Dissertation proposal defense · August 13, 2026

Pedagogical Friction in the Age of Generative AI
and Tertiary Algorithmicity

A qualitative-dominant convergent mixed methods study

Micah J. Miner

National Louis University · Ed.D. in Curriculum, Advocacy, and Policy

Chair: Dr. Terri Jo Smith · Committee: Dr. Ruben Puentedura and Dr. Blanca Gamez-Djokic

The opening tension

A finished artifact no longer reliably shows the thinking behind it.

Two submitted artifacts

what an evaluator can see

The route each one took

drafted, reconsidered, revised

produced in one pass

This is an evidence problem before it is a misconduct problem.

The institutional moment

Adoption is running ahead of institutional clarity.

68.9%

of public K–12 teachers reported AI tool adoption in the fall 2025 RAND/Gallup data.

12.5%

reported a school AI policy that was both present and clear.

Author-weighted descriptive estimates from RAND/Gallup public-use data — contextual indicators, not proposed-study findings. Weighting and noncomparability cautions: Miner (2026a), doi.org/10.5281/zenodo.21152544

The media-ecological claim

The shift is from algorithmic selection to algorithmic production.

This is not a claim of inevitability. If AI’s effects were determined, pedagogical friction would be pointless.

Stages 04 and 05 are the researcher’s extensions of Ong (1982/2002) and Stalder (2016/2018). Distinct from existing tertiary-orality scholarship: that work asks what happens to oral expression in digital environments; this asks whether human expression is required at all.

Talk track

Bridge from the previous slide. Adoption and policy describe how fast schools are moving. They do not explain what is actually changing underneath. That requires naming the media condition.

Ong explains how changes in dominant media reshape consciousness and culture. Algorithmic secondary orality names the social-media condition in which human content is selected and amplified algorithmically. Tertiary algorithmicity names the generative condition in which production itself also becomes algorithmic at scale.

Do not present this sequence as inevitable technological progress. It is a periodizing analytic frame with porous boundaries and hybrid practices.

If asked, is this just tertiary orality? A body of work across linguistics, media ecology, library science, and sociology has proposed tertiary orality for digitally mediated expression: Mayer (2009), Turner and Allen (2013), Soffer (2020), Heyd (2021), Ryu (2023), and Cordón-García and Muñoz-Rico (2024). That literature already recognizes interactivity, voice interfaces, and human-machine discourse. What it does not do is make the automated generation of extended symbolic content the organizing principle of its periodization, or develop the consequences for educational evidence of learning, authorship, and accountability. The distinction is not speech versus writing. It is between environments where algorithms mainly circulate human expression and environments where they can also generate the artifact.

If asked about determinism. The framework is a response framework. Naming friction as something educators design asserts agency within a powerful default rather than claiming the default is destiny (Feenberg, 2002; Hutchby, 2001).

[Sources] Proposal, Chapter Two, Media Ecology and Ong's Developmental Account; Guarding Against Technological Determinism; Prior Extensions and the Limits of Tertiary Orality; Algorithmic Secondary Orality and the Transitional Stage; Tertiary Algorithmicity and the Generative Condition; Ong (1982/2002); Stalder (2016/2018).

Noetic displacement and the bypass of cognitive labor

AI can route around the formation the assignment was meant to produce.

Task Encounter a problem

Learner process Interpret · struggle · articulate · revise

Artifact Submit a fluent product

Generative AI

Noetic displacement AI substitutes for the cognitive work that builds understanding
Rhetorical saturation Generated language weakens exchange and obscures the source of a claim
Existential abstraction The artifact separates from anyone willing to own and defend it

Unproductive success: correct-looking performance without the struggle that performance normally indicates.

Talk track

This is not the claim that every use of AI eliminates learning. It names a possibility that educators increasingly need to diagnose. A learner may complete the task and receive credit while the intended noetic, rhetorical, or existential formation happens elsewhere, or not at all.

These three pressures are the hinge of the argument. Each names a condition of Ong's account under strain, and each has a matching form of friction on the next slide. Name them here so the framework reads as a response rather than an assertion.

Unproductive success is adapted from Kapur's fourfold distinction among productive success, productive failure, unproductive failure, and unproductive success. It is the diagnostic term the study asks participants to recognize in practice.

