CTL’s Evolving Guidelines for Dealing with chatGPT (last updated Feb 6, 2023)
[Note: this page was summarized as a handout for our chatGPT discussions on February 6th, 8th, and 10th.]
The following evolving guidelines are designed to help you address the challenge of teaching in the age of chatGPT while still emphasizing student learning and allowing for meaningful student assessment. Please treat these as preliminary guidelines as Brandeis develops its official policies.
Faculty members will differ on the extent to which they want to use or restrict the use of chatGPT in their courses. We recognize that every course is different and that every discipline will have unique concerns. Please reach out to the CTL at firstname.lastname@example.org with any specific issues you face or your feedback on these guidelines.
What does chatGPT struggle to do?
While chatGPT is pretty good at creating cohesive writing, computer code, and images, they aren’t able to:
- describe how (and why) a student revised their essay from their first draft to final drafts
- reflect on how an essay made use of feedback provided during peer review sessions
- refer to comments brought up during your class discussion
- refer to personal anecdotes
- refer to course materials
- effectively use sources
- provide accurate citations, a works cited list, or an annotated bibliography
- include sounds or video
- collect data or do interviews
Moreover, chatGPT often returns made-up sources or made-up quotes and often contradicts itself, such that users must always evaluate and verify any outputs generated by chatGPT. (As the program’s developers note in their blog, “ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers.”)
We therefore suggest…
1. When designing individual assignments:A. Incentivize the process, rather than just the final written product. Include the steps and habits of mind that are associated with deep learning and critical thinking in your discipline in their assignment. Have due dates for individual elements that precede the final submission. For example, ask students:
- First, to write a bullet-pointed outline with a thesis statement
- Next, to provide detailed notes on sources (research articles, literary critiques, original documents, etc.)
- Next, ask for a first draft / first best attempt.
- Next, have students provide each other with feedback via peer review.
- Then ask for a student to submit a final draft and a reflective paragraph/essay describing how their paper evolved throughout the process and made use of their peer’s feedback.
B. Assign more personalized writing, the more personalized the better. For example, ask for short, personal response essays to weekly readings.
C. Require that your students refer to material specific to your class, such as in-class discussions, LATTE discussions, or other unique materials.
D. Reference current events in your writing prompts/essay topics. For example, ask students to apply a concept or topic for your course to a recent event or discovery.
E. Ask students to use evidence and cite their sources in their papers.
F. Incorporate peer review of drafts and ask students to write reflective paragraphs about how they made use of their peer’s feedback as they finalized their papers.
G. Develop assignments that require original data collection and analysis through interview, observation, fieldwork, archival research, or other methodology.
H. Assign your students to create multi-modal essays that require sound, images, and video.
2. When thinking about your overall course structure and assessment practices:A. Consider lowering the value of any single homework assignment by offering more frequent, lower-stakes in-class or homework assignments. Students are more likely to consider cheating on higher-stakes assignments.
For example, avoid homework assignments that are worth more than 25% of a student’s final grade.
B. Consider short writing assignments during class (these can eventually become incorporated into students’ longer papers or can be stand alone assignments).
C. Consider how chatGPT intersects with the goals of your course.
D. What intellectual skills do you want your students to develop? Are there ways they can practice and develop those skills without AI assistance?
E. What types of assignments are necessary to measure the extent to which students have learned the desired skills and concepts? Are there ways to ask them to demonstrate their progress towards learning these skills and concepts without AI assistance?
F. Consider changing the modality of your assessments. Can some written assignments be turned into oral presentations or podcasts?
3. Talking with your students about the value of writing in your course and discipline
- Some students need a reason to want to do their own writing (rather than have chatGPT do it for them). Find opportunities to explain the value of writing in your course and discipline and to engage your students in discussions about the value of writing.
- Make clear that while the written piece they produce matters, the most important aspect of writing is that it facilitates and enables their learning.
4. Addressing AI tools in your syllabus
Different faculty will have different expectations about whether and how students can use AI tools, so being transparent about your expectations is essential. If you want to forbid using AI tools, be explicit about this on your syllabus. If you allow these tools, but want them to be acknowledged in the student’s work, explain that on your syllabus and in class.
Here is some language you may consider including in your syllabus or using to discuss chatGPT with your students:
- It is important to remember that chatGPT and other AI tools are not a replacement for your own critical thinking and original ideas. The ultimate goal of this course and any tool used to submit work is to enhance your own learning and understanding, not to undermine it.
- As a college student, it is your responsibility to maintain the highest standards of academic integrity. Representing work generated by artificial intelligence as one's own work is considered to be academically dishonest. This includes (a) ensuring that all work submitted for grades is your own original work, and (b) properly citing any sources that you use.
- Having AI write your paper constitutes plagiarism. If the source of the work is unclear, I may require you to meet with me to explain the ideas and your writing process.
- If you consult with other students or use any sources on an assignment, report this in the work that you turn in. Do not generate new content with prompt-based AI tools like ChatGPT or CodePilot without permission from instructors unless specifically allowed by the assignment. (Using, for example, Grammarly as a language aid is OK.) Instructors reserve the right to request an oral explanation of answers.
- If you have questions about what is permitted, please reach out to me.
- Please see here for more examples of possible language you can include in your syllabus regarding AI tools.
5. Regarding AI detection tools:A. While there are emerging tools designed to detect AI-produced content (including Hugging Face, GPTZero, CrossPlag, and Turnitin), we recommend against relying on them for several reasons:
These tools are not perfect.
- Their false positives may cause faculty to falsely identify a student’s own work as being AI-generated. Chasing down these false leads can waste time and inspire unnecessary mistrust between faculty and students.
- Their false negatives may cause faculty to falsely ascribe an AI’s work to the student.
B. These tools are generally only effective when the entire piece was produced by AI, but if the student has included some AI-generated work interspersed within their own words, it is much harder for these tools to detect.
C. Students are already coming up with ways to beat AI-detection tools: one prominent approach is to ask one AI-generating tool (e.g., chatGPT) to create the first draft of an assignment, and then to ask another AI-generating tool (e.g., CopyGenius and Quillbot AI) to rephrase each paragraph, making it much harder to detect.
D. AI-generating tools will evolve to evade AI-detecting tools, and then AI-detecting tools will evolve to get better at identifying AI-generated work. The arms race to detect AI-generated work and the time faculty spent staying current on the best AI-detection tools may have a higher cost than the benefit these tools provide.
E. Some AI-generated tools, such as chatGPT, may develop “watermarks” that allow the text they generate to be easily identified, but as AI-generating tools proliferate, some students will migrate to tools that don’t include these watermarks when they feel incentivized to do so.
In sum, tools that claim to detect AI-generated work are easy for students to work around if they want to, and it can be easy for faculty to invest large amounts of time into policing their students instead of investing time into fostering student learning. Rather than relying on AI-detection software, we recommend designing assignments that foster student learning that cannot be replicated by AI-tools (such as in-class writing; peer review; small, frequent, personal essays; essays that cite scholarship in the field or classroom discussions; projects that involve interviews or original data original; essays that cite current events; assigning multi-modal assignments, podcasts, explainer videos, or any of the other principles outlined above.)
In addition, some faculty may also want to use chatGPT in their course. For example, one can imagine that students can use chatGPT to apply concepts from class to generate initial ideas for papers or research projects, or to analyze and refine early drafts. Moreover, incorporating AI tools into one’s course can also provide training and practice for students entering professional environments where the use of AI will be a relevant skill.
- For some preliminary guidelines to consider if you are interested in using chatGPT in your course, click here.