Universities and student groups are raising concerns over proposed changes to Turnitin’s user license agreement that would allow the company to use student work to train AI models.
In a statement to The Journal, Queen’s said instructors use Turnitin Feedback Studio, a popular plagiarism detector, to check students’ work for similarity scores and that students retain the copyright of submitted work under the current agreement.
The University said Turnitin notified it in July that it wouldn’t change the terms of its licence before Sept. 1, 2027, and would provide at least 30 days’ notice before any future changes. The University continues to monitor the issue internationally.
In August, the University of Southampton announced it would immediately phase out of their contract with Turnitin and would not review its agreement after the 2026-2027 academic year.
The issue was over a proposed change to their end user license agreement (EULA) “that would have allowed it to use student submissions to train AI services,” a spokesperson for the University told Times Higher Education.
According to the BBC, several student unions across the United Kingdom, including those at the University of York, Lancaster University, and the University of St. Andrews, have reacted to the proposed change due to concerns about intellectual property rights and lack of meaningful consent, as students must agree to the EULA to submit their work.
Approximately 700 students responded to an Instagram poll by The Lancaster Tab, a student newspaper at Lancaster University. Of those respondents, 95 per cent of respondents opposed Turnitin using student work to train its AI services.
A spokesperson for Turnitin told the BBC that any changes to its policy have been postponed until September 2027. Following the backlash, Turnitin launched a “Trust Centre” on its website to bring “transparency to AI product design and data use,” on Sept. 24.
The webpage compares AI classifiers and generative AI models and explains how Turnitin trains its in-house AI models. It also has a list of all AI and AI-powered features in Turnitin products.
AI Classifiers models categorize or label existing data and are used to assign a numerical score indicating how likely a piece of writing is to be AI-generated. Generative AI, including Large Language Models (LLMs), on the other hand, can “create new content including text, images, audio, code, video, etc.”
The page states that “We do not train any generative AI models, or LLMs. We only train a classifier model for AI detection in our Turnitin Originality license.”
Turnitin Originality uses an “LLM-powered text classification model” to detect AI-generated or AI-modified text, alongside another in-house model that assesses authorship by comparing the text with the user’s previous work. It is available as an add-on to several Turnitin products.
Turnitin uses human written and AI-generated writing datasets to train and evaluate its AI-detection model, so it can learn to tell the difference. The human-written writing is “anonymized and pseudonymized” and drawn from Turnitin’s databases, from different locations and subject areas to minimize bias.
Institutions can disable AI writing detection in Turnitin Originality or configure private repositories to store student work outside the standard Turnitin database, preventing its use in training AI detection models, according to the company.
Turnitin’s AI-writing detection tools were launched in April 2023, following the release of ChatGPT, with mixed reactions.
According to the company, the AI writing detector has a false positive rate of “less than 1% for documents containing more than 20% AI-generated content,” meaning that one out of every hundred human written documents is falsely flagged as AI-written.
It cannot be used to definitively prove academic misconduct but rather provides data for educators to make an informed decision based on institutional policies, according to Turnitin.
AI-detection tools have faced significant criticism for unreliability and bias against people whose first language isn’t English. Several universities strongly discourage or outright ban the use of AI detection altogether, including Queen’s.
“The use of third-party AI-detection tools is strongly discouraged due to their unreliability and the potential breach of student privacy and intellectual property rights,” reads a statement regarding the University’s continued approach to generative AI in education from July 2024.
Tags
All final editorial decisions are made by the Editor(s) in Chief and/or the Managing Editor. Authors should not be contacted, targeted, or harassed under any circumstances. If you have any grievances with this article, please direct your comments to eic@queensjournal.ca.