[Sources] Proposal, Chapter Two, Noetic Displacement and the Bypass of Cognitive Labor; Rhetorical Saturation and the Erosion of Genuine Exchange; Existential Abstraction and the Severing of Claim From Claimant; Definitions of Key Terms; Kapur (2016); Miner (2026b).

The Pedagogical Friction Framework

Three learner-facing forms depend on one institutional condition.

System Infrastructural friction

Policy, leadership, professional learning, assessment, access, time, and tools make the learner-facing forms possible or impossible to sustain.

Each form answers one pressure from the previous slide. The framework is the study’s lens and its object: its usefulness is an empirical question, not a commitment the data must confirm.

Talk track

Noetic, rhetorical, and existential friction are analytically separable but operationally entangled. Infrastructural friction is different in kind. It is not simply a fourth learner experience. It names the institutional conditions that enable or constrain the other three. That relationship, that institutional conditions enable or constrain classroom friction, is the central claim the study examines rather than assumes.

If asked what these look like in a classroom. Each node carries one instance on click. Give them as recognizable cases, not as findings: this is what the form looks like when it is present, which is precisely what the study goes looking for. Noetic — students build a claim from two conflicting sources before any tool opens, and AI is then aimed at that claim to stress it rather than to supply one. Rhetorical — a written argument must be defended aloud to peers who can question it, so revision is triggered by exchange rather than by a rubric checklist. Existential — students mark which passages they will defend as their own reasoning and which came from a tool, then answer for the marked ones. Note that each names a move that preserves the work, not an instance of the pressure; a story about a student who had a model write the thesis is an example of noetic displacement, not of noetic friction.

Infrastructural, if asked. Deliberately not a fourth classroom move, and worth saying so aloud. It is the condition question: whether the grading calendar, class size, and the platform's default settings leave room for a two-day claim-building sequence, or make the fastest compliant path the only survivable one. The asymmetry is the point — the first three are things a teacher can do, and this one is what determines whether they can keep doing them.

If asked about cognitive augmentation. The strongest objection is that AI extends cognition as prior tools did, and that resisting it repeats the error of condemning writing or the calculator (Mollick, 2024; Riva, 2025). The framework concedes that tools extend cognition and that this has happened before. Its reply is narrower: the question is not whether a tool extends adult expert cognition, but whether routine use during formation prevents novices from building the schemas expertise requires. Calculator research complicates the analogy in the objection's favor, since integrated calculator use has generally not hindered mathematical skill development (Ellington, 2003). Concede that, then restate the narrower question: does the tool extend an already developing schema, or substitute for the practice that builds it?

Committee reference. The framework, instruments, and secondary-data materials are laid out on the proposal's digital companion site: minerclass.github.io/pedagogical-friction/

[Sources] Proposal, Chapter One, Conceptual Framework; Chapter Two, The Pedagogical Friction Framework; Counterarguments and the Limits of the Framework. Original framework by Miner.

The equity boundary on the framework

The question is not whether school should be difficult.

Preserve

Productive friction

Effort that builds understanding, judgment, authorship, and capacity.

Situated professional judgment

Remove

Exclusionary friction

Barriers that block access without serving the intended learning.

The same AI use may change both kinds of friction at once. No policy resolves this in advance, which is why the study asks how educators actually draw the line.

Talk track

This slide is the constraint on everything before it. A defense of friction becomes a defense of inequity if the study does not distinguish learning-producing difficulty from needless barriers. Every dimension on the previous slide is subject to this boundary.

The distinction is situated. The multilingual-learner paradox is the clearest case: a student writing in an additional language who uses AI translation may bypass an exclusionary barrier while also complicating the relationship among language choice, voice, reasoning, and authorship. No mechanical resolution is possible. The analytic question is what work the tool removed and whether that work served the learning goal.

Dolmage (2017) argues that what higher education calls rigor has often functioned as gatekeeping. Annamma et al. (2013) show how race and disability intersect so that what counts as productive struggle for one student is documented as deficit for another. This is why the framework cannot resolve the boundary in advance, and why the dissertation investigates how educators draw it in practice.

[Sources] Proposal, Chapter One, Statement of the Problem and Conceptual Framework; Chapter Two, Productive and Exclusionary Friction and the Equity Boundary; Annamma et al. (2013); Dolmage (2017); Kasneci et al. (2023).

