Academic Honesty and Integrity

Strategies to Manage ChatGPT

Now that you’ve had some time to learn about AI-assisted writing engines, I wanted to share some information about how to manage them in the short term. This challenge will change significantly in the long term, as the engine learns and develops and as programs (from Turnitin and others) come to market to address and combat it. In short, the challenge before us this semester will be different from the challenges we will face next fall and beyond.  

Here is what I’m planning to do in my class this semester:

  • Move shorter writing exercises (Discussion Board posts, short response writing, etc) to in-class writing. Right now, ChatGPT appears to limit itself to shorter responses to prompts. It’s unusual for it to produce anything over 300 words to a single prompt. It can be prompted to produce more, however, by writing “Continue,” but it’s unclear how novel those subsequent responses are or if it is just regurgitating the information it produced initially. I know that this won’t work for everyone, but, where feasible, this small adjustment can help limit AI-generated work in those shorter responses where it appears to be most lethal to authentic engagement.
  • I will require that students produce multiple drafts of longer writing assignments and will evaluate those drafts for increasing competency and sophistication. In terms of quality, ChatGPT’s responses are, for the most part, static. There isn’t an “improve this draft” function, yet. By requiring multiple drafts, I’ll be able to see if a student incorporated feedback and improved their writing (still a hallmark of authentic written work). Again, this won’t work in every class and setting. However, anything you can do to promote the idea of writing as a process (and not merely a product that must be churned out), will lead to more authentic work and to students seeing the value of writing as a recursive, intellectual activity.
  • Use Turnitin, but not in the way we typically do: I’m planning to feed my longer prompts (larger writing assignments and research essay prompts) into ChatGPT, and then submit those responses to Turnitin using the “Test Student” function in Canvas. I’m doing this because I think what ChatGPT produces to a given prompt isn’t that novel from user to user. My suspicion is that if a student uses ChatGPT using my prompt, some portion of their writing will be identified as similar to the version Turnitin now has on file via my “Test Student.”
  • In situations where I suspect a response is not authentic, I will use GPTZero (LINK), the program created by a Princeton student that attempts to identify whether a text was created by AI (READ ABOUT THAT HERE), to attempt to collect more information. This approach has its flaws. The program can be confusing and difficult to use, the results are not definitive and it’s unclear how it arrives at the conclusion it does. However, I see this tool as something I would use IF I suspected a piece of being inauthentic, but BEFORE I met with the student to discuss the concern. I think it can be part of your evidence/information collection process, but should not be the only action you take to determine the piece’s authenticity.
  • Lastly, I am going to talk with my students about AI-assisted writing and why the university considers using this technology in graded work to be academic misconduct. I am not concerned that this is teaching them about it. They already know it exists. Instead, I think that this conversation is an important opportunity to explain my expectations for their work, for us to have a dynamic conversation about the world we’re entering into, and to reinforce the value of writing as an intellectual and educational activity. Am I going to dissuade every student? No, but making the expectation clear will reinforce the students who are going to do it the right way and make the conversation easier for both of us for the student who tries to do it the wrong way.