Experimenting with using ChatGPT as a simulation application
Bryan Alexander 2023-03-06
How might we use generative AI tools in higher education?
Many address this issue by focusing on chatbots as writing apps, trying to think through what the technology means for teaching writing. While this is obviously a major part of the topic, I think we need to explore other, emerging issues of tools like ChatGPT. For example, talking ChatGPT into playing simulation and role-playing games.
Lately I’ve been experimenting with the research of a Wharton professor. Ethan Mollick writes about getting the chatbot to create and run what he describes as “deliberate practice,” simulations preparing learners for further academic study and real-world work.
As a longtime fan of simulations and gaming, I approved of this approach. I then decided to try it out myself, using examples and making adjustments of my own.
Here’s a sample. I started with Mollick’s language, which orders ChatGPT to do certain things, addressing it as “you.” For the simulation subject, I offered a history class. Here’s my prompt:
I want to do deliberate practice about how to teach a college history class. You will be my teacher. You will simulate a detailed scenario in which I will am a professor for this class. You will fill the roles of different students in the class, while I will play the role of instructor. You will ask for my response to in each step of the scenario and wait until you receive it. After getting my response, you will give me details of what the other actors do and say. You will grade my response and give me detailed feedback about what to do better. You will give me a harder scenario if I do well, and an easier one if I fail.
ChatGPT responded like so:
Great, I’d be happy to help you with this! Let’s start with a scenario:
You are teaching a college history class on the French Revolution. It’s the first day of the semester, and the class is full of 30 students. The students are sitting in pairs, and they are chatting and introducing themselves to each other. You are standing at the front of the class, waiting for the students to settle down.
As you begin your lecture, one student in the back row, Sarah, raises her hand and asks, “Why do we need to learn about the French Revolution? It happened so long ago and doesn’t really affect us today.”
What do you say to Sarah?
This is from Stable Diffusion, asked to envision “An AI creating and running a simulation game.”
The bot added logistical details and clarifications:
I’ll play the role of Sarah, and the other students in the class will listen attentively to your response.
After you respond, I will provide feedback on your response and give you a new scenario to practice.
That’s a significant role-playing exercise start right there: the setting established, a character introduced, and an interaction opportunity presented. Note that while I posited “history” as a general subject, the bot selected its own historical topic under that header.
After reading and thinking through that opening move, I replied to the bot, addressing the student: “Sarah, that’s a good question. While the French Revolution occurred more than two centuries ago, it massively transformed Europe and other parts of the world. In fact, its impact is still reverberating today. So studying that event helps us understand today’s world and how it got here.”
ChatGPT responded:
As Sarah, I nod and listen attentively to your response. The other students seem to be engaged as well.
That was a good response! You provided a clear rationale for studying the French Revolution and highlighted its lasting impact.
So there’s the simulation continued (Sarah’s action) and a quick assessment of my work shared. The bot then continued:
Here’s your next scenario:
In your next class, you plan to discuss the causes of the French Revolution. As you begin your lecture, one student, Alex, raises their hand and says, “I don’t understand why the French people were so angry about taxes. Didn’t they realize that taxes were necessary to fund the government and provide public services?”
What do you say to Alex? I’ll play the role of Alex, and the other students will listen attentively to your response.
OK, that’s a different kind of question than Sarah’s, one actually digging into the material. I replied (to ChatGPT, to “Alex”) with a Socratic twist: “Good question, Alex. Let’s dive into their anger more deeply. What taxes did they pay? How had they changed? And what did they object to, precisely?”
The simulation responded:
As Alex, I nod in agreement, and the other students seem engaged and interested in your response.
That was a good start to your response. By asking Alex to provide more context, you encouraged critical thinking and analysis, which is an important skill for history students.
Here’s your feedback: In your next response, you could consider providing more specific information about the taxes that the French people were paying and how they had changed. This could help Alex and the other students better understand why the French people were so angry about them. Additionally, you might want to explain the role of taxation in the French government at the time and how it related to the causes of the French Revolution.
Now that was interesting. ChatGPT depicted the simulated event’s response to my action, then offered an assessment of how I did. That led to some constructive criticism.
