Writing Case Studies Using Generative AI: Interactive Role Play 

Created by Firefly: Virtual simulation world that is futurific

This article is a two-part series. Click here to read article 1, Writing Case Studies Using Generative AI: Intimate Debate Case Study.

Some fear the robot apocalypse, but you might want to roll out the red carpet for our new robot overlords! In a world where educators are increasingly saddled with large classrooms, greater responsibilities, and fewer resources, the ability to interact one-on-one with each student is rapidly disappearing. Yet, research shows that one-on-one tutoring is one of the most effective approaches to helping someone learn (Bloom, 1984; Chi et al., 2001). 

This is where generative artificial intelligence (GenAI) can help. GenAI tools like ChatGPT can gauge a student’s level of understanding and use it to customize a unique learning journey adapted to their needs. As a GenAI tests a student’s thinking, it can instantly respond with personalized feedback to guide student thinking and improve learning. 

In part 1 of this two-part article series, we examined how educators can use GenAI to rapidly design case studies that are used as pedagogical tools in the classroom. These cases were created with the assistance of GenAI, but remained under the control of the educator. In part 2 of this article series, we will use GenAI to create interactive role plays that adapt to each student, providing customized feedback that directs the narrative of the case based on each student’s responses. This is a new type of case study, reminiscent of the “choose-your-own-adventure” book series. This type of adaptive, personalized case study wasn’t possible before the development of GenAI. Here, the educator abdicates control since each student will go on a different journey with an infinite number of options and outcomes, but these are responsive to each student’s preferences and needs.  

What is an interactive case study? 

The idea for an interactive case study comes from a series of posts on the blog, Benjamin Breen’s Res Obscura. Take a moment to read the blog and try it—it’s fun! 

Breen developed a prompt that his students enter on ChatGPT (or other GenAI chatbot). The prompt asks the GenAI to create a simulated world (usually set in medieval times as the plague rolls into town – Breen is a history professor), where the student becomes a character. Each student must navigate this environment, avoiding disease and social unrest. Along the way, they make decisions that affect the hero’s journey and their outcome. There are 10 decision points throughout the story, each one impacting the direction of the narrative. It’s a role-playing game (an RPG in gamer terms), inspired by games like Dungeons and Dragons. Each time the simulation begins anew, a unique environment is generated and a new character is also created with characteristics that can affect their survival and the story. 

As part of the learning activity, students experience this virtual environment and then submit two things: a link to the unique virtual world they navigated (the URL to the ChatGPT conversation) and a reflection on the experience. In the reflection, students identify and describe historical aspects of the story that were accurate and aspects that the GenAI got wrong. It’s a fun way to exercise critical thinking and apply what they know about history. 

Breen’s role plays are case studies. They are narratives grounded in real-world scenarios that ask students to evaluate a complex situation and make decisions. Breen has assigned this activity because it aligns with his course’s learning goals. It’s not hard to imagine that this approach to creating a case study could be co-opted for other disciplines and for other learning objectives. Figuring out how to do this is what we will do later in this article. 

The difference between traditional and interactive case studies 

Traditional case studies are static. They are the same for every student. There is an advantage to that: everyone in the class is exposed to the same information and this serves as a common fodder for thinking through a problem. It also makes it easier to engage in class discussions where each student’s perspective can be analyzed in the context of the story with which everyone is familiar. 

The interactive role play, on the other hand, is dynamic and interactive. While the learning outcomes can be the same for every student, the experiences will be different for everyone. Such cases give educators the opportunity to customize each student’s learning journey based on their preferences and/or their learning needs. Interactive cases are responsive to each student’s choice, providing instant and customized feedback to their choices and responses. Such personalized learning is difficult to achieve with traditional (static) case studies.  

