Brainstorming and Style with ChatGPT
Submitter: J. Seth Lee, Slippery Rock U
——————————————————
The experiment:
This activity had three goals: introduce ChatGPT as a potential writing tool, discuss how it could work as a brainstorming resource, and prompt discussion about style and the need to revise AI generated text.
Classes began this activity by dividing up into groups of five. After signing up for a free OpenAI account, each group received a rhetorical technique (e.g. humor, narration, reiteration) which they needed to define for the class using their textbook. That done, each group choose from a series of questions I provided for the activity. These included such questions as “What is the correct orientation for toilet paper?” and “What is the greatest movie of all time?” Each group then formed a working thesis answering their question before turning to ChatGPT to brainstorm. For example, one group asked ChatGPT to “Help argue toilet paper should be placed over the top using humor.” The groups then copied and pasted the question they asked ChatGPT, its response, and their group’s substantive revision into a GoogleDoc.
Prior to the next class, I lightly edited the GoogleDoc’s formatting to create a worksheet containing all the groups’ prompts, ChatGPT suggestions, and the groups’ revisions. During class each student received a copy of the worksheet and spent approximately 20 minutes reading and annotating the AI generated responses and the group written revisions. As a class we then discussed the differences in style between AI and human generated text.
Results:
I did not include a formal assessment of this assignment aside from class engagement. Based on class discussion, I believe the assignment was a success. Seeing multiple examples of AI-generated text, students rapidly picked up on the blandness of the text generator’s style. One student described it as “soul-less” and another noted that ChatGPT tended to use the same sentence structure repeatedly. This led to further discussions about the human-written revisions.
In most cases, those revisions were substantive enough to avoid concerns about misusing ChatGPT. Serendipitously, each class had at least one revision that reused too much of the the generated text than was responsible. In two of the four classes, students did pick up on the revisions that were problematic. The other two classes did not, which required a good bit of prompting on my part before they saw the problem.
I will almost certainly use this assignment again with some tweaks. One of the rhetorical techniques I assigned was narrative. The groups given that technique all ended up with a text generated story, but the amount of time set aside for the assignment did not give them time to revise substantially. I will also turn the assignment into a three-day sequence for a fifty-minute class. The first day we’ll create the worksheet. Day two will focus on style, and the final day will focus on revision. Covering style and revision related to this activity in one class period felt rushed.
Relevant resources: I have linked a sample worksheet from this assignment here: https://shorturl.at/bdOV3. Student names are redacted for privacy.
Contact: joshua.lee[AT]sru[DOT]edu

Leave a Reply