Research Article
When social interaction becomes an algorithm: AI-mediated education, intercultural competence, and the risks to democratic schooling
Synthesis: Pliogou and Tromara (2026) argue that generative AI and adaptive learning platforms do not simply add a tool to the classroom but change the relational conditions under which learners meet knowledge and each other. Drawing on Seyfert's notion of algorithmic sociality, they contend that prediction, personalization, filtering and automated feedback can reduce the ambiguity, unpredictability and productive tension that social-emotional learning and intercultural competence depend on. Their four-part argument is that AI-mediated education is a form of algorithmic mediation rather than a neutral enhancement, that it can narrow the range of perspectives and disagreement learners encounter, that individualized optimization can undercut collective meaning-making, and that these systems belong inside democratically governed education rather than in place of human relationality. The paper is a critical-conceptual synthesis, not a systematic review, and presents no new data.
Key Findings
- Algorithmic mediation is the right unit of analysis. The claim is not that AI is harmful, but that prediction, filtering and automated feedback reshape the conditions of educational encounter.
- SEL and intercultural competence share a relational foundation. Both develop through encounters with difference, negotiation of meaning and the capacity to stay open to perspectives that cannot be predetermined.
- Fluency and personalization can remove productive friction. Systems that smooth interaction may also remove the ambiguity and disagreement through which perspective-taking and interpretive flexibility develop.
- Individualisation can weaken collective meaning-making. Optimizing interaction for each learner risks constraining the shared dialogue that democratic schooling treats as constitutive of learning.
- Not all educational AI is equivalent. The authors exclude translation and accessibility tools from their argument, because those can support participation without restructuring educational relationships.
- The proposed position is supplementary, not substitutive. Generative AI should sit inside democratically governed frameworks with collective oversight, ethical accountability and participatory design.
The argument in four moves
The first move reframes adoption as mediation. Relational accounts of education treat meaning as co-constructed through dialogue, clarification, disagreement and shared inquiry; generative systems insert a layer that can shape which voices are available, which interpretations are surfaced and how quickly a question is answered. The second move follows from that: if systems privilege dominant patterns of knowledge production, some epistemic positions become more salient than others, and marginal perspectives become less visible to learners who have fewer resources for finding them.
The third move is about optimization. Adaptive platforms and personalized feedback are usually justified by efficiency, but the authors argue that the same mechanisms can individualise learning to the point where the collective negotiation of meaning — the part democratic education treats as formative rather than incidental — is thinned out. The fourth move is normative: because these are questions about the purposes of education, they belong to democratic governance, not only to procurement. The paper aligns itself with calls for participatory design, collective oversight and ethical accountability in educational AI.
What the paper does and does not claim
The authors are careful about scope in a way that opinion pieces often are not. They limit the argument to generative AI and adaptive learning environments, explicitly excluding machine translation and accessibility tools from the claim that AI restructures educational relationships. They also describe the method honestly: a critical-conceptual synthesis drawing selectively on literature about algorithmic mediation, SEL, intercultural competence, relational pedagogy and democratic education, rather than a systematic review, and one that does not attempt to establish prevalence or overall effect.
That honesty is also the paper's main limit. The mechanisms it identifies are plausible and theoretically grounded, but no study here tests whether reduced ambiguity actually impairs SEL or intercultural development, and the empirical work cited concerns AI-supported communication and language learning rather than measured competence outcomes. Read as a set of design and governance questions rather than as findings, it is a useful counterweight to adoption arguments that treat personalization as an unqualified good.
What this means for practice
- Instructors. Preserve tasks where students must disagree, negotiate and interpret an ambiguous situation; those are the conditions the paper argues AI-mediated personalization erodes.
- Curriculum designers. Ask what a personalization feature removes as well as what it adds, particularly where the outcome is perspective-taking or intercultural competence rather than content coverage.
- Administrators. Treat decisions about adaptive platforms as governance questions with a democratic justification, not only as efficiency or procurement decisions.
- Educational technology developers. Design for plural perspectives and visible disagreement rather than for smooth answers; the risk described here is a design property, not an inevitability.
Limitations
- No new data. The paper is a critical-conceptual synthesis and does not test any of its proposed mechanisms empirically.
- Literature used selectively. The authors state that sources were chosen in relation to the conceptual argument rather than through systematic review criteria.
- Mechanism is inferred. The claim that reduced ambiguity impairs SEL and intercultural development is a theoretical inference, not a measured effect.
- Scope restricted by design. The argument covers generative AI and adaptive learning environments only, and the authors explicitly do not generalize it to translation or accessibility tools.
Citation
Pliogou, V., & Tromara, S. (2026). When social interaction becomes an algorithm: AI-mediated education, intercultural competence, and the risks to democratic schooling. Frontiers in Psychology, 17, 1906935.