2025 ASEE Annual Conference & Exposition

Electric Circuit is a core course in Electrical Engineering and serves as a prerequisite for many advanced courses. A key aspect of teaching and learning in this course is the step-by-step demonstration of how to solve circuit problems. When students face challenges in class or with homework, the most effective teaching method is to walk them through problem-solving steps, helping them understand the material through direct application. This approach not only teaches complex topics but also strengthens students’ problem-solving skills. However, some students may not have sufficient time to get help from the instructor or TA due to the limited resources or other reasons.

Simulation software like Multisim and SPICE are widely used in electric circuit course teaching. However, these tools lack step-by-step guidance and explanation of the approaches and formulas used to solve the problem. While the use of AI in teaching and learning remains a debated topic in academia, many students already turn to AI for help before seeking assistance from the instructor or TA. Therefore, it is important to understand AI's strengths, weaknesses, and limitations so that students can benefit from the latest advances while minimizing potential biases and errors in AI models. It is also important for instructors to develop policies and guidelines for the appropriate use of AI in the class.

This study aims to evaluate ChatGPT 4o in the following five categories: 1) its ability to read and understand circuit diagrams presented in images, 2) the interpretation of the circuits, 3) its ability to break down procedures into steps, 4) its ability of selecting appropriate formulas, and 5) its equation and numerical calculation skills. A total of 72 problems (4 problems in each chapter) were selected from the 12th edition of “Electric Circuits” by Nilsson and Reidel. A category is classified as “helpful” if ChatGPT 4o provides >80% correct answers (at least 58 out of 72), or “not helpful” otherwise. An error is only counted once in the category where it first occurred (single-count error) to avoid over-penalizing a chain of dependent mistakes. To ensure consistency, all the problems were presented to ChatGPT using the same prompt.

The results of the study indicate that ChatGPT 4o performed well in reading and understanding circuit diagrams (category 1), procedure decomposition (category 3), and formula selection (category 4). It also performed reasonably well in calculations (category 5), approaching the 80% "helpful" threshold. However, it performed poorly in circuit interpretation (category 2), which caused most of the final answers incorrect.

OpenAI announced on 09/12/2024 that its new large language model, o1, "ranks in the 89th percentile on competitive programming questions (Codeforces), places among the top 500 students in the US in a qualifier for the USA Math Olympiad (AIME), and exceeds human PhD-level accuracy on a benchmark of physics, biology, and chemistry problems (GPQA)." The study did not include OpenAI o1 Preview because it currently cannot process images. An expanded study, now in progress, involves using ChatGPT 4o for image interpretation and then forwarding the interpretations to the OpenAI o1 model. The expanded study is expected to be completed in two months. If the o1 model demonstrates improved reasoning and calculation capabilities, the use of AI to support Electric Circuits instruction appears very promising.