Analysis of ChatGPT-5's scientific explanation ability in solving direct current circuit problems

Abstract

This study examined GPT-5's ability to construct scientific explanations when solving direct-current (DC) circuit problems. The analysis focused on whether the model could provide correct claims, relevant evidence, and logically connected reasoning grounded in physics concepts. A qualitative descriptive design was employed, with GPT-5 accessed through ChatGPT Plus as the unit of analysis. The model responded to six image-based multiple-choice items adapted from a basic DC-circuit assessment. The items addressed electric current, potential difference, and electric power as represented by bulb brightness in series and parallel circuits. To standardize the elicited responses, each item was presented in a new conversation together with the same structured prompt requiring a claim, evidence, and reasoning. The responses were evaluated using an analytic rubric with a maximum score of five per item. GPT-5 obtained 30 out of 30 points, corresponding to 100% across the six items. Its responses consistently selected the correct option, applied appropriate equations and circuit principles, and connected the evidence to the claim through coherent reasoning. These findings indicate that, under the specific prompting conditions and limited item set used in this study, GPT-5 demonstrated strong scientific explanation performance in basic DC-circuit contexts. Nevertheless, the findings should not be generalized to other physics topics, prompt formats, or AI systems without further investigation.