Bad AI Writing Advice

There is so much bad advice for teachers out there concerning how to use AI in the classroom. Some of the worst advice surrounds AI use for writing assignment, and most of that bad advice is rooted in a fundamental misconception of what the purpose of the writing assignment might be.

I have a prime example here. It comes from Dan Sarofian-Butin who is a Full Professor in, and Founding Dean of, the Winston School of Education & Social Policy at Merrimack College, which is itself a pretty snappy little college just a stone's throw from Boston. His CV is a hell of a lot more impressive than mine. Nevertheless, I take considerable exception to his advice about the use of ChatGPT or similar LLMs in student writing assignments.

His piece is entitles "Teachers: It's time to make friends with AI" recently on eSchoolNews (though he has put out other pieces of a similar bent).

He opens by noting that the norm (at least in high schools) is that "AI is the mortal enemy of classroom teachers," while he wants his students "to use AI every day in class and for every assignment." He later describes a gap between "cognitive autonomy" (no AI) versus "cognitive outsourcing" (just have AI write the paper for you). Right off the bat, I feel that he's skipping over a continent's worth of middle ground, but okay. What should we be using our new best friend?

Sarofian-Butin offers an interesting taxonomy of the various degrees of having AI part of the process, noting minor and major amounts of "cognitive offloading." And that's useful because that's the territory where all of these discussions need to happen. Nobody (well, hardly anybody) is seriously arguing to have students just let the AI do it all, and folks who are anti-any-AI-at-all aren't going to be part of the conversation. For everyone else, it's an exercise in line drawing--which parts of the writing process can or should involve an LLM?

Sarofian-Butin has some answers. I don't much like any of them.

Sarofian-Butin thinks AI can be used as "scaffolding," particularly with the business of getting the writing started. He says that students used to come to his office unsure how to start their papers and he would spend 15-20 minutes brainstorming and prodding and pushing. But now...

Today, I teach my students a set of in-class AI prompts, based on a standard model of supporting writing, on how to brainstorm, focus, and develop their ideas. “I didn’t really know where to start,” wrote one student at the end of last semester, “and ChatGPT helped me think about questions, and I was able to start planning what I wanted to do based on the different options.” Another student wrote, “I started off with pretty much no idea and was able to use ChatGPT to find a topic that I’m interested in and I’m working with it to narrow it down.” When I now meet with students, our conversations are so much more productive, as we now have a focus.

Which doesn't sound so much like brainstorming as just generating a list of ideas from which the student can choose. I have had my share of those 15-20 minute sessions with students, and I am having a hard time imagine how one does that in a way that puts the work on the student, that helps them probe their own interests and half-formed ideas aided by what you know of the student and what you can see in their face and voice as they discuss--how do you do all that if you are a computer that has zero perception of the student themself?

But Sarofian-Butin sees even more involved roles for the AI. Some of his topics are complex. So many variables, so much ambiguity, so many ways to define the issue. They're, you know, hard.

I therefore teach my students another set of AI prompts to help them see what good thinking about such issues looks like. This is formally known as a cognitive apprenticeship: “one needs to deliberately bring the thinking to the surface, to make it visible, whether it’s in reading, writing, problem solving.” AI is so good at doing this by walking students step-by-step through its output.

And now alarm bells are ringing, because AI is NOT so good at walking students through its output because it does not "think about" ideas in any human sense of the word. It cannot "bring the thinking to the surface" because it is literally not thinking at all. And some of the other tasks that Sarofian-Butin assigns to his composer's apprentice--

Seeing AI offer suggestions for a thesis statement or a paper outline in real-time, with explanations, is incredibly helpful. “The outlines,” one student commented, “helped me from getting too stuck on small details and reminded me to think about the big picture.”

He also suggests that AI might help students can find answers to "am I making the right argument" and again, an AI does not know anything about how good your argument is or is not.

He reports that a student said that they know that ChatGPT is there to use as an assistant rather than a replacement.

Bad AI writing instruction advice all suffers from the same problem-- it presumes that the only purpose of the writing is to create the final product, an artifact to be handed in. As long as you have a final artifact to deliver to your professor, then the process is of secondary importance.

No. We can say that we want every player on the football team to log an hour in the weight room three days a week. But that's hard, and the players are reluctant, and they're not sure they can manage it, so they go to the weight room and someone else puts the weights on, and someone else lifts the weights, and another person lowers the weights back down, and then the player fills out his log, and that final product, that log-shaped artifact is perfect and exactly what the coach asked for--except that it's not.

Writing is about making thinking manifest. Many of the problems Sarofian-Butin is address with AI are thinking problems, not writing problems. So what happens when we outsource the thinking parts of writing?

I'm trying to figure out what a Sarifian-Butin student has actually done. The student selected a topic from an AI-generated list, picked out an AI thesis "example," followed the AI generated outline, made AI-suggested improvements, all while reading AI-generated "explanations" of the AI "process " (that are not actually a real explanation of how a real human might have done it).

What has the student gotten from this process? What mental muscles did they develop? What critical parts of the writing process did they complete beyond filling in the blanks laid out by someone else? How can one know if they have used the AI as a crutch or had it carry them entirely? How is this superior to, say, watching someone else write an essay while explaining what they are doing? What problem is this solving (beyond a time-sucking parade of wobbly students asking for 15-20 minutes of advice, which is not a student problem)?

How is any of this better than leaving them to struggle on their own?

Yes, I know-- left to their own devices, they will produce some really terrible essays. Believe me-- I may not match Sarofian-Butin's credentials in any other way, but after 39 years in a high school English classroom, I will bet I've read far more terrible writing than he has. And not once did I think, what this student needs is something that can do all the hard part for him. Did I think some could, would, and did benefit from human-to-human tutoring? Absolutely--but that involves a human being who can read them, hear them, respond to them, draw them out and sense when to back off.

The thing about those terrible essays is that you don't get students to do better by doing the hard parts for them. They have to struggle and work and you have to coach and cajole and hold hands and kick butts and let them find their own voice and their own way.

This is at the heart of most student endeavors. I was a yearbook advisor for ages, and there is no question that they best way to get a good yearbook is to shove the kids out of the way and do it yourself. What do they get from that? Not a damned thing, but the book would look good. You could have a much more beautiful prom if you let adults do the decorating.

And you would get much better student writing if you didn't leave it to students.

But the product is not the point. The struggle, the growth, the learning, the human interaction, the heavy lifting is the point. Trying to reduce student involvement in the process gets a better product, but that can't be the whole point. Everything in education would run so much more smoothly if not for all the children.