Banning student AI use will let original thinking flourish again

Cheating at colleges and universities has always been a problem but with the pervasive use of artificial intelligence at American colleges it is now out of control.  Anecdotes and data about the ubiquity of cheating using AI abound.

According to a widely discussed report, the rate of cheating by submitting AI-written assignments in the US is double what it is in the UK, where this comparison might serve as a harbinger of trouble ahead. 

Before I describe my own experience teaching undergraduate and graduate classes in this new era, some brief background on US university structure might be useful.

In our arrangement of shared governance, the curriculum is the purview of faculty, while the administration holds the purse strings. Increasingly, however, the administration is dictating course content and design, requiring the integration of AI into all classes. Of course, AI is already incorporated into many of our student resources, including learning management systems such as Blackboard and Brightspace, which accompany classes with electronic syllabi, course materials, readings, tests, assignments and assessments. 

Universities here have been providing students and staff with university-funded premium AI accounts. Although AI might be appropriate in some disciplines, all platforms are not the same. In this article I am focusing on ChatGPT and Google Gemini since they are the most widely purchased, used and abused generative AI tools in universities across the US.

AI has been aggressively embedded in all classes, coursework and degree programmes. At many universities, faculty members are not yet privy to what exactly this administrative overreach entails.

In this environment, last semester I attempted to teach courses on writing and in literature.

My creative writing class, which I subtitled “From Hand to Hybrid,” adhered to administrative directives emphasising AI. But I ended up conducting an unintended experiment on the fly.

My Janus-faced class started out device-free and later integrated AI. In the first half of the semester, students put away phones, laptops and tablets and used only printed texts, in-class handwritten exercises, peer workshops and spoken performance in a shared physical room.

My old-school classroom eliminated the need to police ethical violations – but it did more than that. Without electronic distractions, students created a community where they were focused and supported.

In the second half of the semester, I followed institutional guidelines on devices and AI. My assignments asked students to collaborate with AI by writing an essay by hand in class. For homework, they typed these drafts verbatim into Gemini and experimented with AI to refine elements of their work. They went on to merge the results and their original essay into a consolidated piece for peer workshops. Students turned in their handwritten, pre-AI draft, transcript of prompts and interactions with AI, hybrid draft, notes from peer discussion and entire revision histories. The final paper was followed by a self-reflection on collaborating with AI. 

I had assumed that students would be enthusiastic about working with previously banned devices and AI yet after the first few assignments they rebelled, pleading for a return to our analogue classroom. I needed no convincing.

Students preferred the difficult, uncomfortable work of wrestling with their drafts without electronic assistance. In the first half of the class, they had practised deep analytical and critical skills, solving complex problems independently or through peer review and critique. When we pivoted to electronic devices and AI, they complained that AI was stunting their intellectual growth. As New York Times columnist Bret Stephens argued in his opinion piece tellingly entitled “I’m Begging You: Never Write with A.I.”, [e]ven though AI will soon write better than we do, writing with AI is “mentally enfeebling – an escalator toward a result when you really need to make a daily habit of taking the stairs...[writing] compels thought”.

In our come-to-Jesus moment, my students objected that AI was doing too much of their thinking for them and that our written work and discussion in the first half of the course had helped them flex their thinking muscles. Observing that relying on AI causes cognitive decline, recent Stanford graduate Theo Baker argued that messiness and complexity were valuable: “[I]n the classroom, difficulty is often precisely the point. Sure, a robot can lift 600 pounds much more easily than I can – but that doesn’t much help me if I’m trying to work out.”

When we returned to our device-free classroom, it once again became a vital space, not only for intellectual life but for connection.

The nature of creative writing might allow for this analogue experience but what about other courses?

I tried to adapt a similar method in a large film course, allowing no devices in class and constructing a syllabus that scheduled lectures, discussions and film viewings. I did not, however, disclose the names of specific films because that would lead the witness. I would show half a film in one class. In the following class I would lecture about film techniques students should notice and then hand out a sheet of viewing notes to be filled out while watching the rest of the film. This allowed students to scrutinise movies without being predisposed by researching interpretations.

Disappointingly, most students cheated on homework, submitting critiques that were written by AI. Perhaps their minds needed a rest after all that thinking in class. 

I recently attended a faculty meeting about what has been euphemistically labeled “outsourced” work. I am beginning to agree with a colleague who objected, “Homework is no longer possible.” AI detectors are still inadequate. Sure, within the past few months detectors such as Pangram have been rapidly improving and platforms like Claude have been starting to follow the EU AI Act, applying watermarking globally. They are, however, constantly weakened by humaniser programs in what could be an ongoing battle.

Although I’m experimenting with homework assignments that lean heavily on process and tracked changes, many instructors avoid AI with prompts that require personal response. This approach might be useful in general education classes, which usually involve presentism or judging the past through the lens of modern culture. For example, a prompt might ask how The Adventures of Robinson Crusoe resonates with students personally. Their reflection, which is apt to lean into their own feelings of solitude and loneliness, is probably without academic rigour.

This might be all well and good in entry-level courses. In upper-level and graduate literature classes, however, I am contemplating introducing the methodology of historicism and reception theory, using archival research and platforms that are behind paywalls, inaccessible to AI. Having examined a text’s original reception without judging it by today’s morals and standards, students would go on to consider its present reception. Their research for this component would be grounded in reception theory, primarily based on Hans Robert Jauss’ Toward an Aesthetic of Reception (translated by Timothy Bahti and introduced by Paul de Man, 1982).  Here Jauss argues that a text has a changing "horizon of expectations” and that readers’ reactions are inextricable from their horizon of time.

The contrast between critical responses when a work was first published and present reactions would demonstrate that the text is noy static. As Stanley Fish has long observed, meaning resides in the action between the reader’s moment and the text. The comparison between contemporary and current reception could make us consider our own values not as a permanent reality but as a product of our own time, highlighting changes in the way we think about parts of our culture including language, morals, art, social structures, politics and psychology.

It could be argued that students need to navigate AI as they enter the job market; however, if they focus instead on original thinking and analysis, they would be able to evaluate technologies ethically without duplicating AI literacy that students learn in other courses.

I have dared to flout university mandates because, as a full professor, I am shielded by tenure, which protects freedom of speech (tenured faculty have, however, been unfairly terminated under false pretexts when they criticised their institutions.) Vulnerable junior faculty without tenure ignore university guidelines at their peril.

This leads me to a modest proposal: “eat the babies” and kill AI in the classroom. Many other solutions to the AI cheating problem have been offered but, for now, I am continuing to develop instructional strategies that do not involve technology. When AI wags education, thinking atrophies.

 Deborah D. Rogers is professor of English at University of Maine.