Who Makes the Meaning? Interpretive Authority in AI-Mediated Education

Synthesis: Porath (2026) argues that the central question of AI in education is not adoption or prohibition but who holds interpretive authority — who defines problems, evaluates responses, and determines when understanding is achieved. Drawing on dialogic pedagogy, self-regulated learning research, and Bronfenbrenner's ecological systems theory, the article introduces two conceptual tools: the Interpretive Authority Matrix, mapping four non-hierarchical configurations (human-dominant, AI-assistive, AI-directive, dialogic partnership), and the Ecological Authority Integration Model (EAIM), which explains how five ecological layers — metacognition, classroom interaction, institutional policy, community structures, and sociocultural narratives — shape authority shifts. The frameworks reposition AI integration as an epistemic design challenge: classrooms can move across configurations within a single task, AI-assistive use is not automatically shallow, and AI-directive use — however sophisticated — can suppress productive struggle and metacognitive development. As a conceptual article it offers diagnostic lenses, not empirical evidence.

Summary

When generative systems participate in teaching and learning, the distribution of interpretive authority — whose interpretation governs the learning episode — becomes a design variable rather than a given. This article asks where epistemic agency resides in AI-mediated learning and builds two frameworks for analyzing and designing that distribution.

The Interpretive Authority Matrix operationalizes dialogic theories of authority (Bakhtin, 1981; Matusov, 2009) as a diagnostic that maps each instructional episode across four configurations defined by the relationship between human interpretive agency and AI directive influence: human-dominant (the learner defines problems and judges understanding), AI-assistive (AI supports while the human remains the primary interpreter), AI-directive (AI-generated responses shape interpretation; prompts begin to direct the argument), and dialogic partnership (meaning emerges in exchange between human and system). Critically, the Matrix does not rank the configurations: classrooms move across them within a unit or even a single task, AI-assistive use is not automatically shallow, and human-dominant instruction is not automatically agentic — it identifies patterns to be examined, not a progression to be climbed.

The Ecological Authority Integration Model grounds those shifts in Bronfenbrenner's (1979) systems theory across five layers: (1) the individual metacognitive layer — AI use supports or bypasses planning, monitoring, and evaluation; (2) the interactional microsystem — each prompt-response exchange as a dialogic encounter; (3) the classroom/exosystem — participation structures, assessment design (process-over-product emphasis strengthens evaluative control), and policies (transparency and disclosure-oriented policies reinforce human-centered authority; detection-focused policies risk normalizing substitution); (4) the mesosystem — teacher education, professional culture, institutional leadership; and (5) the sociocultural macrosystem — cultural narratives about automation, productivity, intelligence, and authorship. Ethics functions as a cross-cutting design principle across all layers, and the article grounds its normative commitments in the ISTE Standards.

Key Findings

• The Matrix reframes the AI debate as an authority question. Rather than whether AI is allowed or prohibited, the framework asks who leads the intellectual work in each episode — making authority shifts visible and examinable instead of treating AI use as a binary.
• Four configurations, no hierarchy. Human-dominant, AI-assistive, AI-directive, and dialogic partnership coexist and interleave; the article explicitly warns against reading AI-assistive as shallow or human-dominant as agentic, and against treating the Matrix as a ladder.
• AI-directive use carries specific epistemic costs. When AI significantly shapes interpretation, it can reduce productive struggle, narrow interpretive diversity, and limit development of metacognitive regulation — capacities associated with learning and growth — even when the use appears sophisticated.
• Policy design is an ecological lever. Transparency-, disclosure-, and reflection-oriented policies reinforce human-centered authority, while detection- or productivity-focused policies risk normalizing substitution; unequal access further shapes who gets to exercise interpretive authority at all.
• The design goal is symbiotic partnership, not restriction. The aim is to strengthen human epistemic agency through partnerships that expand reflection, metacognition, and knowledge construction rather than replace human judgment — with implications drawn for instruction, formative assessment, teacher education, leadership, and ethical integration.
• AI-assisted authorship disclosed. The author states that portions of the manuscript were developed through dialogic engagement with ChatGPT (GPT-5.2) and Perplexity as "epistemic partners," with all theoretical framing, interpretation, and final language decisions made by the human author — modeling the disclosure the article's own policy layer advocates.

What this means for practice

• Instructors. Use the Matrix as a lesson-planning diagnostic: for each activity, ask which configuration it invites and whether that distribution is intentional — a task can slide from AI-assistive to AI-directive mid-assignment without anyone deciding to move it, quietly removing the productive struggle that develops metacognitive regulation.
• Formative-assessment designers. Emphasize process over product — making reasoning and revision visible strengthens learners' evaluative control; product-only assessment rewards AI-directive configurations by default.
• Teacher educators. Build the Matrix into preservice training so new teachers enter the profession with a vocabulary for diagnosing authority shifts rather than binary allow/prohibit instincts.
• Administrators. Audit policy through the ecological layers: detection-first policies normalize substitution, while transparency-and-reflection policies keep interpretive authority human-centered; pair any AI policy with access equity, since authority is unevenly distributed when tools are.
• Leaders. Treat macrosystem narratives (automation, productivity, authorship) as design constraints: institutional culture shapes which configurations feel natural before any tool is chosen.

Limitations

• Purely conceptual: the Matrix and EAIM are theoretical constructs with no empirical validation, classroom data, or measured effects on learning in this article.
• The frameworks are descriptive-diagnostic, not predictive — they offer lenses for examining authority configurations, not evidence about which configurations produce better outcomes in which contexts.
• The Matrix's four configurations may not exhaust real classroom patterns; the article offers them as a starting vocabulary rather than a complete taxonomy.
• Normative commitments rest on the ISTE Standards and cited learning-science literature (dialogic pedagogy, self-regulated learning, formative assessment), whose applicability to AI-mediated contexts is argued rather than tested.
• The author's own use of AI tools in drafting the manuscript is disclosed, but the article does not examine how that partnership maps onto its own framework — the disclosure is procedural rather than analytic.

Citation

Porath, S. (2026). Who makes the meaning? Interpretive authority in AI-mediated education. Educational Considerations, 51(3). https://doi.org/10.63315/20915