Online programming education, especially in MOOCs, faces challenges in scaling personalized feedback, ensuring reliable assessment, and keeping learners engaged. This dissertation explores how advanced AI, particularly large language models, can overcome these issues and enhance the effectiveness of online programming courses. Specifically, it makes several contributions: first, through a comparative analysis of two MOOC coding platforms, it identifies design trade-offs that enable scalable, interactive programming instruction, informing best practices for platform selection and development. Second, a large-scale experiment on automated feedback reveals that providing detailed formative hints did not significantly improve student performance over basic correct/incorrect feedback, highlighting the need to balance feedback richness with scalability. Third, the work introduces an AI-driven unit test generator powered by GPT-4 that automatically produces comprehensive test suites for programming exercises, substantially increasing assessment coverage and instructor efficiency. A pilot study shows that these AI-generated tests are thorough, error-free, and well-received by educators, improving the reliability of auto-grading at scale. Fourth, two novel AI “micro-apps” are developed: one that automates multiple-choice question creation, tripling instructor productivity while maintaining question quality, and another that delivers personalized recap quizzes to reinforce student learning. These tools demonstrate AI’s potential to reduce instructor workload and enhance learner engagement, though content accuracy and oversight remain important. Finally, by benchmarking GPT-4 against thousands of MOOC students, the research finds that the model often matches or exceeds average learner performance on programming tasks (especially in Python courses). This underscores both the immediate capabilities of AI and its limitations on problems requiring broader context or multimodal understanding, suggesting that course designs must evolve accordingly. Collectively, the findings show that AI technologies can significantly improve the scalability, feedback quality, assessment depth, and overall learning experience in online programming education, paving the way for the next generation of AI-enhanced educational environments.
Mohamed Elhayany (Thu,) studied this question.