A.I. Writing in the Classroom

What We Can do about A.I. Writing in the Classroom

The following points of advice have been synthesized from other sources that have developed practical strategies for instructors to avoid students defaulting to A.I. writing software at the expense of their own learning.

1. Require students to cite verifiable sources and use quotations.
2. Analysis of / responses to details from multimodal material such as images, audio, or video.
3. Analysis of longer texts (that can’t fit into the limited prompt window for ChatGPT and the like) or more recent publications that have not been integrated into the language model.
4. Writing that draws on domain-specific concepts & frameworks from your particular class.
5. Assignments that ask students to articulate nuanced relationships between concepts or ideas.
6. Develop assignments with iterative steps / multiple drafts and review checkpoints. This will not only side-step students feeling that they have to “rush” to complete a draft on the dreaded “night-before” the assignment is due, it will also help them genuinely learn writing as a process.
7. Assign short written reflection components to essays, which require students to articulate their decision-making process behind their writing.

 

What We Should Not Do about A.I. Writing in the Classroom

• Require handwritten assignment submissions. Writing by hand is difficult / impossible for students with certain disabilities and students who use voice-to-text software. You may consider by-hand writing as an option, but requiring it abridges good universal pedagogical design.
• Adopting surveillance tools. For-profit software that records students’ entire writing process are incredibly intrusive and potentially exploitative in how they collect data on students.
• Using software that purports to “identify” A.I.-generated writing as a “plagiarism checker.” Generally, it’s better to trust students to submit their own work rather than presume every student wants to cheat. Running every student submission through an “A.I.-detection” plagiarism-checker is not only technologically cumbersome and time-consuming – because LLM discourse is just remixed human-generated language, it is difficult (if not impossible) to verify whether a document is A.I.-composed – current software options are rife with false positives & negatives.