Thomas Kuhn

In the [YOU] on AI Field Guide

The cycle begins with the observation that software development has crossed a threshold that feels qualitatively different from previous tool upgrades. Kuhn explains why the feeling is accurate. What changed between the CLI and the GUI, between Fortran and Python, between procedural and object-oriented programming, were tools within a single meta-paradigm: the paradigm of manual code authorship, in which the fundamental activity was the translation of human intention into machine-executable instructions through the medium of formal language. The natural language interface changes that assumption at the root. The human no longer learns the machine's language. The machine learns hers. This is not a transition within the meta-paradigm. It is a replacement of the meta-paradigm itself.

The senior engineer in Trivandrum who oscillated between excitement and terror over five days of training was experiencing what Kuhn called the gestalt switch in real time. On Monday he perceived the AI tool as an assistant. By Wednesday the perception had reorganized: he was seeing a replacement for eighty percent of the activity that had defined his professional identity for fifteen years. Kuhn documents this reorganization across dozens of historical cases and calls it exactly what it is—not a change of opinion but a change of world. The same data, reorganized into a different pattern. The duck becoming the rabbit. And the crucial feature: once you see the rabbit, you cannot unsee it.

Kuhn also explains the discourse that [YOU] on AI documents: the camps that form, the positions that calcify, the people who debate a technology they have not yet seriously used. The defenders of the old paradigm are not being irrational. They are doing what practitioners have always done when anomalies mount: defending the framework that defines their competence. The converts are not being naive. They have completed the gestalt switch and genuinely cannot understand why anyone would cling to the old framework. The silent middle is suspended in the disorientation that Kuhn documented as the characteristic phenomenology of crisis: the moment when neither paradigm is fully present and the practitioner must hold both in an unstable superposition, which is exhausting in a way that neither commitment to the old framework nor conversion to the new one would be.

The framework also predicts a feature of the current moment that other analytical lenses miss: the incommensurability of the debate itself. When a senior developer insists that real programming requires understanding the machine at a low level, and a natural-language builder insists that directing a tool to produce a working result is real programming, neither is making an empirical claim resolvable by more data. They are expressing paradigmatic commitments. The same word—“programming”—indexes different conceptual structures. The resulting miscommunication looks like disagreement. It is, in Kuhn's precise sense, incommensurability, and it will not be resolved by better arguments. It will be resolved by the generational replacement that renders the debate moot.

Origin

Thomas Samuel Kuhn was born in Cincinnati in 1922, trained as a physicist at Harvard, and came to the history of science through a single pedagogical assignment: he was asked to teach a course on the history of science to humanities students. The assignment required him to read Aristotle's physics carefully for the first time, and what he found astonished him. Aristotle was not a bad physicist who had made mistakes that Newton corrected. Aristotle was operating inside a completely different conceptual framework, in which the questions physics addressed, the standards by which answers were evaluated, and the very meaning of terms like “motion” and “force” were organized differently. To understand Aristotle, Kuhn had to set aside the Newtonian framework entirely and enter another world. That experience produced the central observation of The Structure of Scientific Revolutions.

The book appeared in 1962 as Volume II, Number 2 of the International Encyclopedia of Unified Science and sold millions of copies—a peculiar success for a monograph in the philosophy and history of science. Its core argument was that scientific communities are organized around paradigms: shared exemplars, problems, methods, and standards that define what counts as a puzzle and what counts as a solution. Within a paradigm, practitioners do what Kuhn called normal science—demanding, often brilliant, but framework-bound puzzle-solving that does not question the paradigm's foundations. Anomalies accumulate. At first they are absorbed by ad hoc adjustments, the equivalent of Ptolemy's epicycles. Eventually the adjustments become more complex than the theory they are meant to save, and a crisis emerges. The crisis resolves not by the gradual refinement of the old framework but by its replacement: a new paradigm that reorganizes the field's puzzles, methods, and standards in a way the old paradigm cannot accommodate.

Kuhn spent the rest of his career—at Berkeley, Princeton, and finally MIT until his death in 1996—refining and defending his most controversial claim: incommensurability. He did not mean that practitioners in different paradigms cannot communicate. He meant that certain key terms change their meaning across paradigm boundaries in ways that make direct translation impossible without remainder. “Mass” in Newtonian mechanics and “mass” in Einsteinian mechanics are not synonyms operating in different contexts. They are different concepts sharing a name—and the difference is invisible in casual conversation, visible as a source of genuine misunderstanding when the conversation turns to foundations.

Aristotle

Key Ideas

Normal science and the paradigm. Kuhn's first claim is that most science most of the time is not revolutionary. It is the demanding, productive, cumulative puzzle-solving that takes place within an established framework. The framework—the paradigm—defines the relevant puzzles, the acceptable methods, and the standards by which solutions are judged. Expertise, in Kuhn's account, is paradigm-relative: the Ptolemaic astronomer who could predict planetary positions to within arc-minutes possessed genuine expertise, demonstrably useful and hard-won, that became irrelevant—not wrong, but irrelevant—when the paradigm shifted. The analogy to coding expertise in the age of large language models is exact.

The accumulation of anomalies. Paradigm shifts do not begin with the new paradigm. They begin with failure. Anomalies are observations that the existing framework cannot accommodate. They are first absorbed by ad hoc adjustments; the adjustments multiply; eventually they become more complex than the theory they are meant to save, and practitioners lose confidence in the framework itself. The AI anomalies—from Copilot's autocomplete through ChatGPT's interview performance through the December 2025 Google engineer demonstration—followed this pattern. Each individual anomaly was accommodatable. The accumulated pattern was not.

Crisis and the gestalt switch. The crisis phase of a paradigm shift has a specific phenomenology: not the gradual revision of belief but the sudden reorganization of perception. Kuhn compared it to the duck-rabbit gestalt switch. The same data, suddenly organized into a different pattern. One moment you see the duck; the next moment the rabbit; and once the switch occurs, you cannot un-see the new pattern even if you can, with effort, switch back. The engineer in Trivandrum who saw an assistant on Monday and a replacement on Wednesday was experiencing a gestalt switch. The speed of the AI transition amplifies its pathological consequences: the switch that Kuhn documented taking decades in the history of science is now taking days.

Incommensurability. Across paradigm boundaries, key terms change their meaning in ways that prevent complete translation. The practitioner who insists that “real programming” requires machine-level understanding and the practitioner who insists that directing an AI tool is programming are not disagreeing about an empirical fact. They are operating within different conceptual frameworks that assign different meanings to the same word. This incommensurability explains the frustration of the AI discourse—why better arguments and more evidence do not resolve the debate—and it predicts that the debate will be resolved not by persuasion but by generational replacement, exactly as Kuhn documented in every previous paradigm shift.

The textbook problem. After every paradigm shift, the textbooks are rewritten to present the history of the field as a smooth, cumulative progression toward the current paradigm. The discontinuities are smoothed. The losses are omitted. The practitioners who were displaced are absent from the narrative. Kuhn argued that this retrospective reconstruction serves the social function of legitimizing the current paradigm—and systematically misinforms the next generation about the structure of the transitions they will face. The AI discourse is already engaged in textbook revision, presenting the history of computing as a smooth arc from assembly to natural language. The revision conceals the genuine losses at each transition and misinforms everyone about the structure of the one they are currently inside.