This weekend, I reached out to check on a student who had received some strident feedback from me. Thanks to specification grading, the student had earned the grade she desired on the final project of the advanced course. For this project, she had constructed an argument in support of generative AI in ESOL pedagogy. The project, however, was not intended to be an argument, but rather a literature review and exploration.
I address genre-mismatch in the project development schedule. In this case, however, I did not anticipate my own response to the framing of the student’s argument: I disagreed with every point in the essay. The cherry-picked evidence was thin (at best), and the language of the essay unfairly positioned AI-skeptics like myself.
My feedback totaled 870 words—longer than this blog. And as I reviewed it, I heard my own frustration. There was little (if any) mitigation in my comments:
It's a false choice. As I said earlier, my concern has absolutely nothing to do with whether or not students cheat with it.
Wait—there are a couple of studies. That's not enough to make this sweeping generalization.
There are plenty of tools with games and quizzes—even digital ones—that don't rely on generative AI.
I disagree.
My tone was clear.
Recently, our writing fellows group collaborated with me to research how linguistic choices—from pronouns to punctuation—impact the perceived tone of feedback. Our initial results suggest that mitigated statements and questions (i.e., with modals such as might or could) were perceived more positively, as were statements with we or you (in contrast to I statements, which were perceived as having a more negative tone by students). And while questions in general were perceived more positively, sentence fragments followed by questions were more likely to be perceived as negative or neutral.
My feedback to this student fell squarely into the “negative tone zone.” Eight of my comments were question fragments, like these:
Communicate? With a machine? Not a human? Really?
Our tone study did not address question marks at the beginning of a comment, or the repetition of question marks, like these instances from my feedback:
??? Sorry, but it cannot be "real life" if one half of the conversation is carried about by a machine. A simulation? Perhaps. But let's call it what it is: a simulation.
?? How do we know this?
I suspect that such examples, had we included them in our pilot study, would have been rated generally negative; I hear my frustration shouting in these comments.
So, I reached out to the student: we had had a strong working relationship, and I did not want to jeopardize that. I pointed out that I did not regret the content of my comments, only the manner in which I had given them. This paper came at the end of several hours of responding, and the topic touched a nerve, as it positioned me personally in a way I did not like.
Now some might suggest that my evident frustration is a strong reason for having generative AI compose responses to writing: AI isn’t subject to response fatigue or personal affront.
However, I would argue the opposite: generative AI cannot and should not respond to papers such as these, nor should I let it “improve” my response by making it “sound objective.” Yes, I need to monitor the tone of my feedback. But my student also needs to know how her work is perceived by a human—a tired human, a human who may disagree, a human who ultimately cares deeply about her and wants her to excel. The email that I sent to check on my student opened space for additional dialogue. Would that have happened if AI had responded or modified my comments?
There are critical questions to be asked about generative AI in language classrooms (and in pedagogy overall), what it means to be a teacher, what it means to be a language user, and what it means to learn. Had I decided that feedback on this paper could best be given by generative AI, those questions would not have been asked.
Would I revise some of my comments—and eliminate some of those irritated question marks—if I could? Yes. But in the end, I would not give the job of responding to generative AI; my student is worth my feedback, even if flawed. At least it’s human.
