Train TFs, Not LLMs

A few hours before her problem set is due, my friend slams her notebook closed in frustration and prompts Claude to clarify the intricacies of probability theory. She says it’s a better teacher than her real one.

I can’t blame her. Though I’m a vocal opponent of artificial intelligence in education, it’s difficult to encourage students not to use a technology that provides endless support for learning challenging concepts in a time crunch. But using AI as a teaching assistant doesn’t address the root of Harvard’s pedagogy woes. The College must instead invest in training teachers so students can fully participate in its learning community.

During my time at the College, I’ve witnessed my peers increasingly turn to AI for help with their problem sets, writing assignments, and study tools. Their motive is often not a desire to take the easy way out or avoid the struggle of in-person learning, but instead stems from a genuine lack of support.

The fault lies with the system. Our course assistants and teaching fellows are too often undertrained and overworked. Artificial intelligence provides the capacity, expertise, and endless availability that teaching staff simply do not have.

Indeed, one course syllabus this semester told students that using AI as a “personal 24/7 tutor” might constitute a healthy use of the tool, while having it complete their problem sets was “likely not.”

But even this limited use should not be what we strive for.

Information seeking is crucial for learning and cognitive development. When students offload that mental labor to generative AI, they might attain a basic understanding of the concepts needed to succeed in their assignment or pop quiz — but research shows the tools ultimately hurt their learning when students are forced to think on their own.

A crucial part of the college learning experience is struggling through materials on one’s own and with peers — and then turning to human guidance to provide valuable instruction.

A recently published report from MIT puts it well: “Education is a cultural practice through which students learn to make meaning, exercise judgment, form identities, and participate responsibly in community.” An educator’s role is to facilitate this practice. Excellent teachers have long-term impacts on student success, improving not only students’ test scores but their lifetime income and standard of living.

Yet our current TF system fails to consistently contribute to student success. As the Editorial Board has rightly argued, our teaching fellows sometimes have limited experience with the content they are teaching, and even less experience with how best to teach it. Good teaching is a learned skill — one that requires both practice and guidance.

While resources for improving TF training exist, they’re clearly insufficient. The Derek Bok Center for Teaching and Learning coordinates a mandatory teaching conference for all TFs, as well as professional development opportunities. But the required training is a mere two hours, which cannot possibly provide our future instructors with an adequate introduction to the details of lesson design, classroom management, and effective pedagogy.

The College should instead require more rigorous training for all instructors — focusing specifically on teaching difficult content effectively so that students can learn material in class, section, and office hours — not from ChatGPT.

I can testify firsthand that there’s a dramatic difference in course quality, student engagement, and need for artificial intelligence support in classes taught by well-trained teachers.

My freshman year, I took a General Education course examining the American education system. Students came from disciplines across the College — but unlike other classes known for sleep-inducing lectures and overlong readings, our 9 a.m. lectures were filled with passionate discussion and genuine engagement.

Why? Nearly every member of the teaching staff had a background as an educator or was pursuing a degree in education. As a result, our lectures included collaborative discussion exercises, and our sections employed innovative teaching methods. The success of this course should serve as Harvard’s example.

It’s unsurprising that trained educators are simply better teachers. Of course, not every department’s teachers can learn the minutiae of good teaching as extensively as academics who study education itself — but if we can bring even a fraction of that knowledge about effective pedagogy to the rest of Harvard, we would have increased participation, more motivated students, and a student body that better knows how to learn together. While it would take time and effort, more in-depth teaching training for both undergraduate course assistants and graduate teaching fellows would be well worth it.

It is not students’ fault if they are driven to use AI as a learning aid. And it is not our instructors’ fault that the University inadequately invests in their training and development.

Harvard’s high-caliber students, faculty, and research should be matched by exemplary pedagogy. To truly enable an ideal living and learning environment, Harvard must practice what it preaches and help our course staff learn to teach.

Victoria L. Dolan ’28, an Associate Editorial Editor, is an English concentrator in Kirkland House.