Through Kuhn’s Lens

The Policy Contradiction

July 11, 2026 | 2593 words

Through Kuhn’s Lens: The Policy at War With Itself

A university this spring published a fourteen-page academic integrity policy. Paragraph three prohibits “the unauthorized use of generative AI in any submitted work.” Paragraph seven, on the same page, “encourages students to develop responsible AI fluency as a core professional competency.” Nothing in between tells a student where the first sentence ends and the second begins. The document does not resolve the tension. It does not seem to notice it.

This is not a drafting error. It is a pattern. A corporation blocks employee access to public chatbots on the network while its own internal memo requires staff to complete onboarding for a company AI assistant. A school district forbids AI in assessment and mandates AI literacy in the curriculum that leads to the assessment. The prohibition and the mandate share the same letterhead, sometimes the same paragraph, and they cannot both be followed cleanly at once.

Institutions are calling this “a balanced approach.” Some call it “a paradigm shift in governance.” The task here is to take those claims seriously enough to test them — and Thomas Kuhn’s history-and-philosophy of science supplies the instrument. The question is not how schools or firms should handle AI. That framing already assumes the institution knows what AI is to it. The question sits earlier, in the domain of AI’s social life: what frame is the institution reading AI through, and what does a self-contradicting policy reveal about the condition of that frame?

The Frame Doing the Reading

Start with a plain observation. An institution does not respond to a technology directly. It responds through a governing frame — a shared set of assumptions about what kind of thing the technology is, what problems it poses, and what a solution looks like. Kuhn called this shared frame a paradigm: the set of commitments that decides, for a community, what counts as a real problem and a real solution.

The institutions issuing these policies inherited a frame built for a different class of tool. Call it the regulable-tool frame. In this frame, a technology is an object with a defined function. Its use can be authorized or unauthorized. It sits on one side of a line the institution draws, and compliance means staying on the correct side. This frame has governed institutional technology policy for decades. It handled the calculator, the spreadsheet, the internet-in-the-classroom, the personal phone. Each arrived, each was disruptive, each was eventually sorted into “permitted here, forbidden there.”

Under this frame, policy-writing is what Kuhn called normal science — puzzle-solving inside an accepted frame. The rules of the game are not in question. The puzzle is only where to draw the line. A university does not ask what is a calculator to us. It asks in which exams may it be used. That is a tractable puzzle. The frame supplies the tools to solve it, and the solution, once found, generalizes. Everyone knows what a banned calculator looks like on a desk.

The self-contradicting AI policy is what happens when this puzzle-solving stops producing solutions. Kuhn’s diagnostic battery gives us several instruments to examine why. Three of them fit this case. The essay will run those three and leave the others in the drawer.

First Instrument: Anomaly

Kuhn’s concept of anomaly reframes the contradiction as something other than incompetence. An anomaly is a fact the governing frame cannot digest — an observation the paradigm’s tools were not built to process. In The Structure of Scientific Revolutions, Kuhn shows that anomalies rarely announce themselves as crises. They appear first as things that don’t quite fit, that get set aside, that generate awkward local fixes.

Generative AI presents the regulable-tool frame with a specific anomaly. The frame assumes a technology can be located on one side of a use-line. Authorized or unauthorized. Present or absent. But generative AI does not hold still on one side of that line. It is a tool a student uses to cheat and a skill the same student is required to master. It is the thing the network blocks and the thing the onboarding memo mandates. The competence the institution wants to build and the misuse the institution wants to prevent are the same activity, distinguished only by an intention the policy cannot observe.

This is what the regulable-tool frame cannot digest. It was built to sort tools by their presence, not by the invisible purpose behind their use. A calculator on a desk during a closed-book exam is a violation you can see. A student who used a chatbot to structure an argument, then wrote the argument herself, presents nothing to see. The frame’s central mechanism — draw a line, check which side the tool is on — has nothing to check.

Watch what the contradictory policy actually does with this anomaly. It does not resolve it. It holds both incompatible facts at once. The prohibition paragraph treats AI as a tool that can be banned. The mandate paragraph treats it as a competence that must be built. The document asserts both because the frame can digest neither. This is Kuhn’s key insight into anomaly: an anomaly is not first experienced as a decision to be made. It is experienced as a strain the existing frame absorbs by improvising. The contradictory policy is that improvisation, written down.

