Prototype of an Automatic Problem Generation System for Statistical Learning Using Generative AI

Abstract

In this study, we developed a prototype system that used generative AI based on large language models to automatically generate practice problems, primarily for university statistics courses, and evaluated its effectiveness for practical application. Many word problems in statistics require not only numerical data but also consideration of the underlying context. Manually generating a large number of such problems with contextual data requires significant effort. Therefore, we propose a system that utilizes generative AI to automatically create a large number of word problems that incorporate relevant context and data. Specifically, this paper focuses on statistical hypothesis testing practice problems, examining the prompts given to the AI and evaluating the appropriateness of the generated problem statements, sample data, solutions, and explanations. Based on these evaluations, we assess the overall effectiveness of the system.