How to teach college #23: AI proficiency

I wrote this twenty-five months ago tomorrow, and a lot has changed since then. This term it's my goal to create, in every class I teach, an opportunity for my students to use AI apps. I want any of them who've been shying away to have a first encounter and get a feel for how they work and what the possibilities are. The rest of this article lays out what I've deployed so far.

Chiefly I can report that my old way of assigning and grading papers, described here, is dead, and I'm piloting its replacement in my upper division listening class. Their assignment is to do a literature review, which means they pick at least three studies from peer-reviewed journals and follow a single idea through the findings of the three, showing how the idea developed, was fleshed out, was redefined, or in any other way was advanced by the studies. Until now, they turned this in as an essay, but at this writing we're in the middle of the next generation of the assignment.

First, I had them deliver their literature review to the class as an oral report. I gave them a recipe for what must be included in the oral report, and at the end, they fielded questions from their classmates and from me. This stage of the assignment is my opportunity to gauge how well they understood the studies, and how sound a job they did of tracking the idea they chose through the studies' contributions. But, importantly, as they delivered that oral report, they wore my wireless lapel microphone and I recorded them. I then made the recording available to them, and now stage two is underway.

Second, they will now generate a transcript. There are probably several online apps that can do this, but I've seen good results from Riverside.fm. Students will then go over the output very carefully and correct any transcription errors. They then save the result as a PDF on a thumb drive, along with the studies they reviewed, and they're ready for stage three.

Third, they upload the transcript and studies to the LLM of their choice. In class I demonstrated what I had in mind on ChatGPT, but I'm confident it could work on other platforms. I described to them the prompt I playtested with, but I did not give it to them verbatim, because I said I wanted them to grasp how to author a prompt, and I wanted them to play with their LLM, rewording and fine-tuning their own prompt until it did what they wanted it to. Here's what I used in my playtest:

I am a university student in an upper division communication studies class about listening. I was assigned to write a literature review. The topic I chose was [topic]. I read three studies and delivered an oral report. The oral report is in the PDF titled "[name]." The other three PDFs are the studies I read and reviewed. Please take my oral report and turn it into an essay. I have attached the studies to add context to what I said, but please make the essay match, in written form, what I said. Please also generate APA citations for the three studies.

I showed them that the first report it generated had a number of problems, so I had to point out those problems and prompt ChatGPT to generate the report a second time. I emphasized that they will be responsible for what they turn in, so they may need to prompt it repeatedly, pointing out the things that need fixing. They then give the paper another pass, adding details that matter, tweaking the wording to make it their own voice, and they turn it in, along with the corrected transcript of their report.

My big finish was to tell them that if they hear anyone say "Oh, I never write papers -- I just have ChatGPT do that," the person is announcing to the world that they're an idiot. If, on the other hand, they say "Oh, I never start papers on my own -- I always have ChatGPT do the first draft," then that's someone who gets it.

To date, that's the AI inclusion that I've built out the most, but I'm doing things in other classes as well. For my conflict class, I've taken all the archived columns from this feature and transferred them to minimalist PDFs. I let students pick a column that appeals to them, and in class we upload it to ChatGPT for interactive roleplays. The student gets to decide which party to the conflict they want to play, and in the prompt we ask ChatGPT to behave in particular ways so the student can practice specific conflict skills.

And in my public speaking class, earlier this week I turned on ChatGPT voice chat told it I was about to give an impromptu speech, and please give me a critique afterward, as soon as it hears me say the word "Finished." I then asked students for an impromptu speaking topic, which they supplied, and I delivered the speech, ending with "Finished!" ChatGPT immediately produced a quite reasonable round of feedback. It also helpfully supplied a teachable moment when it said I had plentiful and effective eye contact -- since the tablet wasn't turned toward me, and had no way of knowing the amount or quality of my eye contact, I reminded the class that LLMs say what will probably fit the specific situation, not necessarily what is correct.

I imagine some of my educator colleagues are shaking your heads in horror at some of what I've described. Your complaint might be that I'm giving up on teaching writing, and I suspect some of my ancestors from generations back might be comparably horrified that I can't start a campfire or butcher my own food. The inescapable fact is that these tools exist, and the students I'm teaching right now have, for the most part, no idea how to use them. And it's abundantly clear to me that if they still don't when they graduate, the obstacles to employment and career advancement will be forbiddingly steep. The oral report gives me confidence that they're reading carefully and synthesizing what they read, and the trick of converting one's spoken thoughts to written work is one I want them to have available in their bag of tricks.