexity under AI-supported writing. Teaching in Higher Education.