One of the central problems generative AI creates for schools is that it changes the relationship between academic performance and learning process. A student can now produce a fluent summary, polished essay, organized explanation, or plausible argument without necessarily doing the interpretive and authorial work those products used to signal.
That does not mean every use of AI is harmful. It does mean educators need better language for asking what kind of work was supported, what kind was bypassed, and what kind of difficulty should remain.
This week in my Summer Research Series, I am focusing on the distinction between productive friction and exclusionary friction. It is one of the most important distinctions in my dissertation work because it keeps a friction-centered approach from collapsing into a simple defense of difficulty for its own sake.
Why Friction Matters
In the context of learning, friction is the effort, resistance, delay, uncertainty, or struggle through which students develop understanding. It might involve rereading a difficult passage, retrieving an idea before checking notes, revising a weak claim, defending an interpretation in discussion, or explaining why evidence supports a conclusion.
This kind of struggle is not a defect in the learning process. It is often the learning process.
Learning-science traditions have used different language for this idea: productive failure, desirable difficulties, retrieval practice, germane cognitive load, the generation effect, schema reconstruction. Across those traditions, a common insight emerges: durable understanding usually requires some form of effortful engagement. Students do not simply receive understanding. They build it by working through resistance.
Generative AI complicates this because it can remove resistance at the exact point where resistance may be educationally valuable. It can summarize before the student has wrestled with the text. It can draft before the student has formed an argument. It can explain before the student has attempted retrieval. It can polish before the student has revised. It can answer before the student has learned to ask a stronger question.
When that happens, the result may be what my qualifying paper describes through Kapur’s language as unproductive success: work that looks successful on the surface but does not necessarily carry the learning process underneath it.
But Not All Friction Is Good
The danger is that an argument for preserving friction can quickly become an argument for preserving inequity.
Schools have a long history of calling things “rigorous” when they are, in practice, exclusionary. Some forms of difficulty do not build understanding. They block access. They reward students whose language backgrounds, bodies, schedules, cognitive profiles, prior schooling, or social conditions already match the assumptions built into school systems.
That is why the distinction matters.
Productive friction is difficulty that builds capacity. It helps students develop understanding, agency, authorship, judgment, transfer, or intellectual resilience.
Exclusionary friction is difficulty that blocks participation without meaningfully serving the learning goal.
The cleanest way to test which is which: removing productive friction would let a student skip learning; removing exclusionary friction would let a student show learning they already have.
The same task feature can function differently for different learners. A requirement to compose in English may be productive friction in a language-development task. It may be exclusionary friction in a science assessment where the goal is conceptual understanding rather than English composition. A timed handwritten response might support retrieval practice in one context, but measure handwriting speed, anxiety, disability, or test familiarity in another.
This is where generative AI creates a real pedagogical paradox. A translation tool, speech-to-text tool, summarizer, or drafting assistant might remove an unnecessary barrier for one student while bypassing essential learning labor for another. The tool itself does not settle the question. The learning goal, learner context, and task design do.
A Better AI Question
Many school AI conversations still begin with a compliance question: Did the student use AI?
That question may be necessary for some policy and assessment contexts, but it is not sufficient pedagogically. A friction-centered approach asks something more precise: What did AI bypass?
Did it bypass an exclusionary barrier that prevented the student from participating, demonstrating knowledge, or engaging with the actual learning target? Or did it bypass the productive struggle through which the student would have developed the knowledge, skill, judgment, or authorship the task was designed to cultivate?
This question shifts the conversation from tool use to learning design. In my dissertation framework, that matters because generative AI can affect each of the conditions that make learning possible:
Noetic friction: the internal struggle of thinking, retrieving, synthesizing, and revising.
Rhetorical friction: the social struggle of articulating ideas to other people, responding to critique, and participating in real dialogue.
Existential friction: the vulnerability of standing behind one’s own claims, voice, and intellectual commitments.
Infrastructural friction: the policy, assessment, leadership, professional learning, and governance conditions that make the other forms possible or impossible.
The productive/exclusionary distinction cuts across all four. A district policy might preserve valuable learning conditions, or it might create confusion and compliance burden. An assessment might make student thinking visible, or it might punish students for barriers unrelated to the learning goal. A classroom AI rule might protect authorship, or it might deny access to a student who needs support to participate.
What This Means for Educators and Leaders
For teachers, the practical move is to make the purpose of difficulty explicit. Before deciding whether AI should be allowed, limited, sequenced, or prohibited, ask:
- What is the learning goal?
- What kind of student effort does that goal require?
- Which parts of the task are meant to build capacity?
- Which parts may be blocking access without serving the goal?
- Where could AI support access without replacing the core learning work?
- Where would AI produce surface success while bypassing the intended learning?
One sequencing rule keeps this honest: a friction-preserving practice should pass the exclusion test before it passes the pedagogy test. If a “rigorous” task quietly locks out students with processing differences, language backgrounds, or access needs, it is not rigorous — it is a barrier wearing rigor’s clothes. Check for exclusion first; defend the pedagogy second.
For school and district leaders, the same distinction belongs in AI governance. Policies that focus only on cheating, detection, privacy, and acceptable use are incomplete. Those issues matter, but they do not fully address the learning processes at stake. AI guidance should also help educators decide when to preserve, reduce, redesign, or sequence friction.
That requires institutional support. Teachers should not have to make these decisions alone, classroom by classroom, without shared language or leadership backing. If schools want assessment to focus on visible thinking, process evidence, oral explanation, peer critique, and revision, then schedules, grading expectations, professional learning, curriculum resources, and policy language have to support that shift.
The Calibration Task
The goal is not to make school harder. The goal is not to make school frictionless. Both are too simple.
The goal is calibration. Protect the friction that builds learning. Remove the friction that blocks access. Redesign the task when the same difficulty does both. Sequence AI use so students encounter the right kind of struggle before receiving the right kind of support.
This matters especially in K-12 because AI governance cannot be separated from equity, accessibility, special education, multilingual learning, assessment design, and teacher professional judgment. A rule that looks clean on paper can become exclusionary in practice. A permissive stance that looks student-centered can quietly erase the learning labor students most need.
Productive friction helps students become more capable. Exclusionary friction mostly proves who could already pass through the system. The future of AI in schools will depend, in part, on whether we can tell the difference.
You can work through the distinction yourself in the interactive Friction Lab: https://minerclass.github.io/friction_game-/
Read the full Summer Research Series: https://micahminer.com/summer-research-series/
This post is part of my Summer Research Series, where I translate pieces of my qualifying paper and dissertation proposal into public-facing reflections for educators, school leaders, and researchers. The larger project examines how K-12 educators and institutional leaders make sense of pedagogical friction under conditions of generative AI, and how that sensemaking might inform AI governance, assessment, and instructional design.
