Through Kuhn’s Lens

The Expertise Inversion

May 31, 2026 | 2716 words

Through Kuhn’s Lens: The Expertise Inversion That Isn’t One Thing

A faculty member watches a sophomore open a chatbot, type three sentences, and produce a literature review summary in under a minute. The faculty member, who has spent fifteen years reading the literature being summarized, did not know the tool could do this. In that moment, the watching teacher feels the floor tilt. The student is faster. The student is fluent. The student, it seems, is ahead.

This is the scene that travels under the banner “expertise inversion.” It is real. People are reporting it this week, and the reporting carries a charge — the sense that something foundational has flipped. But the column’s first job is to slow the scene down and ask a flat question: fluent with what, exactly?

Because the sophomore in that scene demonstrated one thing. She generated a working output quickly. She did not demonstrate that she could tell whether the summary was accurate. She did not demonstrate that she knew which of the cited papers were real and which were fabricated. She did not demonstrate that she could detect where the model had smoothed a genuine scholarly disagreement into a false consensus. These are different competences. Treating them as one — calling all of it “AI fluency” — is the move the column refuses from the start.

The Conflation Hiding in the Headline

“Students more fluent with AI than faculty” is a sentence that does a lot of quiet work. It names a single quantity, fluency, and ranks two groups on it. But fluency-with-a-tool is not one skill. At minimum it splits into several: prompting (getting the model to produce what you want), tool-switching (knowing which model or feature fits the task), and error-detection (knowing when the output is wrong). These do not travel together. A person can be excellent at the first and helpless at the third.

The data this week makes the split visible. In one survey of student AI use, a large majority reported using generative tools regularly for coursework — the figure runs north of 80 percent in several recent samples — while a much smaller fraction, often under a third, reported any confidence in judging whether the output was correct. That gap is the whole story. The students are fluent at production. They are not, by their own report, fluent at verification. The headline averages these into a single ranking and loses the thing that matters.

Thomas Kuhn’s machinery is built for exactly this kind of loss. His central concept, the paradigm — the shared set of model problems and accepted solutions that a community uses to recognize what counts as good work — comes in two senses, and he spent years separating them. In The Essential Tension, in the essay “Second Thoughts on Paradigms,” Kuhn pulled apart the loose sense (paradigm as a whole worldview) from the tight sense he cared about: the exemplar, the concrete model case a community points to and says, that is what competence looks like. The exemplar is where the real disagreement lives. And in the expertise-inversion story, the two communities are pointing at different exemplars while using the same word.

Two Exemplars, One Word

Watch what the faculty member and the student actually point to when each says “competence.”

The student points to the working output. The summary appeared. It reads well. It would pass. The exemplar of competence, for this community, is successful production under time pressure. The model case is the seconds-long generation that solves the assigned task.

The faculty member points to a different scene entirely. She points to the moment of catching the error — the recognition that one of the cited papers does not exist, that the smoothed consensus is false, that the summary has confidently described a debate it does not understand. The exemplar of competence, for this community, is the catch. The model case is the judgment that the output is wrong before anyone acts on it.

This is what Kuhn called incommensurability — two communities using the same words to mean different things, with no neutral language standing above both to settle the difference. He sharpened the idea late in his life. In The Last Writings — Incommensurability in Science, Kuhn located incommensurability not in grand worldview clashes but in vocabulary: communities partition the world differently, so that a shared term marks different things on each side. “Fluency” is such a term here. When the student says she is fluent and the faculty member says she is not, they are not disagreeing about a fact. They are reading the same event through different exemplars of what fluency is.

The vendor framing exploits this gap. Calling students “AI natives” flatters the production exemplar and buries the verification one. It tells the reader that being fast is being skilled. The institutional counter-framing does the opposite mischief: it flatters faculty as holders of irreplaceable judgment, which papers over the inconvenient fact that many faculty cannot operate the tool well enough to know what it is even producing. Both framings serve power. Neither serves the reader, who wants to know which competence is actually inverted and which is not.

Is This Normal Science or a Revolution?

