2024/11/09 by Christopher D. F. Honig · 1 voice
Computer Science · Social Sciences · #Education and Technology Integration #Teaching and Learning Programming
paper · pdf · doi:10.14742/apubs.2024.1487
openalex created_date 2024/11/09 · openalex publication_date 2024/11/09 · openalex updated_date 2026/05/21
Generative AI is increasingly used in higher education, prompting the need to effectively test, refine, and robustly evaluate its educational impact. Tools like Generative Pre-trained Transformers (GPTs) enable conversational interactions that support Constructivist learning designs: active learning strategies allowing students to engage deeply with course material through reflection and discourse. However, in our initial trials with an AI-tutors, students reported greater learning utility from non-Constructivist activities. This underscores the importance of not only developing new GPT tools but also educating students on their optimal use, designing assessments to foster appropriate learning strategies, and refining the tools themselves. We also explored self-directed learning scores but found no significant correlation with AI-tutor use-strategies, and only a slight preference for AI-tutors among highly self-directed learners. When using retrieval augmented generation (RAG) most learners could not distinguish between different large language models (LLMs) implying cheaper but refined models may be appropriate. Finally, we find AI-tutors offer a compelling value proposition to universities; student’s perceive value of the AI-tutor exceeds the associated compute costs of running the AI-tutor. Similarly, students tend to prefer AI-tutors over similarly priced teaching alternatives.