Generative AI as a Mediational Agent: Rethinking Learning in Sociocultural Theory

Synthesis: Warschauer, Tate, and Ritchie (2026) argue that generative AI challenges a foundational distinction in sociocultural theories of learning — the separation between mediational means and social interaction. Traditional tools such as language, writing, and educational technologies mediate human activity, while learning and development arise through social participation; generative AI complicates this because it both mediates activity and generates context-sensitive, contingent contributions that shape ongoing interaction. The authors propose the concept of the mediational agent: a responsive but non-accountable system that mediates human action while contributing explanations, critiques, questions, and suggestions to learning activity — occupying a hybrid space between a tool and a social partner. From this, they derive five human-first habits of participation (primacy of human cognition, purposeful engagement, supervisory agency, epistemic vigilance, reflective self-Regulation) intended to preserve learners' judgment, agency, and responsibility in AI-mediated learning.

Why "tool" and "collaborator" both fail

The essay's starting point is a conceptual gap in how education discusses generative AI. Sociocultural theory has long distinguished mediational means (language, writing, technologies) from social interaction (participation with teachers, peers, communities). Two dominant framings of GenAI are each insufficient:

AI as a tool underestimates its interactional influence. Because GenAI generates contingent, context-sensitive responses, it does not behave like a passive instrument; it actively shapes the trajectory of the interaction — explaining, critiquing, questioning, suggesting, and revising in ways no previous educational technology did.
AI as a collaborator risks attributing intentionality, accountability, and community membership that AI systems do not possess. A collaborator is accountable to shared norms, occupies a position in a community of practice, and shares a history; an LLM has none of these — its apparent participation emerges from probabilistic language generation rather than intentional engagement.

This mirrors the wiki's ongoing debate between treating AI as a neutral tool versus as a quasi-partner in human–AI collaboration — and the paper offers a theoretically rigorous middle path.

The mediational agent concept

The proposed concept resolves the dichotomy. A mediational agent:

mediates human action like a tool (extending the learner's activity), yet
participates in interaction by generating contingent, responsive contributions — explanations, critiques, questions, and suggestions — like an interlocutor, while
remaining non-accountable — it does not possess the intentionality, social membership, or responsibility of a genuine community member.

Crucially, the paper argues this is a hybrid form of participation, not a midpoint on a continuum between tools and people. Generative AI's novelty lies not in its intelligence but in its participation: it performs functions that historically belonged to opposite sides of a conceptual distinction. Activity systems that include mediational agents are organized differently, and learners must continually decide whether to accept, reject, revise, interrogate, or redirect AI-generated contributions — judgments that become part of the learning activity itself.

The developmental paradox

Generative AI is exceptionally effective at improving immediate performance (a product: fluent prose, working code, plausible explanations) but learning depends on process — generating ideas, wrestling with uncertainty, evaluating alternatives, revising understanding. Systems designed to reduce cognitive effort therefore create a developmental paradox: the same capabilities that improve immediate performance may reduce opportunities for intellectual growth if they replace rather than support learners' own thinking. This is the paper's link to the Reducing AI Misuse concern about performance–learning gaps and to productive struggle.

Five human-first habits of participation

Because a mediational agent is responsive but non-accountable, the educational task is neither "use the tool well" nor "collaborate productively" — it is to regulate the relationship between one's own thinking and the system's contributions. The paper proposes five dispositions (not a fixed Pedagogy, and discipline-dependent in expression):

Primacy of human cognition — develop ideas, interpretations, or problem framings before turning to AI, creating the intellectual foundation needed to evaluate subsequent AI contributions rather than simply adopt them.
Purposeful engagement — direct AI use by human intentions, using it strategically to extend, challenge, or refine one's thinking rather than to replace it.
Supervisory agency — because the system generates contributions without responsibility for consequences, learners remain accountable for every decision about whether and how those contributions enter their work.
Epistemic vigilance — fluent language and confident explanations must never be confused with truth or disciplinary validity; learners continually interrogate AI claims, corroborate evidence, and evaluate reasoning by disciplinary standards.
Reflective self-Regulation — because AI can obscure the boundary between human and machine contributions, learners consciously monitor their own understanding, identify what they genuinely learned, and evaluate how AI shaped their thinking.

These habits are not generic digital competencies; supervisory agency in historical inquiry differs from software engineering, and epistemic vigilance takes different forms in scientific modeling than in literary interpretation. The common principle is a commitment to preserving human judgment within activities that increasingly include mediational agents.

What changes for the learning sciences

The concept redirects research attention from "does AI improve outcomes or productivity?" toward how patterns of participation change when responsive but non-accountable systems enter everyday learning. It reframes longstanding questions: Scaffolding (mediational agents generate support dynamically without the social relationships that gave Scaffolding its developmental significance); distributed cognition (how should cognitive responsibility be distributed when artifacts contribute to reasoning itself?); and new empirical foci (how learners negotiate authority, responsibility, and judgment during interaction). Whether the term "mediational agent" endures matters less than whether the idea lets the field notice phenomena existing categories leave unexplained.

Implications

The mediational-agent framing offers a principled theoretical vocabulary for discussions that otherwise oscillate between techno-optimism (AI as tool) and alarmism (AI as deceptive collaborator). It is a significant conceptual contribution to the Learning Theories and sociocultural literature in AI education, aligning with posthumanist critiques while ultimately prioritizing human meaning-making and responsibility. For educators, it argues that the object of design should be the habits of participation learners develop, not just the tool's capabilities — a framing that complements the "scaffold, do not substitute" principle and the wiki's broader concern with preserving human agency and judgment in AI-mediated learning. The concept also enriches Framing AI Use For Students by grounding why a human-first framing of AI matters theoretically, and connects to AI Sycophancy via the risk of non-accountable persuasive outputs.

Connected Concepts

Connected Articles

Learning With Machines Toward A Theory Of Epistemic Co Agency — Learning with Machines: Epistemic Co-Agency
AI Cognitive Partner Co Regulation Learning — AI as Cognitive Partner and Co-Regulation
Voicu AI Interpretive Cognition Ssh 2026 — AI-Mediated Learning and Interpretive Cognition
Chatgpt Critical Creative Thinking Review — ChatGPT and Critical/Creative Thinking
Posthumanist AI Literacy 2025 — Posthumanist AI Literacy
Substitution To Scaffolding AI Harm Cycle 2026 — From Substitution to Scaffolding

Citation

Warschauer, M., Tate, T., & Ritchie, D. (2026). Generative AI as a Mediational Agent: Rethinking Learning in Sociocultural Theory. EdArXiv preprint.