A Note from Douglas Dechow
This conversation draws together four writer-professors who are thoughtfully considering what the now-ubiquitous use of generative AI (genAI) means for the academic discipline and practice of creative writing. They represent different teaching experiences and institutional contexts that span the field but share concerns about what’s happening and what’s next.
Shortly before we had this conversation via videoconference, the Association of Writers & Writing Programs (AWP) released the results of the survey conducted to understand how members were thinking about and using genAI. The majority who responded were very concerned about the rise of genAI, and exceedingly few were enthusiastic about the technology. More recently, as we were finishing the edits on this conversation, our professional organization released its “AWP Recommendations Regarding Generative Artificial Intelligence,” which members can access on the website. So, at this point, if you are teaching creative writing and somehow remain unaware of the transformative changes to teaching and learning that have resulted from the introduction of genAI to university classrooms, it’s likely by choice.
Douglas Dechow
I’m excited to have this conversation about genAI with the four of you, not only because I’m a publishing writer myself but also because my PhD in computer science bears more than ever on my role as a librarian who supports the research efforts of university faculty and students. It strikes me that your four different perspectives intersect and diverge in important ways that can help other writers and professors make their own decisions and form their own thoughtful stances about tools like ChatGPT.
While writers are concerned about changes to writing and publishing practices, the results of the AWP survey indicate that navigating genAI as a teacher is the most pressing issue for members. In talking with faculty across disciplines as an associate dean in a university library, I recognize that creative writing faculty share some concerns with other faculty, but also orchestrate their classrooms in ways that value originality and the self. Let’s start by talking about our students and how we’re seeing them interact with genAI.
Hannah Grieco
I teach both creative writing and academic writing at the university level. I am not teaching advanced writers who have established writing practices and established narrative voices. I work with students from a wide variety of backgrounds, usually eighteen-to-twenty-year-olds. They are anxious. They are panicked and not sure if they can complete the work. They don’t feel confident, especially if there are language issues or disability or learning need issues. They have been trained by other professors or prior high school teachers that it’s better to use this tool to create polished work. But they don’t know how to use a tool like that to tell their own stories. They’re just putting in the assignment, and ChatGPT shoots something out. They copy and paste what it gives them, and so they’re missing that component of reading and writing, and researching and revising.
Lillian-Yvonne Bertram
I’m really interested in this question of anxiety: anxiety around learning and performance. Why would you, a student, choose to speak a different way in a different context, and why do you not need to speak the same way in every context? In terms of a kind of legibility, a kind of clarity, or a kind of appropriate speech in certain contexts, students who are anxious about getting that right will use AI to do that.
Hannah Grieco
I say to them, “I can teach you how to polish your work. I can teach you grammar,” if that’s applicable, which may or may not be the case, depending on the class. “I can help you understand how to analyze literature in a way that’s meaningful to you. And I can help you with the hard stuff that’s panicking you, the stuff that’s causing all this learning anxiety. But I can’t support you if you’re not giving me your story or opinion to start with.” I want to hear about their lived experiences. Across the board, they are really surprised when I want to know what they actually think and what their opinions are.
Dechow
You’re pointing to the risks of genAI when writers are developing their own voice and the loss it poses for individuals telling their own stories, which I think are related in fairly complicated ways. This reminds me of a social media post that basically said, over the next few years large language models are going to be flattening everything because they’re going to be condensing all of human language into a stream of probabilistic responses. We’re going to force a lot of sameness onto ourselves. Looking at YouTube and other online sites, already brimming with genAI slop, I sense a loss of variation and, perhaps, a pressure to fit in.
Bertram
Our students are not just using language models for “writing” this flat text, but they are ingesting so much of this flat text. We are all ingesting so much of what might be called a house style or lowest common denominator textual language that it becomes ingrained. So, that’s what we’re used to reading. That’s what we’re used to interacting with. We think we haven’t been reading generated text until ChatGPT appeared, but we actually have been reading generated text for a very long time. And now it’s everywhere. Everything that we write comes with a suggestion for norming it to some kind of legibility. We’ve already adopted a house style, but that house style is theoretically our style too.
AI—language models and predictive text—are giving you solutions for every context, or what they perceive to be solutions for every context, which presents writing or speaking as a problem that needs to be solved, which is not the case. What you’re texting a friend isn’t a problem that needs to be solved, the email isn’t a problem that needs to be solved. A poem isn’t a problem that needs to be solved.
