On the intentional preservation of productive noetic, rhetorical, existential, and infrastructural resistance in an era where frictionless automation has become the pedagogical default.
This companion can be read straight through, but the construct is easier to use when the path matches the task: committee skim, classroom design, equity check, or theory lineage.
Something has shifted in classrooms, and it is not primarily about cheating. It is about a deeper and subtler condition in which students produce work that appears to demonstrate understanding but is not accompanied by the cognitive reconstruction that understanding requires. The essay is coherent. The citations are formatted. The argument flows. And the student, asked to explain what they wrote, cannot. This paper names this condition unproductive success, and treats it as the defining pedagogical problem of the generative AI era.
Manu Kapur's work on productive failure taught the field that slowing initial performance — letting students struggle before instruction — produces more durable learning. The inverse condition is its shadow: performance without the struggle that gives performance meaning. Generative AI, operating as what we might call tertiary algorithmicity — algorithmically-created and algorithmically-curated symbolic content — makes unproductive success not an edge case but the path of least resistance. The output arrives fully formed. The schema was never built. The learner has been bypassed.
The fluency of the output is the problem. The output looks like learning whether or not learning has occurred. — Central framing of the pedagogical friction construct
The argument for friction is not new. It draws on three intellectual resources that rarely meet in the same argument and that arrive at compatible conclusions from different starting points. Learning scientists found the argument through the cognition of struggle. Bernard Stiegler found it through the pharmacology of technics, the philosophical recognition that every technical system is simultaneously poison and cure. Media ecologists found it through the environment of symbolic production. Generative AI is the first condition under which all three resources converge on the same practical demand: preserve the resistance that learning requires, while refusing the conclusion that resistance is futile. Designing friction away is designing learning away.
Cognitive disequilibrium is the engine of learning. Growth occurs when existing mental schemas prove inadequate and must be reconstructed through effortful engagement. Struggle is not a regrettable side effect of learning; it is the mechanism by which learning happens.
From this tradition comes the concept of desirable difficulties — interventions that slow performance now to strengthen retention later — and the taxonomy of productive failure that makes unproductive success visible as its inverse.
Technical systems are never neutral, and they are never simply good or bad. Drawing on Plato's treatment of writing in the Phaedrus, Bernard Stiegler argued that every technology functions simultaneously as poison and cure, what he called the pharmakon. The same generative system that displaces synthesis can, under intentional design, be turned back toward the cultivation of the capacities it threatens to erode.
This tradition supplies the qualifying paper's response to a foreseeable objection. If Baudrillard is right that simulation has reached the point where meaning is lost and resistance is reabsorbed by the system it opposes, then pedagogical friction is naive. Stiegler's pharmakon refuses Baudrillard's totalization without dismissing the diagnosis. Friction-centered pedagogy becomes the structured cultivation of the curative dimension of generative AI against the toxic default the technology produces when left to its own tendencies.
Communication technologies are not neutral tools. They are environmental forces that restructure human consciousness, memory, and social relation. What a technology makes easy, it makes common; what it makes hard, it makes rare; and over time, what is rare becomes forgotten.
Applied to generative AI, this tradition asks what cognitive and communal capacities are gained or lost when symbolic production is outsourced to algorithmic generation — and holds that the answer is never only "gained."
What is given to the student frictionlessly has not been learned; it has merely arrived. — A synthesis of the three resources, applied to generative AI
A century of thinkers whose work, in retrospect, was preparing the ground for this moment. Click any dot to open the card; color indicates the tradition. The sequence is not a procession toward a predetermined conclusion — several of these scholars would have disagreed sharply with one another — but the through-line is visible enough to be named. Each contributed a piece of what eventually became the case for friction.
Pedagogical friction is the framework this paper offers as a synthesis of the three resources. It is not a checklist or a technique. It is a claim about which forms of resistance education cannot surrender without surrendering the conditions of durable learning. The four dimensions name the locations at which friction must be preserved: inside the individual learner's cognition, within the dialogic space of the classroom, in the student's posture toward their own claims, and across the institutional systems that make the first three possible. Click each quadrant to expand.
The framework begins from the qualifying paper's three characteristics of tertiary algorithmicity. Noetic displacement names the substitution of external generation for the learner's own cognitive labor. Rhetorical saturation names a communicative environment filled with synthetic discourse whose origin and dialogic status become uncertain. Existential abstraction names the separation of symbolic production from lived experience, ownership, and accountability. Pedagogical friction is the proposed response to those three threats.
The framework becomes useful only when it can inform judgment in specific classroom moments. The scenarios below are composite cases drawn from common situations that district leaders, teachers, and deans report encountering. Each scenario offers four possible teacher responses. There is no single correct answer — each option is defensible under some framework — but each also makes visible trade-offs across the four dimensions of friction. The feedback shows what a given choice preserves, diminishes, or collapses. Judgment is the work.
Any framework that valorizes resistance risks becoming a justification for practices that exclude rather than develop. This distinction is the most important caveat in the entire framework. Productive friction builds cognitive, rhetorical, and civic capacity; it is the struggle that produces the learner. Exclusionary friction is a barrier that prevents participation without meaningfully building capacity — the impossible deadline, the unaccommodated test, the unexplained rubric, or the language requirement that locks out the very students whose thinking the classroom most needs.
Friction is productive when removing it would shortcut a learner past the cognitive or civic capacity the task is meant to build. Friction is exclusionary when removing it would allow a learner to demonstrate capacity they already possess but are being prevented from showing. The difference is not always obvious in the abstract, and several practices sit genuinely in between. The sorting exercise that follows is designed to surface that ambiguity rather than resolve it.
The final paper sharpens this caveat through disability studies, DisCrit, and the English Learner paradox. Generative AI can reduce exclusionary barriers for multilingual learners, students with disabilities, and students who lack certain forms of academic support. It can also flatten voice, reward standardized expression, and create new forms of algorithmic code-switching when students must translate their thinking into forms a model recognizes. The question is never simply whether AI removes friction. The question is which friction it removes, for whom, and at what cost to authorship, understanding, and access.
The question this framework leaves on the table is not whether generative AI belongs in education — it is already there, and will not leave. The question is what a school, a classroom, a district, or a teacher is willing to surrender in exchange for the convenience it offers, and what must be actively designed to survive its presence. Friction is a design stance. The alternative to designing for it is not neutrality; it is designing against it by default. Every rubric, every policy, every prompt, every workflow is already answering the question one way or the other. The only choice is whether to answer it deliberately.
To design without friction is not to design for ease. It is to design for the dissolution of the learner. — The closing claim of the framework