Summary
GenGenAI creates opportunities to teach expert problem-solving skills in reasonably complex real-world contexts.
Students still need a foundation of basic physics knowledge to interact productively with GenGenAI.
A two-part structure appears most effective: concepts first, complex problem-solving later.
Background
I have been exploring the use of context-rich real-world problems in physics teaching for more than 17 years. These problems are open-ended and “ill-structured,” meaning that they lack complete information and do not have a predetermined final answer. Students must make assumptions, identify relevant concepts, and gather appropriate information. This kind of work mirrors how experts actually approach problems and has many benefits, but students also find it challenging and often intimidating.
Students’ difficulties typically arise from:
(a) limited context-specific knowledge,
(b) uncertainty about which concepts apply in unfamiliar situations,
(c) lack of strategy or not knowing how to begin, and
(d) gaps in basic physics knowledge.
Another issue is that context-rich problems do not fit well into most exam-based assessment structures. As a result, my past implementations relied on a few relatively straightforward scenarios where the context-richness was limited.
Why engage students in real-world problem questions?
The underlying goal is to help students develop adaptive expertise—the ability to use their learning beyond the narrow confines of textbook-style questions. Research from the Wieman group at Stanford has analyzed expert problem-solving across STEM fields and identified 29 decisions that experts consistently make (Price et al., 2021). Burkholder and colleagues (2022) showed that a subset of 15 of these decisions can be practiced meaningfully in first-year physics courses. These decisions involve finding and evaluating information, identifying useful concepts, planning solutions, and evaluating results.
This approach contrasts sharply with typical textbook problems, which are highly structured and provide all necessary information. Students’ only decisions then are to choose formulas and substitute numbers. Burkholder et al. (2020) instead emphasize the process: their problem-solving template gives significant credit for reasoning and decisions, and only a small fraction for the final numerical answer.
What are the barriers?
Although the decision-focused approach has clear instructional value, I had concerns about how well it could be implemented in my courses. My questions centered on familiar student difficulties:
- How do students acquire enough basic physics and math skills to work effectively with open-ended problems?
- How do they look up relevant information if the context is unfamiliar?
- How do they identify which physics concepts are applicable when the situation looks entirely new?
Indeed, Burkholder et al. found that students often struggle to construct an appropriate predictive framework—the set of models, choices, boundaries, and information that guides the solution.
There is also the practical issue of fitting such open-ended work into an exam-oriented assessment structure typical of first-year science courses.
How does GenAI help?
Consider a potential case study: an autocrane topples while lifting a car from a ditch. Students receive only a photo and must determine whether the operator used reasonable judgment. A simple web search will fail unless students already know technical terms like boom length, outrigger, or counterweight. Without that vocabulary, students stall out before any physics reasoning can begin.
This is where GenAI provides a major opportunity. Students can ask in natural language, “Why do autocranes topple?” and receive key terminology and conceptual scaffolding. ChatGPT-4 can generate an initial physics model, including assumptions. That model may be incomplete or partially incorrect, but this creates an ideal opening for students to evaluate, critique, and refine a model—core expert skills.
GenAI therefore lowers the barrier to getting started. Students can form initial ideas without needing extensive prior content knowledge, allowing them to focus on evaluating information, selecting relevant physics concepts, and engaging in sensemaking. This aligns with findings from Wang et al. (2024), who showed that ChatGPT-4 can help students identify relevant knowledge and construct the beginnings of a useful predictive framework.
With GenAI in place, I can imagine course modules in which students practice prompting the GenAI, evaluating models, deciding what information is needed, and checking whether results make sense. At the same time, these activities help students build adaptive expertise and confidence in tackling novel problems.
How could GenAI be implemented in class?
My current courses use online edX materials that include about 400 questions per course, supported by an open textbook. Roughly half of these questions are reading quizzes or homework; the rest are discussed in class. This provides structured practice with the basics: definitions, core concepts, representations, and the physics-related math needed for more advanced work.
I envision freeing up at least four lectures to introduce GenAI-supported case studies. The displaced material will move into tutorials or homework, and a small number of rarely assessed topics will be trimmed.
The skills practiced in case studies—evaluating information, making decisions, planning solutions, and sensemaking—are at the top of Bloom’s taxonomy. Vantage College’s small class sizes make it an ideal environment for a pilot. Effective implementation will require the instructor to guide motivation, facilitate small-group work, and experiment with suitable assessments (e.g., group presentations and reflective writing).
The course would initially maintain traditional exams to ensure alignment with other sections. Two smaller assessments should be enough to support and assess the new skills:
(1) a group component on the midterm, and
(2) a final mini-project where students analyze a real-world scenario and reflect individually on their learning.