The outcome could only be a flattening and losing cultural uniqueness. At the same time, that cultural uniqueness still exists, but it’s going to be normed out for particular kinds of roles and readership. That’s something that we need to be really clear about with our students. We need to be clear about what’s happening, and that comes down to the rhetorical moment.
Anna Leahy
This flattening of individuality is a key concern for me. In science, for instance, repeatability is really important. But in creative writing, repeatability is not important, except in the sense of patterns of language. Donald Hall’s criticism of creative writing as an academic discipline was that it would lead to a flattened McPoem. I never understood that, because the workshop, to me, was a way to avoid sameness. You have fifteen different perspectives on the poem, which makes it possible to understand what other people are doing and distinguish yourself. So, I find these questions of probabilistic flattening really interesting and at odds with our discipline.
We don’t teach the same way as even our literature colleagues in our English departments. We’re using different approaches. We’re seeing text and language differently than our colleagues, and so I think it’s really important that we’re having this conversation in the context of AWP. I don’t know that this conversation about flattening and individuality would be relevant in other disciplines on campus.
Bertram
There’s also Mark McGurl’s The Program Era, which examines how creative writing programs have influenced the style of contemporary literature, particularly fiction. Some of these changes are new forms of writing and styles, but ones that have the watermarks of the workshop and literary marketplace.
Leahy
Indeed, these references suggest that our current worry about systematizing creativity isn’t new.
Nora K. Rivera
I agree with all of this. I’m constantly reminding people that what we see in these large language models, the outputs, is a reflection of what we have been perpetuating. Let’s talk about the standardization of academic writing in high school. We still have standardized testing. Students have been taught in K–12 that good writing looks like this, and they have to follow certain guidelines to get good grades on what they write. And now we’re appalled to see that genAI technologies are spitting out the same thing that we have been teaching students to do for years. So, this is very interesting.
Having guidelines is not a bad thing, but we also have to understand that language is fluid. If we are to see writing as a tool of how we communicate, we have to understand that we don’t always communicate in perfect ways. Especially in creative writing, right? Creative writing is supposed to reflect some part of our reality. Having rules for writing depends on the purpose. It depends on the audience as well. We don’t always write for the same audiences or for the same purposes, but genAI is very deterministic. They’re following guidelines, but according to what? Some of you teach second-language learners, and they might understand writing differently. We might think that what these technologies give us is a representation of good writing. And when we say “we,” we’re talking about a group of people, not always representative of all groups of people.
Grieco
My students come from many different backgrounds and many different skill levels in terms of writing and reading and learning. We often now, very early on in a class, talk about the use of Grammarly because a lot of my students will use it. They’ll write the essay, and then they’ll use Grammarly to “fix” what isn’t working. But I notice it doesn’t sound like them. And if I run it through a checker, it’ll throw every AI flag.
So, I’ll just ask them, “Did you write this? It’s feeling a little like ChatGPT. Tell me what’s up.” And they’ll say, “No, no, I wrote it. It’s just that I used Grammarly to smooth it out.” I don’t ask them not to use Grammarly. I use it sometimes too. But I ask them to interrogate what Grammarly is doing. For instance, where does Grammarly get the idea of what a good sentence sounds like? How does Grammarly choose what rules to use? Where does it profoundly change what they’ve written? And why is it doing that? Who is it trying to make them sound like? They’ve never thought that way before, and they’ve never been asked to ask those questions about their writing.
I tell them that I probably accept 25 percent of the changes that Grammarly wants to make, because I definitely misspell things and forget a comma. But I want to sound like me! And I want them to sound like them. This is true whether it is a personal essay, a short story, a rhetorical analysis, a research essay. My students taught me that Grammarly was changing their writing. I was using Grammarly myself without thinking. But that’s deeply problematic when it comes to personal voice development.
Leahy
I’ve started to look at Authorship mode in Grammarly, which tracks the student’s process, including how much is pasted in and when it was pasted in, in the timeline of writing. That seems a great opportunity to talk with students about their writing process and talk with a class about the different ways that they’re approaching the writing process and their individual voices. It can also be used as policing to check when they have pasted in huge portions of text that might be genAI, but the more important thing, pedagogically, is that we’re thinking about these tools as ways to talk with students about their writing process.