Interactive role plays aren’t perfect as pedagogical tools. GenAI is noteworthy for its tendency to invent facts or link concepts inaccurately. In interactive cases, the GenAI is in control of the narrative, with no chance for the educator to correct inaccuracies during the experience. Thus, it is vital to remind students to turn on their skepticism radar while using the tool. In addition, GenAI sometimes inexplicably breaks down during such prolonged interactions failing to complete the case (for example, while it may work for 10 students, for the 11th student the tool may stop asking questions after the fifth decision point). Such technical mishaps leave students on their own to assess and fix the bug (i.e., to determine that something went wrong and to restart from scratch). Finally, if students complete such an assignment, there is a chance they could “game the system” by repeating the interaction several times until they obtain a scenario that is easiest to analyze and submit for a better grade (though it could be argued that having students engage in a pedagogical activity several times is a win!). 

How to design an interactive case study 

The approach that Breen took in designing his interactive role play was to write a prompt that specified all of the parameters for the interaction—the learning goals, the settings, the goals of the narrative, the nature of the interactions, the rules of the game, the style of the communication—and let the GenAI use that to generate the interaction anew each time.  As might be guessed by the range of information it contains, this prompt is long. In fact, it can fill an entire typed page. It takes a bit of time to craft it, but once written, it is the seed that can grow an infinite number of different worlds. 

So, how does one go about crafting such a prompt? Here, I will propose three different approaches that I have used, along with templates to help educators try this out. 

Method 1: Build from Breen’s prompt 

The first method is perhaps the easiest one to implement. GenAI novices may want to start here, but it is also the one that produces the output least likely to hit the mark. The strategy is simple: Use Breen’s prompt to seed the process and ask ChatGPT to emulate it, but for your own learning outcome and your course topics. 

The prompt you enter into ChatGPT may look like this, where the text in [BOLD] should be modified to fit your context. 

Help me write the prompt for a role-playing simulation game that will help [undergraduate students in a molecular biology course] learn about [ethics in their profession as molecular biologists]. The role-playing simulation game should be inspired from the following history simulator prompt. 

[Then paste the prompt for the history simulator, as per the wording in Breen’s blog post—he offers several versions at the bottom of his post in three Google Docs.]  

The GenAI should return a prompt that you can enter in a new conversation to create a simulation suitable for your learning content. 

Here’s an example of one such interaction. In this case, ChatGPT didn’t propose a prompt but rather launched immediately into the interactive case study for my class. You can see the prompt provided to ChatGPT at the top along with the interaction that followed. This example converted Breen’s history game into an ethical dilemma in molecular biology…but you can spot that Breen’s historical game makes an unexpected appearance into the modern-day story every once in awhile. 

Method 2: Write your own prompt 

The second approach, which is more likely to yield a suitable output, is to write your own prompt. The prompt should be written as though you are asking the GenAI tool to interact with you in the way you intend for the role play (since students will later copy this prompt into the GenAI to generate the interaction). To do this, you must provide instructions about: 

  • The nature of the interaction, namely that you would like the GenAI to create an interactive role play where the narrative pauses at decision points to ask for directions about what to do next. 
  • The goal of the interaction, which is likely to match your learning objective. 
  • The context for the role play, providing as much detail about who, what, where, and when as possible. 
  • The objective of the role play (i.e., what are the characters trying to achieve in this role play?). 
  • Describe the rules governing the role play. For example, consider the number of decision points, whether the GenAI should provide you with options (and if so, how many?), or leave your responses open-ended, and how it should respond to your input (e.g., should the narrative simply resume with your choice in mind, or should it provide feedback on your choice?). 
  • Specify that the GenAI should wait for your response after each decision point before proceeding. When this instruction is left out, the GenAI will sometimes write the entire case study without waiting for input! 
  • If there is any text of information that you would like the GenAI to provide as part of the interaction, you should specify it. For example, do you want the interaction to begin by providing students with brief instructions about the rules of the game? Or perhaps you have a specific way in which you would like the GenAI to wrap up the interaction. For example, I like to add a disclaimer that ChatGPT tends to err and that it is the responsibility of the student to be skeptical of its information. 
  • You may wish to specify that although the GenAI should inspire itself from real events, if it cites any material, it should not make up any source (Note that this instruction will be observed by ChatGPT 4.0, but ignored by ChatGPT 3.5). 
  • Provide information about the intended length of the role play. This can be provided in terms of time spent to complete the experience, or by the number of decision points and the amount of narrative in between. 
  • Specify the desired writing style, such as humorous, formal, engaging, etc. If you would like the GenAI to include simulated artifacts such as memos, state this. 