Research questions

The study follows meaning from experience to institutional action.

Participant meaning leads. Framework concepts sensitize attention without prescribing findings. Select a question to trace the evidence that answers it.

Exact research questions

RQ1. How do classroom-facing educators, building-level administrators, district or system-level leaders, and adult university students make sense of the ways generative AI changes the effort, dialogue, authorship, judgment, and participation involved in teaching and learning?

RQ2. How do educators and school-system leaders make sense of and respond to the policy, professional-learning, assessment, access, and governance conditions surrounding generative AI?

RQ3. What language, assessment expectations, instructional practices, professional supports, and leadership approaches do educators and school-system leaders identify as important for preserving forms of friction that support learning while reducing unnecessary barriers?

[Sources] Proposal, Chapter One, Research Questions.

Research design

Qualitative interpretation leads. Quantitative evidence adds context.

QUAL

Situated meaning

Interviews, explanations, documents, and open responses provide the primary evidence.

quan

Supporting pattern

Closed-ended survey items and national datasets describe context without overriding human accounts.

Concurrent collection Separate analysis Interpretive integration

Why convergent, not sequential: both kinds of evidence are needed about the same phenomenon in the same policy moment, and AI guidance is changing fast. The design seeks understanding, not causal effects or statistical generalization.

Participants and sampling

Four vantage points reveal where friction is experienced and governed.

6–8

Classroom-facing educators

Tasks, student work, feedback, and assessment

3–4

Building administrators

Instructional coherence, supervision, and family expectations

3–4

District or system leaders

Policy, infrastructure, governance, and support

≈4

Adult university students

Retrospective learner perspective, contributing directly to RQ1 and contextualizing RQ2 and RQ3

Survey: ≈200 K–12 educators and leaders, nonprobability frame that may overstate AI familiarity and perceived support. Recruitment priority: educators supporting multilingual learners and students with disabilities.

Evidence architecture

Each source answers a different weakness in the evidence.

Primary qualitative evidence

Semistructured interviews · card-sort explanations · institutional documents · open-ended survey responses

Supporting quantitative and structural context

Closed-ended educator survey · NCES School Pulse Panel · RAND educator datasets

Separately labeled nonparticipant comparison

AI-generated texts, agentic artifacts, and process traces

No source validates or overrides another.

Analysis and integration

Integration preserves difference instead of forcing agreement.

QUAL

Two-cycle interpretive codebook thematic analysis

quan

Descriptive and exploratory statistics. Below 80 shared-core responses, frequencies only.

Joint displays + narrative weaving

Convergence Expansion Divergence Silence

Framework concepts are revisable sensitizing codes, not mandatory categories.

Talk track

Qualitative themes are developed first through engagement with participant language, context, and disagreement. Quantitative sources are analyzed within their own limits. Joint displays then make the relationship among sources visible. Divergence is retained as a finding rather than treated as error.

If asked what happens if the survey underperforms. The primary statistical record is maintained in R, with selected headline estimates reproduced in jamovi as a verification check. If the achieved shared-core sample falls below 80, analysis is limited to descriptive frequencies, role-based patterns where appropriate, and narrative integration. Where role-based branching reduces usable sample size, the study reports descriptive evidence rather than unstable subgroup claims. Item-total and internal-consistency estimates are reported only as preliminary evidence of item behavior, never as validation of the framework as a scale.

Don't-know responses on infrastructural items are interpreted substantively, not just excluded. A high don't-know rate on policy or guidance items is itself a signal of weak infrastructural clarity.

[Sources] Proposal, Chapter Three, Qualitative Analysis; Quantitative Analysis; Mixed Methods Integration; Braun and Clarke (2006, 2021); Fetters et al. (2013).

Rigor, power, and ethics

The study makes the researcher’s investment visible and challengeable.

“The framework’s utility is treated as an empirical question, not a commitment the data must confirm.”

Inductive questions before framework-laden prompts
Reflexive and analytic memoing across the full study
Negative-case analysis and explicit invitations to disagree
Member checking for accuracy and interpretive clarification
Sole analyst by design; peer debriefing on ≈20% of transcripts for challenge, not intercoder agreement
IRB approval, consent, de-identification, and secure storage

Talk track

My professional role and public technoskeptical position are sources of insight and bias. The design does not claim neutrality. It makes interpretation traceable through memos, an audit trail, disconfirming evidence, and structured challenge. Primary data collection occurs outside my own district.

