Student Perspectives on Using a Large Language Model (LLM) for an Assignment on Professional Ethics

TL;DR

This paper investigates how computing master's students perceive and experience using a Large Language Model (LLM) as a discussion partner and writing aid in a course module on professional ethics in Sweden. Grounded in Lennerfors et al.'s ethical process and an autonomy-matrix framework, the assignment centers on a CHI 2024 case (Andreas' dilemma) analyzed with various ethical theories, with the LLM providing explanations and prompting critical reflection. Findings indicate that LLMs can enhance knowledge and broaden perspectives while presenting limitations such as potential inaccuracies and overreliance, and that instructors must explicitly teach LLM use and critical evaluation. The study offers practical guidelines for educators to design ethics education that leverages LLMs without replacing human judgment, with implications for curriculum development and assessment practices in computing education.

Abstract

The advent of Large Language Models (LLMs) started a serious discussion among educators on how LLMs would affect, e.g., curricula, assessments, and students' competencies. Generative AI and LLMs also raised ethical questions and concerns for computing educators and professionals. This experience report presents an assignment within a course on professional competencies, including some related to ethics, that computing master's students need in their careers. For the assignment, student groups discussed the ethical process by Lennerfors et al. by analyzing a case: a fictional researcher considers whether to attend the real CHI 2024 conference in Hawaii. The tasks were (1) to participate in in-class discussions on the case, (2) to use an LLM of their choice as a discussion partner for said case, and (3) to document both discussions, reflecting on their use of the LLM. Students reported positive experiences with the LLM as a way to increase their knowledge and understanding, although some identified limitations. The LLM provided a wider set of options for action in the studied case, including unfeasible ones. The LLM would not select a course of action, so students had to choose themselves, which they saw as coherent. From the educators' perspective, there is a need for more instruction for students using LLMs: some students did not perceive the tools as such but rather as an authoritative knowledge base. Therefore, this work has implications for educators considering the use of LLMs as discussion partners or tools to practice critical thinking, especially in computing ethics education.