Kuhn’s most demanding distinction separates normal science — puzzle-solving inside an accepted frame, where the rules of what counts as a solution are not in question — from revolutionary science, where the frame itself changes and what counts as a problem gets redefined. In The Structure of Scientific Revolutions, he insisted that most scientific work is normal, not revolutionary, and that the word “revolution” is badly overused. The column inherits that insistence. The week’s framing — “inversion,” “hierarchies flipping” — invites the revolutionary reading. The framework’s job is to test it, not absorb it.

So apply the distinction carefully. Suppose what has inverted is who is faster at the tool. The student prompts quickly; the faculty member fumbles. This is a change inside the existing frame of expertise. The frame still says: knowing means understanding the domain deeply, and the tool is an instrument in service of that knowing. On this reading, the student holds the instrument better, the way a younger researcher might run the new software faster. The hierarchy of who knows the domain has not moved at all. The faculty member still holds the deeper knowing. What inverted is instrument-handling — a new puzzle-solving skill inside the old paradigm. That is normal science. It is not a revolution. It is a generational shift in tool dexterity, of the kind every new instrument produces.

Now suppose something stronger. Suppose the student’s production fluency is starting to redefine what counts as the problem. Suppose the question “did you understand the literature” is being quietly replaced by the question “did you produce an acceptable summary.” If the standard of competence itself migrates — from understanding-the-domain to producing-acceptable-output — then the frame is changing. That would be revolutionary in Kuhn’s sense: not a faster solution to the old problem, but a reframing of which problem matters.

The data does not yet show the second thing. It shows the first. The verification gap — production fluency high, error-detection confidence low — is precisely what you would expect if students had gained instrument-handling skill without the frame of expertise moving. They produce faster. They cannot tell when the production is wrong. The deeper knowing that catches the error still lives where it lived. The hierarchy of domain-expertise has not inverted. Only the hierarchy of tool-speed has.

This matters because the colloquial reading — “the paradigm has shifted” — would license a conclusion the evidence does not support: that students now hold the expertise and faculty hold an obsolete craft. Kuhn’s tools block that conclusion. Most candidate shifts are normal science wearing revolutionary clothing, and this one is dressed for the part. The clothing is the word “inversion.” Underneath, the body has barely moved.

The Anomaly the Phrase Papers Over

Kuhn used anomaly to mean an observation the current frame cannot comfortably account for — the kind of stubborn misfit that, if it accumulates, can push a community into crisis. In The Structure of Scientific Revolutions, he showed that anomalies are usually noticed late, because the frame trains people not to see them. And he warned that confident, settled language is often where the anomaly is hiding.

“Expertise inversion” is confident, settled language. What anomaly is it hiding?

The anomaly is this: fluency-with-a-tool and expertise-in-a-domain were never the same thing, and the phrase “expertise inversion” only works by treating them as the same thing. The real misfit the data surfaces is not that students surpassed faculty. It is that we conflated two competences for so long that we now cannot tell which one moved. When typewriters arrived, no one said the typist had inverted the author. The competences were obviously distinct. With AI tools the distinction blurs, because the tool produces content, not just formatting — and content looks like the product of knowing.

That is the anomaly. The tool generates outputs that wear the costume of expertise. A fluent summary looks like understanding. So when a student produces one fast, the production reads as expertise, and the faculty member’s slowness reads as obsolescence. The verification gap is the tell that the costume is just a costume. The student who cannot judge the output’s correctness has not acquired the expertise; she has acquired the ability to summon its appearance. Under a third of students, by the survey figure, claim the judgment to tell appearance from substance.

This is why the column distrusts the headline. “Students more fluent than faculty” does not record an inversion of expertise. It records the moment when a tool made the appearance of expertise cheap, and a community started mistaking the cheap appearance for the expensive thing. The anomaly is the conflation, and the confident phrase exists precisely to keep us from seeing it.

Why the Copernican Case Sets the Bar

Kuhn’s worked example of a genuine reframing is the one he wrote a whole book about. In The Copernican Revolution, he traced how the shift from an Earth-centered to a Sun-centered cosmos was not a single observation but a reorganization of what counted as a problem, an explanation, and a fact. The same data — the same planetary positions — got read through a new frame that made different things obvious and different things puzzling. The revolution was not that someone computed faster. It was that the questions changed.

