(posted on behalf of @Sergio Rodriguez-Garnica)
For many PhD students starting to teach, AI can feel like a bit of a headache. Here’s how I’ve been approaching it (just in case it gives you some ideas).
In individual assessments (like exams), limiting AI makes sense. You want to be sure students are building the fundamentals themselves.
But once they’re working on assignments, projects, or presentations at home, that approach becomes harder to enforce. Realistically, they will use it anyway. In those stages, AI can actually be useful.
When students are working on assignments, I try to encourage them to use AI as a kind of sparring partner. They upload their work and ask it to challenge them: point out weak arguments, question assumptions, suggest tough questions. It shifts the focus from just finishing slides to actually being able to defend their reasoning.
Then in class, I do what they should have already done themselves. I ask AI to generate difficult questions about their work and critique it, and I pose those questions directly. If they’ve prepared well, they can answer. If not, it becomes obvious pretty quickly, and that usually sticks for next time. In my experience, once you do this once, they expect it the next time and come better prepared.
Assessment then becomes less about the final output and more about whether students can defend their work and respond to questions they should have anticipated. That’s usually where you see if real learning has happened.
Used this way, AI doesn’t reduce effort, it just shifts it. Less time drafting, more time thinking, questioning, and defending ideas. And that feels much closer to what we’re actually trying to teach.
Curious to hear what you think.