Univeristy of Pennsylvania
Close

Asking the Right Questions: Using Generative AI to Advance Skills in STEM

Large, empty lecture classroom.

09Jul’24

AI Stories, Feature Stories, Spotlights

July 9, 2024 by Erin Bartnett

When Masao Sako was planning his Intro to Programming and Data Science course for spring 2024, he settled on one goal: by the end of the semester, he wanted students to be able to conduct meaningful data analysis and data science. This was an ambitious target for the students in the class, who would generally be freshmen with no coding background. He decided that “getting help from ChatGPT was probably the fastest and most effective way for students to quickly learn how to code.” Sako had used it in his own work and discovered it could be incredibly useful if you knew how to ask the right questions. 

Bhuv Jain came to similar conclusions, but rather than teach students in his Solar Systems, Other Planets, and Life on Planets course how to code, he wanted to figure out how to use ChatGPT in group assignments to give introductory students the opportunity to take a “one-time deep dive” into more complex topics. For starters, nuclear physics. 

Both Sako and Jain have used Generative AI tools in both independent assignments and group work to help students with a wide range of knowledge accelerate their learning, specifically in introductory STEM courses.  

"Getting help from ChatGPT was probably the fastest and most effective way for students to quickly learn how to code."

Using AI to Help Students Code

In order to know what the right questions were, Sako believed it was crucial for students to understand the basic structure and terminology of Python: “If they don’t know what an array is, they can’t ask how to use it.” Sako began by teaching students the basic structure and terminology of Python, emphasizing the step-by-step approach to solving a problem.  

Once students understood how to write the initial steps of an algorithm, they could start using ChatGPT to code. However, Sako insisted that students remain informed users. He told students: “Whenever you copy and paste code from ChatGPT, make sure you understand every single line that it’s giving you. Don’t just blindly use it, because if you do, you may end up being okay, but you will never know.”  

Using ChatGPT helped students learn how to code, but it also helped Sako assess whether students were grasping the core concepts of the class. Initially, Sako said, “[students] were confused and not quite asking the right questions. But after a certain number of tries, or probably dozens of tries, most of the students got it and now they know what questions to ask.” The transition was swift: “after four weeks, they were writing pretty sophisticated Python code.” 

"The purpose of this activity, Jain explained, was to help students 'develop their ability to interact with ChatGPT as you would if you had a one-on-one office hour or recitation.'"

A Deep Dive on Nuclear Physics for Non-Majors  

“When you get heavy elements deep inside a star, why isn’t the whole periodic table equally represented? Why are carbon and silicon and iron on our planet so abundant?” These were the questions Bhuv Jain’s students answered together in small groups, using ChatGPT as a guide. After working together, students submitted the questions they posed to ChatGPT on Canvas, and shared what they had learned with one another in a broader class discussion to align on what they were learning. The purpose of this activity, Jain explained, was to help students “develop their ability to interact with ChatGPT as you would if you had a one-on-one office hour or recitation.” 

Jain also used ChatGPT in a previous course on machine learning and the natural sciences. Similar to students in Sako’s course, students used ChatGPT to help advance their coding skills, and Jain recalled how using ChatGPT in groups accelerated students’ ability to do a problem that would normally take a couple of hours. He saw the value in using ChatGPT “both as an accelerant, and for deeper dives where students can have a moment of insight or a moment of grappling with something that's more technical than they think they're prepared for.” 

Group work offered students an opportunity to do that grappling. Ideally, Jain said, students working in groups not only learn the material faster, but also teach one another: “[If] they do it together, they may develop some understanding together while they’re interacting with it. If one person’s typing the question and getting a response, the other might be like, ‘Oh, that’s what it meant. Maybe you can ask this question next.” 

CETLI Resources  

If you are interested in learning more about whether or how to use Generative AI tools in your own teaching, visit CETLI’s web resource, Generative AI & Your Teaching, sign up for the Seminar, Using AI Productively for Teaching, or email us to brainstorm about specific questions or general approaches you have for your own courses. 

To learn more about how professors in German, art history, and nursing are using generative AI in their classrooms, read the Feature Story, Student as Sorcerer.