How AI Undermines Harvard’s Grading Reform

Harvard students are very good at optimizing.

We compare course evaluations, memorize which departments give out the most A’s, and quietly shape our schedules around what will maximize grade point average with minimal stress. This instinct appears cynical, but it is more rational than anything: in a system where grades are inflated and expectations are predictable, optimization works.

That system is now changing. The University has begun to crack down on grade inflation, leading to a near-seven percent drop in A grades last fall. Reducing inflated grades is an admirable goal: it disincentivizes chasing courses rumored to be lenient and pushes students toward classes they actually want to take.

At the same time, Harvard has yet to decide on a shared approach to artificial intelligence. While the University has indicated that it plans to incorporate student input on new practices and standards, students currently must navigate a patchwork of course-specific policies. These shifts are often discussed separately, but they are inseparable — high variance between course policies on AI or grading, which directly affect difficulty, can lead students to try to optimize their courseload.

Consider grading. In a report last semester, Dean of Undergraduate Education Amanda Claybaugh described rampant grade inflation, with more than 60 percent of grades given to undergraduates being A’s. When nearly everyone earns an A, grades stop communicating rigor and growth, and course selection stops communicating intellectual risk-taking.

But grading reform does not operate in a vacuum. Uncertainty around AI undercuts progress on the issue. Right now, individual instructors have broad discretion over how AI may be used. Some courses ban it outright. Others allow limited use or encourage experimentation. Enforcement varies widely; some professors require version histories or drafts, while others admit they would struggle to detect AI use at all. Like grading, outcomes increasingly depend not on institutional norms but on course-specific design.

When grades were predictably high, ambiguity around AI mattered less. But as grading becomes more rigorous, unstandardized AI rules give students looking for the easy way out another path. Students will need to put in genuine effort in courses with strict AI enforcement, while others can advance more easily in classes where expectations are vague or unenforced. The result is a system where students are evaluated not just on learning, but on how well they interpret or avoid unclear rules.

Faced with stricter grading and ambiguous AI policies, students may respond by minimizing risk: choosing familiar topics or avoiding courses with in-person exams. This strategy is based on fear and ease, not learning. It narrows intellectual exploration and discourages students from taking classes that challenge them in new ways.

This is not to say that students should never consider which course structures and assessment styles allow them to learn most effectively. In fact, as optimization becomes less reliable, alignment becomes more important — when defined correctly. A student who thrives in discussion-heavy seminars with extensive feedback may struggle in a lecture course graded on two high-stakes exams, and vice versa. Choosing courses that fit one’s learning style can improve learning.

That’s where alignment diverges from optimization. Optimization asks which classes are easiest or safest. Alignment asks how a student actually learns best. The former protects GPA at the expense of growth. The latter can enhance learning and performance, especially in a system where grades are no longer inflated across the board.

Students will, of course, always be drawn to optimization. They still have concentration requirements to fulfill. GPA still matters for fellowships, graduate school, and jobs. Nothing can eliminate the pressure to protect one’s transcript. Nor can AI policy be truly universal — a class built around revision and feedback demands a different approach to AI than one centered on in-class performance or timed assessments. And, of course, some classes will always be easier than others.

But within those constraints, students usually have meaningful choices: lecture or seminar, exam-based or paper-based, collaborative or independent. Those choices matter more now than they did before.

This moment therefore demands intentionality. Students can no longer rely on departmental reputations or historical grading curves alone. Instead, they should ask: How is work evaluated? How much is feedback emphasized? What assumptions does the course make about AI use?

It also demands transparency from the administration. If Harvard wants grades to communicate values rather than obscure them, it must clarify how AI fits into academic work. Grading reform without AI guidance risks fostering a system that rewards caution and conformity instead of engagement.

For students, the takeaway is clear. The best strategy is no longer optimization. It is alignment — not with ease or ambiguity, but with how we learn best. That is how we protect both our education and the meaning of the grades meant to represent it.

Catherine E.F. Previn ’27, an Associate Editorial editor, is a Government concentrator in Lowell House.