.png)
Why are we interested in AI?
AI-generated code is reshaping the educational landscape, especially in programming courses. With tools like large language models (LLMs) becoming increasingly sophisticated, educators are facing a critical moment of adaptation. According to Becker et al. (2023), the sudden rise of these technologies "presents both opportunities and challenges for students and educators," and there's an urgent need to reassess traditional teaching methods to accommodate this shift.
LLMs, such as Copilot ChatGPT, can outperform most novice programmers in solving typical assignment questions—especially well-structured ones. While this might seem advantageous, it also raises concerns about how students learn and are assessed. Many educators have been facing this fundamental question for a while: Should they allow the use of LLMs, or should they ban them entirely? If allowed, how can educators ensure that everyone has access, and can use LLMs correctly? The difficulty lies in balancing student learning with the ubiquity of these tools, which are now integral to many coding environments.
Navigating the Challenges of AI in Learning
One of the biggest challenges with LLMs is that they can solve assignments more effectively than most students, particularly when tasks are well-defined—something educators aim for to ensure clarity. This raises the concern: how do we ensure students are truly learning if AI does most of the work? Traditional assignments may no longer provide an accurate measure of student comprehension.
To counter this, some institutions are returning to in-person exams and proctored assessments, ensuring students can demonstrate their knowledge without AI assistance. However, this doesn’t address the fact that LLMs are now ubiquitous in industry, integrated into most IDEs and widely accessible through platforms like ChatGPT. Banning them outright isn’t realistic, nor does it prepare students for the real-world use of these tools. Instead, the focus should shift to assessments that encourage learning while allowing for responsible use of AI.
LLMs also produce code that can be more advanced than what’s expected at a beginner level, potentially leading students to rely too heavily on AI-generated solutions. As Petrovska et al. found, first-year students often praised ChatGPT-generated code for efficiency but noted missing elements like comments and proper variable naming. More advanced students found it helpful but believed it was more useful for small tasks than complex projects. This highlights the need to ensure AI doe
