The review arrives at a moment of extraordinary turbulence for education systems worldwide. Since the explosion of generative AI tools such as ChatGPT, debates about academic integrity have dominated institutional agendas, from universities drafting new assessment policies to examination boards in West Africa catching candidates who used AI to answer exam questions in the 2023 WASSCE tests. Yet the authors argue that the fixation on cheating misses a far more structural problem: the very architecture of educational AI reflects the economic, infrastructural, and epistemic conditions of the wealthy countries that build it, and is being deployed into contexts radically unprepared to interrogate it.

Technically, the problem begins with data and infrastructure. Large language models and adaptive learning platforms are trained on corpora dominated by English and a handful of other high-resource languages, embedding the cultural assumptions of their training data into what appears to be neutral technology. Low-resource languages, including many spoken across Africa and Southeast Asia, are poorly represented, which means generative tools perform worse for learners who most need affordable educational support. The review notes that linguistic challenges in generative AI translate directly into pedagogical disadvantage, because students are often forced to learn through the linguistic and cultural frames of former colonial powers rather than their own.

Infrastructure compounds the linguistic gap. AI-powered education presumes reliable electricity, broadband connectivity, modern devices, and institutional technical capacity. The differential levels of economic development, digital infrastructure, cultural norms, and technological capacity across countries mean that the integration and impact of AI in education vary enormously between systems. In urban centers of the Global North, AI tutoring is layered onto already well-resourced classrooms. In rural communities of sub-Saharan Africa or Southeast Asia, even accessing a digital platform can be impossible, so the technology widens rather than narrows existing divides. The review characterizes this as a mechanism through which AI deepens the processes of educational inequality rather than resolving them.

The authors ground their analysis in the intellectual tradition of critical social theory, drawing on thinkers such as Herbert Marcuse, whose critique of one-dimensional society warned that advanced technology can suppress critical consciousness, and Pierre Bourdieu, whose concepts of cultural capital illuminate how dominant groups convert their advantages into seemingly meritocratic success. Viewed through this lens, educational AI is not a neutral delivery mechanism but a cultural artifact that carries the values, priorities, and worldviews of its creators. When a curriculum platform decides what counts as knowledge, which language is standard, and which forms of reasoning are rewarded, it exercises a form of power that students and teachers rarely see, let alone contest.

This invisibility is precisely what makes algorithmic systems so potent in educational settings. Black-box models, the review observes, are increasingly shaping decisions about admission, assessment, and learner support without transparent accountability. Prior research on algorithmic injustice has documented how commercial systems can exhibit accuracy disparities against marginalized groups, and the review extends that concern to classrooms: when biased systems mediate learning, they do not merely make errors, they systematically misrecognize certain categories of students. The danger, in the critical theoretical framing, is a new one-dimensionalism in which students learn to work with machines they cannot question, and educators become facilitators of opaque systems rather than agents of critical pedagogy.

The review also confronts what it identifies as colonial legacies embedded in technological innovation. Concepts such as cognitive imperialism and the cognitive empire describe how dominant knowledge systems marginalize indigenous epistemologies, and the authors argue that AI risk reproducing this pattern at unprecedented scale. Global AI governance conversations, they note, remain dominated by wealthy nations and multinational corporations, while African voices, values, ethics, and worldviews are largely absent. Referencing the African Ubuntu perspective, which emphasizes communal personhood and relational ethics, the review contends that alternative ethical frameworks are not merely cultural curiosities but substantive resources for rethinking how educational AI should be designed, governed, and evaluated.

The empirical picture assembled across the reviewed literature is sobering. Studies of AI deployments in Africa document persistent barriers of cost, connectivity, and policy capacity. Research from India highlights both the promise and the difficulty of AI in inclusive education. Analyses of government AI readiness across Latin America and the Caribbean, and of digital inequality between urban and rural communities in Southeast Asia, reveal the same structural asymmetry: AI readiness correlates strongly with existing wealth. Meanwhile, the hype surrounding AI in education and development often outruns the evidence, with discursive enthusiasm obscuring the material conditions required for meaningful implementation. The review’s synthesis suggests that AI in education currently follows the contours of global power rather than correcting them.

Yet the paper is not a rejection of educational technology. The authors explicitly acknowledge that AI offers genuine opportunities for teaching and learning, and their concern is not the technology itself but the conditions of its design and deployment. Their central recommendation is that AI developers adopt inclusive design from the very start of the application lifecycle, ensuring that educational tools are accessible, adaptable, and affordable to the economies of the Global South. That means designing for low-bandwidth environments, supporting low-resource languages, involving local educators and students in development, and pricing models that do not exclude the majority of the world’s learners. It also means building AI literacy among teachers and students so that users can critically evaluate rather than passively consume algorithmic outputs.

The review’s most pointed conclusion is political: Africa and other developing countries must not merely be sites of AI deployment but key actors in shaping the educational futures these technologies enable. Incorporating African expertise into global AI governance, the authors argue, is essential to confronting colonial legacies in technological innovation and ensuring that AI advances democratic empowerment rather than digital dependency. In an era when governments, universities, and technology firms are racing to embed AI into every layer of education, the study offers a timely corrective: the question is not only whether AI can improve learning, but who decides what it improves, whose knowledge it carries, and whose futures it shapes. The answers, the researchers insist, must include the billions of learners the technology claims to serve.

Subject of Research: The negative social impacts of artificial intelligence in education, including inequality, power relations, and cultural marginalization, analyzed through critical social theory

Article Title: A Systematic Review of the Dark Side of AI in Education: A Critical Social Theory Perspective

Article References: Adu, K. K., & Owusu-Agyeman, Y. (2026). A Systematic Review of the Dark Side of AI in Education: A Critical Social Theory Perspective. Frontiers of Digital Education, 3(2), Article 10. https://doi.org/10.1007/s44366-026-0084-0

Image Credits: AI Generated

DOI: 10.1007/s44366-026-0084-0

Keywords: artificial intelligence, education, Global South, digital inequality, critical social theory, AI governance, academic integrity, Africa, generative AI, digital dependency, algorithmic bias, inclusive design