Back...

Smart talk: Reflections on utilising large language models (LLMs) to advance inclusive assessment practices

Warwick Von Hagen

Learning Technology Unit, STADIO Higher Education, KwaZulu-Natal, South Africa

ORCID: https://orcid.org/0009-0001-3362-1853

Jolanda D. Morkel

Instructional Design Unit, STADIO Higher Education, Western Cape, South Africa

ORCID: https://orcid.org/0000-0001-8807-2700

Qwanita C. Burger

School of Education, STADIO Higher Education, Gauteng, South Africa

ORCID: https://orcid.org/0009-0007-4441-8440

Abstract

Despite the increasing employment of large language models (LLMs) in higher education assessment, their contribution to inclusion remains under-researched, particularly in the Global South. This study investigates how academics in a large South African private higher education institution, where the adoption of Generative Artificial Intelligence (GAI) occurred early, utilise LLMs for inclusive assessment practices. Using collaborative autoethnography, the authors documented and discussed their engagement with ChatGPT-4o and other LLMs while designing assessment tasks, creating rubrics and formulating feedback. Transcripts of three recorded conversations were analysed inductively by applying Braun and Clarke’s thematic procedure, after which they were interpreted using Batchelor’s typology of voices as being practical, epistemological and ontological. Nine interlinked themes emerged - three for each of the voices. Practically, LLMs enabled rapid iteration, creative artefact generation and closer alignment between assessment briefs, rubrics and feedback. Epistemologically, this entailed dialogic prompting to achieve reflection and triangulation, with the model acting as a Socratic peer reviewer. Ontologically, persona-based prompts facilitated co-creation with simulated students, supported linguistic differentiation and cultivated student agency via self-assessment tasks. While participants remained critical of algorithmic bias, over-reliance and the risk of “cookie-cutter” assessment tasks, they concluded that LLMs, when used reflexively and responsibly, can promote inclusion by extending Universal Design for Learning (UDL) principles.

Keywords: Higher education, generative AI, inclusive assessment, large language models (LLMs), Universal Design for Learning (UDL)

How to cite this article: Von Hagen, W., Morkel, J.D., & Burger, Q.C. (2025). Smart talk: Reflections on utilising large language models (LLMs) to advance inclusive assessment practices. African Perspectives of Research in Teaching and Learning, 9(6), 836-850.

DOI: https://doi.org/10.70875/v9i6article13