Investigsting student use of Copilot for object-oriented programming

Abstract

Generative artificial intelligence is gaining popularity and becoming more accessible to students in higher education. Students learning programming can use code-recommender systems such as GitHub Copilot to help them code. Educators worry students might become over-reliant on these types of systems. To gain more insight into how GitHub Copilot is used in higher education this study aims to find typical behavior of students while they use Copilot to solve an object-oriented programming problem. A think-aloud study was conducted to observe students while using Copilot. Observations focused on identifying whether students used fundamental programming skills and how this affected their performance. I found that when students planned their solution well they used Copilot as a beneficial tool. While students who did not plan their solution did not benefit from using Copilot.