Humanities academics are not addressing AI where it matters – in assessment, according to Gonzalo Villanueva (Uni Melbourne) who reports, “the main governance challenge lies less in declaring policy than in redesigning assessment at scale.”
Four years after ChatGPT hit the market, he warns assessment in Group of Eight arts and humanities courses, “remain highly exposed and vulnerable.”
“The central finding is straightforward: even in Australia’s most well-resourced universities, published assessment remains heavily concentrated in unsupervised, writing-intensive formats whose conditions of completion have been transformed by advanced GenAI systems.”
“Substituting a viva or supervised in-class task for a research essay is not an equivalent swap of instruments; it alters what learning can take place and what students can subsequently demonstrate.”
Nor is the challenge amenable to edicts.
“Universities may articulate sophisticated positions on acceptable GenAI use, disclosure, detection, and academic integrity while leaving practical redesign to unit coordinators working within inherited assessment templates, curriculum approval processes, workload constraints, and uneven local support.”
Dr Villanueva analysed 15,587 annual Arts and Humanities unit records and 53,915 assessment items across Australia’s Group of Eight universities from 2022 to 2026 to create AI Vulnerability Index that measures how Gen AI could (not necessarily did) assist with a specific assessment task. Scores are set separately for each year, according to changing AI capacity.
Rather than a way to check for AI, he suggests the Index can “make GenAI’s role explicit and ensure there is enough evidence for the claims attached to grades and accredited outcomes.”
“Treating GenAI vulnerability principally as an enforcement problem misreads the issue at this scale. Detection, disclosure, and misconduct adjudication remain part of governance, but they cannot carry reform where exposure is widespread and embedded in everyday assessment practice,” he writes.
He finds five assessment types most at risk this year take-home essays, research reports/literature reviews, unsupervised on-line quizzes, reflective journals/essays and unsupervised creative writing/ digital artefacts.
But while Dr Villanueva does not quote Corbin et al’s celebrated AI assessment as wicked problem paper, it seems he shares its potent mix of realism and optimism, “the consequential work lies in redesigning assessment items, programme structures, and verification practices so that the evidence universities certify matches the learning they claim to assess.” .