Rivera
You mentioned Grammarly, and I myself, as a bilingual instructor, use genAI tools to double-check my grammar and my spelling.
On that note, sometimes we understand writing as a two-fold tool, something that has content and form. We think of form as the convention, the grammar, and it doesn’t have to be. You can give a whole class on form without discussing grammar. So, I personally don’t assess grammar anymore. I assess other things, but I do give feedback on grammar if grammar is an issue. Why? Because there are tools for that, right? I teach rhetoric and composition, and I also teach technical writing or technical communication, so I assess based on what I’m teaching. If it is a rhetorical analysis, then I’m going to assess the argument, the disciplinary evidence, the connections to the real world.
Leahy
I appreciate this connection you’re making between what we’re teaching, what we’re asking students to do, and what we’re grading them on. I sense that students are more likely to use ChatGPT and other available tools to summarize their reading than they are to use ChatGPT to do the writing for them. Writing feels more personal, and Hannah and Nora are talking about ways we encourage students’ personal stakes in what we ask them to write, but what are their stakes in reading? Are students taught to read for pleasure and to get pleasure out of reading? I think that’s been largely lost in the academic context, though there are certainly students who enjoy reading and often find their way to creative writing and literature classes. I talk with students about reading selfishly, and I teach close writerly reading. They’re reading not only to enjoy the poem or story or, in a literature class, to pick out the theme, but also to learn models themselves that they can decide to adapt or play with.
This moment is an opportunity for us to rethink the kinds of things we’ve been teaching and . . . to consider which things we might want to let go of.
Also, Nora and Lillian-Yvonne, I appreciate that you pointed to the large language models as reflecting back what we have been teaching and writing, though in a supernormed version. This moment may offer an opportunity for creative writing—and rhetoric and composition as well—to recognize the ruts that we got into because we took some things for granted. This rise of genAI tools could shake things up in a good way. Whether or not any individual uses a particular tool, I think this moment is an opportunity for us to rethink the kinds of things we’ve been teaching and get back to our learning priorities to consider which things we might want to let go of and which things we need to look at and emphasize in ways that we lost track of.
Dechow
Lillian-Yvonne, you currently teach computational writing techniques to graduate students in an MFA program, which seems a different context than the experiences Hannah and Nora have identified with undergraduates. These MFA students presumably come to you with critical thinking skills developed in an undergraduate degree. What have you observed about how students are using genAI in creative writing? And which of those things do you lean into, and which do you try and redirect into more productive uses?
Bertram
My students typically do not use genAI. Maybe they’ve asked ChatGPT for a recipe or something like that. But my experience with them is that they come in highly skeptical and apprehensive of genAI and large language models, as they should be. There’s a lot to be skeptical about.
Leahy
I’ve been teaching graduate students too, and I think the graduate and undergraduate students differ in many ways. So, we should keep in mind course level and program level. Also, we’re not through the four-year cycle since tools like ChatGPT scaled up and became widespread for our students. Though the pandemic made students especially used to looking at their screens for their learning and interactions, there’s a divide at the undergraduate level right now. GenAI is being picked up quickly by undergrads without really thinking about what that means. Graduate students seem to bring more skepticism to making academic work easier.
Bertram
I teach computational techniques in creative writing because there is a long history of experimental and conceptual analog computation in creative writing and in writing in general. We can go back thousands of years and look at how people have generated text using computational means. A lot of that includes so-called randomization or stochastic text, which for a long time has been an interesting and very fruitful method of generating text or putting together text in interesting ways. If we go back to Jakobson’s defamiliarization and make-it-new, how do you disrupt and break apart language and sort of put it back together again in interesting ways when you only know one way of using language? That’s where computational techniques can come in. Students are interested in how they can create constraints and programs for themselves. I call them small language models that they can manipulate and that they have a lot of control over. And we do talk about and use genAI as examples. It all gets prefaced with a lot of critical work. Here’s what a language model is, here’s what it does, here are some of the issues with it, and why would you want to use this in the first place? What is the difference? What do we value about writing that we don’t get from a language model? Does a language model actually write?