Notice what this reading refuses to say. It does not say the policy authors are foolish. Kuhn was consistent on this. Normal science practitioners are not stupid when their frame strains; they are doing competent work with tools that have reached their limit. In The Essential Tension, Kuhn stresses that the tension between tradition and innovation is productive precisely because practitioners cling to the frame — and should, until it genuinely fails. The policy author drawing a line between permitted and forbidden AI is doing exactly what the frame instructs. The instruction has simply stopped yielding a coherent line.

Second Instrument: Incommensurability

The contradiction lives inside single documents. But it also lives across the communities that produce them. Here Kuhn’s concept of incommensurability does the work: two camps using the same words to mean different things, so they talk past each other without noticing.

Inside a single institution, two discourse communities are usually writing the AI policy — sometimes literally in adjacent offices. Call the first the integrity community: registrars, assessment officers, academic-standards committees, compliance and legal. Call the second the capability community: instructional-design teams, the provost’s innovation office, the corporate learning-and-development function, the people whose job is workforce readiness.

Both communities use the word “responsible.” Both use “integrity.” Both would tell you they support “AI literacy.” And they mean incommensurable things by each.

For the integrity community, the model case — Kuhn would call it the exemplar, the concrete example from which a community reasons — is the fabricated submission. The essay the student did not write. The report generated wholesale and passed off as work. From that exemplar, “responsible AI use” means use that leaves the authorship intact. “Integrity” means the work represents the person. AI literacy, to this community, means knowing where the forbidden line is and staying behind it.

For the capability community, the exemplar is the unprepared graduate. The employee who cannot use the tools the workplace now assumes. From that exemplar, “responsible AI use” means fluent, disclosed, effective use. “Integrity” means honesty about method, not abstinence from the tool. AI literacy, to this community, means hands-on capability — the very activity the integrity community is trying to prevent.

These are not two positions on a shared scale, where one wants more AI and the other wants less. They are two frames reasoning from different exemplars toward incompatible pictures of what the good outcome is. When they co-author a policy, each writes its own paragraph faithfully. The integrity community writes the prohibition. The capability community writes the mandate. The word “responsible” appears in both and papers over the seam. The document reads as balanced because both authors used the same vocabulary. It functions as contradictory because they meant different things by it.

Kuhn’s late work, collected in The Last Writings — Incommensurability in Science, sharpened this point. Incommensurability is not total failure to communicate. It is local — a breakdown at specific terms, where a word carries different taxonomies for different speakers. That is exactly the pattern here. The two communities agree on most of the language of institutional life. They diverge precisely at “responsible,” “integrity,” “literacy” — the load-bearing terms of the AI policy. The divergence is invisible because the words are shared. The policy is at war with itself because its authors were, and neither noticed the language had split beneath them.

This matters for the reader being managed by such a policy. When an institution presents its contradictory document as “a balanced approach,” the incommensurability lens lets you see what actually happened. Two communities did not reach a balance. They failed to notice they were speaking different languages, and stapled their translations together. “Balanced” is the word for a compromise deliberately struck. This was not struck. It accreted.

Third Instrument: Crisis, and the Discipline of Not Over-Reading

Now the hardest instrument, and the one that requires the most restraint. Kuhn has a name for the state between a working frame and a new one: crisis. Crisis is when anomalies have accumulated to the point that the community’s confidence in the old frame wavers. Puzzle-solving falters. Practitioners begin proposing incompatible ad hoc fixes. The frame has not broken — nothing has replaced it — but the sense that it works is gone.

The self-contradicting policy looks like a symptom of crisis. Kuhn describes crisis-period science as marked by exactly this proliferation of incompatible articulations, a willingness to try anything, a blurring of the rules. The prohibition-plus-mandate has that texture. It is an ad hoc fix. It holds two incompatible commitments because the frame that once adjudicated between them has lost its grip.

But the column’s discipline demands reluctance here. Kuhn was insistent, in The Structure of Scientific Revolutions, that crisis is a specific condition with specific markers — not a name for any confusion. The danger is over-reading. A muddled document is not automatically evidence of a frame in crisis. It may be evidence of two offices that did not talk to each other. It may be evidence of legal caution — the prohibition as liability shield, the mandate as strategic signal, both cynically retained because deleting either creates exposure. That is not crisis. That is administrative hedging with the old frame fully intact.