Why a sole analyst. Chapter Three names the choice directly. Braun and Clarke distinguish coding-reliability, codebook, and reflexive thematic analysis. This study uses an interpretive codebook approach: the framework supplies revisable sensitizing codes, the analysis stays interpretive, and quality is pursued through challenge rather than through agreement statistics. Themes are reasoned interpretations developed from participants’ situated accounts, not fixed categories waiting to be discovered (Crotty, 1998; Guba & Lincoln, 1994). A second coder trained into my codebook would demonstrate the success of that training, not the validity of the interpretation. Say codebook, not reflexive; the distinction is Chapter Three’s own.

What the debriefer sees. De-identified material drawn from approximately 20% of interview transcripts, including material from every participant group, together with the evolving code definitions and selected analytic memos. The focus is codebook clarity, disconfirming evidence, and interpretive challenge. The debriefer is asked to produce alternative readings, not to confirm mine. Peer access follows the approved confidentiality and data-security procedures.

If asked why approximately 20%. It is not a reliability threshold and is not doing the work a coverage percentage does in a coding-reliability design. It is set so that material from each participant group reaches the debriefer, which keeps challenge possible across classroom educators, building administrators, and system leaders rather than concentrated in whichever group I found easiest to interpret.

If asked whether codebook clarity is agreement by another name. No. Clarity asks whether a definition is legible and applicable by someone other than its author. Agreement asks whether two coders independently converge on the same assignment, and reports a coefficient. The first is a check on the instrument. The second is a claim about accuracy this study does not make.

If the challenge comes from the quantitative side. The survey and secondary-data strands carry their own quality logic and report it on their own terms. Reliability reasoning does not transfer across strands in a convergent design. Each is judged by the standards of its tradition, and integration happens at the level of findings through joint displays.

[Sources] Proposal, Chapter Three, Researcher Positionality, Power, and Ethics; Researcher-Controlled Analytic Workflow; Qualitative Analysis; Trustworthiness, Validity, and Reliability. Braun & Clarke (2006, 2021); Crotty (1998); Guba & Lincoln (1994); Lincoln & Guba (1985); Saldaña (2021).

Agentic AI boundary

A system response is not evidence of agentic execution.

Any agentic claim requires a prespecified, bounded multistep task with enabled and captured planning, tool use, execution, and process traces.

Comparison begins only after themes from human accounts are finalized
No participant data supplied to any system
No actions that modify external data, systems, accounts, or affect people
Tools, permissions, memory, network access, approvals, stopping conditions, failures, and outputs documented
Materials interpreted only as system-, configuration-, task-, and time-specific nonparticipant artifacts

The result is an exploratory capability snapshot, not evidence of classroom deployment.

What the study can contribute

The framework earns value only if it helps interpret practice and survives contradiction.

Scholarly

Examines whether a media-ecological and learning-science framework helps explain K–12 sensemaking.

Practical

Offers language for distinguishing support, augmentation, bypass, and exclusionary barriers.

Policy

Connects AI governance to the conditions of learning, not compliance alone.

Evidence that does not fit will revise, qualify, or extend the framework.

Timeline

Approval starts the IRB and instrument sequence, not data collection.

No participant data is collected before IRB approval. Once collection begins, instruments remain stable unless a documented IRB-approved amendment is required.

Talk track

This slide answers the practical question behind the vote: what happens next, and in what order. Stage two is the one committee members can most usefully shape, because cognitive interviews and procedural rehearsal are where their objections to specific items can still change the instruments cheaply.

CITI human-subjects certification is current and filed with the IRB materials, so stage one is IRB submission and instrument work rather than training. If a committee member reads stage one differently: the proposal's Anticipated Timeline still lists CITI completion among the stage-one activities, and Ethical Considerations states it must be completed before the proposal defense. It has since been completed, which is why this slide says filed. The chapter text is behind the fact, not in conflict with it.

Cognitive-interview respondents are recruited outside the main study and are not eligible for it. Their feedback is stored separately from main-study data.

If feasibility forces a reduction, the documented order is: agentic systems, tasks, or runs first, then institution-specific document collection, then survey subgroup scope. The interview and card-sort core is scaled only after those. Any change is documented and approved through committee and IRB procedures.