I would hesitate to say that it “writes.” I hesitate to say that we collaborate, or that we coedit, or that we coauthor with a language model. I don’t think those terms apply, and I try not to use them. I also don’t call it AI. We call it a language model, because AI obscures what’s actually going on. There are different kinds of language models. You can make one without any computation whatsoever, without using a computer. When students use a language model, they can see how quickly the language model breaks down. Or how quickly the limits of it are reached. How quickly it starts to do something that—Hannah, you talked about this—is not like their voice at all. They can see the difference, or how quickly they find themselves being defensive with it. So, part of my approach to teaching it is always critical, and it always comes after we learn the history of generated text. It’s very easy for us to believe that this is the first generated text, that the first era of generated text that we’re encountering is GPT-based models, which isn’t true. There’s this long history of generated text that has a lot to do with how we store and process data and understand text as data. Artists‚ researchers, poets, writers, since they first got their hands on a computer, have been like, “Ooh! And can it write a poem? Wouldn’t that be cool?” They went in that direction, experimenting with how we can make computers write things, but we wouldn’t have language models if we didn’t have this prehistory that had a lot to do with art and poetry and poetics.
Grieco
I’m teaching younger college students, primarily. Even in advanced workshops, I’m working with juniors and seniors. It’s such a different situation, and for anyone who listens to you, Lillian-Yvonne, it’s very clear the critical thinking that’s happening, the tremendous deep dive that’s happening here, that you’re requiring your students to do. For my students, their instinct is to put a prompt into ChatGPT and to copy the output, because why wouldn’t they? Anyone who says that these students are cheating, and that’s it, is missing the point. We all try to make things easier, we all look for shortcuts, we all want to feel good about our work.
Rivera
Sometimes students don’t know that what they’re doing is considered cheating. So, it’s also important to teach them, to actually do exercises with them so they can see, “this is something that you could do, and this is something that you probably don’t want to do in my class.” At this point, right now, almost every professor has some sort of rules for how to use these technologies.
I was teaching a class called Technical Writing, in which about half of my students were software engineering majors, and the other half were creative writers who wanted to have a plan B, just in case creative writing didn’t work. The software engineers were very comfortable with using these technologies. They knew how they worked, but they have used genAI mainly for coding, for synthetic language. The creative writers didn’t know if they wanted to use it. “No,” they said, “I’m a little iffy about it.” It was a very interesting class, where we discussed what these technologies are, what they can do, what is it that they are not doing well right now. We experimented with image generating, prompt writing, and embedding visual elements into writing because technical communication and technical writing involve a lot more than just alphabetic writing.
The background of the students had a lot to do not just with whether they felt comfortable using it or not, but also how much faith they put in these technologies. The software engineers— because they were so used to using these technologies to write code and to work with very deterministic outputs—didn’t like to question outputs or the ethical implications of genAI. We worked a lot on that because they thought there’s a developer, the database, the training method—and that’s it. So, they thought, “It’s not my responsibility if the output is wrong.” However, the creative writers usually questioned, “No, no, no, wait a minute. No, it’s not just my responsibility as a user. It’s also the responsibility of everything that comes with whether my output is ethical or not, or is cheating or not.”
Dechow
You’re pointing out that STEM students, particularly computer science students, don’t seem to be particularly concerned about what went into making the large language model, what it was being used for, what came out of it, whereas the creative humanities students were incredibly concerned. I’ll share an anecdote from decades ago, back when I was in graduate school. For computer science, these kinds of machine learning techniques were just getting popular, and we were still twenty years away from transformers and large neural networks. Someone came to the university to interview for a job, and he was using a technique called collaborative filtering. If you’re not familiar with that, it’s what underlies the recommender systems that we see on sites like Amazon that recommend what you should look at next, or on Netflix, what you should view next. As he was talking about it, I pointed out that he had outlined a system that could be used for political manipulation. It could increase fundraising for certain perspectives or be used to share information representing certain perspectives. And his answer was “That’s not my problem.” I was totally stunned. I looked around the room, and everybody was wondering why I was asking such a stupid question that had nothing to do with this person’s talk. One of the reasons that I’ve made such an effort on our campus to get these ideas of artificial intelligence literacy in front of arts and humanities students is that I believe that they’re going to be more critical about these tools and how they’re used in the future.
Bertram
Yes. This is why the humanities have to be at the center of AI development and not at the periphery. If we could trust technologists and companies to consider these issues, we wouldn’t be having this conversation, right?
These tools are not just benevolent gifts. They’re products that are sold to us as tools that we need even though nobody asked for them.
