A Case for Human Evaluation of Student Work

The first time my students saw the ChatGPT tab on my browser as I connected my laptop to our interactive whiteboard, they shrieked. “Oh, so you recognize that icon?” I teased them. “You’re familiar with it yourselves?”

I told them I use AI all the time as a teacher: to help me generate images to illustrate a complex piece of text, to pull and define vocabulary words they might need to understand for an essay, to track curriculum standards. They were sensitive to the potential hypocrisy: After all, I had told them that I expected them to complete their own writing assignments without the use of AI.

I affirmed that I do expect them to write with nothing more than their own brains because, as I had discussed with them before, when it comes to learning writing, I most value the process, not the product. To paraphrase science-fiction writer and computer programmer Martin L. Shoemaker, “Your teacher doesn’t care about your essay. They care about you.”

And then I told them that, although I am curious and optimistic about AI’s potential in schools (my master’s degree is in foresight!), I have made a solemn commitment to never use it to evaluate their work. It’s my own tired, aging, human eyes on their papers (or the screen showing their papers). Whatever feedback they get is the reaction of my biological neural synapses to what they produced with their own.

The Modern Language Association, the scholarly authority on language and literature, recently published a statement in support of sustaining the millennia-old practice of growing writers through human conversation:

“The purpose of assessment in language, literature, and writing courses is to provide feedback on how students are developing as writers, readers, speakers, and thinkers. Effective feedback is both formative and summative, but most important, it is centered on communication. Communication, education, and assessment are human-centered activities, conducted for human-centered purposes.”

Shaping instincts

But it’s not just a matter of principle. Students—my own and others—seem supportive of this policy of relationship-centered feedback. Knowing that they are writing for a human audience helps them buy into the demanding labor of putting words on a page and the vulnerability that comes with trying something you are not yet good at.

Beyond that, AI just isn’t very good at assessing something as nuanced and complex as communication. After AI was used to score written responses on our state English I exam, districts appealed to have human scorers reevaluate the work—and saw a significant discrepancy. According to OpenAI’s own research, ChatGPT was not able to achieve a passing score for the AP English Language and Literature exams, so I certainly do not feel confident in its ability to evaluate for that assessment.

But most of all, I know that doing the hard work of analyzing and evaluating my students’ writing myself has shaped me into a capable and effective teacher. Over the past 20 years, the opportunity to read thousands of essays, journal entries, and stories from the young people with whom I share my classroom has helped me to hone my instincts in understanding what and how they are thinking. The hours I have spent assessing their work and responding with feedback have been my richest source of professional development. I’m a better teacher when I am diagnosing my students’ needs rather than outsourcing them to an algorithm. The same “skill atrophy” that makes an AI shortcut a poor choice for students applies to teachers, too.

I understand the temptation. When you’re staring at a (virtual) stack of essays, and you have an inbox full of emails to respond to, meetings to attend, duty to supervise, and lessons to plan, it can feel hard to give up such a significant block of time. In an era of ever-increasing demands on teachers’ limited working hours, here are some ways that hold the line on my promise to grade student writing myself.

Setting expectations

I frame the issue for my students. They know that they are getting a human reader, and they know that that often comes with a delay. Providing narrative remarks for a round of essays for almost 200 students can take weeks. When I’m aiming to return their work sooner, sometimes I just score using the rubric instead of providing individual responses.

I read while they write. As soon as my first class of students has submitted an assignment, I begin to evaluate their submissions while later periods in the day are working on the same task. This allows me to immediately see if I need to adjust my instructions and get a pulse check on students’ success with the objective.

I use a comment library to expedite redundant work. I was growing weary of typing the same feedback, over and over, when I discovered that our school’s learning management system (Canvas) had a feature to allow me to quickly stamp frequently used comments. Even without that feature in a platform, keeping a typed log of comments to quickly copy and paste not only speeds up the process of grading but also can give teachers a sense of where groups of students are struggling.

Speaking of which, I frequently give classwide feedback. As I’m scoring, I keep track of the issues that I see (and my suggestions for improvement), and then I do a lesson focused on those areas as the next step in the writing cycle. Not every piece of feedback is applicable to every student, but if it’s common enough to appear as a pattern, it’s likely a challenge that students could encounter in future assignments.

Teachers make thousands of choices in a single day. I would argue that one of the best ones they can make is the choice to evaluate student work themselves, with their own human minds.