ChatGPT Writes Answers
Comments at the Pitt Faculty Senate Plenary on April 4, 2023 and in honor of Dave Bartholomae
I promise I say some things about AI here, but give me a minute to get there….
The following is a transcript of comments I delivered today as part of a panel on Generative AI, “Unsettled: Frames for Examining Generative Artificial Intelligence,” sponsored by the Pitt Faculty Senate and organized by the Senate President Robin Kear (flyer). It was a great array of Pitt faculty across the campus speaking on various aspects of AI. One key exigence of this talk is that the Chancellor and Provost are contractually required to be there the whole 2 hours(!). It’s also open to all faculty and students at Pitt.
What did I want to say to that audience? (with a big caveat that both our chancellor and provost are leaving their positions at the end of the year….) As I was drafting the remarks, I kept steering away from a more technical angle, away from my research on AI and writing, and toward my role as the Director of Composition here at Pitt. I kept thinking about first year writing. And when I think about first year writing, I think about Dave Bartholomae.
Dave was an endowed chair in Composition at Pitt, a giant in the field of Composition, a former chair of our English department (14 years!), Pennsylvania Professor of the Year (2014), a great colleague and mentor and teacher. His article, “Inventing the University” (1986), which describes the way first year students have to imagine the university in order to reproduce its discourse, is likely the most cited piece in the field of Composition. He was a big man in his thinking and his presence and his influence on the English department and the Composition program at Pitt. Yet he was kind and took time to be supportive of me as junior faculty. He taught first year writing throughout his career. In his last year teaching at Pitt, he requested to teach Workshop in Composition in order to return to the course and the students who helped to launch his career. He was still and always learning from students.
Here he is as a kids’ soccer coach on Mr Rogers’ Neighborhood in 1989, a role he was so proud of he had it in his faculty bio.
Dave passed away this morning. We knew it was coming—he’d been in hospice for a few weeks. I’d recently sent a note to his wife to read to him about how he influenced and supported me as a colleague. His influence is all over the argument I make here. And the weight of making these comments on the very day that he passed—well, it’s a lot. Colleagues say he would have loved it. I’m humbled.
Dave always thought of me as a tech person, but I like to think that, near the end of us working together, he understood the value of having a tech person in the humanist camp. I tried to carry out that role today.
Chat GPT Writes Answers
My research is at the intersections of computation and writing, on computer programming, literacy, generative AI, and automated writing across historical periods. But I also direct the Composition program here at Pitt, and it's with the frame of undergraduate writing, particularly in the first year, that I want to deliver my remarks today.
I want to make a case for leaning into the writing process, leaning into engaged writing pedagogy, despite and even because I believe generative AI will be part of our writing landscape from now on. The question driving my argument here is: What is writing for?
First, a background to my frame: The Composition program at Pitt serves our first year students and offers gen ed writing classes for students across the university. We teach over 7000 undergraduate students a year in that model, plus a thriving major, and graduate students and collaborations across English and the university as well. We work closely with the Writing Institute and the Writing Center, which support writing across our campus.
Pitt's Composition courses focus on students, along with the texts they produce. We focus on students’ writing process, what they think and grapple with as they write, how they change and grow in that process. In first year writing, we sometimes call it “writing for critical inquiry.” Our first course goal for the class is this:
Engage in composing as a creative, disciplined form of critical inquiry.
In this course, you’ll compose as a way to generate ideas as well as explain them. You’ll form questions, explore problems, and examine your own experiences, thoughts, and observations. Investigating a multifaceted subject, you’ll be expected to make productive use of uncertainty as you participate in sustained scrutiny of the issues at hand.
While many universities’ first year composition courses focus on argument, our model at Pitt asks students to generate ideas, form questions, examine their own experiences, and make productive use of uncertainty.
Making productive use of uncertainty runs counter to what large language models such as ChatGPT represent, and even what ChatGPT says explicitly about the value of writing: to explain, to argue, to persuade. ChatGPT is—infamously—never uncertain. It responds with confidence if it’s right and even if it’s obviously, tragically wrong. More importantly, it has no relationship to what it means to be uncertain, to inquire, to examine its own experiences. It has no stakes in what it writes.