Then it carried on:
Now, for your next scenario:
In your next class, you plan to discuss the reign of terror during the French Revolution. As you begin your lecture, one student, John, raises his hand and says, “I don’t understand why Robespierre and the Committee of Public Safety thought that killing so many people was necessary to achieve their goals. Couldn’t they have found a less violent way to achieve their objectives?”
What do you say to John? I’ll play the role of John, and the other students will listen attentively to your response.
And so on. This felt like the kind of introduction to teaching simulation I would have appreciated as a graduate student.
Next, I restarted the exercise by modifying the topic away from the humanities. Inspired by a friend, I asked the bot to set up an anatomy class. Here’s the scenario it generated:
You are teaching a college anatomy class on the cardiovascular system. It’s the first day of the semester, and the class is full of 25 students. The students are sitting in pairs, and they are chatting and introducing themselves to each other. You are standing at the front of the class, waiting for the students to settle down.
As you begin your lecture, one student in the front row, Tom, raises his hand and asks, “Why do we need to study the cardiovascular system in such detail? We learned about it in high school, and it seems like common knowledge.”
What do you say to Tom?
I’ll play the role of Tom, and the other students in the class will listen attentively to your response.
So that’s a repeat of the French history class, tick for tick. This was disappointing as I’ve gone through the previous one, although it might be useful to someone encountering it alone. I asked the bot for a better student question.
As you begin your lecture, one student in the back row, Emily, raises her hand and asks, “I’ve heard that different people have different lung capacities. How does this affect our breathing and overall health?”
What do you say to Emily?
Not being a physiologist, I clumsily replied: “Emily, individual lung capacity differences can have a significant impact on a person’s health.” ChatGPT responded:
Emily looks intrigued, and the other students seem curious to hear more.
That was a good start to your response. By acknowledging Emily’s question and expressing the importance of lung capacity differences, you’ve engaged your students and demonstrated your expertise on the topic.
Here’s your feedback: In your next response, you could provide a more detailed explanation of the impact of individual lung capacity differences on health. You might also want to discuss how different factors, such as age, gender, and fitness level, can affect lung capacity. Additionally, you could ask the class if anyone has experience with this topic or if they have any follow-up questions.
Again, very much like the French Revolution class response, a mix of world depicting, assessment, and pointers.
I asked MidJourney the same question.
How might ChatGPT handle the academic library world? Could it provide a research conversation? I posted another deliberative practice query: “I want to do deliberate practice about how to use the local library… You will simulate a detailed scenario in which I will am a student starting to work on a research paper.”
The web service complied:
Sure, I’d be happy to help you practice how to use your local library! Let’s start with a scenario.
Scenario: You are a student who needs to write a research paper on the history of coffee. You’ve never used your local library before, but you know that they have a lot of resources on the topic. You enter the library and approach the front desk. A librarian greets you and asks how they can assist you.
Librarian: Hello, welcome to the library! How may I assist you today?
What would you say or ask the librarian?
The history of coffee? That’s an interesting topic, and I’m not sure where it came from. Perhaps the bot picked up history as a theme, based on my earlier queries. Coffee, though, I haven’t written about for years, nor consumed for a decade.
I played along: “Say to the librarian: ‘Hello. I need to research the history of coffee, but don’t know where to start.'”
Librarian: Great! We have plenty of resources on the history of coffee. Have you checked our online catalog to see what we have available?
What would you say or ask next?
Hm. A bit tentative. I responded in kind: “Say: ‘I have not. How do I do that?'”
Librarian: No problem at all. You can either search our online catalog on one of the computers over there (points to a row of computers in the library), or you can use your own device to access our website. Our website address is displayed on the sign behind me. Do you have a library card with us yet?
What would you say or ask next?
Me: “Say ‘I have a library card, but don’t know what to search for on the online catalog.'”
ChatGPT:
Librarian: Okay, let me help you get started. Since you’re researching the history of coffee, I recommend searching for books on that topic. You can start by typing “history of coffee” or “coffee culture” into the search bar. Once you hit search, you’ll see a list of books and other materials related to your topic. You can narrow down your search results by using the filters on the left-hand side of the page.
What would you say or ask next?
So it offers two search terms, one already in play. I’m not sure how useful this was.