Future Developments
GenAI continues to advance rapidly, and its capabilities are reshaping what it means to do physics in an educational context. This raises fundamental questions:
- Why should students practice traditional textbook-style problems?
- What role should GenAI play in our instruction?
- How do we assess students’ physics skills?
My recent experience implementing a concepts-first approach (Webb & Paul, 2023) informs my answers. In a 200-student PHYS 118 section, front-loading concepts made early lectures more fluid and reasoning-focused. However, when we revisited longer problems at the end of the term, it became clear that many were overly specialized or exam-driven rather than conceptually meaningful.
Thus:
- Traditional textbook problems remain useful for building foundational math and physics skills—but they do not need to be overly complicated. Problems that enhance conceptual reasoning are the most valuable.
- GenAI is well-suited for context-rich case studies. These require students to evaluate models, make assumptions, and engage in expert-like thinking. These skills represent genuine disciplinary expertise.
- Assessment must reflect what students actually do in the course. A realistic structure blends traditional exams for fundamentals with project-based assessment of expert decision-making in real-world situations.
Assessment Ideas
To align with course goals, assessments should include:
- Assessment of fundamentals, mainly via quizzes and exams.
- Assessment of expert problem-solving skills, ideally during the final tutorial or in a dedicated session.
A structure with one midterm, one final exam, and several quizzes seems appropriate. Group quizzes work well in Vantage College but require class time.
Current Experience
Goal: explore whether context-rich problems supported by GenAI and peer review can be integrated smoothly into a first-year physics course.
I implemented eight examples of varying complexity. For three of them, I used UBC’s ComPair tool for assessment. Students worked in small groups (2–5 students) to solve a context-rich problem and uploaded a scanned solution. They then (individually) reviewed two solutions from other groups, comparing them and explaining which one they considered stronger and why. Each solution typically received four peer reviews. The intent was to help students recognize important features of physics models and to see how different groups organized and communicated their solutions.
What I learned:
The context-rich problems and ComPair activities fit naturally into the course structure. The total time investment amounted to about 10 lecture and tutorial hours (roughly 14% of class time), which did not disrupt the broader curriculum. The main trade-off was shifting some traditional problem-solving questions to homework and removing a few low-value items. All core physics topics remained in the course.
Students were able to construct simple models, but they often missed important elements in more complex scenarios. Even when they understood the situation visually, many needed additional scaffolding—especially when the key quantity required to solve the problem was not obvious (e.g., stopping distance in the “cell phones and traffic” example). The timing of these activities mattered: context-rich problems were most effective when they followed soon after the relevant concepts were introduced.
The ComPair tool itself worked very well, but the depth of students’ peer-review comments was mixed. Most comments were brief and somewhat superficial, and some students appeared hesitant to offer constructive criticism. Stronger incentives or clearer expectations may help encourage more thoughtful peer review. Monitoring or accountability measures might also be useful. The superficiality may partly reflect that this component contributed only a small, easily earned portion of the overall grade.
Overall, I consider the initial implementation a success. We demonstrated that these activities can be integrated into first-year courses without major disruption. Although much can be improved—both in scaffolding and in encouraging deeper engagement during peer review—the experience showed clear value in having students model real-world situations. The next steps will involve refining the activities and gathering more systematic evidence of student learning.
References:
Price, A. M., Kim, C. J., Burkholder, E. W., Fritz, A. V., & Wieman, C. E. (2021). A detailed characterization of the expert problem-solving process in science and engineering: Guidance for teaching and assessment. CBE Life Sciences Education, 20(3), ar43-ar43. https://doi.org/10.1187/cbe.20-12-0276
Burkholder, E., Miles, J., Layden, T., Wang, K., Fritz, A., Wieman, C., et al. (2020). Template for teaching and assessment of problem solving in introductory physics. Phys. Rev. Phys. Educ. Res. 16, 010123. doi: 10.1103/PhysRevPhysEducRes.16.010123
Burkholder, E., Salehi, S., Sackeyfio, S., Mohamed-Hinds, N., and Wieman, C. (2022). Equitable approach to introductory calculus-based physics courses focused on problem-solving. Phys. Rev. Phys. Educ. Res. 18, 020124. doi: 10.1103/PhysRevPhysEducRes.18.020124
Wang KD, Burkholder E, Wieman C, Salehi S and Haber N (2024) Examining the potential and pitfalls of ChatGPT in science and engineering problem-solving. Front. Educ. 8:1330486. doi: 10.3389/feduc.2023.133048
Webb DJ and Paul CA (2023) Attributing equity gaps to course structure in introductory physics. Phys. Rev. Phys. Educ. Res. 19, 020126.