I’m trying to understand where the notion that constantly monitoring students and subjecting them to real-time feedback is a benefit to learning came from, because I gotta tell you, from my perspective as a teacher of writing, it’s nuts.
Let me count the ways (and whys):
1. Real-time monitoring and feedback increases anxiety surrounding learning-related activities.
I don’t know about you, but I absolutely do my best at learning something unfamiliar when I get the sense that I’m constantly being monitored and judged, particularly in public or social situations like a classroom.
And yet, in K-12 education we have things like classroom data walls, or real-time data monitoring apps like ClassDojo where students can see exactly where they rank every moment of every day broadcast on a screen. Nothing anxiety producing about that.
Facial recognition software used to monitor student moods is apparently right around the corner.
2. The brain is not a muscle.
Believing that monitoring students has a benefit essentially requires us to believe that all learning most effectively happens through a behaviorist model, and well, that just ain’t true.
Video games are a great example of a world governed by real-time feedback, a purely behaviorist playpen. If you do the right thing, the game continues. If you do the wrong thing, you lose. Ultimately, your brain is quite literally trained to do the “right” things.
For discrete skills which rely on muscle memory, real-time feedback may be a benefit. If you’re learning to shoot free throws, a knowledgeable coach who can correct a problem with form in real-time (“You’re not following through.”) is a benefit.
But not all skills are the same, the brain isn’t a muscle, and unlike a discrete skill like shooting a free throw or tying a knot, there’s no real benefit to two human beings learning the exact same thing in the exact same way.
And even for skills which require muscle memory, having some time to go off and be alone while practicing free of judgment and monitoring has a benefit.
For example, when learning to play the guitar, it’s useful to have some periods of real-time feedback where a teacher may be able to correct a flaw like a bad hand position, but you also need to go lock yourself in your room and practice, likely making a bunch of unpleasant noises in the process. Imagine trying to do this while being constantly reminded that your noises really are unpleasant.
And as I’ve argued previously, learning to write isn’t a skill like trying to learn an instrument. It’s practice at becoming not just a musician, but a composer.
Writing is thinking, writing is thinking, writing is thinking, and sometimes when we’re working on our thinking, we have to be left alone.
3. Real-time feedback does have some (very limited) applications in learning to write.
There are parts of the writing process where real-time feedback, even unsolicited real-time feedback may be useful. I would often kick off research intensive projects by having students do their research in class, where they can get feedback from me and their peers in real time about the quality and potential relevance of what they’re finding as they search.
If a student says they’re having a hard time finding good sources, it makes sense for me to intervene immediately by helping them craft a better search term, or showing them how to limit a search to a more promising universe of sources.
But that particular step in research is a discrete skill where immediate feedback is helpful. When it comes to the much more complicated reading and evaluation of sources, real-time feedback would short-circuit an important part of the writer’s necessary critical thinking. There is no straight line through that part of the process.
To make knowledge, the writer needs to wander through the woods on their own until they find a way out. That’s learning.
4. Not all feedback is created equal, especially the kind of feedback which can be given in real-time.
Even where real-time feedback may be useful, we must also be extremely cautious with the type of feedback we give. I’ve often set aside class periods for students to simply come and work on their writing, particularly during the drafting stage. I’ve found it’s a great way to get students over any hesitation or tendency towards procrastination.
It’s also a chance for some real-time feedback, but importantly, all feedback they receive must be both judgment and answer free. I do not want students worrying about performance as measured by a grade while they’re drafting and developing a piece. I particularly don’t want to make choices for them which are their responsibility.
Even when a student asks me a straight-up “Should I do this or should I do this?” question, I only answer with a question in return: I don’t know. What are you trying to do? Who is your audience? What do you think are the differences between your two choices?
My goal is to send them back into the type of thinking writing requires within a full rhetorical situation involving audience, message, purpose, and genre.
5. AI-driven real-time feedback is the worst feedback of all.
Computers can’t read; they can only count.
Algorithm-driven feedback on writing is driven by an adherence to writing formulas which are inherently incompatible with helping students learn to think like writers. An algorithm can see that there are only four paragraphs in what is supposed to be a five-paragraph essay and tell students something is missing. While pointing this out may help a student score better on a standardized exam, it is not helping them learn to write because it is preventing them from thinking like writers. In fact, it’s actively harming their development as it causes them to view writing as something formulaic and standardizable, which it most definitely is not.
Yes, privileging these bad assessments is the root problem here, but a thought-free embrace of ideas like real-time feedback being an unalloyed good allows these bad AI products to gain traction, while also drawing resources and money away from what we know works.
The ultimate goal of feedback is to make the instructor as unnecessary as possible. Writers will hone their messages based on their understanding of what they’re trying to achieve inside a rhetorical situation when considering of audience, message, purpose, and genre. The real-time feedback becomes an internal reflective process that’s part of the writer’s practice.
Essentially, I know that a student is doing well at learning to write when, as the instructor, I fade further and further into the background.
By the end of a course, I hope to disappear altogether.
 For funsies I’m going to note some things that an AI-based feedback algorithm might flag or fail to understand in order to demonstrate the limits of automated real-time feedback. In this case, the spellcheck identifies the misspelling but because I am not a student and have the freedom to write in the manner I desire, I have overridden this alert in the name of establishing my particular (perhaps peculiar) authorial voice.
 There will never be an algorithm which would correctly identify the use of sarcasm here, which my human audience easily picks up.
 More sarcasm, undetected by my theoretical AI writing feedback machine.
 “Incorrect,” says spellcheck, ignoring the virtues of my folksy charm.
 My AI writing feedback machine just told me I can’t start a sentence with a contraction. It’s particularly put out by starting a new paragraph with a sentence that starts with a contraction. I told it to get over itself.
 Starting a sentence with a contraction again. One thing you’ll note about my blog posts is that I tend toward very short paragraphs, and I often break a paragraph right before a sentence which really punches my claim home. It’s deliberate and tailored to a medium where skimming may be happening. I don’t want the audience who is reading quickly to miss the punch. An AI writing feedback machine would tell me I can’t have a paragraph that’s a single sentence long.
 Smoke is now coming out of the AI writing feedback machine because I am ignoring its warning that I have repeated the same phrase three times in a row. Does not compute! Does not compute! I’ll stop with this; you get the idea. (I can’t do a footnote in the footnote, but an AI writing feedback machine would flag that use of “you” as incorrect, even though it’s purposeful.)
 Because of standardized tests, students are essentially trained to work in a single genre, a genre which doesn’t exist outside of standardized tests. An AI feedback machine could not even handle the nuances between an essay and a blog post, where audience expectations for correctness and completeness are different. It definitely cannot handle any form which benefits from figurative language of any kind. Standardization only becomes a virtue because it makes evaluation more efficient, be it done by human or algorithm.
 The AI feedback definitely can’t appreciate the sheer artistry of concluding with this sentence, where the author literally disappears after saying his goal is to disappear, and I’m gilding the lily pointing it out, but it’s my blog, and I’ll footnote as I want to. (An AI would also have a very hard time picking up the fact that that last sentence is referencing Lesley Gore’s “It’s My Party and I’ll Cry if I Want to,” because as I think I’ve mentioned, computers can’t read.)