Most AI commentary about creative writing starts with a panic: the machine will replace the artist. The course designed by Juliana Spahr and Stephanie Young at Northeastern University’s Oakland campus starts with a different question. As Spahr puts it in the Northeastern Global News profile of the class, the goal is “to start thinking about what AI can do and what it can’t do.” That distinction, grounded in a semester of hands-on experimentation, is more useful than any theoretical debate about artistic agency.
The class, called Writing Creatively in the Age of Artificial Intelligence, is not a how-to-prompt course. It is a laboratory where students examine generative AI as a pattern-recognition engine that produces “endless variations on a sentence,” in Spahr’s words, but fails at deeper mimicry. Spahr notes that the machine can “sort of explain how a Raymond Carver short story is made, though it’s not great at that either. It’ll do a pale imitation of Carver, or Ray Bradbury, or Gertrude Stein. It can get to the surface of things.” That surface-level facility is exactly what makes the tool seductive and what the course is designed to expose.
The pedagogical frame matters. Young, an associate teaching professor of English and the W. M. Keck Foundation Professor of Creative Writing, grounds the curriculum in the avant-garde tradition of the Oulipo, the French literary movement from the 1960s that built new forms by imposing rigid constraints. Oulipo writers eliminated entire letters or bound themselves to numerical systems, arguing that aesthetic freedom comes from restriction, not disinhibition. Young sees AI as a contemporary constraint machine. “The tools change, but the questions don’t,” she says. “Constraint, randomness and collaboration have always been part of how writers make meaning. AI just makes those dynamics more visible.”
This is a genuinely useful framing for the AI industry. Most discussions of creative AI focus on capability benchmarks or output quality. The Northeastern course shifts the lens to process. Students are asked to interrogate where they feel authorship slipping, where they can identify the signature of their own voice, and how much labor creativity still demands even when a machine is involved. Those are not abstract questions. They are the same questions every professional writer faces when using GitHub Copilot, Claude, or ChatGPT for drafts.
The course also examines copyright lawsuits, consent, and how “voice” is constructed. Young and Spahr look at how chatbot voices have been flattened to avoid harm, resulting in a “strange, heterogeneous tone.” That observation has direct implications for how AI companies train their models. The pursuit of safety through tonal averaging produces outputs that are technically inoffensive but artistically dead. The course gives students the vocabulary to articulate why that matters.
Student reactions in the article are instructive. Ryan Huang, a freshman biology major, says AI provides “a more fleshed-out version of your own thinking” but should not be used to make the art itself. Tiffany Lee, a business administration major who took Young’s class in 2025, says the course “completely changed his perspective on the use of AI.” These are not the reactions of people who have been told to fear the machine. They are the reactions of people who have been taught to use it critically.
What is missing from the article is any discussion of the economic context. The students are learning to write alongside machines that are already being deployed to replace entry-level copywriting, content generation, and even some forms of journalism. The class teaches critical literacy, but it does not appear to teach negotiation. A writer who understands exactly what AI cannot do is better positioned to charge a premium for that gap. That is a market signal the curriculum could amplify.
The most important line in the article belongs to Spahr: “This is where we are. We can’t roll it back. So the question becomes: what do you do with it?” That is the correct stance for an AI-native education. The course does not indulge in techno-utopianism or Luddite rejection. It treats the machine as a collaborator with known limitations, and it teaches students to recognize those limitations through practice rather than theory.
For AI builders, the lesson is uncomfortable. The course demonstrates that creative professionals are not going to be impressed by surface-level fluency. They care about voice, constraint, and the labor of decision-making. A model that cannot explain why it chose one word over another, or that cannot sustain a consistent authorial tone across a long piece, will always be a tool rather than an artist. The Northeastern class is training a generation of writers to demand more from the technology than it can currently deliver.
The course also raises a question for AI companies that build creative tools. If the best pedagogical use of generative AI is to expose its limitations, what does that say about the product roadmap? The industry is racing toward models that can produce publishable fiction. The Northeastern class suggests that the more interesting application might be models that can explain their own creative decisions, or models that can collaborate under explicit constraints. That is a harder engineering problem than scaling parameters.
Spahr and Young have built something rare: a curriculum that treats AI as a serious creative partner without pretending it is a human. The students leave with a sharper sense of their own agency, not a diminished one. That is the outcome the AI industry should want. A generation of writers who understand exactly what the machine is doing, and exactly where it falls short, is a generation that will build better tools.