Lesson Planning, Iterative AI Prompting, and Rubber Ducks
I’m facilitating a number of virtual faculty learning communities this spring on teaching and generative AI. After each session of one of these learning communities, I take some time to write a recap of the discussion and collect the various resources shared during the discussion. I do this primarily as a service to the learning community participants, but also to help myself keep track of which discussions happened in which learning communities!
Many of the recaps I write detail the questions, ideas, and examples shared by learning community participants, and I wouldn’t be comfortable sharing those beyond the learning community. However, there are some recaps where I’ve generalized a bit from our discussions and written something that might be more widely useful. Today, I’ll share one of those recaps, drawn from a learning community I’m leading that is reading Teaching Effectively with ChatGPT by Dan Levy and Angela Pérez Albertos.
This is currently my favorite book in the rapidly growing space of books about generative AI and teaching and learning. One reason I like Levy and Albertos’ book is that it’s so very practical. Each chapter is full of very concrete ideas for using ChatGPT in teaching and learning, with lots of sample prompts and sample outputs. Most of the examples are drawn from the actual teaching practice of Levy and Albertos or from colleagues across higher ed.
The other reason I like the book is that the first half or more is devoted to using AI as a kind of instructional assistant, that is, using AI to help write lesson plans and design activities and build rubrics. I have found that when working with faculty and other instructors about the potential role of AI in their teaching, it’s a lot easier to start with their own uses of AI than with how their students might use it. When you start with student use of AI, you get all kinds of questions about academic integrity and unauthorized aid. Those are important questions, but I find that we can discuss them more productively after we have a better sense of how generative AI works, which we can attain when we talk about our own potential uses of these tools.
It was during a discussion of that part of the book, specifically the chapters on using ChatGPT for planning new lessons and designing pre-class activities, that a couple of colleagues in this reading group shared ways they have used AI recently in their course prep. One colleague described using ChatGPT to help write a new role-playing activity for her course based on a case study. The chatbot wasn’t good at role-playing itself, since it tended to make up things that weren’t true about the real-world events in the case study, but my colleague found it very useful in helping her plan the in-class role-playing activity. Another colleague mentioned his use of ChatGPT to put together a new lesson plan on a particular topic. He found that ChatGPT provided useful ideas and resources and was able to generate an outline for a slide deck, which he then turned into an actual slide deck using PowerPoint’s import-from-outline tool.
Those examples led into a good discussion about prompting strategies that I’ll try to summarize here, using some jargon I’m still learning myself. Let’s say you’re trying to get ChatGPT to do something like those instructors did. You have some prompting options:
- Ask ChatGPT to generate the lesson plan or activity or whatever you’re after. If this is your first prompt and you’re not providing an example of what you want, this is called zero-shot prompting. For some tasks, ChatGPT will do fine, but for most tasks, the output won’t be what you want.
- Maybe you have an example handy of the activity or case study or whatever you want ChatGPT to produce. If you give the AI one or more illustrative examples, this is called few-shot prompting. Even one example is likely to produce better outputs than a zero-shot prompt.
- If you’re not happy with the output you have at this point, you can give ChatGPT some feedback and ask it to revise its work. It will be “happy” to do so. You might go back and forth several times in this process, which is called iterative prompting.
Option 3 is my usual way of interacting with AI chatbots, and one reason I like it is that the back-and-forth helps me refine and articulate my own thinking about whatever task is at hand. It’s a little like talking about your lesson planning with a colleague; you might not take all their advice, but just having the conversation can help spark your own thinking. I recently learned a fun term for this: rubberducking. The idea is that if you explain your thinking to some animate object on your desk (like a rubber duck, if you have one handy), you can help yourself solve a problem. In this case, however, the duck can talk back and maybe share a useful idea or two.
In our session, my teaching center colleague Jess Taggart shared a piece of advice that was new to me and that I’ll definitely use in the future: At the end of an iterative prompting session, if you have the result you want from the chatbot, ask the chatbot to write a prompt you can use in the future to accomplish the given task. You’re likely to get a prompt that integrates elements of the entire back-and-forth, one that you can copy and paste in later AI chat sessions. This is called meta-prompting, and Jess learned about from in this newsletter.
One trick I did know about was to use someone else’s well-developed prompt. In Chapter 4 of their book, Levy and Albertos recommend this lesson planning prompt designed by Lilach Mollick and Ethan Mollick. That prompt will ask you questions about your lesson plan, inviting iterative prompting. You can read my experiment with the Mollick prompt here. Other canned prompts might not be so interactive, but might also yield useful results.
During the session, another teaching center colleague, Adriana Streifer, made the point that even if generative AI can save you time as you plan your lessons and assignments, it can’t make decisions about what to value or prioritize in your teaching. That’s part of why I prefer iterative prompting, since the back and forth often clarifies my own values and priorities. I keep finding myself thinking about that idea of rubberducking, that coming up with prompts for ChatGPT requires me to reflect on my teaching and the teaching choices I want to make.
Photo: “Funky Duck,” Funkyah, Flickr (CC)