In an age of artificial intelligence, educators across the University of Toronto are encouraging students to work with the technology - not to obtain easy answers, but to teach themselves how to ask better questions.
U of T students are expected to complete assignments without outside assistance unless instructed otherwise. However, several professors say certain AI tools can have a role in enhancing learning by acting as round-the-clock personal tutor, a sounding board for half-baked ideas or a patient conversation partner for language learners.
The tools can prompt questions, provide feedback and help students puzzle through problems, sharpening their thinking rather than supplanting it. The harder part is training students to assess what comes back. In other words, the real test is whether students can hone the skill that sets them apart: their judgment.
U of T News spoke to four professors about how they're tackling the technology this school year.
Rethinking what counts
Safieh Moghaddam (supplied image)
The AI response Safieh Moghaddam put in front of her linguistics class read like it could have been cribbed from a textbook. When she pointed out it was completely wrong, her students' jaws dropped.
"They see how AI responses are so polished and so confident, and they think that polished means correct," says Moghaddam, an associate professor, teaching stream, in linguistics at U of T Scarborough. "But that doesn't necessarily mean the reasoning is good, or that the student understands it."
Puncturing that assumption is one aim of "Teaching in the Age of AI," a toolkit she developed for instructors. It started from a question she kept hearing, and asking herself: Should students be allowed to use AI or not?
The problem was more complicated than whether to ban AI or embrace it. She had to rethink her own assumptions about learning - and let students be part of the answer.
"I've moved away from policing and detecting," she says. "I want to have open conversations and give students a say in their own learning."
In her upper-year courses, that starts on day one, when the class co-creates the AI policy they will be held to all term. Last winter, they settled on allowing AI for brainstorming, structuring and editing - but the underlying ideas had to be their own.
That distinction is partly why Moghaddam is moving away from assessments that judge only the final product. She sometimes asks students to submit outlines, early drafts and reflections that show the thinking that went into their work.
"It's about making selected parts of the learning process visible," she says. "The technology will change. What matters is helping students continue to think critically, evaluate information, make judgments and use AI in ways that support those abilities."
A 24-7 tutor
Tovi Grossman (supplied image)
Tovi Grossman built an AI tutor that can help with any question his computer science students might have about the course. But it won't give them the answer.
Instead, LearnAid breaks the question into steps and puts the thinking back on the student, working from the course's own lecture notes and materials rather than the open internet.
"It helps guide the student towards the answer themselves, much like a TA or professor would during office hours," says Grossman, a professor of computer science in the Faculty of Arts & Science.
LearnAid's design is rooted in Grossman's research on what happens when students hand their problem-solving to the AI instead of doing it on their own. Students who copied and pasted AI-generated answers did measurably worse on final tests, but those who used AI tools to work through problems scored higher than peers who hadn't used AI at all.
The system is meant to reach students who might be reluctant to seek extra help in person. In one pilot, women made up a minority in the class but used the tool at roughly twice the rate of the men.
Grossman is also preparing his students to enter a field being rebuilt around AI.
"Workplaces are going to expect graduates to understand how to use AI tools," Grossman says. "The more we can expose students to using AI responsibly, the better."
Learning how to look
Wei-Han Vivian Lee (supplied image)
Associate Professor Wei-Han Vivian Lee knows that AI can produce more images in minutes than architects can draw by hand in a week. But in her view, more options don't necessarily mean better ideas.
"I'm trying to warn students about what it means for there to be an abundance of imagery coming at you and a scarcity of judgment to match it," says Lee, director of the master of architecture program at the John H. Daniels Faculty of Architecture, Landscape, and Design.
Students in any of her courses are welcome to use AI as a study partner to clarify a concept, troubleshoot software or talk through ideas. But first-year students aren't allowed to submit work created by AI since they don't yet have the skills to see its distortions - and how that shapes what gets built.
"All drawings are loaded," Lee says. It takes time and training to scrutinize what images emphasize, what they erase and whose interests they serve.
Once students have that grounding, the rules shift. With Lee's approval, third-years can use AI in their design process, but they must account for why and own the result.
In her upper-year studio, Lee has her students act as editors, sifting through dozens of AI-generated visualizations to pick out the ideas worth salvaging while discarding the rest.
Lee expects her students will be using AI for the rest of their careers. But she worries the technology is accelerating a trend: glossy renderings that show a sunlit building but not what it's trying to do. Her goal is to equip students with a critical eye to tell the difference.
"The most precious thing they have is their judgment," she says. "Even when AI is a tool, the student is the decision maker."
Hola, AI
Pablo Robles-García (supplied image)
For many language learners, the hardest part isn't memorizing verb tables. It's finding the confidence to actually speak. Pablo Robles-García, assistant professor of Spanish at U of T Mississauga, says AI can help students practise with a fluent, non-judgmental partner rather than stumbling through conversations with fellow beginners.
One challenge in language classes is that students' skills can vary widely within a single course, which makes it hard to pair students with someone at their level. Microsoft Copilot lets them carry on conversations at their own pace, adjusting to their proficiency and waiting patiently as they search for the right words.
The only drawback, Robles-García jokes, is that AI "doesn't know when to stop" - it keeps the conversation going long after a human classmate would have called it quits.
He also uses Copilot to support writing. Using tailored prompts, students feed sentences into the tool and receive comments on the type of issue - grammatical, structural or practical - rather than a ready-made correction. They revise and resend until they reach a final version, and then fold those exchanges into a portfolio he reviews during the term.
The approach gives students instant feedback, but not without flaws. Most models are primarily trained in English, Robles-García notes, so AI is more prone to "hallucinate" in Spanish, insisting on corrections when it's actually in the wrong.
"AI can help you, but the learning has to come from you."