We’ve been talking about language models and Grammarly as tools, which is one way to see them, but I don’t want us to lose sight of the fact that these are actually products. They are made for the largest applicable audience because it is a product, so it needs to serve the widest audience. They’re not designed to be niche tools. I don’t use Grammarly, and maybe there’s a make-it-weirder mode. But out of the box, that’s not what it’s for, because that’s not going to make it successful. A chatbot that goes a little wonky is not a successful consumer-facing product, right? These are products, and I think we should, very importantly, make sure our students know that. These tools are not just benevolent gifts. They’re products that are sold to us as tools that we need even though nobody asked for them.
They are products trained on, sort of, what we’ve taught people or how we’ve taught. And that’s not exactly true. When we interact with ChatGPT or a Grammarly or any kind of interface, we are not interacting directly with the raw architecture that is the probability engine. We are interacting with an interface that has a lot of other tertiary programs on top of it and a lot of other filters. We’re not getting just us and the transformer. It is everything that OpenAI has put on top of it. There are limits to what it will allow you to generate. That’s also something that I talk about with students. You’re not talking with the model. You’re not meeting the model. You’re meeting a program that is a lot of things happening when you engage with it.
I approach it from what I’m calling a stance of intellectual self-defense. The more that we understand what’s actually happening with a language model, the more students—if they’re going to use it—have the necessary awareness and understanding to interrogate it so as not to lose sight of the fact that, when you use OpenAI, you’re using a product. When you use Grammarly, you’re using a product. A product that is not made specifically for you. It’s made for a whole bunch of people and other things. I don’t think it’s anathema to a creative writing program. I think it’s actually very important that we incorporate these elements as well.
Leahy
You have me thinking now about ethical responsibility, which is something I thought about as a writer and talked about with students long before ChatGPT. When Doug [Dechow] talked about AI literacy with the MFA students I teach, he pulled up the Rolling Stone article about the women at Google who tried to be whistleblowers about language model bias. That problem of bias resonated so strongly with my students that, even if they saw a benefit to using genAI, they became extraordinarily hesitant to move forward with it because of the bias that’s built in. They don’t want to perpetuate harm via language model bias.
Also, I know several students who have taken environmentalist stands on other things, and they feel that not using language models is an extension of their existing political stance on the environment.
Bertram
Teaching the ethics of AI is why the humanities have to be centered, which is very difficult to make happen in industry or on a college campus. Who asked for a bespoke campus ChatGPT? Did the students? Did the professors? Who decided that that was a good idea?
My approach, regardless of what level I’m going to teach, is that a language model is an exercise in bias, first and foremost. Bias is what it does best. It’s biased towards a certain kind of speech, a certain kind of tone, and that is a tone that works as a product, that can be sellable. To “de-bias” it, or to build the model in an ethical way, you’d have to start over. And nobody is going to start over because of the amount of investment that it takes to train a model, the computing power, the financial investment—nobody’s going to do that. If you don’t level the foundation, everything from then on up is going to be a little wonky. No matter how many shims you put in something, it was built on a crooked foundation. And that’s where we’re at.
For the creator who says, “That’s not my problem,” that’s willful. Maybe it’s not your problem, but it’s not a neutral statement. As we tease this out for our students who are creatives and our students who are not, none of these decisions are without context, and that context is biased. A zip code is not a socially neutral thing. That’s critical thinking. That’s how to ask the appropriate questions. To me, critical thinking is the hardest thing to teach. How to teach students to ask a good question? And how to question something and to keep following that? Critical thinking is getting technologists to understand that none of these things are politically neutral or socially neutral, which they do, and if at this point they don’t, it’s willful.
It’s hard to put proliferation back in the box, and that’s not going to happen, but we have to really, really argue and be loud about humanities being centered in these conversations on our campuses.
Leahy
I agree. For me, in terms of the bias related to large language models, the timing is especially heartbreaking because of the efforts to demolish diversity, equity, and inclusion across the country. We’re quashing variety and inclusion in so many other ways, and then using this new‚ scaled-up tool—this commercial product—to probabilistically reinforce the status quo, and its disparities are particularly problematic. This confluence makes it even more important that the humanities be actively engaged in campus conversations and policymaking.
Grieco
The humanities are getting reduced at my school and elsewhere, though. There’s not even an English major at my institution anymore. There are very few creative writing classes. There’s very little happening. I teach all those classes under the humanities department and its major. So, that’s a related part of a very demoralizing situation.
That being said, what the humanities offers is this introduction to critical thinking. Most students are not coming into my classroom with this skill. They’re not questioning ChatGPT. Students who consider themselves environmentalists weren’t taught to think that way in high school. They’ve been using this technology since it became widely available. The only thing they question is whether it is cheating or not.
Bertram
Who is being cheated? If you send the robot to the gym for you, you’re not getting the workout. So, am I the teacher being cheated? No. The effect it has on me is ultimately very little, right? Who’s actually being cheated in that situation? Which I know is the students. It takes work to get students to even ask that question.
Grieco
Absolutely. It takes work to get professors to ask that question too, because professors are taking this very personally a lot of the time. They are cheating on me, and there’s this self-righteousness about it.
I just want to know from students: Is this you? I’m not asking them to think at that moment about the environmental impact of using ChatGPT. I’m just asking, “Have you had this lived experience?” It really shifts my focus. I never fail them on an assignment they’ve initially used generative AI on. I always sit and talk with them and ask them, “What are you panicked about?” Usually, something is going on outside of class. Or they haven’t read a book since fifth grade. Or maybe the student pretended to write essays using ChatGPT the whole time they were in high school. And I’m horrified about all of these scenarios, but I’m glad the student told me.
Rivera
That some professors take cheating so personally makes me think of how language is attached to authority and power. My work is interdisciplinary, and I was, at one point, a high school teacher as well. I’ve taught in different departments, not just in English. I taught Spanish, and I taught in business and IT. It’s interesting to see the power dynamics within English and Spanish departments. We give value and power to writing in particular.
When I taught high school, they used to send us to trainings and professional development workshops, and they would always make us do activities and write in groups. Every time I was in a group with teachers from different disciplines, they would ask, “What do you teach?” I teach English. “Oh my gosh! I don’t want to write,” they would say. They felt intimidated about writing next to me because I was a writing teacher. We might also consider the reasons why some writing professors might feel intimidated or personally attacked when students plagiarize, intentionally or not.
Dechow
This discussion of changing understandings of cheating and the roles of authority in relation to genAI seems an opportunity to rethink outdated or static understandings of writing. How does our evaluation of the teaching of writing translate into a writing course itself? And how might genAI tools or products figure into that classroom?
Rivera
Maybe we’re starting to think that overrelying on essays and research papers might not have been the best idea. Even in nonwriting courses like linguistics, sociology, psychology, and in STEM, the instructor says, “Write an essay about blah-blah-blah.” If I am teaching a rhetorics course, I work with rhetorical analysis, but a rhetorical analysis does not have to look like an essay.
In my cultural rhetorics class last semester, we included a little bit of AI into rhetorical analysis. I addressed at the beginning of the course the appropriate uses of genAI in my course. I did a few activities with them so that they could understand biases and hallucinations. In a story mapping project, the students needed to do research on three places that they felt connected to, making those connections to their own context meaningful, and they needed to do some historical research and work with historiographies. I asked them to interview storytellers so that another person could tell stories of those places. They embedded disciplinary evidence from within [the] rhetoric and composition field. They worked on ideas from working with genAI. Some of them used these tools to generate photos. Some of them used them to generate questions for their interviewees. Then they put it all together in Google Earth. They did awesome presentations in which they discussed their personal context.
I can’t say that nobody used ChatGPT to cheat or to come up with everything, but the majority of the students presented unique and very personal rhetorical analyses. I don’t think that they overrelied on technologies. I could see who was engaged through the writing, through what they produced, and how personal they got. I could also see what was meaningful to them, or when they wrote something that they were interested in. I saw a big difference between writing about something that they’ve been impacted by—they put a lot more effort into—and writing something very generic without a soul. Do I see generic, soulless writing sometimes? Yes, sometimes. But did I see that before when I taught high school with no genAI technologies? Yes! I honestly haven’t seen a huge difference in cheating or plagiarizing over the years.
Dechow
We’ve been talking a lot about teaching and student use of genAI, but I’d also like to think about faculty use of these products and the institutional contexts at stake.
Leahy
As my department prepared for a program review this summer, I asked Copilot to look at the catalog copy of our MFA program. I asked it first to look for what the catalog copy suggested about our strengths and weaknesses. Even though it didn’t read it in a superintelligent way, I sense that it captured something about how a prospective student might look at the curriculum and make assumptions about what our priorities were or what they would get out of the program. That was an interesting and helpful use of technology because it wasn’t my insider perspective.
