BEYOND DETECTION: A HUMAN-CENTERED APPROACH TO INTEGRITY IN THE AGE OF GENAI
TL;DR
This preliminary study explores a human-centred approach to assessing suspected GenAI-assisted writing and examines how human-led evaluation practices offer a more context-sensitive alternative to automated detection tools.
Abstract
Tensions surrounding academic integrity have intensified as generative artificial intelligence (GenAI) tools become increasingly sophisticated and widely available. Traditional language assessment practices were not designed for contexts in which students can easily generate highly polished written work using AI, yet existing GenAI detection tools have been shown to be unreliable, often biased toward particular writing styles, and vulnerable to false accusations. As a result, educators are left in an ambiguous space, where they must make high-stakes judgments about the authenticity of student writing without accurate detection tools, and in the absence of clear institutional policies, inconsistent responses can inadvertently create opportunities for misuse. Research underscores the importance of explicit, transparent guidelines that specify acceptable uses of GenAI. This preliminary study explores a human-centred approach to assessing suspected GenAI-assisted writing. Using a dataset of post-secondary students’ essays collected through a digital language assessment platform and AI-generated essays produced by ChatGPT 4.0 and 5.0, the study examines how human-led evaluation practices offer a more context-sensitive alternative to automated detection tools. The analysis also considers how clearly defined policies could support the prevention of student misuse and ensure more consistent, equitable decision-making.
