Advancing Global South University Education with Large Language Models

TL;DR

This paper addresses the Global South higher education quality gap and faculty workload by exploring large language models as learning assistants in Telkom University's courses. It proposes a tripartite learning interface and a controlled-action research design with two experiment types (assignment-based and exam-based) across five courses to empirically evaluate impact on learning quality and instructor workload. The study provides a concrete experimental framework, data-logging interface, and organizational model for scaling AI-enabled education in resource-constrained settings. Significantly, it highlights potential improvements in student motivation and learning efficiency while cautioning about ethical, privacy, reliability, and cost considerations and pointing to soft-skill development as a key future direction.

Abstract

In recent years, it has been observed that the center of gravity for the volume of higher education has shifted to the Global South. However, research indicates a widening disparity in the quality of higher education between the Global South and the Global North. Although investments in higher education within the Global South have increased, the rapid surge in student numbers has resulted in a decline in public expenditure per student. For instance, the student-to-teacher ratio in the Global South is significantly higher compared to that in the Global North, which poses a substantial barrier to the implementation of creative education. In response, Telkom University in Indonesia has embarked on an experiment to enhance the quality of learning and teaching by integrating large language models (LLMs) such as ChatGPT into five of its courses-Mathematics, English, Computing, Computer Systems, and Creative Media. This article elucidates the ongoing experimental plan and explores how the integration of LLMs could contribute to addressing the challenges currently faced by higher education in the Global South.

Figures (4)

• Figure 1: A comparison of the total number of students (in millions) and public spending per student (in USD) in the Global South. While the student population continues to grow, public spending per student has not kept pace, resulting in a widening gap between the two. The data in this graph is based on the 2022 HESA report on global higher education institutions, students, and funding hesa2022world.
• Figure 2: Five courses have been selected for the pilot study to integrate LLM applications into the learning process. These courses represent a typical computer science curriculum, encompassing both theory/deductive and practice/inductive approaches. Vertically, these courses span different levels: foundational courses (Mathematics and English), specialized courses (Computing and Computer Systems), and creative courses (Creative Media).
• Figure 3: Two types of experiments: assignment-based and examination-based. The first utilizes assignments to evaluate students achievement, whereas the latter conduct examinations. Both methods are applied in the experimental and control classes.
• Figure 4: A learning assistant application is needed to facilitate the tripartite interaction between students, LLMs, and educators. This allows educators, including lecturers and teaching assistants, to monitor student-LLM interactions and intervene when necessary. The application also features a knowledge base for recording course-specific information, a question bank, and an assignment repository. Additionally, the problem solver module helps guide students through step-by-step activities required to solve complex problems.