I also asked it if it could point to courses to consider deleting. It chose courses where it found redundancies, and sometimes it was right, and sometimes it was wrong. When asked what to keep, it pointed to courses with practical, skills-based applications and to several signature courses that distinguish our program from other programs. We don’t often reevaluate our program ourselves with a critical eye to ask, What is the rationale behind this? What is the internal logic of how these courses work together, and how are students actually moving through the program? It was interesting to use a language model to look at the structure of the program from a perspective that didn’t have a lot of detail about how the program works or have information about who teaches which course and how it’s taught. That way of looking at a program might be more like how an administrator looks at a program, with less connection to it and with a more corporate approach.
In creative writing, there’s also something important about community that we need to keep in mind. Writing is done in isolation. I sit down, I type on my computer. But the class and the program asks us to be writers in community, and that seems to me important. Our pursuits as writers are validated—we’re hanging out with people who say, “Writing is a cool thing to do and a worthwhile way to spend our time,” which is not necessarily celebrated in the larger world. We’re learning together, and this goes to the anxiety about grades too. In a workshop, because of the emphasis on revision and the workshop method, it’s okay to fail. And that failing is part of the learning process. I don’t know that students get that kind of communal failure-and-support mechanism in other areas of campus, so I can understand turning to ChatGPT as a fail-safe.
I want my students to make mistakes. And then to recover from them, and then to try again.
Grieco
I think that’s at the heart of this. I want my students to make mistakes. And then to recover from them, and then to try again. I am the mother of three teenagers of my own who are all terrified of making mistakes. This is a pervasive thing. The use of ChatGPT can make it even harder to trust that process of making a mistake and then recovering from it.
One of the first things I introduce in my 100- and 200-level classes, especially if they’re students I haven’t taught before, is that we’re just gonna screw up constantly. And that’s part of the creative process, or part of the research process, or part of fill-in-the-blank process in the classroom. That messiness gives them permission. In whatever field they’re in, this ability to make mistakes and learn, that’s why we workshop in all of my classes, whether it’s creative or academic writing. Of course, I’m a believer in a very different form of workshop than I was taught growing up. So, anyways, just thank you all so much for this.
Leahy
Hannah, you brought up that you’re an adjunct. Lillian-Yvonne and I have tenure, and Nora’s starting a new tenure-track job. If I were an adjunct, I think I would be using genAI more, because it would allow me to increase my hourly wage. The structure of academia has exploited half the people teaching in our classrooms, so genAI efficiency has made me think about the adjunctification of academia in a different way too. This goes back to it being a product, and it’s also about how we’ve set up the structure of labor as well as anxieties for faculty and for students.
Grieco
No one is adjuncting because they’re making any money. I do other work in addition to adjuncting. It’s all about the hustle at this stage.
When I first heard that my students’ high school teachers were grading their essays with ChatGPT, I was so upset about it. But they’re also teaching five classes, all with many essays to grade!
Dechow
Despite the challenges you’ve all touched upon, genAI has taken hold. You’ve been incredibly thoughtful about outlining serious questions that writing professors need to continue discussing as well as pitfalls we’ve seen in these first two years of widespread use by students. Before we close, I’m wondering how we might think about the potential of genAI in the context of the discipline.
Bertram
As someone who practices computational text generation, as someone who has used language models to generate text, I don’t want to lose sight of the fact that there is interesting potential. In using these models, we can think about process in our own writing. I’m not even concerned about the output. Not using the output of language models, but using them as a type of process interrogation and a type of reflection.
I don’t ban AI use. I can’t imagine how that would even work. But it’s something that we have to reckon with. It doesn’t make sense to say it’s the death of creative writing or the death of poetry or the death of the author, because these things, if you go by any of the news, have been dying for a very long time. It’s stuff to be fearful of, absolutely, but there are ways to approach it, to work with it, in a nuanced and informed way.