[Sources] Proposal, Chapter Three, Anticipated Timeline; Instrument Validation Strategy; Research Design; Ethical Considerations.

The decision before the committee

You are not asked to approve the framework. You are asked whether this design can examine it credibly.

01 Human meaning remains primary.

02 Equity constrains every claim about productive struggle.

03 The framework remains open to revision.

The aim is not to slow learning. It is to protect the work through which learning becomes durable, accountable, and human.

Closing

State the ask plainly. The committee is not being asked to accept tertiary algorithmicity or pedagogical friction as true. It is being asked whether the research questions, sampling, evidence boundaries, analysis, and safeguards can examine the framework credibly, including the possibility that the evidence revises, qualifies, or extends it.

The study keeps human sensemaking primary, keeps equity inside the central distinction rather than beside it, and preserves the possibility that the framework must change.

Companion materials for the committee

In this repository: Proposal Defense Studio (claims, traceability matrix, challenge deck), the printable one-page research matrix, the printable committee brief, the expanded explorer, and the shared intellectual history.

Published and citable: Miner (2026a), K–12 teacher AI evidence arc, 2024–2026, Zenodo, doi.org/10.5281/zenodo.21152544, documents the weighting behind the slide 3 estimates. Miner (2026b), i.e.: inquiry in education, 18(1), Article 4, is the peer-reviewed statement of tertiary algorithmicity and unproductive success.

Companion sites: minerclass.github.io/pedagogical-friction/ presents the framework, instruments, and secondary-data materials and is cited in the proposal Overview. minerclass.github.io/dissertation-proposal-studio/ is the intellectual-history experience. minerclass.github.io/dissertationquestionsbeta/ previews the browser-local interview and card-sort instruments described in Chapter Three; it is a draft instrument preview and collects nothing until IRB approval and consent procedures are final.

[Sources] Miner dissertation proposal, Chapters One through Three, controlling copy dated July 24, 2026.

Pedagogical Friction in the Age of Generative AI
and Tertiary Algorithmicity

A finished artifact no longer reliably shows the thinking behind it.

Adoption is running ahead of institutional clarity.

The shift is from algorithmic selection to algorithmic production.

AI can route around the formation the assignment was meant to produce.

Three learner-facing forms depend on one institutional condition.

The question is not whether school should be difficult.

Productive friction

Exclusionary friction

The study follows meaning from experience to institutional action.

Qualitative interpretation leads. Quantitative evidence adds context.

Situated meaning

Supporting pattern

Four vantage points reveal where friction is experienced and governed.

Classroom-facing educators

Building administrators

District or system leaders

Adult university students

Each source answers a different weakness in the evidence.

Integration preserves difference instead of forcing agreement.

The study makes the researcher’s investment visible and challengeable.

A system response is not evidence of agentic execution.

The framework earns value only if it helps interpret practice and survives contradiction.

Approval starts the IRB and instrument sequence, not data collection.

You are not asked to approve the framework. You are asked whether this design can examine it credibly.

Everything the committee can open during questions.

Present without touching the mouse.

Pedagogical Friction in the Age of Generative AI and Tertiary Algorithmicity

A finished artifact no longer reliably shows the thinking behind it.

Adoption is running ahead of institutional clarity.

The shift is from algorithmic selection to algorithmic production.

AI can route around the formation the assignment was meant to produce.

Three learner-facing forms depend on one institutional condition.

The question is not whether school should be difficult.

Productive friction

Exclusionary friction

The study follows meaning from experience to institutional action.

Qualitative interpretation leads. Quantitative evidence adds context.

Situated meaning

Supporting pattern

Four vantage points reveal where friction is experienced and governed.

Classroom-facing educators

Building administrators

District or system leaders

Adult university students

Each source answers a different weakness in the evidence.

Integration preserves difference instead of forcing agreement.

The study makes the researcher’s investment visible and challengeable.

A system response is not evidence of agentic execution.

The framework earns value only if it helps interpret practice and survives contradiction.

Approval starts the IRB and instrument sequence, not data collection.

You are not asked to approve the framework. You are asked whether this design can examine it credibly.

Pedagogical Friction in the Age of Generative AI
and Tertiary Algorithmicity