By now, it is clear that instructors can no longer simply require students to write essays. ChatGPT can do it for them. In fact, ChatGPT can generate a textbook without too much work. I recently generated a full 300-page textbook in a day. Further, I generated twenty humanities courses in two weeks. In the following, I describe how editing a ChatGPT-generated textbook can be used as a central pedagogical approach when teaching a class.
ChatGPT cannot, as of yet, generate valid and useful citations for what it writes. However, Sam Altman has demonstrated future versions that can do just that. It is more difficult for it to weigh the evidence and annotate the citations, allowing the reader to decide for him or herself how much to believe the author. (Don’t hold your breath on this state of affairs remaining, but this is where we are now.)
Process for teaching with a ChatGPT-generated textbook
Instructors can take advantage of ChatGPT’s ability to generate extremely useful text by this using the following process:
- At the beginning of the year, the instructor should work with ChatGPT to generate a textbook.
- Put the document in a wiki (e.g., MediaWiki or WikiDot) with each topic/chapter in a separate file. Both platforms track revision history so that everyone can see who made which changes. This allows the instructor to give the proper credit, if nothing else.
- Give the students edit rights to the chapters.
- The major assignment for the year is for students to edit a specific chapter and add citations and annotations to everything that they can throughout. Think of law review articles in which most pages are over half footnotes containing the citations and their associated text. Require students to validate (or refute) every important claim or fact in a chapter.
- By the end of the semester, you will have a well-vetted book with citations and clarifications. Students would have a much deeper understanding of the following:
- The strengths and weaknesses of ChatGPT,
- The topic being studied because they would have to do in-depth research (with whatever tools are appropriate),
- Research skills (using whatever technologies and approaches that are appropriate at the time…and this will change) because they would have to learn to weigh the strength of the evidence that they find on a wide variety of topics, and
- Writing skills because they would have to explain their findings related to evidence on dozens of points throughout a chapter. This would have to be monitored as AI tools advance but, for now, AI such as ChatGPT would not be of much help here.
- At the end of the semester (and maybe with a midterm, as desired), test the students on the textbook they’ve edited.
Standing on the shoulders of giants
I took inspiration for this idea from Brian Coppola, a fellow Thurnau Professor from Michigan as well as the CASE/Carnegie U.S. Professor in the Year in 2009. For the UM introductory chemistry class, he had the students write the textbook and then tested them on it. His students had to do research and run experiments in order to generate the course textbook. Here we are combining the strengths of ChatGPT at generating explanatory text with students’ abilities to explain topics to their fellow students. This process of generating explanations is also well-known to be a great way to learn a topic; thus, it’s a great pedagogical approach to incorporate.
Faculty and administrators should embrace ChatGPT’s ability to generate well-written and mostly accurate explanatory text. Editing a ChatGPT-generated textbook, as I have described here, is one way to rely on this ability and not run away from technical advances that cannot and should not be stopped.
