Consider revising your assignments to disincentivize shortcuts, to make learning goals transparent, and to prioritize student voice and ideas.
There is no way to make an assignment completely GPT-proof, but there are three key paths available to educators: a) crafting assignments that make using GPT less useful and effective, b) foregrounding WHY we are doing assignments and what students will lose from outsourcing their learning to a chat bot, and c) changing assessments so that they privilege process over product and individual creativity.
Crafting assignments that make ChatGPT less effective and useful
Make your assignments very course-specific, drawing on information that has come up in discussions, data that students have had to gather themselves, or concepts that have emerged from the particular trajectory of your syllabus.
Consider testing your assignments on ChatGPT or related technologies to see what it does with your prompts/assessments.
Link class work with current events or contemporary issues, since training data always lags behind (for example, ChatGPT relies on data from 2021 and before).
Scaffold your assignment with multiple steps and checkpoints for feedback: i.e., proposal, bibliography, presentation, draft, etc. You may even consider integrating some in-class checkpoints (i.e., give a 2-minute verbal overview of your paper, brainstorm or peer review with a partner, or write your developing thesis on an index card).
Ask for “why” statements and/or metacognitive reflection. If you have an assignment to produce a bibliography, have students add a sentence explaining how each work helped them with their project. If you want students to write discussion questions, have them also state why they think their questions are the most important ones for the class to address. Ask students to include written reflections on their research processes, or to write letters of advice to next semester’s students.
Note that these practices will in general make your assignments stronger. Adding in scaffolding, metacognitive exercises, and making your classes’ relevance visible, are all good teaching practices. In this case, we have the opportunity to become better teachers in response to these emerging technologies.
Some caveats:
At this point, Reed is not recommending the use of detection programs for identifying AI-produced writing, which can often yield false positives or be biased against English language learners. Cultivating your own understanding of your student’s voice and/or adding in checkpoints to long-term assignments are likely the best detection tools available.
With AI and opportunities for academic misconduct more readily available, faculty may be tempted to increase the logistical rigor of their courses, which does not always increase the cognitive rigor of the course (which is likely the true goal). Aim for assignments that promote cognitive rigor rather than logistical rigor. Craft your course such that it requires students to think deeply without imposing unnecessary logistical barriers.
While you may want to integrate more in-class writing and/or oral assessments, please keep in mind that these practices may not be ideal for many of our students and especially for students who need accommodations. In-the-moment writing and quizzes can be useful, especially for low-stakes assignments, but most scholarship suggests that untimed assessments enable deeper learning, more accurately reflect student understanding, and are more equitable.
Foreground your own “why.”
There have always been shortcuts available to students who privilege outcomes over the process and practice of learning. Likewise, students who don’t believe in the relevance of their work, or who lack confidence in their own abilities, are more likely to engage in academically dishonest behaviors. We need to be able to show students what they will gain from the hard work of doing, and they need to see and understand the benefits of learning.
Explicitly link your assignments with your course objectives.
Consider explicitly modeling what LLMs could do in your course, and have a class discussion about what would be gained and lost from using them.
Students need to have and maintain confidence in their own writing and thinking. Affirm their learning and their knowledge, and link what you’re doing in class with other places where this learning will be valuable.
Get Creative: Prioritize student voice and ideas.
It is quite likely that the abilities to generate new ideas, to think creatively and analytically, to collaborate with others, will become the most valuable and important abilities in the emerging world of AI. Small colleges like Reed are perhaps especially well-positioned to cultivate those skills. We have the opportunity to get creative with our assignments, and to think expansively about how to enable students to demonstrate their learning.
Is a traditional essay the right format for asking students to demonstrate their learning? Why? (State this why clearly in your assignment instructions). If not, what other options might be available for students? (examples might include making podcasts or short videos, developing lesson plans, writing op-eds, designing a museum exhibit, creating a webpage, making a video game….) James Lang offers this list of ideas.
Consider allowing students to draw on personal experience or integrate elements of a personal essay.
Think about how to privilege process over product. Develop assignments and checkpoints that emphasize the steps to achieving a desired outcome more than the outcome itself.
Are you considering using AI as a tool in your classroom?
First, make sure you understand the technology and that you are taking ethical and privacy concerns into account, and make sure that the tools you choose are accessible to all your students. (This may require using free versions of generative AI). Then, here are some ideas: