Don’t Kill the Take-Home Paper

The advent of artificial intelligence might radically transform education at Harvard. Let’s not let the take-home paper be one of the casualties.

Concerns about AI-enabled cheating have triggered proposals to dramatically overhaul take-home assignments — including by permitting AI use, changing how they are graded, or even turning our back on them entirely. However, in the form of the take-home paper, instructors retain an avenue to elicit and evaluate thoughtful, long-form student work — an increasingly precious commodity in the age of AI.

It is true that students sometimes face a perverse incentive to use AI instead of doing homework honestly — and policing forbidden AI use outside of class is difficult. But constructing an “AI-proof” paper assignment may not be as Herculean a task as the defeatists would claim.

In my experience in the social sciences, chatbots more often than not break down somewhere in the paper-writing process. At first, they might stick tightly to the academic literature — so tightly as to produce a thesis that does little but reformulate existing work. Then, when prompted to complicate such an argument, they might veer off the beaten track entirely and conjure up a claim wholly detached from the scholarly conversation that the paper is meant to engage with.

Thus, the newly in vogue phrase “AI-proof assignment” need not exclude all out-of-class work. Rather, it can include papers that require students to use the literature and materials presented in class as a jumping-off point for a genuinely original contribution to scholarship.

For a concrete example, take the final project for my freshman-year Expository Writing 20: “Personhood in U.S. Constitutional Law” class. Students had to take a two-pronged approach: first, conduct secondary research into an issue from one of the Supreme Court decisions discussed in class, and second, formulate a hypothesis exploring the role of that issue outside class context. Any student-selected topics had to be approved by the instructor. Despite rapid advances in capability, I doubt AI could complete those tasks in conjunction at A-quality.

On the other hand, some instructors are prone to simply asking students to make their own prompt in relation to some topic of interest from the readings — and in doing so open the door to AI-generated work of satisfactory quality. The fewer specifications a paper must meet, the fewer hoops AI must jump through, and the less effort a work-skirting student needs to make.

And while my proposed “AI-proof” design does reduce student autonomy in the writing process, students could still retain some choice within specific parameters, as well as to decide on the form their argument takes. Moreover, students seeking a higher degree of authorial independence can continue to turn to avenues like the thesis, as they have for generations.

Assignments designed in my proposed “AI-proof” manner serve a practical as well as pedagogical function. If a student outsources their work to AI, they are clearly sacrificing any meaningful feedback on their understanding on the altar of grades. But once a student receives a B-minus on an AI-generated midterm paper, I would bet they’ll feel a new incentive to produce original work on future assignments. If that incentive is not a deep intellectual curiosity, so be it — the means be damned, so long as the ends can be secured.

Certainly, even under this proposal, longform writing will not look exactly as it did before students discovered ChatGPT. AI can be of immense help in areas that traditionally required human work, such as finding sources. But so long as those changes do not disrupt the fundamental function of the take-home paper — forcing writers to meditate, as a fellow editor puts it, “on difficult concepts over long periods of time,” among other things — then they do not disrupt the core educational value.

Still, it is worth acknowledging that not all take-home work can survive. We can save papers, but in a world of ubiquitous AI access, short reflection or summary assignments must go. Fortunately, the new grade cap provides a second incentive to instructors — even beyond AI considerations — to make such assignments, which ridicule the intellectual capacities of hard-working students, more rigorous, or slash them entirely.

On the other hand, even if AI gets better at writing literature-grounded, original academic work, there is still hope for take-home papers. Additional assessments like oral exams could ensure students are familiar with the process and reasoning underlying the work they submit — without eliminating the work itself. And even if cheating at home becomes possible again, at least students who do care about learning will gain a greater reward from doing their work properly. They will benefit from a more rigorous assignment.

Students deserve the opportunity to demonstrate their competence. To that end, instructors should continue to assign rigorous papers where they are already AI-proof, and replace work where it is facile or easily automated.

After all, many of us are eager to vindicate the human mind in the face of rapid technological advancements. We just ask our teachers for a little bit of help.

Michael Isayan ’29, a Crimson Editorial editor, is a joint concentrator in Government and Economics in Cabot House.