On the first day of Dr. Robert Goehlich’s space tourism course, students are introduced to Max Mustermann, a generative AI assistant that can do everything from providing deadline updates to explaining complicated concepts and offering feedback.

Thanks to Max’s success, Goehlich, an adjunct assistant professor with Embry-Riddle Aeronautical University’s Worldwide Campus, is pioneering ways to incorporate artificial intelligence into his course, which asks students to grapple with the space tourism industry and its economics, operations, safety and marketing. While most people interact with generative AI through Anthropic’s Claude, OpenAI’s ChatGPT, or Google’s Gemini, these models can also be used to create customized chatbots.

In his upcoming class, Goehlich’s students will engage with four new AI-powered avatars, as well as Max. The space tourism industry is still emerging, and chatbots, Goehlich said, “integrate naturally” into the course’s pivotal “simulation and role-playing scenarios.” Students will also have access to a distinct “AI editor chatbot,” which will not generate text as the commercial products do but will act as a coach, Goehlich said.

“For me, this is a game changer,” he said, “as there is a lot of concern in academia about students using ChatGPT.”

To create the bots, Goehlich collaborated with Dr. Alyssa DeNaro, previously an AI expert with the Worldwide Campus Division of Academic Innovation and now the administrative fellow for AI Initiatives in the Office of the Provost. Over a year ago, Embry-Riddle’s information technology team worked with DeNaro to create a bespoke chat builder for the university. Now, more than 50 chatbots are used in courses offered by Embry-Riddle Worldwide, the university’s online campus, which serves more than 20,000 students around the globe.

“In an asynchronous, online environment,” DeNaro said, “the chatbots are an extra tool to give students an interactive and engaging experience with their course content.”

The first new chatbot in Goehlich’s space tourism course is a human resources manager named Elena Moonwalker. It interviews the students, who must clarify and expand on their responses just as they would in a real hiring process, Goehlich said.

A different, more sophisticated chatbot simulates a project kick-off meeting, where students must pitch their space tourism projects to three stakeholders.

“While the AI chatbot Max typically waits for questions, the others — the HR manager and the three stakeholders — actively ask questions and help guide the discussion,” Goehlich said.

To build the kickoff meeting chatbot, Goehlich began with his original chatbot, Max, and its knowledge base, which consisted of the course’s curricula. He created a draft master prompt, including the three stakeholders’ character traits, behaviors and domains.

Dex Demand, one of the avatars, focuses on market considerations and pricing. Another, Dr. Risk Null, examines risk management, safety and regulatory compliance. And Brand Nova addresses marketing issues, such as customer perception and experience.  

Where Max is friendlier, the stakeholder avatars are more demanding, Goehlich said. “They are a little harsher and more direct, and they also ask for deliverables.” 

DeNaro, who said that the stakeholder meeting is the first Embry-Riddle chatbot to have three avatars in a single experience, refined the prompting framework to make the chatbot interaction more robust, including the timing, sequence and flow of the stakeholder questions. The prompt for it ran to more than four pages.

Through extensive tweaking and testing, they instructed the chatbot on appropriate types of questions and feedback. The most difficult part, DeNaro explained, was to make the timing clear, including “the stages of interaction and when and how to move on.”

When asking questions, each avatar will stick to its assignment, but “will shift dynamically according to what the student is presenting,” DeNaro said, “taking turns, responding and playing off one another.”

To add to the immersive experience, Sabrina Kissen, who works in instructional design and development, created visual bios for the avatars, including headshots and job descriptions.

Once students complete the stakeholder meeting, they will save the transcripts and study them with their classmates, Goehlich said.

“This type of structured use of AI does a lot of really nice things for the student experience,”  DeNaro said. “The first is giving them the opportunity to act out some sort of element that they’ve learned in class conceptually and apply it in a realistic scenario.” The other is for the students, many of them mid-career, to build “AI literacy” during their education.

While many classes are now experimenting with AI, it is still unusual for students to engage with multiple custom chatbots in a single course.

“The major upgrade,” Goehlich said, “is a distinctive simulation in which five AI avatars, students and the instructor interact within one learning experience.”