Core Concepts
Generative AI can be used to provide students with personalized and contextualized reflection triggers during collaborative problem-solving activities in computer science education, to enhance learning.
Abstract
The paper presents a novel approach to leveraging Generative AI, specifically ChatGPT, to provide students with dynamic and personalized reflection triggers during a collaborative SQL optimization task in a cloud computing course. The key contributions are:
Technical contribution: Development of a prompt engineering approach to elicit contextually appropriate suggestions of alternative code contributions from language models as reflection triggers.
Learning resource development contribution: Enhancement of a platform for online collaborative software development with dynamic support for reflection during the learning activity.
Learning research contribution: Testing the impact of the LLM-constructed reflection triggers on student learning in an online collaborative SQL optimization activity, along with a thorough analysis of their impact.
The authors designed five types of reflection triggers targeting different learning objectives related to data types, indexing, and denormalization in database optimization. These triggers were personalized using prompts to ChatGPT to generate alternative solutions based on the students' current work. The triggers were then scheduled and presented to the students during the collaborative activity.
The results show that the reflection triggers affected how students spent their time, with a negative impact on early task completion but a positive impact on the more difficult later tasks. However, the analysis did not find a significant difference in learning gains between the experimental group that received the personalized reflection triggers and the control group. The authors discuss several areas for improvement, including enhancing student engagement, improving the prompting context, and enhancing the readability of the reflection triggers.
Stats
The study involved 34 students in a Cloud Computing course, with 22 in the experimental condition and 12 in the control condition.
Out of the 34 students:
31 (91.17%) completed task 1 (19 in the experimental condition and 12 in the control condition)
29 (85.29%) completed task 2 (17 in the experimental condition and 12 in the control condition)
20 (58.82%) completed task 3 (3 in the control condition and 17 in the experimental condition)
Quotes
"An advantage of Large Language Models (LLMs) is their contextualization capability – providing different responses based on student inputs like solution strategy or prior discussion, to potentially better engage students than standard feedback."
"Recent advances in Generative AI (GenAI) and Large Language Models (LLMs) have enhanced AI capabilities for the evaluation of multimodal student input and real-time feedback, which has provoked intensive exploration of the space of application possibilities."