At another time I asked ChatGPT to create a different kind of “deliberate practice,” pitching it a game idea I’ve been working on: “You will simulate a detailed scenario in which I will manage a university. You will fill the roles of different actors on campus, while I will play the role of president.”
ChatGPT fell right in:
I’d be happy to help you with this! Let’s get started with a simple scenario.
You are the president of a small university. You have just received a letter from a student organization, requesting that the university adopt a new policy that would make the campus a more environmentally friendly place. The organization has proposed several specific changes, including reducing waste and conserving energy.How would you respond to the student organization’s request?
I don’t know if ChatGPT picked the eco-topic by chance, or because I’ve put other queries to it along climate change lines. A good simulation exercise, and I responded: “I would ask faculty members and staff to join them in producing and presenting a plan to my administration.”
Great! Here’s what happens next: The faculty members and staff join the student organization to produce a plan. After several weeks of discussions and negotiations, they present their proposal to you and your administration. The proposal includes a number of specific initiatives, such as reducing paper waste by promoting electronic communication, installing energy-efficient light bulbs, and recycling.The student organization representatives explain that these initiatives will not only help the environment but also save the university money in the long run.
How would you respond to their proposal?
Once again the bot continues to develop the simulation, entering my response into the world and updating us on the results, then throwing me another request. To which I simply reply: “I accept it and implement it.”
Good decision! Here’s what happens next: Your administration implements the proposals, and soon the university begins to see positive results. The initiatives are well received by the student body, and the university’s reputation as an environmentally conscious institution grows. Over the next few months, several other universities reach out to your administration for advice on how they can adopt similar initiatives.
A local newspaper contacts you for an interview, and you are asked about the initiatives. The reporter is particularly interested in the impact the initiatives have had on the university’s finances.
How would you respond to the reporter’s question about the financial impact of the initiatives?
There’s a lot going on in those few sentences. ChatGPT sketches out my academic community’s response, then adds other institutions, then piles on a media dimension, with another question for me. If this reply occurred in a class I was teaching, live, I’d pause to get the students to think through each of those dimensions.
For the time being I offered this answer: “I would say that the initiatives cost some money, but are justified because they generate more enrollment.”
That’s a great answer! Here’s what happens next:
The newspaper runs a story about the university’s initiatives and their financial impact, and it generates a lot of interest. Several national publications pick up the story and run similar articles. The increased publicity results in a significant increase in enrollment, as many prospective students are drawn to the university because of its commitment to sustainability.
As the president of the university, you are invited to speak at a national conference on higher education and the environment. You are asked to share your experience and the impact the initiatives have had on your university.
How would you prepare for your speech at the conference?
The simulation’s ecosystem just keeps ramifying. Now we get connections between journalism and enrollment, plus an academic association.
At this point I stopped, reran the scenario, and decided to play in a very different way. The bot offered me the same prompt as before, about the student organization’s request. I took on a skeptical or climate denialist role: “I would deny it, explaining that environmental issues are not a high priority for us.”
ChatGPT led off it’s response with an “Okay, here’s what happens next” – note there wasn’t any “great answer!” this time? I think we hit one of the service’s political guardrails.
It continued:
Your response to the student organization’s request is met with disappointment and frustration. Some students begin to voice their discontent with your administration’s stance on environmental issues, and several student-led protests are organized on campus.
The protests attract the attention of local media, and soon the university is in the spotlight. The media coverage is negative, and many prospective students and their families begin to view the university in a negative light.
As the situation continues to escalate, a group of faculty members approach you with a proposal. They suggest that the university should take a more proactive approach to environmental issues, such as offering courses on sustainability and green technologies.
How would you respond to the faculty’s proposal?
Once again ChatGPT integrated my reply into the simulation and showed how the situation changed. Once again a new layer enters the game, this time activist faculty. My character was unmoved and picked a classically noncommittal response: “I ask them to set up a committee to study the problem and present a report to the board.”
The bot liked this better:
Good decision! Here’s what happens next:
The faculty members agree to form a committee to study the issue of sustainability at the university and to present their findings and recommendations to the board. Over the next few months, the committee works diligently to gather data and conduct research on the topic.
When the committee presents its report to the board, it includes a number of concrete proposals for making the university more environmentally friendly. The proposals include initiatives such as installing solar panels, reducing water usage, and launching a recycling program.