    This information will take about a page. You may write these instructions in one paragraph in a conversational style, or you may consider copying the above text in a text-editing software and replacing each bullet point with your own entry and then pasting that text into the GenAI as a prompt. 

    Alternatively, you may use the following template if the nature of the interaction you envision aligns with the following description. This was written to create ethical dilemmas in bioengineering where students were given five decision points, each with four options about their course of action. If this scenario works in your context, simply replace the elements in [BOLD] for your own situation. 

    Start a detailed, interactive case study on [an ethical dilemma] within the field of [bioengineering]. The case should unfold continuously over several decision points, delving deeply into [one ethical situation]. With each new piece of information revealed, ask me to make a choice that influences the direction and outcome of the scenario. 

    The case study should include: 

    1. Introduction: Begin with a brief introduction to [the ethical dilemma], setting the stage for the case study. You may also want to provide brief instructions for me about what I am expected to do and the goal of this simulation. 
    2. Decision points: Present me with a series of [five] decision points as the case progresses. The first question should give me options and ask for my preference for the topic of the case study. Each question after that should match a decision point in the case related to an [ethical dilemma]. You should wait for my response after each decision point. Each decision point should offer [four] distinct options to choose from. The outcome of each choice should directly influence the next stage of the case.  
    3. Continuous narrative: Ensure the narrative flows logically from one decision point to the next, gradually revealing more complexities of [the ethical dilemma]. 
    4. Conclusion: Provide a final summary that reflects on the consequences of my choices throughout the case study and that surfaces [some of the values that I may have manifested in making the choices.] If a similar [ethical dilemma] occurred in real life, this example should be briefly outlined and cited. Do not make up sources.  

    The case study could explore topics like [gene editing, cloning, stem cell research, GMOs, research ethics more broadly such as conflicts of interest, animal and human research ethics, or plagiarism or academic integrity, or any specific aspect of bioengineering that presents a significant ethical challenge]. The goal of this simulation is for me to [articulate my values and how they might influence my choices in an ethical dilemma]. 

    The writing should be [casual and engaging, aiming to immerse me as a character in the ethical complexities of the case]. You may assign me a specific role or persona in the case, to help me take on a perspective as I consider [the ethical dilemmas]. This interactive experience should not take more than [10 minutes] to complete, offering a concise yet insightful exploration of [bioengineering ethics]. 

          Method 3: Ask ChatGPT to improve your prompt 

          The third method is perhaps the most time-involved approach, but not surprisingly, it also yields the best results. Here, you provide all of the instructions for the interaction that you did in Method 2, but you ask ChatGPT to transform those instructions into an effective prompt for a GenAI tool. This is done by adding the instruction to create a prompt that students can enter into a GenAI tool to result in the interaction described below (and then you enter the prompt you created in Method 2).  

          A prompt can look like this, where you may want to customize the words that appear in [BOLD] for your context: 

          I would like you to help me design a prompt that I can give to my [first year undergraduate science] students to input into [ChatGPT]. Once they enter the prompt, [ChatGPT] will work as an interactive simulation, helping students think through [the steps in designing a scientific experiment].  

          [Provide your instructions for the interaction, as per Method 2] 

          Remember that I am looking for an effective prompt that students may give [ChatGPT] to accomplish the interaction described above. 

