Generative AI as a Pedagogical-Cognitive Integration Agent: A Theoretical Model for Interdisciplinary, Multidisciplinary and Transdisciplinary Learning

Integrative learning requires learners and teachers to coordinate disciplinary concepts, methods and standards of evidence, yet such coordination is cognitively demanding and institutionally difficult. This conceptual article develops a model of generative artificial intelligence, particularly large language models, as a pedagogical-cognitive integration agent. The model was constructed through a structured, concept-driven synthesis of three literature domains: integrative learning, human-AI collaboration in education and generative AI. Concepts were compared according to the integrative demand addressed, the locus of activity, the division of human and computational labor, and the conditions required for epistemically defensible use. The resulting model comprises three interdependent axes: cognitive mediation, pedagogical co-orchestration, and systemic design alignment. Unlike a taxonomy of AI functions, the model specifies a recurrent process in which institutional conditions shape pedagogical design, teacher and learner interaction activates cognitive integration, and human evaluation feeds back into subsequent design. Generative AI contributes candidate connections, translations and syntheses, while teachers and learners retain responsibility for disciplinary grounding, verification, interpretation and judgment. Six revised vignettes illustrate both productive uses and predictable tensions, including misleading metaphors, superficial synthesis, bias and overreliance. The article defines the model's assumptions and boundaries and advances six propositions for empirical testing. Its contribution is therefore conditional: generative AI may support integrative learning when embedded in human-led cycles of inquiry, validation and revision, but fluent output alone does not constitute integration.