GenAI systems have become everyday tools in higher education within just a few years. Beyond generating text and images, large-language-model services are increasingly used for coding, data analysis, and the creation of assignments and tests. Meanwhile, most students already rely on genAI in their learning. This raises a key question for statistics education: to what extent is students’ genAI use conscious and critical, and how can statistics courses support responsible, learning-oriented use? This presentation introduces two student project assignments embedded in introductory statistics courses, designed with two aims: to deepen disciplinary knowledge through the applied use of statistical methods, and to develop the ability to use genAI tools consciously, critically, and responsibly. In Statistics I, students worked in groups. They selected three previously solved practical exercises from the semester—one from descriptive statistics, one from time series, and one from index numbers. Because these tasks had been completed in class, students had a reference solution for comparison. Within the project, groups solved the selected tasks using one or more genAI systems and then evaluated the outputs. In their presentations, they documented which systems they used, the prompts they formulated, and the procedures they applied to verify correctness. A typical observation was that students primarily compared AI-generated results to the in-class solutions in a purely numerical way. Many groups simply checked whether the computed values matched; if they did, the solution was considered correct, and if they differed, the answer was judged incorrect. Because most groups did not verify the solution process itself, cases occurred where AI produced the correct numerical result via an incorrect derivation, and conversely, students also labeled solutions as incorrect when they were in fact correct but used a different method than the one presented in class. Another common strategy was to copy the solution produced by one AI system into a second AI system for “validation,” which often accepted the first system’s output regardless of whether it was correct. The project therefore served well to demonstrate that the use of generative AI systems is only safe when users themselves understand the methods and are capable of critically interpreting the steps of the solution process. In Statistics II, the project supported exam preparation. Students selected two content areas: (1) sampling, estimation, and hypothesis testing, and (2) association analysis and regression. Using genAI, they generated new theoretical questions, short quiz-style items, and calculation tasks with proposed solutions. They then evaluated the corretness of the questions, computations, and the validity of interpretations. Compared with the first project, students more frequently checked not only the numeric answers but also the solution process, including attempts to solve tasks independently and compare results. AI. For theoretical questions, they relied on lecture notes, Google searches, and other generative AI systems for comparison. Overall, students concluded that genAI can support practice and self-assessment, but it cannot substitute for solid theoretical understanding or independent computation and interpretation. GenAI use is especially risky for learners without sufficient statistical foundations to verify outputs meaningfully.
