Beyond Automation: AI as a Pedagogical Mediator in Collaborative Learning

Synthesis: Niari (2026) advances a conceptual framework that reconceptualizes AI as a pedagogical mediator in collaborative learning — not a tool, tutor, peer, or automated orchestrator. Drawing on socio-cultural theory, distributed cognition, connectivist perspectives, and sociomaterial/posthuman theory, the paper argues that AIED's dominant instrumental and individualistic framings are theoretically misaligned with collaboration understood as a socially mediated, co-regulated process. The framework distinguishes three overlapping layers of AI mediation — interactional, epistemic, and regulatory — and contends that agency, authority, and responsibility are dynamically redistributed across human and non-human actors without displacing learner or teacher agency.

Key Findings

• From automation to mediation. The paper critiques the instrumental orientation of AIED (personalization, automation, instructional efficiency) that privileges individual outcomes over social and relational processes, arguing that generative and interactive AI participates directly in educational discourse and therefore cannot be treated as a neutral tool.
• The pedagogical mediator construct. Mediation is defined as an agent (human or artificial) that actively shapes learning by structuring interaction, Scaffolding sense-making, and supporting AI Regulation in Education; it is distinct from orchestration (coordination/management) and automation.
• Three layers of AI mediation. (1) Interactional mediation shapes the organization and flow of collaborative discourse (prompting participation, foregrounding underrepresented contributions, highlighting convergence/divergence); (2) epistemic mediation influences the construction, evaluation, and circulation of knowledge — generating explanations, modeling reasoning, summarizing ideas — without adjudicating correctness; (3) regulatory mediation supports planning, monitoring, and evaluation of joint activity, contributing to socially shared regulation without centralizing control.
• Redistribution, not displacement, of agency. AI mediation entails a dynamic renegotiation of agency and epistemic authority across sociotechnical systems rather than a zero-sum transfer; AI participates asymmetrically and does not claim epistemic parity with learners.
• Boundary work against slippage. Unlike tutors (individualized instruction), peers/teammates (epistemic parity), coaches (individual optimization), or orchestrators (centralized control), mediators emphasize participation, relationality, and shared responsibility, keeping AI subordinate to teacher professional judgment.
• Ethics and AI Governance as constitutive. Equity/epistemic inclusion, transparency and pedagogical interpretability, teacher professional agency vs. technocratic governance, and responsible data stewardship are framed as integral to design — calling for "pedagogical governance" rather than regulatory compliance alone.

Conceptual Contribution

The paper positions AI as an active participant in the orchestration of interaction, epistemic sense-making, and shared regulation of collaborative activity, informed by socio-cultural theory, distributed cognition, connectivism, and sociomaterial/posthuman accounts of distributed agency. It explicitly critiques teammate/peer framings (e.g., AI speakers in Jigsaw groups, Large Language Models (LLMs) writing partners) for obscuring asymmetries of responsibility, accountability, and ethical agency. Within a mediation framing, AI structures the conditions of collaboration rather than substituting for human interaction, preserving the dialogical and co-regulated nature of collaborative learning theory and the epistemic sovereignty of learners and teachers.

What this means for practice

• Designers. Build AI as a mediator of collaborative discourse rather than an orchestrator of it: prompt participation, foreground underrepresented contributions, and flag convergence and divergence instead of managing turn-taking centrally.
• Designers. Let AI generate explanations, model reasoning, and summarize ideas without adjudicating correctness, and make that pedagogical reasoning visible so teachers can interpret and adjust it.
• Designers. Support planning, monitoring, and evaluation of joint activity without centralizing control — keep shared regulation from collapsing into dashboards, alerts, and nudges that substitute algorithmic authority for student co-regulation.
• Designers. Treat equity, transparency, and responsible data stewardship as design requirements rather than compliance afterthoughts — scaffold equitable participation in the discourse the AI mediates and disclose what learner data the system collects.
• Researchers. Study mediation as redistribution rather than delegation: measure where Learner Agency and epistemic authority move among learners, teachers, and AI, and test empirically whether teacher judgment and learner epistemic sovereignty survive the shift across the interactional, epistemic, and regulatory layers.

Limitations

• This is a conceptual synthesis with no sample, intervention, or comparison group, so the three mediation layers — interactional, epistemic, and regulatory — remain theoretical constructs awaiting operationalization and measurement.
• Its critique of teammate and peer framings leans on one cited design-based case (Lee et al., 2025, embedding AI speakers in Jigsaw groups) rather than a systematic review of outcomes, and that study itself reported student ambivalence about accountability, epistemic parity, and trust.
• The boundary distinctions between mediator and tutor, peer, coach, or orchestrator are argued conceptually; nothing in the paper shows that these roles are separable in deployed systems or that teachers and students can recognize which one they are working with.
• The framework's theoretical bases (sociocultural theory, distributed cognition, connectivism, and sociomaterial/posthuman accounts) differ in their commitments about where agency resides, so the central claim that AI redistributes rather than displaces authority remains contested rather than demonstrated.

Citation

Niari, M. (2026). Beyond Automation: AI as a Pedagogical Mediator in Collaborative Learning. Umanistica Digitale, (24).