A Metacognitive-Cultural Model for AI-mediated intercultural learning

The integration of artificial intelligence into higher linguistic education raises questions about how multicultural competence develops when students interact with AI systems that simultaneously represent multiple cultural perspectives and reproduce Western cultural bias. This study examines how such interaction affects the development of multicultural competence among language faculty students. The research proposes and empirically tests a metacognitive-cultural model (MCM) describing this process through four links: stereotype deconstruction, perspectivisation, cultural modelling, and reflexive integration. A mixed 3×3×3 factorial design with repeated measures was implemented with 180 first- to third-year students from three Russian linguistic universities. Participants were assigned to experimental (AI with metacognitive scaffolding), comparative (AI only), and control (traditional instruction) groups. During an eight-week intervention students interacted with a GPT-4-based chatbot. Data were collected using adapted Byram questionnaires, professional intercultural cases, log-file coding, stimulated recall interviews, and implicit association tests. Students in the experimental group demonstrated greater gains in multicultural competence (+12.6 points) than those in the AI-only (+5.9) and control (+3.1) groups (Cohen’s d = 0.92). The largest between-group difference occurred at the reflexive integration link (φ = 0.41). Mediation analysis showed that 43.6% of the intervention effect was explained through activation of MCM links. Students receiving metacognitive scaffolding also identified AI cultural bias more frequently (69.8% vs. 30.4%). The results suggest that AI interaction combined with metacognitive reflection can support critical cultural awareness in language education.