Leahy
I agree. At this stage, I am most interested in the ways that we can talk about language models as an opportunity to reinvigorate what we already do well. I’m less interested in using them than I am in using this moment to think about what creative writing as a discipline really is and wants to accomplish, what our learning priorities are. For a long time, in order to secure our place in the academy, we had to prove that creative writing can be taught, and so we diminished the role of talent. But I think we need to go back and talk about talent because it’s something ChatGPT can’t offer. Talent as it relates to individual interests and individual skills, and how those interests and skills create a feedback loop that involves curiosity, experimentation, diligence, and a sense of accomplishment. If we can create that dynamic cycle in the student, then the student has more stake in what they’re doing. Talent and failure and learning and skills and interests are all interrelated. Instead of cheating or the death of the author, I hope that’s what ChatGPT gives us an opportunity to discuss over the next several years.
Dechow
I think we’re all interested—some with excitement, some with trepidation—to see how these possibilities will unfold over the next few years. I hope the questions you’ve posed and the ideas you’ve shared here help others think about how we are each shaping that future of creative writing every day. This conversation is one that should spark others across the discipline and reshape our policies and practices across institutions. Creative writing is continually evolving and rethinking itself, and this conversation about genAI can be part of that ongoing evolution.
Lillian-Yvonne Bertram is an African American writer, poet, artist, and educator who works at the intersection of computation, AI, race, and gender. They are the author of Travesty Generator (Noemi Press), a book of computational poetry longlisted for the 2020 National Book Award for Poetry. They are the recipient of a National Endowment for the Arts poetry fellowship. Their other poetry books include How Narrow My Escapes (New Michigan Press / DIAGRAM), Personal Science (Tupelo Press), a slice from the cake made of air (Red Hen Press), and But a Storm Is Blowing from Paradise (Red Hen Press). Their most recent full-length poetry book, Negative Money, was published in 2023. Their chapbook, written with AI, is called A Black Story May Contain Sensitive Content and won the 2023 New Michigan Press / DIAGRAM chapbook contest. They direct the MFA in creative writing program at the University of Maryland and are a 2024 Foundation for Contemporary Arts poetry grant recipient and 2024 Deutsch Foundation Rubys Artist Grant recipient. They are coeditor with Nick Montfort of the recently released anthology Output: An Anthology of Computer-Generated Text, 1953–2023.
Douglas Dechow is the associate dean for Library Research & Data Services at Chapman University. Previously, he was a research computer scientist contractor at the Fermi National Accelerator Laboratory. Dechow is the coauthor or coeditor of five books, including Generation Space: A Love Story, Intertwingled: The Work and Influence of Ted Nelson, and Your Craft as a Teaching Librarian: Using Acting Skills to Create a Dynamic Presence. He held a visiting writer fellowship at the American Library in Paris in 2016. His writing has appeared in TheAtlantic.com, Air & Space Magazine, Fifth Wednesday Journal, Curator, and Aviation History Magazine.
Hannah Grieco is the author of First Kicking, Then Not, published by Stanchion in August 2025. She writes a literary column for Washington City Paper, edits prose at a variety of small presses and literary journals, and teaches writing at Marymount University. Read her work in The Washington Post, The Independent, Al Jazeera, Huffington Post, Brevity, CRAFT Literary, Wigleaf, Poet Lore, Fairy Tale Review, and more. Find her online at HGrieco.com and on most social media at @writesloud.
Anna Leahy’s books include the poetry collections If in Some Cataclysm, What Happened Was:, and Aperture and the nonfiction book Tumor. Her work has won awards from Mississippi Review, Los Angeles Review, Ninth Letter, and Dogwood and appears at Aeon, Atlanta Review, The Atlantic, Bennington Review, BuzzFeed, Poetry, Scientific American, The Southern Review, and elsewhere. She has been a fellow at MacDowell, Joshua Tree National Park, and the American Library in Paris. She teaches at Chapman University, where she edits the international Tab Journal and is the former director of the MFA in creative writing and founded the health humanities program. Read more at AMLeahy.com.
Nora K. Rivera is an assistant professor in the technical communication and rhetoric program at Texas Tech University. She is also guest faculty in the Department of Languages at the Universidad Autónoma Benito Juárez de Oaxaca in Oaxaca, México. Her research centers on writing, technical communication, cross-cultural UX, translation, cultural rhetorics, and emerging technologies. She is the author of The Rhetorical Mediator: Understanding Agency in Indigenous Translation and Interpretation Through Indigenous Approaches to UX (2024). She currently collaborates with an Indigenous organization from México on an AI and translation project. Learn more about her work at NKRivera.com.