It was a seemingly simple Grade 11 assignment. Design a presentation on healthy eating for kids. Sixteen-year-old Marissa got to work.
She did some online research, gathered the sources she planned to credit and then submitted a 308-word paper containing a wholesome meal plan that recommended apple slices and oatmeal. It was worth 15 per cent of her mark. But according to her school, those 308 words did not come from the mind of a student; they were generated by artificial intelligence.
“Your assignment has been highlighted as it does not meet the academic integrity standards we aim for ...
specifically regarding text generated by AI,” TVO ILC, Ontario’s largest online school, noted in an email to Marissa. “ILC takes academic honesty very seriously.” Marissa, who received a zero, insists she did not cheat.
“The accusation was shocking,” says the teen from Cobourg, Ont., whose last name the Star has agreed not to publish. “She’s scared of being accused of it again,” says her mother, Karin Lowe, who has demanded an investigation.
“To get falsely accused of something like this, that’s major.” But TVO ILC, which did not respond to specific questions about this case, insists in earlier correspondence with Lowe that its AI detection process is 98 per cent foolproof: “The decision is final.” In an era when students are newly exploring all that generative AI can do and educators are still trying to figure out how to handle this disruptive technology, such standoffs over academic honesty are proving inevitable, tricky and potentially damaging.
Since , students have embraced the use of deep-learning models to create original content — for schoolwork, and 82 per cent of those surveyed admit they’ve claimed such material as their own. In the face of that reality, some educational institutions have sought out an ever-growing offering of detection tools that boast an ability to analyze linguistic features and identify patterns to determine what is human-made and what has been spit out by a computer. But it’s a promise under scrutiny: .
Even due to “its low rate of accuracy.” AI detectors are a “super attractive fiction,” says , who is sympathetic to the desire of educators to go after what looks like an easy solution. “But to use that technology, knowing that you will falsely accuse students who did nothing wrong — because it’s mathematically inevitable that you will falsely accuse students — that’s just monstrous.
” Marissa left her bricks-and-mortar school at the end of Grade 9, driven away by bullying and social anxiety. In the fall of 2023, she enrolled in a publicly funded online learning partner of the Ontario Ministry of Education that grants high school diplomas and offers credits with no set classes or semesters. Marissa has since completed three courses and is doing final exams in three more, including “Raising Healthy Children,” the social science credit affected by the cheating allegation.
“She was excited to take the course because she can’t wait to be a mom,” says Lowe. “It is something that really interests her.” Generally a B student, Marissa had an 80 per cent course average heading into the healthy-eating assignment.
She submitted her paper for marking in early August. As per , the assignment was run through Turnitin’s text-matching software, colloquially known as a plagiarism checker, which compares uploads to a massive database of published work and calculates a similarity score. Since , it has been widely adopted.
The American edtech company says on its website that it serves 20,000 institutions worldwide. This includes Canadian universities, colleges and school boards, like the Toronto District School Board, where teachers have had the option to for at least a decade. TVO ILC says it has used Turnitin since 2022.
Last year, to its existing products, initially touting a 98 per cent confidence in flagged materials. Turnitin did not answer Star queries about whether Marissa’s work could potentially be a , but a spokesperson said the company “supports the responsible use of AI in education which is achieved through the visibility that AI-writing detectors can provide. We provide resources to help educators leverage our AI detector in a way that both honours a culture of academic integrity and protects students from undue harm.
” , there is a gap in school guidance about what responsible AI use looks like. The Toronto Catholic District School Board and the TDSB say they are developing policies. The TCDSB would not confirm whether its teachers use plagiarism or AI-detection software, but TDSB spokesperson Emma Moynihan said 3,000 teachers have active Turnitin accounts, and the board recently added on AI detection.
“We appreciate the ongoing dialogue around the use of AI as we evaluate its broader impacts on education,” says Moynihan. “We recognize we are in an evolving education landscape, and are committed to ensuring a responsible approach that takes into consideration input from experts and educators.” Meanwhile, post-secondary institutions and Canada, including the and , have already taken a stand against the use of AI detection.
But turning away from detection does not mean backing away from AI itself, says Daley, who points to the work of the as “The most fundamental lesson we can teach young people with this really sophisticated technology is: be an agent in your own life. Think carefully about what it is you want to outsource to AI and what it is you want to keep for yourself,” he says. “Use the technology to empower you, not rob from you.
” Karin Lowe with her daughter Marissa, 16, outside their home in Cobourg, Ont. Marissa was accused of cheating on an assignment for a course in TVO ILC, a publicly funded online learning partner of the Education Ministry. When Marissa got her similarity score back, it said 26 per cent of her paper matched already published material.
There is no specific threshold that written work is expected to meet, according to the TVO ILC website, and Marissa had received scores up to 20 per cent for previous assignments. But this time, Marissa also got a follow-up email from the school indicating AI detection had determined “100 per cent of the submission to contain unoriginal material.” She was given a chance to revise.
Marissa says she “dumbed it down” and corrected a few things: Originally, she had written broccoli “florists” instead of “florets” and had made a smattering of grammatical errors. “Honestly,” says Lowe, “if she used 100 per cent AI, it would have been way better.” For her second attempt, the similarity score came back at 13 per cent, but the paper was still deemed to be 100 per cent written by AI.
Marissa was granted a mark of zero with the warning that any additional occurrences could result in course withdrawal. “I know my daughter is not the first, nor will be the last student to be falsely accused of using AI, and I felt the injustice for her deeply enough to want to speak out,” says Lowe. A Turnitin spokesperson told the Star that “When concerns about false positives arise, educators should continue to engage in an open and honest dialogue with students, relying on their experience and judgment to determine what is or is not acceptable regarding AI use and original thinking.
” In an emailed response to Lowe last month, TVO ILC’s manager of academic services extolled the accuracy of Turnitin and said the assignment had also been reviewed by two teachers. A TVO ILC spokesperson told the Star that teacher markers review Turnitin results and “use their professional judgment to determine if there is an academic integrity concern,” and if there is one, it is reviewed by a TVO academic officer who is also a certified teacher. Of TVO’s 30,000 annual enrolments, fewer than five per cent of learners receive a first flag, according to the spokesperson, and roughly 0.
5 per cent have a subsequent occurrence. For Marissa, the cheating accusation initially chipped away at her confidence. Her mother feared she would be completely demoralized especially after what she had gone through in Grade 9.
“I thought it might put her off track,” says Lowe, “But I told her to make sure (she) finished the year off great” Marissa’s subsequent course work got her marks of 73, 92 and 100 per cent. Next month, after exams, Marissa plans to return to in-person classes at a new high school..
Top
This Ontario student accused of cheating was flagged by an AI detection program. But the software isn't always right
A Grade 11 assignment got a zero after the province's largest online school deemed it to be AI-generated. But experts say AI detection is inaccurate.