Hold the expertise-inversion case against that bar. For this to be a Copernican-grade reframing, we would need to see not faster production but a reorganization of what counts as knowing. We would need the verification competence to be either absorbed by the students or rendered unnecessary — not skipped, but no longer the thing that defines expertise. We do not see that. The verification competence is still the thing that defines expertise; the students simply have not acquired it. The frame has not been reorganized. It has been outrun in one dimension while standing still in the dimension that matters.

The contrast is instructive precisely because it is unflattering to the headline. Copernicus did not give astronomers a faster telescope. He gave them a different question. The AI tool gives students a faster output. That is a real change. It is not a different question. The question — do you understand this well enough to know when it is wrong — is exactly the same question it has always been. The students answer it less often, not differently.

Particularizing: Back to the Sophomore

Return to the opening scene, and look harder. The sophomore produced the summary in under a minute. The faculty member felt outpaced. But notice what the faculty member did next, in the version of this scene that the data supports: she read the summary and saw the false citation. She caught it in seconds — faster than the student produced it. The catch was instantaneous because the deep frame, the domain knowledge, makes the error glow.

So which inversion happened? The student inverted the faculty member on time-to-produce. The faculty member inverted the student on time-to-catch. Both are forms of speed. Only one of them is expertise in the sense the frame has always meant. The student’s speed produced an artifact. The faculty member’s speed produced a judgment. Kuhn’s exemplar distinction lands precisely here: the two communities will never agree on who is fluent, because each points to its own speed as the model case.

And the reader’s interest cuts through this cleanly. The reader does not want to be flattered as a native or managed as a novice. The reader wants to know which competence she actually holds. The honest answer, from the data, is that production fluency is now widely distributed and verification fluency is not. The first is easy to acquire and easy to see. The second is hard to acquire and easy to fake. Any framing that hides this — vendor or institutional — is selling something.

There is a further particular worth holding. Some students do develop verification fluency, and some faculty develop production fluency, and these are the cases where the conflation finally breaks apart cleanly. The student who learns to distrust the model’s citations is not “more fluent.” She has acquired a second, different competence. When she has both, the word “fluency” stops being useful and we should drop it. The clarity is not in ranking the two communities. It is in refusing the single word that pretends they share one scale.

The Evidence Question

The column ends where it always ends: not on what should be done, but on what we would have to observe to know whether the frame had actually moved.

To know that expertise itself — not the tooling around it — had been reorganized, we would need to see the verification competence displaced, not merely unevenly distributed. Specifically:

We would need to observe that the catch — the recognition that an output is wrong — had migrated from the faculty community to the student community, or had ceased to be the thing that distinguishes competent work from incompetent work. Not that students produce faster. That they detect error as well as or better than the people currently positioned as knowers. The survey gap is the measurement to watch. If the under-a-third figure for verification confidence rose to meet the over-80-percent figure for production use, and rose because students could actually catch errors rather than because they grew careless about checking, the frame would be moving. Today the two figures sit far apart, and that distance is the evidence that the inversion is in the instrument, not the expertise.

We would need to observe the exemplar change. Right now the faculty community points to the catch and the student community points to the output. A genuine reframing, in the sense Kuhn drew in The Essential Tension, would show both communities pointing to the same model case of competence — and that shared exemplar would tell us which way the frame had settled. If both came to point at the output and to treat the catch as someone else’s job, the frame would have shifted toward production. If both came to point at the catch and to treat fast production as trivial, it would have shifted toward verification. Either would be a reframing. Neither has happened. The communities still point at different scenes.

And we would need the anomaly to surface in plain language. As long as the phenomenon travels under the confident phrase “expertise inversion,” the conflation stays hidden and the frame looks stable. The day the reporting stops saying “students are more fluent” and starts saying “students produce faster but verify less” — the day the single word “fluency” breaks into its parts in ordinary speech — that will be the visible sign that the community has stopped papering over its anomaly. Kuhn’s account in The Structure of Scientific Revolutions would read that linguistic break as the early tremor before any real reframing.

Until then the honest reading is the austere one. A tool got faster in younger hands. The judgment that catches the tool’s errors did not move. Calling that an inversion of expert