// artificial intelligence // curriculum design
Most school conversations about student AI use begin with a familiar question: Did the student use AI?
Sometimes that question is necessary. Schools need expectations for attribution, disclosure, acceptable assistance, privacy, and responsible conduct. Students need to understand when using a tool violates the conditions of an assignment. Teachers need reasonable ways to respond when submitted work does not represent what a student was expected to do.
But academic integrity is not the whole problem. A student can follow the stated rule and still avoid the intellectual work the assignment was designed to develop. Another can violate an overly broad rule while using a tool to remove an access barrier that had little to do with the learning goal. A polished product can be properly disclosed and still provide weak evidence of understanding.
The integrity question asks about the origin and permitted production of the artifact. The learning question asks what the student can understand, explain, defend, revise, and transfer. Schools need both. They should not mistake one for the other.
Integrity Governs Conduct. Learning Design Governs Development.
Academic integrity frameworks are generally designed to establish whether work is authentic, attributed, and produced under the stated conditions. These are legitimate concerns; a school cannot abandon expectations for honesty because generative systems make authorship more complicated. Yet a conduct framework has limits.
Imagine a teacher allows AI for brainstorming. One student asks a system to generate a complete thesis, organize the supporting reasons, select likely evidence, and propose counterarguments, then rewrites portions and discloses that AI was used for “brainstorming.” The student may have complied with the literal rule, and the disclosure may be accurate, but the tool completed much of the interpretive and rhetorical work the assignment was meant to cultivate.
Now imagine a multilingual learner using translation support to understand the directions for a science explanation. If the assessed goal is scientific reasoning rather than English-language production, that support may remove an exclusionary barrier without replacing the intended conceptual work.
The category “AI use” cannot explain the pedagogical difference between those cases. That is why Week 2 of this series focused on productive and exclusionary friction. The question is not only whether a tool removed difficulty, but what kind of difficulty it removed: did it help the student reach the intended learning, or did it perform the learning labor on the student’s behalf? Academic integrity can tell us whether a rule was followed. Learning design must tell us whether the rule protected the right thing.
How the Work Was Produced Is Not the Same as Evidence of Learning
Generative AI makes it harder to interpret how academic work was produced. Teachers may no longer know how much of a fluent essay, explanation, or solution reflects the student’s unaided production. That uncertainty is real. but how the work was produced and what the student learned are different problems.
Academic integrity asks who or what produced this artifact, and were the rules followed? Evidence of learning asks: What can the student understand, explain, defend, revise, and transfer? A teacher can know exactly how a paper was produced and still know little about the student’s understanding. Students have always been able to imitate models, follow formulas, or assemble correct-looking work without building durable knowledge. Generative AI magnifies that longstanding problem, because it can produce the surface features of competence quickly and convincingly.
This is the illusion of completion; that the work looks finished, but the relationship between the artifact and the student’s learning has become uncertain. The answer cannot be to treat every student as a suspect or turn every assignment into a forensic investigation. Detection-centered approaches place teachers in an exhausting enforcement role while leaving the underlying assessment problem intact. The stronger response is to design learning so that important thinking becomes visible before, during, and after the final product.
Protect Thinking, Judgment, and Authorship
If the final artifact is no longer sufficient evidence, educators need to name the human work the task is supposed to develop. Three categories give a practical start — and each maps to a form of friction worth preserving.
Thinking. What must the student retrieve, interpret, connect, analyze, or generate? Evidence might include annotated sources, initial questions, concept maps, attempted solutions, draft claims, or a short explanation produced before AI assistance begins. The point is not to romanticize paper or document every mental step; it is to preserve enough of the process that teacher and student can see where understanding is developing. This is noetic friction.
Judgment. What choices must the student make, and how will they justify them? AI can produce options, examples, and revisions. Judgment becomes visible when students compare possibilities, identify weaknesses, select evidence, reject misleading suggestions, and explain why one approach is more defensible than another. A student who accepts an AI response because it sounds polished is doing something different from one who can evaluate it against disciplinary criteria. This is rhetorical friction.
Authorship. What must the student be prepared to stand behind? Authorship is more than typing the words; it is taking responsibility for claims, evidence, interpretation, and consequences. Students demonstrate it when they can explain what they mean, respond to questions, revise after criticism, and distinguish their judgment from the assistance they received. This is existential friction: learning asks students not only to produce an answer but to enter a relationship with people who may question, challenge, or misunderstand it.
Thinking, judgment, authorship can be categorized in the noetic, rhetorical, existential. These are the three frictions from Week 2, now anchored to what a task must protect.
Design the Learning Sequence
Once educators name the protected work, AI guidance gets more precise. Instead of asking only whether AI is allowed, ask when it should enter the sequence.
Five moves:
1. Name the capability. What should the student become more capable of doing? Be more specific than “complete an essay” — interpreting conflicting sources, constructing a causal explanation, defending a claim with evidence, applying a relationship in an unfamiliar context.
2. Identify the protected human work. Which stages are necessary for developing that capability? These are the places where premature automation produces unproductive success: a correct-looking outcome without the development the task intended.
3. Check for exclusion. Before preserving a difficulty, ask whether it serves the learning goal for this student. A friction-preserving practice should pass the exclusion test before it passes the pedagogy test. The goal is not to protect difficulty; it is to protect learning.
4. Decide when AI may enter. Before learning, students may need to retrieve, attempt, or question without automated completion. During, AI may help compare explanations, test reasoning, or find gaps. After, it may support revision, transfer, or accessibility, or in other words when students can explain what changed and why.
5. Gather more than the final artifact. The finished product should sit alongside at least one other signal such as a brief oral explanation, an initial claim, a revision note, or a short in-class transfer task. The principle is simply that evidence should match the capability the task claims to develop. (How to build that assessment repertoire is the subject of Week 7, Assessment Redesign in Practice.)
A caution: don’t answer AI uncertainty with surveillance. Capturing every keystroke or treating work histories as permanent behavioral records creates privacy risk, inflates workload, and turns learning into a compliance performance. Process-visible learning should be purposeful and proportionate that are a thoughtful set of meaningful checkpoints, not total monitoring. Students need room to experiment, make mistakes, and develop a voice without every incomplete thought becoming a record.
From the Classroom Problem to the Institutional Problem
Moving beyond academic integrity does not make the work easier. It asks educators to make more contextual judgments about goals, access, sequence, evidence, and authorship; and teachers should not have to make them alone. A single teacher can redesign one assignment’s sequence. Doing it consistently, across a department or a building, is no longer a classroom problem. It requires shared language, task-level examples, protected time, and aligned policy which is where the institutional weeks of this series go to with governance (Week 5) and design (Week 6) and the conditions beneath them.
The central question is more than “What AI use should we prohibit?” It is what human development are we responsible for protecting, and what institutional conditions allow educators to protect it?
That is where the series turns next. The ladder so far:
Week 1 asked why polished AI-supported performance can make learning difficult to see.
Week 2 asked which friction to preserve, reduce, or redesign.
Week 3 asks what human learning must remain visible.
Week 4 asks how educators are making sense of AI’s role — and what they believe it is doing to student learning.
Week 5 asks whether they have the institutional conditions to act on it.
Academic integrity remains part of the picture. It is simply not the whole frame.
Read the full Summer Research Series: https://micahminer.com/summer-research-series/
Related: The Illusion of Completion · Productive vs. Exclusionary Friction · Writing Policy for the Learning Sequence
