Putting the technology into practice


Four quick looks at how faculty members have implemented artificial intelligence in their teaching.

Kas Kasravi: Students’ “level of thinking went up drastically”

Kas Kasravi worked as a vice president and HP Fellow in industry with a focus on analytics and artificial intelligence for more than 30 years before joining UM-Dearborn as a lecturer III in industrial and manufacturing systems engineering in the College of Engineering and Computer Science.

When ChatGPT was first introduced, Kasravi recognized the opportunities it offered for teaching and research.

Kas Kasravi (right), lecturer III in industrial and manufacturing systems at UM-Dearborn, says, “AI is just a tool. It’s not going to become a student; it’s not going to become a professor.” (Photo by Scott C. Soderberg, Michigan Photography)

“I am a strong believer in AI and its applications in education in particular because education, whether it’s research or teaching is all about knowledge — identifying knowledge, transferring knowledge, using knowledge — and AI is about accessing knowledge. So to me, there’s a natural match between those two,” Kasravi said.

“And AI is just a tool. It’s not going to become a student; it’s not going to become a professor. It just makes the process of education more effective, in my opinion, and it does this by shortening the paths to the knowledge that we are seeking, making the process more efficient.”

During the winter 2023 term, Kasravi incorporated AI into his prototype design class. The class required students to write microcontroller code. He said students in the past struggled with the intricacies in writing code and created basic functions for the controllers.

When the students used AI to help develop the code, Kasravi said, they were able to shift the level of abstraction from syntax to logic, focusing more on the outcomes, and their “level of thinking went up drastically.”

Kasravi created custom U-M Maizey tools to serve as an aid for students in an introductory engineering class, a technical communications class and a graduate level course on product development. A survey of his introductory class found that 49% of the students used Maizey to help them prepare for a quiz, and 95% of those who used the tool found it helpful.

Anne Gere: Examining GenAI’s pros and cons

Anne Gere, Arthur F. Thurnau Professor Emerita, professor emerita of education in the Marsal Family School of Education, and professor emerita of English language and literature in LSA, introduced U-M GPT into her classrooms last semester to help students examine the tool’s pros and cons.

“My view is GAI is something students are going to be using in their work lives as they move forward. And part of our job as faculty is to prepare them for that, which means helping them see what it can do and what it can’t do,” Gere said.

“If I don’t bring AI into the classroom, it seems to me that’s an invitation for students to try to use AI in surreptitious ways, and I don’t think that’s healthy for them, or certainly not for their learning.”

While AI can help generate ideas for writing, Gere said, responding to specific assignments — like analyzing characters or comparing narrators in different novels — requires the creativity of in-depth analysis that is beyond AI’s capability. To highlight this, Gere tasked her students with giving AI a writing assignment and comparing it with what they and their peers produced.

Gere said the generated essays were factually correct; however, students saw AI’s limitations because “it doesn’t have a voice. It doesn’t have a personality, and one hopes that students are going to be able to produce writing that has a voice, a way to connect with their audience. AI doesn’t do that,” Gere said.  

Andrew DeOrio: Using U-M Maizey to support students

Andrew DeOrio, lecturer IV in computer science and engineering in the College of Engineering, implemented AI in his courses last fall.

In his advanced course examining how the internet works, DeOrio used U-M Maizey to create an assistant equipped to help students with the course’s most difficult programming assignments. He uploaded lecture slides, transcripts of lectures, tutorials and other materials to build a strong base to support the students.

Andrew DeOrio, lecturer IV in computer science and engineering, says the digital assistant he built using U-M Maizey “was meant to augment existing course resources; it definitely filled that role and was helpful.” (Photo by Robert J. Scott)

“Overall, students found it helpful. And it’s important to note that this wasn’t conceived as a replacement for our existing course resources like office hours with the professor or teaching assistants,” DeOrio said. “This was meant to augment existing course resources; it definitely filled that role and was helpful.”

A potential danger with AI is the risk of “hallucination.” When AI “hallucinates,” DeOrio said, it provides misleading responses that may appear real.

“If you give this thing (AI) to help students and it’s wrong, then you’re helping them learn the wrong thing. So, that was a big concern that we wanted to test,” DeOrio said.

DeOrio and his team measured the success rate and found more than 92% of the AI answers were correct. When incorrect, they found students correctly identified the response as incorrect more than 96% of the time.

DeOrio said he’s working with his team to expand the scope of the AI assistant to bring the technology to other projects and future classes.

Jun Li and Andrew Wu: A tool to “help with the thought process”

Jun Li, associate professor of technology and operations in the Stephen M. Ross School of Business, teaches a core course to undergraduates. With eight sections and more than 600 students in the course, Li said, students can find it difficult to approach instructors for specific questions, even during office hours, which can be cramped and overflowing.

Li shared her struggles with Andrew Wu, assistant professor of technology and operations, who faced similar challenges. They collaborated on an AI tutor through U-M Maizey to give students a personalized source to turn to with questions.

Li saw in-class quiz scores significantly improve after implementing the Maizey tutor. As the quizzes are timed individual in-class quizzes, Li said, this demonstrates that the students are indeed developing a deeper understanding of the material. She said it also helps students who may be too shy to ask a question during class or who want to ask a question in different ways multiple times.

While faculty members may fear students have the potential to abuse AI to obfuscate learning, Li said, Maizey gives faculty the ability to construct strict guard walls.

“One of the key things that I hoped to achieve was that I didn’t want the tool directly spoon-feeding the final answer to the students. That would defeat the purpose of the learning process. But I do tell the AI tutor to help promote independent thinking from the students. So, the tutor’s goal is not to provide a final answer, but actually help with the thought process,” Li said.

Li and Wu have submitted a grant proposal to create an AI tutor that is embedded with underlying material that will cover the more than 30 statistics courses offered at U-M. Creating a unified foundational component will help students across campus who are taking the courses, Wu said, as well as students in more advanced courses who want to brush up on old material.

“I keep telling the students that the really nice thing about generative AI models is that it’s very controllable. … It’s almost like having an extremely knowledgeable TA who happens to be available 24/7,” Wu said.


Leave a comment

Commenting is closed for this article. Please read our comment guidelines for more information.