DYNAMICS OF SUBJECTIVITY IN THE EVOLUTION OF INTERACTION BETWEEN EDUCATOR, LEARNER AND ARTIFICIAL INTELLIGENCE

Abstract

Artificial intelligence (AI) is increasingly embedded in learning ecologies, reshaping how subjectivity – understood as the capacity of each participant to act as an agent in learning – is distributed among learners, educators, and AI. This article advances a tri-subjective perspective and proposes a four-state model of AI participation: S0 (instrument), S1 (assisting facilitator), S2 (active co-agent), and S3 (autonomous mediator). Across these states, we operationalize five components of subjectivity – motivation, activity, reflection, adaptability, and interactivity – and describe conditions that trigger transitions in the distribution of agency. Methodologically, the work combines conceptual synthesis with two applied cases. At the micro-level, we analyse a seminar scenario in which an AI assistant surfaces parallels between two independent learner analyses, thereby mediating dialogic exchange. At the macro-level, drawing on third-party case materials, we examine how an AI “facilitator” sustains group memory and consensus across iterative cycles over time. Results indicate that movement from S0 to S2 tends to expand learner agency when AI feedback is transparent and bounded, educator mediation remains active, and contextual memory is preserved. By contrast, S3 offers powerful personalization but risks over-optimization, educator disengagement, and learner passivity if human framing is not maintained. We distil design guardrails – bounded autonomy, provenance cues, reflective prompts, and orchestration protocols – that help support human subjectivity while leveraging AI’s adaptive capabilities. The contribution is threefold: (I) a state-based vocabulary for analysing AI’s pedagogical roles; (II) a separation between conditional AI “subjectivity” and human agency; and (III) actionable implications for curriculum design and facilitation. Limitations include reliance on conceptual modelling and case-based illustration rather than controlled trials. Future work will translate the model into evaluation rubrics and conduct empirical studies in diverse instructional settings.