Abstract

Generative AI (GenAI) has reshaped higher education. Students increasingly rely on GenAI to understand course material, complete assignments, and prepare for exams; yet educators are concerned that uncritical reliance may bypass the deliberative problem-solving processes that traditional learning aims to cultivate. As an instructor, I see it as my responsibility to teach students to harness AI rather than be replaced by it—using GenAI as a learning aid while emphasizing that disciplinary knowledge remains the essential competence through which humans guide AI and validate its outputs. To operationalize this in a Data Science course, I developed the GDD Triad framework, which posits that effective use of GenAI in technical work requires three competencies. G – Generative AI Proficiency involves selecting tools, crafting precise prompts, and iteratively refining outputs. D – Data Science Literacy refers to understanding the Data Science pipeline (data preparation, exploratory data analysis (EDA), feature engineering, modeling, and evaluation) and the techniques applied at each stage. D – Domain Knowledge denotes familiarity with the application domain (e.g., healthcare, finance), enabling informed decisions about which features are meaningful and how to process or combine them. The framework’s central claim is that prompt engineering alone is insufficient: without the two “Ds,” students cannot formulate methodologically sound prompts or verify whether GenAI outputs are contextually appropriate. The framework was implemented in Fall 2025 with 111 senior undergraduates. Lectures built Data Science Literacy through data cleaning, EDA, predictive modeling, and interpretation, while labs applied the pipeline to UCI healthcare datasets (e.g., heart disease). Students used GenAI to generate Python code, with worksheets guiding them to compose prompts grounded in Data Science principles and the healthcare domain. Challenged to outperform published baselines, their GenAI-assisted models achieved 89–93% accuracy and 92–96% precision in heart disease prediction. To evaluate students’ ability to verify GenAI outputs, written exams used a reverse-engineering approach: students received Python code snippets and composed prompts that would lead GenAI to produce equivalent code. Three levels of understanding emerged. (1) No interpretation: students could not interpret the code; prompts were largely guesses. (2) Coding Literacy without Data Science Literacy: students grasped what the code did syntactically but could not connect it to Data Science concepts. Two sub-patterns appeared—literal restatements (e.g., “compute the mean/median”) lacking analytical intent (e.g., imputing missing values), and misattributions (e.g., using mean/median to address outliers). (3) Data Science Literacy: students interpreted the code and correctly mapped it to the appropriate stage of the pipeline. The GDD Triad, combined with reverse-engineering assessment, extends prior work (e.g., Shanto et al., 2025) and offers a concrete approach for Information Systems courses where concepts are expressed through code. To generalize beyond Data Science, instructors can adapt it by identifying foundational concepts, determining where GenAI can support implementation, and designing assessments that evaluate students’ ability to assess GenAI outputs against disciplinary knowledge. Reference Shanto, S. S., Ahmed, Z., & Jony, A. I. (2025). A proposed framework for achieving higher levels of outcome-based learning using generative AI in education. Educational Technology Quarterly, 2025(1), 1-15.

Download

Share

COinS