Six Things I Learned After a Semester of Lots of AI

Two years ago (plus about a week!), I wrote about how “AI Can Save Writing by Killing ‘The College Essay,'” meaning that if AI can be used to respond to bad writing assignments, maybe teachers will focus more on teaching writing as a process the way that scholars in writing studies have been talking about for over 50 years. That means an emphasis on “showing your work” through a series of scaffolded assignments, peer review activities, opportunities for revision, and so forth.

This past semester, I decided to really lean into AI in my classes. I taught two sections of first-year writing where the general research topic for everyone was “your career goals and AI,” and where I allowed (even encouraged) the use of AI under specific circumstances. I also taught an advanced class for majors called “Digital Writing” where the last two assignments were all about trying to use AI to “create” or “compose” “texts” (the scare quotes are intentional there). I’ve been blogging/substacking about this quite a bit since summer and there are more details I’m not getting to here because it’s likely to be part of a scholarly project in the near future.

But since the fall semester is done and I have popped the (metaphorical) celebratory bottle of bubbly, I thought I’d write a little bit about some of the big-picture lessons about teaching writing with (and against) AI I learned this semester.

Teachers can “refuse” or “resist” or “deny” AI all they want, but they should not ignore it.

As far as I can tell from talking with my students, most of my colleagues did not address AI in their classes at all. A few students reported that they did discuss and use AI in some of their other classes. I had several students in first-year writing who were interior design majors and all taking a course where the instructor introduced them to AI design tools– sounded like an interesting class. I had a couple of students tell me an instructor “forbid” the use of AI but with no explanation of what that meant. Most students told me the teacher never brought up the topic of AI at all.

Look, you can love AI and think it is going to completely transform learning and education, you can hate AI all you want and wish it had never been invented and do all you can to break that AI machine with your Great Enoch sledgehammers. But ignoring it or wishing it away is ridiculous.

For my first-year writing students, most of whom readily admitted they used AI a lot in high school to do things that were probably cheating, I spent some time explaining how they could and could not use AI. I did so in part to teach about how I think AI can be a useful tool as part of the process of writing, but I also did this to establish my credibility. I think a lot of students end up cheating with AI because they think that the teacher is clueless about it– and I think a lot of times, students are right.

You’re gonna need some specific rules and guidelines about AI– especially if you want to “refuse” or “resist” it.

I have always included on my syllabi an explicit policy about plagiarism, and this year I added language that makes it clear that copying and pasting large chunks of text from AI is cheating. I did allow and encourage first-year writing students to use AI as part of their process, and I required my advanced writing students to use AI as part of their “experiments” in that class. But I also asked students to include an “AI Use Statement” with their final drafts, one that explained what AI systems they used (and that included Grammarly), what prompts they used, how they used the AI feedback in their essay, and so forth. Because this was completely new to them (and me too), these AI Use Statements were sometimes a lot less complete and accurate than I would have preferred.

I also insisted that students write with Google Docs for each writing assignment and for all steps in the process, from the very start of the first hint of a first draft until they hand it into me. Students need to share this with me so I can edit it. I take a look at the “version history” of the Google Doc, and if I suddenly see pages of clear prose magically appear in the essay, we have a discussion. That seemed to work well.

Still, sometimes students are still going to cheat with AI, and often without realizing that they’re cheating.

Even with the series of scaffolded assignments and using Google Docs and all of my warnings, I did catch a few students cheating with AI in both intentional and not as intentional ways. Two of these examples were similar to old-school plagiarism. One was from a student from another country who had some cultural and language disconnections about the expectations of American higher education (to put it mildly); I think first-year writing was too advanced and this student should have been advised into an ESL class. Another was a student who was late on a short assignment and handed in an obviously AI-generated text (thanx, Google Docs!). I gave this person a stern warning and another chance, and they definitely didn’t do that again.

As I wrote about in this post about a month ago, I also had a bunch of students who followed the AI more closely than the first assignment, the Topic Proposal. This is a short essay where students write about how they came up with their topic and initial thesis for their research for the semester. Instead, a lot of students asked AI what it “thought” of their topic and thesis, and then they more or less summarized the AI responses, which were inevitably about why the thesis was correct. Imagine a mini research paper but without any research.

The problem was that wasn’t the assignment.  Rather, the assignment asked students to describe how they came up with their thesis idea: why they were interested in the topic in the first place, what kinds of other topics they considered, what sorts of brainstorming techniques they used, what their peers told them, and so forth. In other words, students tried to use the AI to tell them what they thought, and that just didn’t work. It ended up being a good teachable moment.

