How Ontario professors are upending the way they grade students - Toronto Star

Five university professors explain the new ways they're using to gauge students' understanding, and where AI fits in.

Stuart Chambers, a professor of sociology at the University of Ottawa, says in trying to detect the use of AI, “I was becoming too much of a police officer and less of a professor.”

There will be no take-home essays in Stuart Chambers’ university courses this fall.

Tired of trying to detect AI usage, the University of Ottawa sociology professor is replacing the traditional means of assessment with an in-person 75-minute handwritten paper. “I was becoming too much of a police officer and less of a professor. I want to go back to being a professor.”

Chambers’ dilemma is playing out in classrooms everywhere. Since ChatGPT’s public launch in 2022, generative AI has rapidly gained traction: Last fall, 73 per cent of Canadian students reported relying on it for schoolwork, up from 52 per cent in 2023. And while some educators have responded by banning the technology, just over half of Canadian post-secondary students say they are permitted to use it.

Meanwhile, cheating scandals have come to dominate the discourse, from a Western University professor who sought to invalidate an exam over suspected widespread AI misuse, to an American prof whose ruse to hide the word “Madagascar” in a midterm question revealed nearly all students had fed the query to a chatbot.

As educational institutions grapple with how to harness AI ethically without sacrificing critical thinking, classrooms have become petri dishes for a period of intense experimentation in teaching and assessment.

“Our credibility as a university is dependent on how accurately we can assess grades,” says Chambers. “And if we’re not accurate in assessing grades, the diploma is worthless.”

Here’s how five Ontario professors are transforming their practice.

Kyle Smith, a professor in the history of religions program at University of Toronto, has returned to the oral exam in light of increasing use of AI by students.

For the first 15 minutes of each class, University of Toronto Prof. Kyle Smith conducts what he calls “proof of human thinking.”

He hands out a question, typed on paper, based on that week’s reading materials. He recognizes students may use AI to study, but once in his lecture hall, there’s no hiding behind a screen. Students “have to show up in class and, without any sort of device other than a pencil, be able to respond in a coherent way.”

Instead of banning AI or grudgingly tolerating it, the history of religions professor has redesigned how he measures student understanding. “A lot of colleagues maybe still have their heads in the sand and think we can do the same things,” he says, but AI is “impacting post-secondary education, particularly in the humanities, in ways that are unwelcome.”

Instead of a written final, Smith now conducts a “viva voce,” an oral exam in which students meet with him and roll a 12-sided die three times, randomly selecting three of the course’s 12 weeks of topics for discussion. Students must also pose one hard question to Smith to demonstrate they have synthesized the material.

“The vast majority of students have never had this sort of assessment,” says Smith. “The pressure to perform live is a little scary … but the feedback I have consistently gotten is, ‘I was really worried about this; it turned out I was super-prepared, eager for the conversation and it went really well.’ ”

That final is worth 20 per cent; the in-class questions a total of 40. The remaining grade consists of two multilayered, sophisticated take-home assignments in which AI is not just permitted but potentially necessary, enabling students to create such things as a religious shrine using digital tools.

But as AI improves, the balance between human and machine keeps shifting: “I’m reacting to the changes I’m seeing,” Smith says, noting he may need to pivot again this fall. 

Sharon Ferguson, assistant professor in the management science and engineering department at the University of Waterloo, found that having students do group projects resulted in less AI-generated content.

When Sharon Ferguson put her first-year students into randomized groups for projects, she noticed something striking: their finished work contained almost no AI-generated content.

Not because they couldn’t use it. But students who might have happily thrown the assignment into a chatbot found themselves working alongside peers who wanted to do their own thinking. The group kept itself honest.

The result confirmed something Ferguson, an assistant professor in management science and engineering at the University of Waterloo, has already built into her practice: the most effective response to AI isn’t restriction, it’s design.

“A large majority of our students go into tech, and the philosophy in most tech companies is use as much AI as you can, gain as much efficiency as you possibly can with this tool … So, it doesn’t make sense to say you absolutely cannot use it.”

As a researcher of human-AI interaction and in her second year of teaching, Ferguson has always inherently thought about generative AI. Her approach is to teach responsible and effective use while demanding transparency, asking students to sign a declaration and show their chat history. She has also built a grading rubric tied to the exact terminology and concept applications she uses in lectures, making generic AI work easier to spot. And she introduced frequent quizzes with thought-provoking questions that can’t just be memorized.

In foundational year courses, she still does as many in-person, hands-on activities as possible: “If we never let them become experts because they’re never learning, because they are just using AI all of the time, then that’s a very dangerous future.”

