TLDR: A survey of 26 CS students found that AI coding assistants are primarily used for writing code, while AI chatbots are the top choice for debugging. Students across all experience levels prefer online resources (AI or non-AI) over direct human help from peers and instructors for coding tasks.
In an era where generative AI tools are increasingly accessible, their impact on computer science education is a topic of significant interest. A recent study by Natalia Echeverry and Arun Lekshmi Narayanan from the University of Pittsburgh delves into how CS students are personally choosing to use various resources, including AI tools, for their coding assignments and tasks. This user study, which surveyed 26 CS students and practitioners, offers valuable insights into their preferences and practices.
Understanding Student Resource Choices
The research aimed to understand the actual resource selection of CS students for core coding tasks like writing and debugging. Traditionally, students relied on online resources such as blog entries, reference books, documentation, and human help from instructors and peers. However, with the advent of AI tools, the landscape has shifted.
The survey asked participants about their most common resources for writing a 300-line program from scratch. The findings revealed that blog entries, like those found on Stack Exchange, were the top choice. Closely following were AI coding assistants, such as GitHub Copilot. This indicates that while traditional online resources remain strong, AI tools are quickly becoming a significant aid in code generation.
When it came to debugging code, a different pattern emerged. AI chatbots, including popular platforms like ChatGPT, were identified as the most common resource. Blog entries were still a strong second choice. This suggests that students find AI chatbots particularly effective for identifying and resolving issues in their code.
Preference for Online Help Over Human Assistance
A notable finding across the board was the strong preference among respondents for online resources—whether AI-powered or not—over direct human help from peers and instructors. This preference held true for participants with varying levels of coding experience. Even students who reported not being able to write a 300-line program from scratch, or who hadn’t tried, leaned on AI coding assistants, LLMs, and blog entries, rather than seeking help from instructors or peers.
The study also highlighted the widespread accessibility of AI tools. Many AI coding assistants, like GitHub Copilot and Code Whisperer, offer free tiers for educational contexts. Similarly, chat LLMs such as GPT, Gemini, and LLAMA are freely available to anyone with an internet connection, making them easily integrated into a student’s workflow.
Also Read:
- Unlocking Instructional AI: A Tool for Generating Teaching Dialogues
- AI Personalization in Learning: Connecting Academic Content to Career Paths
Implications for Education and Future Research
The insights from this study are crucial for educators and AI tool designers. Understanding how students naturally integrate these tools into their learning process can help institutions develop more effective policies and support systems. Rather than focusing solely on prescriptive uses or instructor views, this research provides a student-centric perspective.
Future work for the researchers includes revising the questionnaire to offer more programming language options and translating it into multiple languages to broaden the survey’s reach. This ongoing research aims to provide even deeper insights into the evolving relationship between CS students, traditional resources, and advanced AI tools. For more details, you can refer to the full research paper available here.
