Replace detector-centred assessment with a four-stage process-oriented model: declared starting point, versioned drafting, process artefacts, and targeted oral verification¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The paper proposes a concrete process-oriented assessment model in which "Students submit a short proposal and annotated source list" and the assessment brief states explicitly which AI functions are permitted, which require disclosure, and which are prohibited. Its component practices—portfolio assessment, supervised drafting, staged feedback, and viva voce examination—predate generative AI but acquire renewed relevance because final-text classification cannot establish authorship, understanding, or misconduct on its own. The model shifts evidentiary weight from product-only classification to documented process and verified understanding.

Design Implications¶

Context¶

Requirements¶

• Assessment briefs must state explicitly which AI functions are permitted, which require disclosure, and which are prohibited
• Policies should distinguish functions such as brainstorming, grammar correction, translation, structural support, and generation of substantive content rather than treating AI use as one category

Constraints¶

• Routine, non-punitive disclosure is necessary but not sufficient: it cannot repair genuine false positives, cannot prevent permitted polishing from producing incriminating scores, distributes its burden unequally across language backgrounds and disabilities, and does not remove the percentage from the assessor's interface

Target Learners¶

• higher-education students
• second-language English writers
• students with disabilities

Target Learning Goals¶

• authentic authorship assessment
• academic integrity without detector scores
• verified student understanding

Affordances¶

• Detectors Infer From Textual Properties Not Provenance

Related Strategies¶

• Assess process over product using staged, connected tasks with iterative evidence
• Use a mixed-methods assessment sequence: diagnostic inventory, structured verification tasks, and discipline-specific critique or workflow-design assignments
• 5-Minute Writing Conferences
• Use weekly oral code review interviews to incentivize students to understand AI-assisted code
• Scaffolded critique-of-AI-output assignment sequence with paired verification of individual understanding
• Pair AI-permitted take-home assignments with supervised modifications, oral defense, prediction, and transfer tasks to verify understanding
• Redesign assessment toward supervised, oral, process-based and authentic formats to verify students' own reasoning under GenAI
• Evaluate GenAI declaration frameworks empirically and pair them with institutional guidance on which tools are in scope
• Use a six-question drafting and audit framework for GenAI assessment rules covering task rules, defaults, disclosure, contribution, verification, and hierarchy
• Stage GAI use in assessment: permit exploration, restrict independent reasoning, require justification
• Assessment design levers that make learning visible and reduce the payoff from generic generated text
• Calibrated disclosure templates scaled to risk of the permitted use
• Two-instrument assessment: AI-free coding quizzes plus verification-gated oral defenses of comment-stripped AI-assisted work
• Institutions should define generative-AI prohibitions by function, distinguish generation from format conversion, and require validated evidential methods before sanctions
• Require process transparency artefacts: process logs, prompt records, reflective commentaries and oral defences
• Provide AI-free proctored spaces so some take-home work is verifiably unassisted
• Apply four operational criteria—fidelity, non-augmentation, traceability, attestation—to identify transcription-only AI use
• Prevent AI-assisted plagiarism through student education, signed declarations, draft submissions, clear guidelines, and close monitoring
• Mitigation strategies for GenAI challenges in PBL (validation, AI-free periods, equity monitoring)

Examples¶

-

Key Sources¶

• Victor Angelier. (2026). The pitfalls of AI detection in academic writing: bias, false positives, and the need for inclusive assessment. AI and Ethics. https://doi.org/10.1007/s43681-026-01376-w