So the two readings must be held apart. Under the crisis reading, the contradiction is a symptom: the regulable-tool frame is failing to process AI, practitioners sense it, and the incoherent policy is the improvisation of a community whose frame no longer works. Under the muddle reading, the contradiction is a byproduct: the frame is fine, the mechanism of line-drawing still commands full confidence, and the incoherence is just poor coordination between offices — the kind of thing a better memo would fix.

The distinction is not academic. It tells you what will happen next. If it is crisis, no amount of better drafting will resolve the contradiction, because the contradiction is in the frame, not the prose. If it is muddle, a single well-run meeting between the two offices produces a clean policy.

The week’s data offers a way to test which reading holds — and this is where a number does real work rather than decorate. When surveys report that a large majority of students now use generative AI in their coursework while their institutions’ policies still treat that use as an exceptional violation to be caught, the gap itself is diagnostic. A recent finding that roughly nine in ten students report using AI tools, set against policies written as if such use were the rare transgression, is not a coordination gap. It is a frame gap. The regulable-tool frame assumes the tool is outside the normal activity and can be excluded from it. The usage data says the tool is inside the normal activity and cannot be excluded without excluding the activity. That is the signature of anomaly overwhelming the frame — the crisis reading, not the muddle.

A second number sharpens the point. Where institutions report that their AI guidance is revised on cycles of months rather than years — repeatedly reissued, repeatedly amended — the churn is itself a crisis marker. Normal-science policy is stable, because the frame yields stable solutions. When the same document is rewritten every term and still contradicts itself, the instability is not in the authors’ diligence. It is in the frame’s inability to produce a settled answer. Kuhn’s account predicts exactly this: in crisis, the rules loosen and the same puzzles get worked and reworked without resolving.

Together these numbers tip the reading toward crisis over muddle — cautiously. They show a frame straining, not a frame that has broken. Which brings us to the phrase the institutions keep reaching for.

Policing “Paradigm Shift”

Some of these institutions describe their contradictory posture as “a paradigm shift in governance.” The claim must be refused, and Kuhn’s machinery shows precisely why.

A paradigm shift, in Kuhn’s demanding sense, is not confusion. It is the opposite of confusion. It is the arrival of a new frame that resolves the anomalies the old one could not, and that reorganizes the whole field around new exemplars. In The Copernican Revolution, Kuhn traces what a genuine shift required: not just dissatisfaction with the Ptolemaic system, but a new organizing conception under which the old anomalies dissolved and new puzzles became worth solving. The shift was a resolution, hard-won and total. It let astronomers stop patching.

The contradictory AI policy is the reverse of this. It resolves nothing. It patches. It holds the anomaly open rather than dissolving it. To call it a paradigm shift is to invert Kuhn’s meaning — to award the prestige of revolution to what is actually a frame refusing to break. The prohibition-plus-mandate is not the new astronomy. It is the last, most strained epicycle of the old one: a device added to keep the failing frame in operation a little longer.

This is not pedantry. The marketing use of “paradigm shift” launders a failure into an achievement. It tells the reader that the institution has arrived somewhere, when it has only failed to leave. Kuhn’s tools let you decline the laundering. Strip the phrase and say what is actually being claimed: we could not make our old frame produce a coherent policy, so we wrote down both halves of the incoherence and named the result progress. Stated plainly, the claim collapses. That is the anti-mystification the framework is for.

What a Real Reframing Would Look Like

If the contradictory policy is not a reframing, what would one look like here? Kuhn’s account lets us specify it, not just gesture at it. A genuine new frame would not draw a better line between permitted and forbidden AI. It would stop treating the line as the relevant object.

The regulable-tool frame asks: on which side of the use-boundary does this tool fall? A successor frame would abandon that question as malformed — the way Copernican astronomy abandoned the question of how many epicycles a planet needs. It might reorganize around a different exemplar entirely. Instead of the fabricated submission or the unprepared graduate, its model case might be the disclosed process — work whose method is made visible, where