AI as a Pedagogical Actor in Multilingual Contexts: A Critical Review of Benefits, Epistemic Threats, and Governance Gaps in Tanzanian Education

Abstract

The rapid integration of Artificial Intelligence (AI) into global education systems presents a paradox of promise and peril, particularly for multilingual societies in the Global South. While AI offers scalable solutions for personalised learning, its deep-seated implications for pedagogical authority, cognitive development, and linguistic diversity remain critically underexplored. This paper addresses this gap by conducting a critical, theory-informed literature review that synthesises scholarly evidence on AI’s pedagogical benefits, structural challenges, and transformative threats, with a specific focus on language education in Tanzania. Drawing on a semi-systematic search of global and Tanzanian scholarly and policy literature published between 2019 and 2025, 50 key sources were thematically analysed through an integrative framework combining critical pedagogy, sociocultural theory, and technology mediation perspectives. The results confirm that AI can enhance personalised feedback and access to learning resources in resource-constrained settings. However, the review critically extends existing discourse by conceptualising AI not merely as a tool, but as a socio-technical and epistemic actor. The analysis reveals significant pedagogical threats, including: (a) cognitive deskilling from learner over-reliance on generative AI; (b) a shift in epistemic authority from educators to opaque algorithmic systems, potentially deprofessionalising teachers; and (c) linguistic homogenisation, where AI models trained on dominant global languages marginalise Kiswahili and indigenous languages, reinforcing epistemic inequalities. The discussion interprets these findings as manifestations of a nascent "algorithmic pedagogy" that is poorly aligned with local sociocultural realities. The paper's primary contribution is a context-sensitive framework for understanding AI's multi-dimensional impact, which foregrounds issues of epistemic justice and linguistic inclusion often absent from mainstream adoption debates. It concludes that realising AI’s potential in contexts like Tanzania requires a paradigm shift from technology-centric adoption to pedagogically-driven integration. Key recommendations include the urgent development of national AI governance frameworks that mandate linguistic inclusivity, investment in localised, low-resource language AI models, and the redesign of curricula to cultivate critical AI literacy among both students and educators, ensuring technology serves, rather than subverts, sustainable and equitable educational development.