The writing for critical inquiry that Pitt’s first year students do isn't about projecting confidence. It’s not about getting points in a row, with a thesis statement that lays them out and a conclusion that restates them. Writing for productive uncertainty isn't that neat or simple. It's based on the idea that students can thrive wrestling with difficult texts, that they can come up with new and important questions about their worlds and their own words. The point isn't what kind of writing students produce, exactly, although writing is the medium students use to pursue their inquiry. The point is about the process they went through to get there. The point is the challenge. Why else do people bike the Dirty Dozen hills of Pittsburgh?
Key to Pitt’s model of writing for critical inquiry is teachers who talk with students, who foster conversations among students about their writing, who read student writing, who host communities of students listening to each other, teachers who serve as authentic sounding boards and audiences for student writing. This kind of teaching works in small classes where students get to know each other and where the teacher knows them, and their writing process. Crucially, there are no right answers at which to arrive.
ChatGPT's model of writing is something else. It is about arriving at answers. And when economists project that large language models will automate the teaching of first year writing, which they have (Felten 2023), then that's the model of writing they're thinking of. Shrink the vision of what writing is for and *then* it can be automated. Writing is for conveying forgone conclusions. It's a conduit for information. The student could automate much of their writing in this model, it's true. And the teacher could automate their comments back. And the student and teacher could then go off and do something else while their AI proxies send writing back and forth, in an infinite ouroboros! This is no one's utopia, at least here at Pitt, I hope.
Our undergraduates don’t want this world either. In a survey I gave to undergrads in our composition courses last term, they overwhelmingly recognized the value of the writing process and commented that using a software platform that generated essays was completely missing the point of their composition course.
But, as we respond to this new reality of large language models that can write, as the model of writing for right answers gets integrated into our writing processors, as the breathless rhetoric about efficiency and amplification of intelligence that accompanies the generative AI discourse gets operationalized in the very materials we use to compose and think with at a research university like Pitt, I keep returning to the model of first year composition here, to the idea of writing for critical inquiry, writing for productive uncertainty.
Large language models like ChatGPT have become exceptionally good at writing answers. And even if they’re not always right just yet, they’re getting better every day. But: writing like that misses the plot.
And so, against this headwind of generative AI, I wonder: what if we leaned into a model of writing for critical inquiry? Let’s remind ourselves—and our students—how writing can help us learn, can help us develop new questions. How it can help us to heal trauma, recognize love, and understand new perspectives.
Leaning into this mode of writing means leaning into the humans who connect to students—our faculty, advisors, and undergraduate research mentors who make the university a place for our students. It means smaller classes taught by teachers paid decently and who are dedicated to supporting and listening to students, to helping them discover their own ideas in writing, helping them to process the accumulated knowledge of generations of scholars before them, probing and even challenging it in order to build on this knowledge.
Again, our undergraduates can lead the way for us here: the National Survey of Student Engagement outlines high-impact practices that connect students to faculty and locations more deeply, situations that support and also challenge students like community work and collaborative research opportunities with faculty. The intensity of writing courses is also what keeps students here at Pitt.
So when the call comes to eliminate the Writing Center because ChatGPT irons out student prose just fine, when a budget model suggests that we could save thousands by raising course caps and eliminating sections of first year writing because students don't really need to learn to write anymore, we must resist that. What is writing for? I hope we will remember that it’s the faculty and the community and the human engagement that keeps students here—and keep us here as faculty as well. We will need to learn to teach by integrating this technology into the way we teach writing, but we can still do so with productive uncertainty.
What is writing for?
ChatGPT answers that it's for informing, persuading, expressing, recording, or entertaining. It doesn't say it's for learning or inquiry or growth or belonging or productive uncertainty or the pleasure of wrestling with difficult ideas. Large language models such as ChatGPT will not produce challenging, thoughtful, innovative humans that Pitt faculty help to nurture now.
As a human writing teacher and administrator, I find the *question* What is writing for? to be more generative than any answer AI or human can give. As I teach and talk with colleagues, I want to keep asking the question and keep finding more reasons to write at the same time that I write and research these brave new technologies.
This is really lovely, Annette. Thanks for sharing it!