          The last sentence reminds the GenAI about the target output since it can sometimes be missed in such a long prompt. Here is an example of such an interaction. You may notice that ChatGPT simplified my initial prompt and organized the information. The latter can make it easier to modify the prompt (to improve it or to modify the role play). Note, however, that this is not the last step. Consider the GenAI’s proposed prompt as a beta-version of the prompt that you will provide to your students. Before that can happen, you should field-test it to see how it performs. 

          For example, using the suggested prompt in the example above, I noticed a few glitches. First, the interaction sometimes didn’t wait for my responses at a decision point and instead continued with the next part of the case study. I therefore improved the prompt by adding the instruction to wait for a student’s response after each decision point before proceeding. In trials, I also noticed that after commenting on the student’s choice for the second decision point (the choice of hypothesis), the interaction stopped, and instead ChatGPT provided information (but asked for no decision) about the rest of the narrative (see example of this glitch here). A quick inspection of the prompt that ChatGPT suggested for this interaction reveals that there are no instructions to ask for a decision for the remainder of the interaction. I therefore added these in. Finally, I customized the introduction language, as I felt the wording applied to younger students than my undergraduate learners. 

          As you may see from this example, the ChatGPT prompt, while organized and easy to work from, can include errors. These can be identified and fixed by trying the role play a few times. 

          Examples  

          What sort of role play interactions can be crafted through such a technique? It may be suitable for any narrative that requires students to make decisions that effect the narrative’s progression. Here are two examples suitable for my undergraduate science courses. 

          Scientific research design 

          One prompt was created to guide students through the design of a scientific experiment. It guides students to consider each element in planning an experiment to choosing a research question to identifying the ways in which the data will be analyzed, visualized, and communicated. The GenAI provides feedback on the strengths and limitations of each choice along the way. 

          The same prompt generated completely different role plays that helped students design experiments to investigate a research question most aligned with their preferred topic: 

          Note that this prompt is not perfect. In all three examples, the section on defining the control group seems a bit amiss and could benefit from some revision. Thus, like any educational materials, there is room to improve the prompt for each term this activity is assigned. Or alternatively, you could ask students to criticize this part of the interaction and propose a better control group. 

          Bioengineering ethical decision 

          The second example targets a very different learning goal; it targets the affective domain. Here the activity is designed to help students examine their values as they navigate complex ethical decisions that are typical of their field (bioengineering). The same prompt resulted in cases about ethical decisions made in the context of: 

          Since the learning objective of this activity was for students to reflect on their values as drivers of ethical decisions, students were first asked to participate in the role play and then reflect on the values that drove their choices at each decision point. 

          Next step: Sharing GenAI cases among educators 

          The two strategies for developing case studies presented in this article series may help educators create new case studies. It takes times—even with the assistance of GenAI—to create a good quality case study. For this reason, educators like to share the resources they develop. 

          The intimate debate designed using GenAI as an assistant could be submitted for peer review and publication in repositories such as the National Center for Case Study Teaching in Science. It would be ethical for the authors to disclose that the cases were written with the help of GenAI. Indeed, at the October 2, 2023 webinar on GenAI in higher education hosted by the Higher Education Strategy Associates, undergraduate panelists said it was important to them that educators disclosed their use of GenAI in the development of resources used in the classroom. Transparency is valued by students as well as educators. 

          As for interactive role play cases, there doesn’t seem to be a clear repository at the moment for educators to share carefully crafted prompts. It is possible that over time online forums led by a community of interactive case study practitioners may emerge to fill this need.  

          Won’t you join our ranks and welcome the robots?  


          Dr. Annie Prud’homme-Généreux is an educational developer and instructor at the University of British Columbia. She is completing a master of online education and has designed and facilitated several workshops to help educators incorporate GenAI tools into their practice. This includes the OER faculty development course Forward Facing Assessments, downloadable free of charge from BCcampus. 

          References 

           Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational researcher, 13(6), 4-16. 

          Chi, M. T., Siler, S. A., Jeong, H., Yamauchi, T., & Hausmann, R. G. (2001). Learning from human tutoring. Cognitive science, 25(4), 471-533.