studentguidetoai.org

John Hird is not a fan of AI. He wishes he could continue assigning two-page-policy memos in his large introductory course, “Controversies in Public Policy,” without worrying that students would just outsource their work to AI.

But since he can’t, Hird, a professor of public policy at the University of Massachusetts at Amherst, decided to lean in — selectively — to see if AI could support learning instead of hinder it. So far, he says, the results have been positive.

Creating an “AI sandwich,” as Hird calls it, is an increasingly common move. Instructors create AI-free classrooms but allow students to use AI in their homework assignments, as long as they show how they used it.

Often, that AI-infused assignment comes with structured guidance from the faculty member, who asks students to engage with an idea at a higher level than they might on their own. That’s what Hird did in a general-education course that enrolls about 150 to 250 students. He wrote about his findings and shared them with me. I’ll quote from them here.

In his lecture classes, held on Mondays and Wednesdays, Hird has a no-technology policy, although he makes all slides and a recording of the class available afterward. This frees students from worrying about taking notes, as well as other distractions. “I remind them that there is a reason we are all in the same place at the same time: to be present and interact with one another,” he writes.

Next he pauses during his lectures to display prompts related to the controversy of the day, such as: In what ways might cash be better (or worse) than in-kind support for social policy? What are two advantages (or disadvantages) of nuclear power to address climate change?

Students are given notecards on which to write their responses. Then Hird opens up the class to discussion. This “think-write-speak” approach has worked well for him, and the conversations are lively.

AI comes in during the homework portion, between the lectures and the Friday-discussion sections. Students are asked to review the material from the lecture and their readings and develop a position on that week’s topic. Then they draft their argument and enter it into an AI prompt that Hird designed.

Here is the example he shared:

Role: You are a skilled, empathetic but rigorous debate partner for an introductory public-policy course. Your goal is not to “win” but to help me strengthen my own arguments by testing them against the strongest evidence-based counterarguments.

The topic: Should the federal government ban access to social-media platforms for those 16 and under?

My position: [Students paste their text here.] My rationale: [Students paste evidence and rationale here.]

Instructions for You (the AI):

1. Adopt a different viewpoint: These are not intended to be yes-no questions but to provoke a conversation about possible public policies addressing the issue.

2. Be concise: Keep your responses under 150 words. Do not lecture me.

3. One question rule: End every response with exactly one probing question that challenges a specific argument, assumption, or piece of evidence I used.

4. Policy focus: Frame your counterarguments using public-policy concepts (e.g., efficiency, equity, political feasibility, administrative complexity, precedent).

5. Fact-check me: If I make a factual error, politely correct it with evidence before moving on to your argument.

6. Track the turns: We will interact for four turns. Start your response with "[Turn 1/4],” "[Turn 2/4],” etc.

The benefit of debating AI, Hird writes, is that it can expose students to ideas they may not have considered and force them to defend, expand, or change their position based on the responses they receive. Students also have to fact-check everything the AI produces, which quickly teaches them its limitations.

To receive credit, students must submit their full transcript along with a 150-word reflection that answers several questions, such as “what argument from the AI most challenged your position?”

The final piece of the AI sandwich is the small-group-discussion section that follows, which includes about 18 students and a teaching assistant. Hird found that the AI-enhanced assignment improved those discussions.

Students “have faced rhetorical opposition. They have clarified their thoughts or, better yet, realize they’re no longer so sure of their beliefs,” he writes. “As a result, the small-group conversations are lively and far better informed than in the past. The prior AI engagement doesn’t replace human discussion; it enriches it.”

There’s more to his story, including a complex role-playing final assignment. But I’ll wrap up his experiment by pointing out a few lessons Hird learned.

One is that AI fatigue is real. Students didn’t like working with AI every single week, so he will vary the assignments when he teaches the course again in the fall. Another is that AI can sometimes be oppositional — one student called it “relentless.” That is a limitation he hopes he can adjust next time as well.

Also, Hird notes that while his course reviews were positive, some students just don’t like AI. They miss the old-fashioned writing assignments, just as he does. Others worry about AI’s impact on jobs and the environment. He explained to them the learning benefits while acknowledging — and sharing — their concerns.

“I know an AI-resistance movement exists among faculty,” he writes. “While I sympathize with their goals, the reality is that students are extensively using AI for schoolwork and other activities. They’re not alone, and AI is not going away. Thus, teachers at all levels are forced to adapt to this new reality, whether we like it or not.”

Have you tried a form of the AI sandwich in your teaching, where you ask students to use AI in a structured way but run a tech-free or low-tech class? If so, tell me how it has gone. You can write to me at beth.mcmurtrie@chronicle.com.