Harvard, Jump off the AI Bandwagon

Harvard students should be role models. Instead, they’re letting models play their roles.

Last Friday, The Crimson published a survey estimating that Harvard undergraduates complete an average of 34.5 percent of their homework using artificial intelligence — in other words, large language models. Furthermore, nearly 40 percent of respondents reported regularly using AI on assignments in ways that seemingly violate course policies, and more than 13 percent reported using AI on between 70 and 100 percent of their coursework.

If one thing is clear from these results, it’s that AI has become a problem. In response, Harvard must shift its pedagogy to one that demands active, intellectual participation from students, rather than allowing AI to replace meaningful work.

The survey results reveal the disheartening reality that a swath of Harvard undergraduates offload their educational tasks to large language models in ways that are explicitly disallowed. More than just a symptom of the times, the statistic represents a moral failing among Harvard students.

By opting to use AI instead of laboring through the material, Harvard students are denying themselves the invaluable opportunity to truly learn. Preliminary research suggests outsourcing meaningful intellectual labor to AI can damage students’ ability to think critically and analyze academic material effectively. And, in doing so, Harvard students are choosing not to make the most of an educational experience that many dream of getting — as the few with this opportunity, we ought to use it meaningfully.

While we wish Harvard President Alan M. Garber ’76 could wave a magic wand to prohibit this kind of conduct, AI is nearly impossible to rigorously police, especially as LLMs — and students’ cheating skills — improve at the speed of Moore’s law.

Because we can’t oust AI from the College entirely, the issue of combating its potentially destructive effect on learning becomes one of incentives. To create an environment in which AI’s inevitable presence does not degrade scholarship, Harvard must incentivize students not to use AI when it’s harmful for learning.

By shifting models of instruction to ones that demand legitimate, monitored engagement with course material, Harvard can reshape students’ academic incentives to prioritize comprehension over optimization. Moving forward, Harvard should implement more in-person exams, physical copies of readings checked for annotations, oral exams, and oral defenses of essays and research to demand that students genuinely understand the work they produce.

Furthermore, professors should consider grading routine work meant more for practice than assessment, like problem sets, on a completion basis, as it’s increasingly difficult to ensure that AI won’t be used outside of the classroom. Completing said psets with AI only hurts the student who will be unprepared for course exams.

Inside the classroom, Harvard should consider instituting broader tech-free rules, such as laptop bans in lecture halls, not only to push back against distractions, but also to engage students in the learning process, strengthening their minds for self-reliance.

Of course, in certain cases, AI can be helpful for learning. This rings especially true in STEM classes, where AI can complete auxiliary calculations, assist in software debugging, and serve as a helpful complement to course assistant instruction. However, we see this as all the more reason to craft incentive structures that encourage AI use when it’s helpful and discourage it when it’s dangerous. The College must adapt pedagogy to match AI’s complex impact on Harvard’s classrooms to keep the tool educationally valuable.

Nevertheless, writ large, we are skeptical of Harvard’s attempts to integrate AI into educational programming. Though it may seem that incorporating artificial intelligence helps “keep up with the times,” Harvard’s core mission is providing a liberal arts education — a goal that is in no dire need of accommodating every last technological innovation.

Technology will come and it will go; until it is abundantly clear that any one tool will advance its mission, Harvard’s job is to emphasize the value of human thought and ensure the highest quality of pedagogy. Doubly so when the risks of AI — to security, the environment, and our brains — abound.

Zooming out, it would be a mistake to view the issue of AI usage on its own.

The survey results evoke many issues being debated at the College, including grade inflation and the pre-professional pressure cooker culture we’ve built. It’s distressing that so many of us feel so distanced from our coursework that we’d choose to rely on LLMs to carry ourselves through it. Beyond pedagogical updates, Harvard is due for a broader reckoning with a culture that prioritizes assignment submission over depth of engagement.

AI will be an inevitable part of our future — it doesn’t need to decide our four years at the College.

This staff editorial solely represents the majority view of The Crimson Editorial Board. It is the product of discussions at regular Editorial Board meetings. In order to ensure the impartiality of our journalism, Crimson editors who choose to opine and vote at these meetings are not involved in the reporting of articles on similar topics.

Have a suggestion, question, or concern for The Crimson Editorial Board? Click here.