This post is the second half of an essay I drafted last spring. The first part, on overarching questions about writing process, can be found here. This part discusses practices I tried in my high school classes last semester, as well as some thoughts moving forward.
If one purpose of the essay in its original form, before it was reduced to assessment and grades, was to think through ideas, then perhaps this is what educators (myself included) should be leaning into as we consider our response to generative AI.
Right now [spring of 2023], I’m experimenting. And I’m asking my students to experiment with me, hoping this will make them feel as though they have a say in the direction of writing in an age of AI. Earlier this semester, I gave my AP Literature class a creative writing prompt. We had just finished a unit on Modernist poetry, so my plan was for students to write their own poem and then analyze three literary devices they used. I found that ChatGPT, given the prompts from the assignment, produced an adequate (if dull) high school-level poem. Its analysis of metaphor, simile, and imagery in the poem it created were also decent.
And so I asked myself why I cared about students writing the poem themselves. Well, I wanted them to play around with language. To think of what they wanted to say and search for the words and images that would help them say it. Or to let words and images meet on the page and ask themselves what the heck they’d just written. I wanted the ones who were obsessed with author intentionality (Did Langston Hughes really mean that? Why did Emily Dickinson put those dashes and capitals in her poem?) to see the messiness of writing. I wanted them to experience how meaning could unfold as they wrote and revised.
So I added a component to the assignment: “Describe your thought process, including dead ends and decisions to revise.” I modeled this for them aloud in class, imagining starting with one prompt and disliking it and moving to another. Looking out the window and pulling imagery into the poem from what I saw. Substituting one word for another one with a better sound or more exact connotation.
I ran this thought-reflection prompt through ChatGPT, asking it to write as though it were a 17-year-old high school student, and got a bland paragraph that lacked detail and read nothing like the inside of even a dull human mind, much less the idiosyncratic personalities of my students.
I’ve been reading through their poems slowly and it’s been oddly enjoyable—odd because while I love teaching, I loathe grading. But this is different. Sometimes the poems themselves are particularly well-written, sometimes they’re not. But regardless of the quality of the poem, the thought-process part of the prompt gives me a window into its composition, all of the hidden work and time that doesn’t appear on the page.
The thought-process reflections made me confront some of my own biases as a teacher. When I read a poem filled with cliches and simple language, I make a series of judgements. The student didn’t spend much time on it. The student didn’t think about the words or image. The student didn’t revise.
The reflections let me see that even students who have a weaker vocabulary or don’t possess the knack of creating unique comparisons can still ask some of the questions that go into a good piece of writing: What word fits best here? Why am I making this comparison?
In short: even if the end product is lacking, the process can be rich and meaningful.
Adding the narrative component to the assignment makes it harder for students to use ChatGPT to write and analyze their own poem. But it’s more than an added step to guard against cheating. It’s a way of shifting our focus from the end product, a way of saying: how you get there matters. In fact, it might be the most important part.
I’m experimenting again at the moment. After formally introducing AI and ChatGPT to my students in March, we started a research unit, which I’m asking them to write as a narrative, starting with their research question and tracing their thought process as they work through a variety of sources (and yes, ChatGPT is on the table as a source—though after seeing its ability to hallucinate and confidently produce fake sources, their interest in using it for this particular assignment has been low). In many ways, it’s a reverse of the typical research paper, which starts at the end of the research process with one’s conclusions and then spends the following pages drawing from sources to support them. The narrative approach emphasizes that one’s conclusions are constantly being refined and challenged, and whatever final conclusions one has are strongest at the end of one’s research.
So far, it’s been messier than the poetry assessment, and I’m not sure if, in the end, it will be a workable model: Will it allow students to embrace the non-linear aspect of the research process and trace how their response to their research question changes over time? Will it give me a clearer window into how well a student understands, not just the topic they’re researching, but also how to find and think about sources? Or will it lead them to more muddled thoughts and tangled reasoning? Will it ultimately be a more tedious form to both write and read?
I’m not sure.
Responding to Chat GPT will also be a process. A long, creative process of trial and error. We will try some things that will not work. We will try some things that work in unexpected ways. We will likely try some things that have unforeseen consequences, good or bad. So we’ll adjust. Tinker. Reconsider.
The process should open us up to broader questions: What, in the end, is a liberal arts education for? Why do we ask students to write—still? What other assumptions of education have become invisible to us over time? And what can we do to prompt ourselves to see them once again?
These are questions I’m still thinking through, as I wrap-up my first semester teaching undergraduate writing at BGSU. And they’re questions I still have for secondary education too. The narrative research papers I tried last spring were somewhat of a flop: students were already familiar enough with research-paper writing that they defaulted to old forms, and I found the narrative ones somewhat more cumbersome to read. It reminded me that there’s a reason we often shed the steps of our process when we get to the end: not because the process didn’t matter, but because not every misstep is an interesting one.
The poetry projects, though, I still regard as a success. They opened my eyes to students’ process in a new way. Perhaps they worked better because the narrative of their process was a distinct component from the poem. Or perhaps because narrating creative choices is easier than tracing one’s research process.
I do still think that there’s a place for narrating process within a final paper, but I’m still not sure what form that should take. In lieu of that, practices like student self-evaluations, reflections, and process logs can place emphasis on writing process alongside of the final product. Next semester, as I teach a first-year research course, I’ll be experimenting with ungrading, to see how shifting my grading practices might de-emphasize the value placed on end-product. As with my previous experiments, I have a lot of reservations about the results. But I think that being open with students about the experimental nature of what I’m trying—what our class is trying, collectively—will be particularly relevant and meaningful in a research course, a way of emphasizing to students the role that ongoing research plays in my own practice as an educator.
In sum: I remain uncertain about how to engage with AI in the classroom, but certain that talking to students about it, voicing that uncertainty and encouraging them to pose questions too, is worthwhile. This, after all, is what my favorite classes did when I was a student: ask big questions and grapple with their elusive answers together.