A lot of my students do not like AI and don’t use it that much. 

This was especially true in my more advanced writing class– where, as far as I can tell, no one used AI to blatantly cheat. For two of the three major projects of the semester, I required students to experiment with AI and then to write essays where they reflected/debriefed on their experiments while making connections to the assigned readings. Most of these students, all of whom were some flavor of an English major or writing minor, did not use AI for the reflection essays. They either felt that AI was just “wrong” in so many different ways (unethical, gross, unfair, bad for the environment, etc.), or they didn’t think the AI advice on their writing (other than some Grammarly) was all that useful for them.

This was not surprising; after all, students who major or minor in something English-related usually take pride in their writing and they don’t want to turn that over to AI. In the freshman composition classes, I had a few students who never used AI either–judging from what they told me in their AI Use statements. But a lot of students’ approaches to AI evolved as the semester went on, and by the time they were working on the larger research-driven essay where all the parts from the previous assignments come together, they said things like they asked ChatGPT for advice on “x” part of the essay, but it wasn’t useful advice so they ignored it.

But some students used AI in smart and completely undetectable ways.

This was especially true in the first year writing class. Some of the stronger writers articulated in some detail in their AI Use Statements how they used ChatGPT (and other platforms) to brainstorm, to suggest outlines for assignments, to go beyond Grammarly proofreading, to get more critical feedback on their drafts, and so forth. I did not consider this cheating at all because they weren’t getting AI to do the work for them; rather, they were getting some ideas and feedback on their work.

And here’s the thing that’s important: when a student (or anyone else) uses AI effectively and for what it’s really for, there is absolutely no way for the teacher (or any other reader) to possibly know.

The more time I have spent studying and teaching about AI, the more skeptical I have become about it. 

I think my students feel the same way, and this was especially true with the students in my advanced class who were directly studying and experimenting with many different AI platforms and tasks. The last assignment for the course asked students to use AI to do or make something that they could not have possibly done by themselves. For example, one student taught themself to play chess and was fairly successful with that– at least up to a point. Another student tried to get ChatGPT to teach them how to play the card game Euchre, though less successfully because the AI kept “cheating.” Another student asked the AI to code a website, and the AI was pretty good at that. Several students tried to use AI tools to compose music; similar to me I guess, they listen to lots of music and wished they could play an instrument and/or compose songs.

What was interesting to me and I think most of my students was how quickly they typically ran into the AI’s and their own limitations. Sometimes students wanted the AI to do something the AI simply could not do; for example, the problem with playing Euchre with the AI (according to the student) is it didn’t keep track of what cards had already been played– thus the cheating. But the bigger problem was that without any knowledge of how to accomplish the task on their own, the AI was of little use. For example, the student who used AI to code a website still had no idea at all what any of the code meant, nor did they know what to do with it to make it into a real website. Students who knew nothing about music who tried to write/create songs couldn’t get very far. In other words, it was not that difficult for students to discover ways AI fails at a task, which in many ways is far more interesting than discovering what it can accomplish.

I’m also increasingly skeptical of the hype and role of AI in education, mainly because I spent most of the 2010s studying MOOCs. Remember them? They were going to be the delivery method for general education offerings everywhere, and by 2030 or 2040 or so, MOOCs were going to completely replace all but the most prestigious universities all over the world. Well, that obviously didn’t happen. But that didn’t mean the end of MOOCs; in fact, there are more people taking MOOC “courses” now than there were during the height of the MOOC “panic” around 2014. It’s just that nowadays, MOOCs are mostly for training (particularly in STEM fields), certificates, and as “edutainment” along the lines of Master Class.

I think AI is different in all kinds of ways, not the least of which is AI is likely to be significantly more useful than a chat box or to check grammar. I had several first-year students this semester write about AI and their future careers in engineering, logistics, and finance, and they all had interesting evidence about both how AI is being used right now and how it will likely be used in the future. The potential of AI changing the world at least as much as another recent General Purpose Technology, “the internet,” is certainly there.

Does that mean AI is going to have as great of an impact on education as the internet did? Probably, and teachers have had to make all kinds of big and small changes to how they teach things because of the internet, which was also true when writing classes first took up computers and word processing software.  But I think the fundamentals of teaching (rather than merely assigning) writing still work.