A student writes, “Increasing effective nuclear charge decreases atomic radius.” It may have been memorized, reconstructed from an electrostatic model, copied from a friend, or polished by AI. The words alone cannot tell me which.
The problem is not new. AI has made it impossible to ignore.
Education has never depended on students inventing their own ideas. No chemistry student originated electronegativity, the mole, or the orbital model. Neither did the teacher explaining them. Learning is largely an act of appropriation: taking concepts inherited from other minds and making them part of your cognitive machinery.
So originality is a poor proxy for understanding.
A student can produce awkward, original language while barely grasping the concept. Another can repeat my exact sentence while understanding the mechanism well enough to reconstruct it, extend it, and recognize its limits. Authorship and competence overlap, but they are not the same variable.
Memory still matters. Experts carry compressed knowledge because reasoning cannot begin from zero every time. “Electron,” “mole,” “oxidation,” and “electronegativity” become cognitive primitives that free working memory for larger operations.
Vocabulary is something like an API for thought. An API lets one system call upon a complicated operation without rebuilding the machinery each time. A student who has internalized “effective nuclear charge” can invoke an entire electrostatic model with a few words. But knowing the command is not the same as understanding what it does. A student may call the function correctly in a familiar problem and still have no idea why it fails in a new one.
The educational mistake was never asking students to remember. It was confusing retrieval with understanding. Memory matters when it becomes available to reasoning.
The better question is: “Under what support conditions did the student demonstrate this capability?”
A student does not possess one fixed quantity called understanding. Performance changes with notes, examples, teacher prompts, collaborators, calculators, and AI. What we observe is a curve across levels of support. Two students may submit equally polished AI-assisted work while occupying radically different places on that curve.
This also clarifies what grades are. An assessment is a probe. The artifact is its signal. A rubric interprets that signal, and a score becomes one local estimate of capability under those conditions. A grade combines readings from several channels—tests, homework, labs, conversation, transfer tasks—through an often implicit weighting model. It is sensor fusion compressed into an administratively convenient scalar.
Grades are therefore neither curve shifts nor stationary truths about a learner. A score estimates position at one moment under one set of supports. Comparable measurements over time estimate movement. Feedback and instruction may help cause that movement. Asking one number to represent both position and motion confuses state with derivative.
The artifact was always an imperfect sensor. AI has increased the noise in certain sensors so dramatically that we can no longer ignore the measurement problem.
That may ultimately be productive, because it forces us toward a more defensible conception of education: “We aren't grading ownership of ideas. We're estimating what intellectual operations a learner can reliably perform under specified conditions.”
And then education has another job beyond measurement: “Move that performance curve.”
Get the student from needing a walkthrough, to using an example, to working with notes, to acting independently—and finally to using tools to reach intellectual states that neither the unaided student nor the tool could reach alone. The endpoint is not independence from assistance. It is agency in the presence of tools. It is learning to rotate the kaleidoscope.
The grade is only a static, imperfect reading by imperfect humans of imperfect sensors. Rotation and translation are better models for education.
Related posts
What Actually Carries Forward
May 4, 2026
The content may fade; what matters is whether students become able to acquire, test, and apply information when the script runs out.
Learning & Moral Geometry
June 18, 2026
Education increases capability, but capability has direction. Skill is magnitude; character governs where that power points.
Thoughts on the web's effects on our minds
July 28, 2008
An early classroom encounter with the same measurement problem: students could copy information into slides without analyzing, distilling, or reasoning from it.
IQ revisited
August 13, 2013
A test may measure something real without measuring something fixed; preparation, education, family conditions, and opportunity all alter the observed score.
Creative combinations: Lennon and McCarthy
July 8, 2014
Creativity appears less as pristine originality than as productive tension: novelty generation disciplined by selection, focus, and craft.
