Andrew Burton is an English professor at Marianopolis, where he has taught business communication and literature since 2010. Earlier this year he published an article about his experience designing student assignments that integrate ChatGPT.

We’ve heard a lot about some of the problems AI has created or has the potential to create in the field of education in terms of academic integrity and plagiarism. What were the initial reactions to this technology among teachers and what convinced you it presented valuable learning experiences for students in your English class?

Many teachers very reasonably moved to entirely in-class assignments. Other teachers turned to AI detection software, although it eventually became clear that these tools aren’t all that reliable.

However, for three main reasons, I chose (and still choose) to authorize and train students to use ChatGPT to help them produce one writing assignment per course.

First, I don’t want students to lose the experience of writing at home without time limits, as practicing some skills isn’t really possible otherwise: e.g. analyzing slowly and deeply, planning elaborately, taking advantage of certain reference materials and revising extensively. Also, take-home assignments better resemble the writing contexts students will experience at other levels of academia (e.g. during graduate studies) and when writing both personally and professionally, and so take-home assignments help students transfer their skills.

Second, students’ use of ChatGPT doesn’t necessarily prevent them from learning. This is partly because large language models have significant weaknesses: they frequently produce factual errors, illogical statements, disconnected sentences, off-topic content, vague language and, especially, writing that doesn’t respect detailed guidelines. Such weaknesses reflect how large language models are what computational linguist Dr. Emily Bender calls stochastic parrots (i.e. amped-up autocompletes): they don’t really know, understand or think. Given the kind of deeply analytical and highly structured writing that I require, I determined that even with ChatGPT’s help students would have to inject a lot of their own thought and effort and so wind up learning. Another reason that students learn while using ChatGPT is that obtaining good output requires writing good prompts, and so students who are working to become strong prompt engineers are also working to become strong writers!

A third consideration is that students need help understanding large language models. These powerful tools were released without warning or a user manual, they’re freely and widely available, they’re rapidly being embedded in office productivity software (e.g. MS Copilot) and they’re easily misunderstood as being the superintelligent AIs of science fiction. In my view, it is therefore urgent for everyone to learn about the nature, strengths and weaknesses of large language models, and I feel that I’m ideally positioned to help my students do just that.

Were you worried that allowing students to use ChatGPT might somehow go wrong?

Yes, quite worried! I therefore hedged my bets: I’ve decided to only have one take-home assignment per course, and I reduced its weight from 25% to 20% of the course grade.

How have you prepared your students to use a large language model, notably in terms of harnessing its strengths and avoiding its weaknesses?

I’ve authored a guide that I assign as a course reading; it explains what large language models are, warns against their shortcomings, discusses best practices (e.g. for prompting) and finishes with examples of relevant prompts. Also, I’ve also developed a 100-minute class in which students gain hands-on experience. This is really helpful: even in October 2023, there were students who had never used ChatGPT before! Finally, I instruct students in employing the Modern Language Association’s methodology for citing “functional uses” of AI, which consists of providing a note explaining how AI was used.

Can you tell us a bit about how your students have used AI and the specific projects they’ve worked on? What are the takeaways?

I’ve now had students complete ChatGPT-assisted writing assignments during three semesters: Winter 2023, Fall 2023 and Winter 2024. In four sections, those assignments have been essays analyzing literature, and in the four sections of my Business Communication course, the assignment has been to produce a resume and cover letter.

In students’ notes citing their use of AI and in their survey responses, students report finding all kinds of ChatGPT use cases: brainstorming topics, explaining concepts, generating example resume bullet points, increasing concision, reducing redundancy, correcting grammar, and on and on.

My broad take-away is that my approach is working. More specifically, students are finding genuinely helpful uses for ChatGPT, and the result is that they receive minor to moderate assistance with some aspects of their work while still having to perform the bulk of it. So, as they always have, my students are learning lots about writing, critical thinking and literature, and now they’re also learning about large language models. Also, the grades for these ChatGPT-assisted assignments are quite normal, further validating my approach.

How have students reacted to using these tools in class or for assignments?

In a Fall 2023 survey, 80% of my students thought that being authorized to use AI was a good idea. When I asked them to explain why they responded as they did, students commented on how receiving this assistance helped them produce better work, how their acquisition of skills and knowledge was enhanced and how they liked being honest about their use of AI. Of the 73% of students who said they used ChatGPT, 88% of them reported that ChatGPT only helped them “somewhat,” “a little” or not at all, and having carefully read the assignments in question, I think those students are right.

Here’s one comment from a student that I particularly like: “What I learned about ChatGPT is that you cannot trust it at all. Therefore, you need to evaluate the answer you get to see if it’s valid/relevant or not and then, you have to make the decision if you should integrate this in your essay.”

Here’s one more: “It should be a tool that should be used more often, as it’s going to be a big part of our lives in the future. So thank you for teaching us how to utilize it.”

What are your views on the future impact of AI in academia and teaching? Is this something that you’ve discussed with colleagues? If so, what is the outlook within the profession?

The education world has only begun to digest the implications of large language models; for example, last fall, most institutions of higher learning still didn’t have a policy on AI use by students (humblebrag: Marianopolis has had such a policy since August 2023).

I’ve been sharing my approach to integrating AI both within Marianopolis and beyond; highlights have included publishing an Éductive article in May and presenting in October at an AI Theme Day put on by the Association des collèges privés du Québec. I’ve received a lot of positive feedback and perceived a lot of interest, I think partly because resorting to exclusively handwritten in-class assignments strikes many as a solution with significant shortcomings.

There seems to be an awareness that the extreme approaches of a) blanket bans on AI use and b) forcing teachers to integrate it are both counterproductive. Rather, each institution and, especially, each teacher need to figure out whether or not to integrate AI, and if so, how that can best be done. Accordingly, I suspect that increasing numbers of teachers will lead their students to learn both about AI and with AI in ways that are appropriate for a given educational level, discipline, assignment type, etc.

Do you have plans to continue or even expand the use of AI in your courses?

I don’t plan to expand my use of AI: I think having one take-home assignment that incorporates it strikes the right balance for me and my students, in large part because I want most of my students’ writing to feature words and ideas that are entirely their own. I plan to continue with my current approach until the day arrives when students can push a button and instantly generate a superbly written, deeply insightful and brilliantly argued essay. Of course, on that day, I may be too busy declaring allegiance to our robot overlords to worry about course design!