To the Editor:

Jeff Schatten’s essay, “Will Artificial Intelligence Kill College Writing?” (The Chronicle Review, September 14), has rung the bell. Indeed, all college faculty need to follow the emergence of new AI writing technologies and their potential impact on college classrooms.

GPT-3 is only one of many language models that are being used to develop writing for specific purposes. And yes, these new AI technologies will write technical reports in the voice of Hunter S. Thompson and even author books of poetry that find publication in 24 hours. There are currently some limitations — you’re not likely to want to read that book of poems — but as Moore’s Law dictates, AI-powered writing will get better and cheaper quickly.

These new AI-powered writing generation technologies are going to change college writing substantially. But they won’t end college writing. Instead, we’re going to need to create some new guard rails for the assumptions we make about writing assignments in higher education. What will that future look like?

First, college faculty will need to examine their purposes in creating writing assignments. Those purposes usually fall into two overlapping camps: writing to learn and writing to report learning. Assignments that feature writing to learn may be short summaries, in-class writing, or research reports. And assignments that use writing to report learning can also include research reports, but also quizzes and tests, with short and long answer writing components.


For writing to learn assignments, teachers need to think more carefully about defining the stages of the writing process and also defining the engagement with AI-powered writing generators. When students are just starting a writing assignment, AI-powered writing generators can help them explore their thinking, formulate ideas, and refine their claims. Tools like Elicit can help students to identify sources for ideas and expand their research. Within current limitations, AI-writing generators such as Open AI’s playground can provide summaries, which can help students clarify thinking during the drafting stages of an assignment. And structured use of AI-powered writing generators can help students identify multiple perspectives on their ideas, help them to diversify their thinking, and develop counter-arguments. Further, AI-powered tools like Fermat can apply a visual layout to the writing process and help students and teachers alike engage AI with defined purposes.

But for writing to report learning, the arrival of AI-powered writing generators means that we can no longer assume that students’ writing represents their own knowledge. Short answer test questions are today much more problematic than this time last year. Faculty will want to think carefully about the chain of writing production, and what technologies were available in that process, before evaluating student writing and assigning a grade. Yes, as Schatten suggests, some faculty will respond by strictly controlling the production for tests and quizzes by using computer labs. In some cases, this will mean faculty will deploy lockdown browsers, such as Respondus, to verify that students are producing their own writing. (How does Respondus accomplish this? Using AI, of course). Other classrooms may see a return to blue books with pen and paper tests. And students will be right to also ask faculty whether comments on their papers were written by the faculty members alone, or in concert an AI-powered writing generator.

Readers should also consider that there are evolving uses of AI in our classrooms already. Packback is an example of one tool that has been using AI for years to improve student responses in online discussion boards and to promote curiosity. Its relatively successful deployment points to the future of engaging AI in writing assignments: carefully defining the stage of the writing process, engaging AI for structured input, and understanding the purpose of the writing task within the overall course framework.

Here at UM, we have established a group of writing faculty who are investigating how to teach intentionally with AI. At this stage, we are gingerly following the path of carefully designing writing assignments with deliberate and defined AI engagements. We are also building room for reflection, and asking our students to reflect on how AI is changing their writing.

We also need to better understand two main categories of AI applications in writing technologies: writing assistants and writing generators. We have been using writing assistants to correct our spellings, identify grammar issues, and suggest the completion of our sentences for years in our word processors and texting devices. But the arrival of AI-powered writing generators — and their power to develop streams of larger and larger chunks of texts — present a different challenge than the gradual evolution of AI-powered writing assistants.


As my colleague Stephen Monroe points out, for those of us who teach writing, this moment is the equivalent of when the handheld calculator arrived in math classrooms. We didn’t stop teaching mathematics. Instead, teachers of mathematics had to think through the role of calculation on the path to teaching quantitative reasoning. And designing those classrooms started with the acknowledgement that math students would live in a world with nearly free access to quick and accurate calculations at their fingertips.

Within AI-initiatives on my campus, we often say that we are educating students to work alongside AI. This is tidy way to summarize the future and it glosses over a lot of hard work ahead of faculty. We are indeed teaching through a fascinating time. The emergence of AI technology is evolving so quickly that researchers in the field cannot keep up with all of the developments.

For most of our lives, we have lived with the assumption that writing represents human thought. When I read Dr Schatten’s essay, I made the (formerly) reasonable assumption that Dr. Schatten himself authored it, and that the editors of The Chronicle verified his authorship before they published it. But we are on the cusp of a profound change, and we can no longer assume that writing is the result of human thought — and it will take time for our habits of mind to adjust.

Some 20 years ago the arrival of Wikipedia was also seen as classroom disruptor. Since then, teaching with Wikipedia has become common in college classrooms, spawned the Wiki Education Foundation, and shaped the breadth and quality of public information for the better. Faculty in higher education are at a similar moment today. Our students now live in a world where Wikipedia is taken for granted as a knowledge utility service. Today’s students will also live in a future with AI engaged writing, but unlike the comparatively stable development of Wikipedia, the rapid evolution of AI technologies makes our near term future in college writing seem challenging indeed.

Robert Cummings
Associate Professor of Writing and Rhetoric


Executive Director of Academic Innovation
University of Mississippi