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

Synthesis: Dawson and Rebello (2026) investigate students' perceptions of generative AI-assisted problem posing, a pedagogical practice in which learners generate novel problems or variations to strengthen transfer and Problem Solving in physics. Using a phenomenological approach, they found that students perceived a positive change in their AI interactions after receiving prompt-engineering training, and held generally positive views of problem posing as a self-study technique, though a smaller subset showed hesitations about using AI. The study suggests structured training can help students use Generative AI productively in physics while mitigating risks, positioning problem posing as a Self-Regulated Learning strategy.

Key Findings

1. A majority of participants (76%) perceived a positive change in their interactions with GenAI after watching the prompt-engineering training video, while 13% reported no significant change — largely students who already had prior experience with Prompt Engineering.
2. Attitudes toward problem posing with GenAI were generally positive (about 68% positive, 17% mixed, 9% negative), with positive students planning to use it for practice, exam preparation, and targeting weak areas in their self-study.
3. Students who held reservations tended to prefer readily available materials, distrust the reliability of AI-generated problems, or hold broader moral and ethical objections to using GenAI for coursework.
4. Across both themes roughly 72% of responses were positive, and most students who perceived a positive change in interactions also held positive views of the technique — reinforcing training on Prompt Engineering as crucial for students who engage with GenAI.
5. Training most benefited novices: students with little prompting experience found their interactions became more conversational, relevant, and closely aligned with their prompts after instruction in techniques such as roleplaying and chain-of-thought prompting.

Problem Posing as a Learning Practice

Problem posing — asking students to generate novel problems or meaningful variations — supports Transfer of Learning and strengthens problem-solving skills in Physics Education. It is challenging for novices, because posing problems involves synthesis at a high level of Bloom's taxonomy and requires thinking about both how a question can be re-framed and the feasibility of the problem in real-world contexts. Because of this difficulty, students benefit from Scaffolding. This study positions GenAI as a tool to provide that scaffolding in ways that maximize benefits and limit risks.

A Transfer-Based Conceptual Framework

The authors adapt their prior Transfer of Learning framework, derived from Jonassen's problem-solving work, distinguishing horizontal transfer — changing a problem's representation or context while keeping the approach the same — from vertical transfer, which alters complexity and requires applying concepts more deeply rather than memorizing key examples. Students pose new problems through both routes, and the framework gives them a theoretically grounded basis on which to design new problems with GenAI as a guide rather than a black box.

Training Shapes AI Interactions

Students were introduced to GenAI-assisted problem posing as a self-study technique and received training on Prompt Engineering techniques drawn from prior work. A phenomenological analysis of N = 49 students' open-ended reflections found that students perceived a positive change in their interactions with GenAI after training. Those with little or no prior prompting experience reported that their pre-training interactions were unfocused, lacked detail, or produced too little change in the problem to be useful, whereas post-training interactions felt more conversational and aligned with the prompt. This suggests that how students are trained to interact with AI shapes both the quality of interaction and their confidence.

Attitudes Toward Problem Posing with GenAI

Students' attitudes toward the technique were generally positive but more varied. Many planned to use the method for practice and exam preparation and to deepen their understanding of concepts, often building new problems from course activities like quizzes and homework. A subset individualized their learning by targeting weak areas and advancing their problem-solving strategies. Mixed students showed interest but cited hesitations, including not fully trusting the generated output and preferring instructor- or course-provided materials. The small negative group opposed not problem posing itself but the use of GenAI for the practice, citing moral and ethical objections and broader impacts beyond education.

Toward Structured Study Techniques

The findings lay a foundation for broader Prompt Engineering training and for incorporating GenAI into structured study practices like problem posing, connecting to Self-Regulated Learning, Motivation, and Student Experience in Physics Education. It complements work on AI-generated solutions by emphasizing learner-generation of problems rather than consumption of answers. The authors note that most students were open to incorporating GenAI into structured self-study, supporting the introduction of GenAI training into curricula while keeping in mind the subset who hold moral objections.

What this means for practice

• Instructors. Teach specific prompting techniques (role assignment, chain-of-thought prompting) before asking students to use Generative AI for problem posing; 76% of participants perceived better interactions only after this training, and novices benefited most.
• Instructors. Assign GenAI-assisted problem posing as a self-study technique: generating new problem variations rather than consuming answers keeps learners in the producer role and supports practice, exam preparation, and transfer.
• Instructors. Give students an explicit variation framework — horizontal change of representation versus vertical change of complexity — so posed problems target transfer deliberately rather than cosmetically.
• Instructors. Offer non-AI alternatives and make ethics discussable, because a minority of students prefer instructor-provided materials or object to GenAI on moral grounds.

Limitations

• Recruitment. The online modality and small course size limited recruitment, and students self-selected into the extra-credit assignment (N = 49 reflections), which may have skewed views toward the agreeable while excluding GenAI opponents.
• Measurement. Reliance on written reflection responses rather than interviews may have contributed to moderate inter-rater agreement (Cohen's Kappa around 0.50–0.53), which is nonetheless in the acceptable range for this kind of Qualitative Research.

Citation

Dawson, L., & Rebello, N. S. (2026). Exploring students' perceptions of using generative AI-assisted problem posing.