Auckland study finds 5 inclusivity problems in GitHub Copilot
A study of 20 students at the University of Auckland found that the AI coding assistant suits some ways of thinking better than others. It also found Copilot did little to help students learn.
Researchers at the University of Auckland posted a study on 30 October 2026 that found five inclusivity problems in GitHub Copilot, an AI coding assistant.
The paper, by Grace Kerr, Aarushi Parasrampuria, Faith Culas and Kelly Blincoe, is on arXiv, a public archive of research papers. Twenty students who knew Java used Copilot to write, debug and refactor code while saying their thoughts aloud. Refactoring means rewriting code to make it clearer without changing what it does. The sessions took place in July and August 2025.
The team used GenderMag, a method for finding software features that work less well for some users. It compares people on five traits, including motivation, attitude to risk and learning style. Some students who wanted to know why Copilot made a suggestion found the explanations long and in a separate panel. Some students blamed themselves when they could not find features such as Chat and Agent mode.
The authors found that Copilot helped students finish tasks but gave little support for learning. Some students accepted code while unsure and then removed it with Undo. Others felt pushed towards one complete solution. Verified students and teachers can get a plan called GitHub Copilot Student free, according to GitHub Education.
The authors suggest short explanations next to each suggestion and optional learning modes with links to read more. They say these ideas have not been tested and that the sample was small and from one university. Their findings reflect the version of Copilot used at the time. It is not known whether later versions behave the same way.