Exploring Students' Perceptions of Using Generative AI-Assisted Problem Posing

Imagine you are trying to learn how to ride a bicycle. You could just sit there and watch someone else ride, or you could try to memorize the exact path they took. But the real magic happens when you start building your own little ramps, designing your own obstacles, and figuring out how to balance on a wobbly surface you invented yourself. In the world of physics education, this act of inventing your own challenges is called "problem posing." It's a fancy way of saying: instead of just solving a math puzzle someone else gave you, you twist the rules, change the numbers, or imagine a new scenario to see if you truly understand how the universe works.

Now, imagine you have a super-smart, all-knowing robot friend (Generative AI) who can help you build these ramps. But here's the catch: if you just ask the robot, "Make me a hard bike ramp," it might give you something weird or useless. However, if you learn how to talk to the robot using a special "secret language" called prompt engineering—where you give it very specific instructions, roles, and steps—it becomes an incredible partner. This paper explores what happens when students learn this secret language. The researchers wanted to know: Does teaching students how to talk to AI properly change how they use it to invent their own physics problems? And do students actually like using this robot friend to study, or do they think it's a method that undermines the learning process?

---

The Experiment: Teaching Students to Talk to the Robot

In this study, researchers at Purdue University invited a group of college students taking a physics course to play a game with Generative AI. The goal was to see if a little bit of training could turn a clumsy interaction with AI into a powerful study tool. The students went through three stages over the course of one week.

First, in the "Before Training" phase, the students were told to use an AI tool of their choice to create a new version of a physics problem they had already solved. They had to chat with the AI, make a new problem, and then write down what they thought about the conversation. At this point, most students were just asking basic questions, like "Give me a harder version of this."

Next came the "Training" phase. The students watched a 15-minute video. This wasn't just a boring lecture; it taught them how to use "prompt engineering." Think of this like learning how to give a GPS very specific directions. Instead of just saying "Drive to the beach," you learn to say, "Drive to the beach, but take the scenic route, avoid the highway, and stop for ice cream." The video also taught them a special way to think about physics problems, helping them understand how to change a problem's context or difficulty in a meaningful way.

Finally, in the "After Training" phase, the students went back to the AI. They were asked to do the exact same thing: create a new physics problem. But this time, they had to use the new "secret language" and thinking skills they just learned. Afterward, they reflected on how different the experience felt compared to the first time.

What the Students Discovered

The results were like watching a group of people go from fumbling with a new video game controller to becoming pro players. When the researchers looked at the students' reflections, they found that 76% of the students felt a positive change in how they interacted with the AI after the training.

Before the training, many students described their conversations with the AI as "basic," "unfocused," or like they were getting answers that didn't quite match what they needed. It was like trying to order a custom pizza by just shouting "I want pizza!" and hoping for the best. After the training, the students felt their conversations became more like a real dialogue. They used techniques like "role-playing" (telling the AI to act like a specific character) and "chain of thought" (asking the AI to show its work step-by-step). One student noted that before, the answers felt less detailed, but after learning the new techniques, the AI's answers became "clearer, more thorough, and better explained the steps involved."

Interestingly, the few students who didn't feel a big change were usually the ones who already knew how to talk to AI or were already using simple tricks on their own. For them, the training just gave names to things they were already doing.

Do Students Like Using AI to Study?

The second big question was: Do students actually want to use this method to study on their own? The answer was mostly a cheerful "Yes," but with a few grumpy "No's" and some "Maybe's."

About 68% of the students had positive attitudes toward using AI for problem posing. They saw it as a way to get personalized practice problems, get tutoring-style feedback, and target the specific areas where they were weak. They imagined using it to prepare for exams or to deepen their understanding of tricky concepts. It was like having a personal coach who could invent a million different drills just for you.

Another 17% of students had mixed feelings. They thought the idea was cool and might be useful, but they were hesitant. Some worried that the AI might make mistakes or give them fake problems, so they would need to double-check everything. Others felt that while it was helpful, they preferred sticking to the official homework and practice tests given by their teachers. They were like kids who liked the idea of building their own LEGO castle but still wanted to make sure the instructions were perfect before they started.

Finally, about 9% of the students held negative views. These students were firmly against using AI for their studies. Their reasons weren't about the quality of the problems; they were about ethics. Some felt it was morally wrong to use AI for schoolwork, while others were worried about the environmental impact of running these powerful computers. They believed that students had managed to learn physics for years without AI, using textbooks and human teachers, and that AI shouldn't be encouraged.

The Big Picture

When you put all the pieces together, the study suggests a clear trend: Training matters. The vast majority of students who felt their interaction with AI improved after the training also ended up having a positive view of using AI for problem posing. Even among the students who were skeptical about AI in general, the ones who saw their interactions improve still acknowledged that learning how to prompt the AI was crucial.

The researchers found that while most students are open to using AI as a study partner, there is a small group who will never use it due to moral objections. However, for the rest, teaching students how to talk to AI properly seems to be the key to unlocking its potential. It turns a tool that might just give you answers into a partner that helps you build your own understanding. The study concludes that schools should consider teaching these "prompt engineering" skills to help students get the most out of AI without losing their own ability to think critically. It's not about replacing the student with the robot; it's about teaching the student how to drive the robot.