Preserving epistemic authenticity: process-oriented assessment in the age of generative AI

Synthesis: Generative AI has made fluent academic prose cheap, and with it the assumption that an unsupervised written submission is evidence of the student's own reasoning. Thapa and Lewis argue that the answer is not better detection but better design. Drawing on Authentic Assessment and Formative Assessment, they set out process-oriented assessment: tasks that make interpretation, justification, and knowledge construction visible over time rather than only at the endpoint. Their organizing idea is epistemic authenticity — the extent to which assessment captures a learner's genuine engagement with reasoning and Evaluative Judgment — and they propose four design commitments: reflective justification, staged task design, dialogic engagement, and evaluative transparency. They are candid that such designs intensify educator labor and can widen equity gaps unless deliberately scaffolded.

Key Findings

1. Detection cannot carry the assessment burden. Tools risk false positives and inconsistent outcomes across writing styles and contexts, raise procedural fairness problems in high-stakes decisions, and trail increasingly capable models.
2. Reactive monitoring crowds out learning. Detection identifies misuse only after it occurs, and relying on it pulls institutional culture toward verification and policing rather than intellectual and ethical engagement with knowledge.
3. The pivotal question is evidential, not technological. The central issue is no longer whether AI-generated work can be caught, but whether assessment can still yield meaningful evidence of reasoning, judgment, and disciplinary engagement.
4. Epistemic authenticity is the paper's central conceptual contribution. It names the extent to which assessment captures genuine engagement with interpretation, evaluative judgment, reasoning, and knowledge construction — a property polished output no longer reliably signals.
5. Process-oriented assessment repositions assessment as part of learning. Built on constructive alignment and assessment for learning, it makes reasoning visible through staged development, reflection, dialogue, and iterative feedback rather than one final product.
6. Four design principles follow. Reflective justification, staged task design, dialogic engagement, and evaluative transparency require learners to explain decisions, articulate interpretations, and show how conclusions emerged.
7. The redesign carries equity and workload costs. Reflective and dialogic tasks can privilege students confident in academic self-articulation, and the relational work they demand intensifies educator emotional labor.

Why detection is not a sustainable response

Universities have largely answered generative AI with detection software, revised integrity policies, and guidelines regulating AI use, and the paper assembles the case against leaning on these. Detection tools risk false positives and inconsistent outcomes across writing styles and contexts, and the decisions they inform are consequential; the rapid evolution of language models also leaves the technology a step behind. Detection is additionally reactive: it identifies misuse only after it has occurred, without engaging Assessment design or learning. Sustained reliance reframes Academic Integrity as policing rather than intellectual and ethical engagement with knowledge, straining the trust and partnership student-centered higher education depends on. The stakes run broader than misconduct — authorship, originality, and knowledge production are all in question — which is why AI Detection is a weak lever for a substantially epistemological problem.

From product to process

Process-oriented assessment shifts attention to how learners develop ideas, interpret information, and construct understanding over time. It draws on constructive alignment, on assessment for learning rather than assessment of learning, and on Authentic Assessment, which requires applying knowledge in meaningful, situated tasks rather than reproducing it. The paper is careful here: authenticity alone is not enough. An authentic task that remains a single unsupervised written product is exposed to the same substitution problem as the essay it replaced. What matters is making interpretation visible — staged development, reflection, dialogue, and iterative engagement with Feedback — so the task also supports Self-Regulated Learning. Generative AI is comparatively weak at exactly these things: contextual judgment, reflective interpretation, and reasoning that evolves in response to feedback.

Epistemic authenticity as the organizing idea

The paper's most distinctive move is to name epistemic authenticity as what assessment should protect under AI-mediated conditions: whether an assessment captures genuine engagement with interpretation, evaluative judgment, reasoning, and knowledge construction. Under product-oriented assessment that authenticity was inferred from the writing itself; generative AI breaks the inference, because a polished response may no longer represent anyone's thinking process. The authors do not conclude that AI must be excluded from learning environments. Assessment must instead evaluate how students critically engage with information, justify interpretations, and demonstrate reflective understanding — capacities that bear on the validity of claims about learning. The challenge is epistemological as much as technological, and it demands more direct evidence of conceptual and evaluative engagement than a final submission provides.

What the design principles require in practice

Four principles carry the argument into design: reflective justification, staged task design, dialogic engagement, and evaluative transparency. In practice, learners explain interpretive decisions rather than only presenting conclusions, submit work that develops across checkpoints, take part in oral or dialogic exchanges about their reasoning, and work against criteria they can apply themselves — the terrain of Feedback Literacy and Self-Assessment. The authors are explicit about the costs. Reflective and dialogic tasks can privilege students more confident in academic self-articulation unless they are inclusively designed and scaffolded, and uneven access to AI tools compounds the existing equity problem. The relational, feedback-intensive work also generates substantial emotional and professional labor in large or resource-constrained contexts — a contradiction with institutional expectations for scalability, standardization, and measurable performance.

What this means for practice

• Instructors. Convert one high-stakes unsupervised written task into a staged sequence with reflective justification at each checkpoint, so students explain decisions instead of submitting only a finished product.
• Assessment designers. Write the four principles into the task brief: checkpoints with reasoning artifacts, a dialogic or oral exchange, and criteria students can apply to their own drafts.
• Administrators. Resource the redesign before mandating it — dialogic, feedback-intensive assessment needs staffing, time, and workload recognition, and detection spending does not remove that obligation.
• Faculty developers. Provide scaffolding so reflective and dialogic tasks do not reward verbal confidence learners were never taught, and treat the shift as curriculum change rather than a task swap.

Limitations

• The paper is conceptual: it reports no participants, no data collection, and no implementation trial, so the four design principles are argued rather than tested.
• Its claims about equity, emotional labor, and sustainability rest on the authors' synthesis of existing scholarship, not measurement; the paper itself calls for empirical work on Student Experience, educator workload, and long-term viability.
• The argument is normative about what assessment ought to do, with no account of how institutions under efficiency and standardization pressure would resource dialogic assessment at scale.
• Feasibility is left undifferentiated across settings — transfer to large lectures, laboratories, studios, or clinical placements is not addressed.

Citation

Thapa, G. C., & Lewis, S. (2026). Preserving epistemic authenticity: process-oriented assessment in the age of generative AI. Assessment & Evaluation in Higher Education, 1-15.