This paper investigates the patterns and trends in higher education in Britain over the last five decades (1972–2025). Drawing on most authoritative national representative data, we examine class, gender and ethnic differences in educational attainment, focusing on access to higher education. The analysis is situated within major sociological debates on education, particularly modernisation theory and rational action theory, which respectively anticipate declining and persistent class differences in educational attainment. The findings indicate widening class differences, a reversal of gender differences, and substantial ethnic variation. This paper then considers the extent to which these findings converge with, or depart from, the existing literature. A central feature of prevailing theories of educational attainment is their emphasis on class-lined inequalities in family income; there is strong evidence that income inequality between classes in Britain has widened over the past 50 years. This growing income gap closely mirrors the class gap in higher education. Building on the most systematic research, the present analysis extends and enriches the stratification tradition in the study of higher education in Britain, both conceptually and empirically.
Students use generative artificial intelligence (GenAI) tools for academic writing, but there is still little understanding of how students can prompt ChatGPT to write literature reviews, how structured prompting frameworks affect their outputs and what the implications are of such use of GenAI in teaching practices. Prompts and literature reviews written by seven students were collected before and after the introduction of three prompting frameworks (COSTAR, POSE, and Sandwich) and analysed. Following the intervention, students’ prompting became more focused and efficient, although they did not use all the elements of the frameworks, except for Sandwich. While literature review scores improved, approaching near-full marks, they lacked critical thinking and originality. Therefore, more training and understanding of prompting frameworks and fundamental knowledge about literature reviews are still needed. Plagiarism checks demonstrate a reasonable degree of similarity, although recurrent citation mistakes were found. These findings highlight pedagogical and ethical risks associated with GenAI use, which may be mitigated by alternative approaches to evaluating literature reviews that place greater emphasis on paper selection and topic choice. Although the small sample precludes generalisation, this study provides an initial exploration of how prompting frameworks influence GenAI-assisted literature reviews and what the ethical and pedagogical implications are.
The rapid proliferation of generative artificial intelligence in higher education has precipitated a new crisis of legitimacy for Indigenous Knowledge Systems. While the utility of large language models is widely touted, this paper critically interrogates the epistemological architecture of these tools, theorising algorithmic colonialism as a curricular mechanism of epistemic erasure in post-#FeesMustFall South African universities. Designed as a conceptual paper, the article extends the boundaries of knowledge by theorising peer-reviewed literature and policy frameworks between 2020 and 2025 to understand the specific threat Generative AI poses to Indigenous Knowledge Systems within African curricula. The principal analytical findings reveal that Large Language Models rely on data extractivism and linguistic hegemony, mechanisms that systematically marginalise Indigenous Knowledge Systems in favour of Western cognitive models. Drawing on decoloniality and critical digital pedagogy, the paper argues that the uncritical integration of these tools into African university curricula risks entrenching a monoculture of the mind where machine-generated consensus displaces localised, context-specific wisdom. The analysis explores the specific mechanisms through which algorithmic bias reinforces historical inequities. Consequently, the paper proposes a robust, operational framework for AI sovereignty within African higher education grounded in data trusts, Reinforcement Learning from African Feedback (RLAF), and algorithmic literacy as counter-conduct. The paper concludes that for the Fourth Industrial Revolution to be emancipatory rather than recolonising, the curriculum must explicitly teach students to audit the colonial logic embedded within the code itself.
This exploratory qualitative study examines how students’ reasoning becomes observable and develops within Socratic dialogues with Riffbot, a large language model (LLM)-based learning chatbot designed to guide students from descriptive and clarificatory questioning toward explanatory and reflective inquiry. The empirical dataset comprised 121 dialogue transcripts generated by 93 students across seven project-based courses at a university of applied sciences. All 121 transcripts were initially reviewed to examine participation, dialogue length, and interactional characteristics. Of these, 72 dialogues containing at least 10 turns were selected for detailed coding and reflexive thematic analysis (RTA), while 49 shorter dialogues were retained as contextual material. The analysis examined student responses in relation to successive chatbot prompts, using theoretically informed reasoning categories to trace interactional patterns within complete dialogues. Coded data were visualized in a chronological matrix to support the interpretation of reasoning trajectories. Descriptive reasoning predominated, while some dialogues showed emerging explanatory and reflective reasoning. Other trajectories remained primarily descriptive, involved repeated clarification, or moved recursively between reasoning categories, often including moments of uncertainty and redirection. These findings highlight the non-linear and interactionally situated nature of learning expressed during Socratic LLM–student exchanges, without establishing independent learning gains. Methodologically, the study demonstrates how systematic coding, chronological visualization, and reflexive thematic interpretation can be combined to make thinking-in-action more visible and analytically traceable. It contributes an empirical account of student reasoning within AI-mediated dialogue and a practical, digitally supported approach to investigating learning-related processes in higher education.
The literature reviewed for this study addresses inclusive-design pedagogy, project quality, and assessment transparency, but does not treat the artifact-level correspondence between a source-located access condition and the response represented for that condition as a separate analytic object. Within this accessibility-focused proof-of-concept, barrier–response traceability is treated as a distinct artifact-level evaluation layer for examining whether a represented design response remains explicitly connected to the source-located access condition it is intended to address. A retrospective multiple-case analysis of 13 group submissions identified 135 problem–response artifact records coded through structured qualitative content analysis, with independent coding used to assess reproducibility. Most records showed direct linkage (124/135, 91.9%), but this concentration is treated as task-conditioned background because the assignment explicitly required problem–response representation. The analytically discriminating evidence comprised nine partial records and two weak/absent negative cases; four candidate incomplete-translation patterns were identified among the nine partial records. These results provide a proof-of-concept demonstration of barrier–response traceability as a domain-specific operationalization of established traceability logic, while transferability beyond the present cohort, course, task, and design medium remains to be independently validated.
