Preserving Authenticity in Student Writing in the Age of Generative AI

Ideogram: student writing synth wave

A huge part of the angst about generative AI centers around the generalized concern that students will use it to cheat. News flash, they will.  

Student will cheat with AI because there have always been students who cheat, this is just a new way for them to do it. Cheating is not new. But, let's put this problem in perspective. Risk taking is normal teenage behavior. All teens take risks, some drive too fast, some jump off roofs into swimming pools, some drink in excess, some lie to their parents, and some turn in work they did not do. Some teens are more cautious and some are less cautious. Brain science, and yes I have read about this with my students, shows us that teens value rewards higher in many risky situations, therefore making poor decisions seem more attractive. 

I am not trying to excuse academic cheating. It's just that I stopped viewing it as a moral failing, and I now see it as part of the risk taking, adventuresome, process of becoming an adult. I would much rather see a teen take a risk academically than jump off a roof, although I do fear that there is a strong correlation between academic cheating and other risky behaviors. 

How to Stop Students from Cheating With AI

Then the next stage of the moral panic for teachers is how to stop students from cheating with AI. Many educators, having only recently found the amazing affordances of digital pedagogy thanks to the pandemic, are now sighing and resorting back to pencil and paper tasks. But, we don't have to go backward. 

I'll lay out some of the things that have helped me teach with digital tools and deter plagiarism of all types, including generative AI. 

When it comes to getting authentic writing from students, proactive is better than reactive. If I assign an impersonal essay and my students never work on it in class, never bring a draft to writing groups, never self assess and revise, and never confer with me about their progress, then of course I am going to see a high rate of students using generative AI to write for them. Students are most likely to plagiarize when they don't feel like they can do the task on their own, don't have support to do the task, and don't care about the topic. 

I have found my best defense against AI writing is showing my students that I know a lot about AI writing. I use ChatGPT in front of them. I talk about the ways I've caught students cheating. I teach them ethical ways to use AI to help themselves without being academically dishonest. 

I make them sign academic honesty pledges before exams and I inject a bit of arrogant humor into those statements. 

When I create a writing assignment: 

Assignments that students can in some way personalize are always better than those that are totally impersonal. When we read Into The Wild, my seniors compare McCandless to themselves or someone they know. When my 9th graders write narratives they create their own super hero version of themselves to use as their main character. When my students research topics in the news, they choose their own topic, based on their own interests. 

Choice and personalization means students are more invested in the work they are doing and the assignment itself won't work out well when fed into an impersonal writing generator. 

I set the assignment up so that each student gets their own copy of a Google Doc. I explain to them that doing their own writing in their own Google Doc is their best option to defend themselves against any accusation of plagiarism because they will be able to use their own revision history to show they did their own work. (Many high achieving students are justifiably worried about being accused of using AI even when they didn't. The high rate of false positives from ineffective detectors has them worried too.) 

I make sure my assignments have appropriate scaffolds. For many students it means giving them outlines or sentence frames. (I gradually withdraw that support as their skills improve.) Support means we do similar tasks multiple times while students improve. It means we do a lot of pre-writing work, designing characters, planning a plot, looking at models. It means we do process work like a lesson on dialog after they have a draft, and then giving them time to go add dialog to their story. 

Support means my students often collaboratively gather evidence for their writing before they get the writing assignment. I wrote about scaffolding evidence gathering here. Then they get a spreadsheet of evidence to help them with their essays. (I want them writing, not pouring through a book looking for a quote.)

Support means I show them how AI can help them ethically. I wrote about the lessons I used to show my students how I can ask it for an outline, a time management plan, and feedback. 

While they work on the assignment

Writing is a process. Teaching writing is a process. I have always wanted my students to spend time drafting in class. I use that time to confer with them about their progress. I use the preview function in Google Drive to flip through their essays and see who is and is not making progress, and then I use that to prioritize my conferences. 

My students bring their drafts to writing groups. I have written about Writing Groups here including a guide for getting them started in your own classroom. 

While I assess student writing

First of all I read it. It amazes me that sometimes, when I catch plagiarism on a small assignment, and I talk to the student about it, their first reaction is often, "I didn't think you'd read that." Somewhere in the education pipeline they had teacher(s) who gave full credit to every student who turned in something and these students developed the theory that their teachers don't really read their work. 

If, in reading it, I see something that does not seem like the student wrote it, then I investigate. I tell my students that the best plagiarism detector is 28 years of being an English teacher. I would add, "Go ahead, make my day," but I don't think they would get the reference. 

My first line of investigation is usually to run it through an AI detector. I like the Brisk extension for Chrome that works in Google Docs. Origin from GPTzero is another extension I like because I can use it in Canvas speed grader for text submissions. Regardless of the result from the detector I still look at the version history to see if large chunks were pasted into the doc. And I can also use the Revision History extension to see some useful stats about the doc. It will also play a video version of the writing process. 

The version history under the docs file menu also can show me what they worked on and when. 

Below is a screen shot of a student essay in development. I used arrows to mark some of the things that I mentioned above. At the top is the Origin writing report button indicating 5,748 edits. A high number of edits indicates authentic work. Clicking that button generates a more detailed report. The beige bar just above the essay is from Revision History. It lists useful stats like writing time etc. The comments on the side are from peers who read the work in a writing group. Students know their peers will call them out if they try to pass off AI writing as their own. Presenting writing to a group means the student was prepared before the deadline and this is another sign of authentic writing. Lastly, on the lower left is the Brisk extension icon. Clicking that brings up four standard options, one of which is to detect AI writing. 

When I find writing a student did not write themselves

Yes, I do all of the above and I still find students who try to pass off writing that is not their own. See above about risk taking. Students with high absence rates are also more likely to turn in things they did not write because they were not in class for all the support described above. 

The response to inauthentic writing depends on a lot of factors, the assignment, the student, the time of year, prior infractions, my best guess about the reason for the in authentic writing. Every single situation of plagiarism is different, and how I handle mine might not be the way you want to handle yours. 

I will share this strategy that I used last year and again this year. It was very effective for showing my students that AI writing is very easy to detect. It helped that this year, the AI writing I used for the lesson was submitted by a student who was in the room. They flagged their own submission as being written by AI without recognizing that it was the work they turned in. 

In the past I have also blogged about an entire unit my team created about Academic Honesty and about plagiarism and tricks for preventing it in general. 

Generative AI does not mean we give up on teaching students to write, but it might mean we need to adapt the way we do that. If you are thinking, "Wow Jen, that sounds like you spent a lot of time on this," well I did. It is literally my job. The things I described above are what I am doing when I teach. So, yeah, most of the time I spend five days a week on it. 

If you need to add more scaffolding, or writing groups, or some other strategy to your writing instruction toolkit, then those things might take you some time. I recommend you check out the resources available through your closest Writing Project.