Rahul Kumar, associate professor in the department of educational studies at Brock University, has started interviewing students as a way to better assess student knowledge in the age of generative AI. 

Rahul Kumar is teaching the teachers. And there’s great responsibility in that.

“Without themselves knowing (AI), how can they guide their students?” says the associate professor of educational studies at Brock University.

When Kumar’s research revealed self-disclosed AI use among post-secondary students had risen from 74 per cent in 2023 to 94 in 2025, he found himself at a critical juncture: “How do we ensure the degrees we are conferring actually ensure people are learning and not subcontracting the work to generative AI?”

His answer was to overhaul his teaching and assessment practices, without banning AI. “It is our responsibility to prepare our students for their future, not our past.”

He asked students to disclose their usage, with no repercussions, and added 15-minute interviews. “If you can have an intelligent conversation and demonstrate your learning, that ought to be sufficient.” 

He likens his approach to “Swiss cheese,” accepting that any single assessment may have holes where students turn to AI. But by stacking several assessments — an outline, a conversation, multiple drafts, a final paper — “you get a relatively solid piece of cheese,” he says, giving him greater assurance his students can apply their knowledge.

This fall, he also plans to introduce debates, grading students on their position and how they listen and respond in the moment.

But these solutions are time-consuming and not always practical, workable for Kumar’s small graduate student cohort, but for a class of 40, “probably not.”

His ongoing research is only deepening his resolve: He’s finding some Ontario teachers personally use AI but ban it in class. “The duplicity of that further gives me confidence that these steps I’m taking are worthwhile,” Kumar says. “Changes in our practice are a must.”

Emily Kuang, an assistant professor in electrical engineering and computer science at York University, thinks the focus shouldn’t be about “how do we make our assessments AI-proof, but how can we make our students more AI- literate?”

When students in Emily Kuang’s third-year course in user interfaces built their own websites with AI last winter, the project average was over 90 per cent. Then came the closed-book, in-person exam, and grades dropped to the 60s and 70s.

That gap was hard to ignore. So this fall, the York University assistant professor of electrical engineering and computer science is making two changes.

Kuang, whose research focuses on human-computer collaboration, says her graduates will be expected to use AI on the job, so the question shouldn’t be “how do we make our assessments AI-proof, but how can we make our students more AI-literate?” Still, she is expanding the declaration form students submit with each assignment, adding more detailed questions about fact-checking and whether results matched what was envisioned. She also plans to share the grade discrepancy with new classes as a warning that “you still need to learn the material.”

She believes foundational coding and debugging skills still matter and need to be mastered without AI assistance, but in the upper-year courses she teaches, the focus shifts to application and judgment. It’s a distinction that shapes her philosophy: “Humans are not going to be replaced by AI, but humans are going to be replaced by humans who know how to use AI. So, the viewpoint I take is we need to prepare them.”

Assessment also looks different depending on class size. In her fourth-year multimedia technology course, with just 20 students, she replaced the closed-book test with in-person presentations. In her third-year course of 120, that wasn’t possible. “The scalability issue is a big one.”

Stuart Chambers of the University of Ottawa said compared to 2023, suspect submissions among students’ take-home essays increased eightfold this winter. 

The first red flag was the sudden appearance of near-perfect papers from a couple of students who had failed the earlier multiple-choice midterm. It was a pattern Stuart Chambers hadn’t seen before. Usually, grades dropped on the take-home essay “because writing is a harder skill,” says the University of Ottawa professor.

That was 2023. This past winter, suspect submissions increased eightfold. When Chambers confronted the students, most couldn’t remember a single detail of their written arguments and admitted using ChatGPT. “The only thing valid on the paper was their name,” he says. “By doing nothing, we’re complicit in cheating, so I decided I had to do something.”

This fall, his students will write closed-book essays under his watch. Chambers spent months testing the format, gauging what students can realistically communicate with pen and paper in the allotted time. He has reduced the weighting from 40 per cent to 30 to account for constraints.

He believes the approach will preserve the writing and critical-thinking skills central to a university’s mission. Worried about AI’s effect on reading too, he introduced a bonus assignment last year, offering five per cent to those who read a book from a list and sparred with him for 30 minutes. It worked well and was manageable with 10 per cent of the class taking him up on it. 

Chambers will be emailing expectations to his incoming students: when it comes to assignments, they’re on their own.

“This,” he says, “is the class where AI goes to die.”

Janet Hurley is a Toronto Star journalist and senior writer covering culture, education and societal trends. She is based in Toronto. Reach her via email: jhurley@thestar.ca.