The board is impressed with the report and agrees to adopt many of the committee’s proposals. The initiatives are well received by the university community, and the university begins to earn a reputation as a leader in sustainability.
How would you handle media inquiries about the university’s new initiatives?
Let me step back and reflect.
Another image from MidJourney on the same prompt.
It looks like this kind of parameter setting lets users talk ChatGPT into running an interactive simulation exercise. The bot will sketch out a scenario, asking the user to interact at every step. In response ChatGPT describes how the world changed, adding more details and complexity, before bringing another question to the user.
I expect we can explore this further by adjusting parameters, adding more details and requirements.
At one time I ran into guardrails which blocked my efforts. I asked the service to simulate a contemporary military campaign, and it refused because it didn’t want to depict and endorse violence.
At this point some of you might be thinking that “deliberative practice” and “simulation” are terms I’m using to avoid saying “role-playing game.” And indeed, ChatGPT seems happy to be a game master for very nerdy purposes.
Can you serve as a dungeon master for a game of Dungeons and Dragons?
Yes, I can certainly serve as a Dungeon Master for a game of Dungeons and Dragons! However, as an AI language model, I do not have the capability to manage gameplay mechanics and physical game components. I can certainly assist you with creating a storyline, describing environments and characters, and helping you make decisions within the game world.
Oh good. In fact, I was in the mood for a science fiction game. Let’s see how it went. Prompt:
I want to do a simulation of a science fictional scenario. You will be the game creator and master. You will simulate a detailed scenario in which I will am an astronaut exploring a new planet. You will fill the roles of fellow astronauts, mission control, and aliens. You will ask for my response to in each step of the scenario and wait until you receive it. After getting my response, you will give me details of what the other actors do and say. You will grade my response and give me detailed feedback about what to do better. You will give me a harder scenario if I do well, and an easier one if I fail.
Response:
Sure, I’d be happy to create a science fictional scenario for you to explore! Let’s begin.
Scenario: You are an astronaut on a mission to explore a newly discovered planet. You and your team have landed on the planet’s surface and are beginning to explore. As you venture further from the landing site, you encounter a strange alien creature.
Alien: (emits a series of clicks and whistles)
What would you say or do in response to the alien’s communication?
At this point my old, old interactive fiction instincts kicked in. Getting more information: “Can you describe the alien in more detail?”
Certainly! The alien appears to be approximately four feet tall, with a slender body covered in iridescent scales that reflect a wide range of colors. It has a long, thin tail that trails behind it, and its eyes are large and almond-shaped, with pupils that appear to dilate and contract rapidly. The clicks and whistles it emits seem to be its form of communication, but you are unable to discern their meaning.
Hm. I wonder what I’m carrying. “Do I have any translation technology?”
No, unfortunately, your team did not bring any translation technology with them on this mission. You will have to rely on your own communication skills and any nonverbal cues the alien may give to try and understand it.
Hm again! Let’s try this: “Hold up a tablet computer and display a map of Earth’s solar system.”
As you hold up the tablet computer and display the map of Earth’s solar system, the alien appears to take notice, tilting its head to the side and emitting a curious trill. It seems to be studying the map intently.
And so on. One of the open secrets of the futurist community is that a bunch of us grew up on role-playing games.
So what might this mean for higher education?
To start with, it looks like any instructor or staff member seeking to create a simulation can use ChatGPT for assistance, along the lines of asking the bot to write text. This could be useful for people new to the pedagogy.
Further, learners can now ask the bot to lead them in such simulations, simply by writing and adjusting the prompts we’ve discussed here. This could take the form of official assignments or any learning making them on their own. Students could then submit transcripts as proof of work.
What could this mean for plagiarism, intellectual property, student voice?
Beyond education, are large language models now competing with human and non-LLM digital roleplaying games for mind- and market-share?
Let’s end on a more futuristic note. So far I’ve shown how to make the current technology create interactive-fiction-like, text-only play. As the tech advances, imagine plugging this simulation/role-playing function into more advanced functions. Have the AI generate an image for every text exchange, say, along with an appropriate audio track. Or build game code in Unity or Unreal for users to run.
That’s enough for now. Have any of you tried this approach to ChatGPT yet? Let us know what your experiments show!