The bottom line
Science advances through long stretches of puzzle-solving inside a shared paradigm, broken by revolutions that swap in an incompatible one. For product, growth or research teams, your framework decides what you measure, and new approaches win by persuading people, not by one proof.
Key takeaways
- A field gets productive once its community shares one concrete model of good work. The shared model ends arguments over basics, so people can dig into precise problems. Ch. II · The Route to Normal Science
- In the first half of the eighteenth century, electricity had nearly as many theories as major experimenters. Franklin's explanation of the Leyden jar, a device built by at least two investigators in the early 1740's, became the common paradigm.
- Sometime between 1740 and 1780, electricians stopped rebuilding foundations and began writing articles for each other, even though few new kinds of phenomena were added.
- Normal work solves puzzles whose answer is expected but whose route is unknown. A solution assured by the model is what lets skilled people commit to hard, narrow problems. Ch. IV · Normal Science as Puzzle-solving
- Through the eighteenth century, attempts to derive the moon's motion from Newton's laws kept failing. Some proposed changing the inverse square law. Scientists kept the rules, and in 1750 one of them found how to apply them.
- Precise expectations are what make a real surprise visible. An anomaly only shows up against a clear prediction, and recognizing it takes time. Ch. VI · Anomaly and the Emergence of Scientific Discoveries
- Roentgen saw a shielded screen glow during routine work and spent seven hectic weeks probing it. X-rays emerged in Würzburg between November 8 and December 28, 1895. Another investigator had seen the glow and discovered nothing.
- Priestley and Lavoisier tested their gas with two volumes to one volume of nitric oxide and misidentified it. When Priestley later used four times the nitric oxide, almost no residue remained.
- Switch frameworks only when the current one keeps failing at problems it once handled. Alternatives are easy to invent but get ignored until a crisis makes retooling worth its cost. Ch. VII · Crisis and the Emergence of Scientific Theories
- After 1890, ether-drift experiments kept failing and theories multiplied, including Lorentz's and Fitzgerald's. Einstein's special relativity emerged in 1905.
- Aristarchus fully anticipated heliocentrism in the third century B.C. It was ignored because geocentric astronomy had no needs a heliocentric system could meet.
- A new framework changes what people see and record, not just how they read the same data. The model chooses the measurements, so the numbers shift with it. Ch. X · Revolutions as Changes of World View
- On at least seventeen occasions between 1690 and 1781, astronomers recorded a star where Uranus must have been. Once Herschel's 1781 object was accepted as a planet, astronomers had several fewer stars and one more planet.
- Chemists stopped writing that two carbon oxides held 56 per cent and 72 per cent oxygen. They wrote that one weight of carbon combines with 1.3 or 2.6 weights of oxygen, and a 2:1 ratio stood out.
- A new approach wins through a growing shift in allegiance, not a single proof. Rivals disagree on problems and standards, so early adopters back future promise. Ch. XII · The Resolution of Revolutions
- Poisson, an opponent, showed that Fresnel's wave theory implied a white spot at the center of a disk's shadow. When Fresnel demonstrated the spot, French resistance collapsed suddenly and fairly completely.
- Copernicanism won few converts for almost a century after Copernicus died. Priestley never accepted the oxygen theory.
though the world does not change with a change of paradigm, the scientist afterward works in a different world.
Thomas S. Kuhn, Revolutions as Changes of World View
1Science only looks cumulative because its textbooks are rewritten
The picture of steady, brick-by-brick progress is produced by sources that get rewritten after every revolution, so the real engine of change, reformulating questions, drops out of the record.
- Textbooks are rebuilt after each revolution, and the rebuilt version erases the revolution itself. Ch. XI · The Invisibility of Revolutions
- Textbooks, popularizations and textbook-style philosophy all describe a settled body of problems, data and theory. Their job is to teach the current result, not to report how it was won.
- When the language, problems or standards of a field change, the textbook must change too. Practitioners then sense their field's history only back to the latest shift.
- Laying out experiments, laws and theories one at a time is good teaching. Combined with an unhistorical tone, it suggests that science grew toward today's goals one fact at a time.
- For teams, polished docs, onboarding decks and retrospectives work the same way. The original record is the only place where the changed questions still show.
- The rewrite follows a pattern. Teach the new vocabulary and standards. Cut what came before. Add back a little history in an intro or scattered hero references. Keep only the past work that looks like a contribution to today's problems.
- A scientist who lived through a revolution has a longer sense of history than the textbook gives. Most practitioners did not live through one.
- Some of the distortion is useful. It keeps a profession focused and secure. The cost is a false image of how discovery works.
- Discarded theories were as scientific as current ones, so being dropped does not make an idea a myth. Ch. I · Introduction: A Role for History
- Close study of Aristotelian dynamics, phlogistic chemistry and caloric thermodynamics found them, taken as wholes, no less scientific and no more idiosyncratic than present views.
- That leaves two options. Either myths come from the same methods as knowledge, or science has held beliefs incompatible with ours. Historians must take the second, and the accretion picture breaks.
- Questions like who first conceived energy conservation get harder to answer as research piles up. That suggests the questions are framed wrongly.
- The newer, Koyré-style history asks how Galileo's views related to those of his teachers, contemporaries and immediate successors, not to modern science. It reads past opinions from whichever viewpoint gives them the most internal coherence and the closest fit to nature.
- Method alone does not pick a unique belief. A newcomer to electricity or chemistry who knows how to be scientific can still reach several incompatible conclusions. The outcome depends on beliefs carried in from other fields, which experiment comes first, and which features of the result seem relevant.
Out-of-date theories are not in principle unscientific because they have been discarded.
- Heroes keep their names while their actual work gets revised to fit the current model. Ch. XI · The Invisibility of Revolutions
- Newton credited Galileo with finding that a constant gravitational force produces motion proportional to the square of the time. Galileo's framework could not even ask that question.
- Scientists rewrite their own stories too. Dalton's accounts of his atomism make his career look linear.
- Galileo's theorem takes that form only inside Newton's dynamical concepts. Galileo's work on falling bodies rarely mentions forces, let alone a uniform one. The credit hides the reformulation of questions that drove the move from Aristotelian to Galilean to Newtonian dynamics. That reformulation mattered more than new empirical findings.
- Dalton gave three incompatible accounts of how his atomism began. All three imply an early interest in combining proportions. Those problems seem to have occurred to him only along with their solutions, near the end of his creative work. All three leave out his real move, importing questions and concepts from physics and meteorology into chemistry.
- Concepts keep their verbal definitions while their meaning shifts with context. Ch. XI · The Invisibility of Revolutions
- Element, time, energy, force and particle are often neither invented nor discovered. The wording stays fixed while the surrounding concepts, procedures and applications change.
- Theories and the facts they fit emerge together from a reformulation. They do not grow piece by piece around facts that were always there.
- Chemistry texts credit Robert Boyle's Sceptical Chymist with the concept of element. Boyle said his definition only paraphrased a traditional one. He used it to argue that no chemical element exists. The wording runs back at least to Aristotle and forward through Lavoisier.
- Boyle's real role was leading a revolution that changed how "element" connected to chemical manipulation and theory. Lavoisier's revolution was still needed to give the concept its modern form.
- Einstein gave "space" and "time" new meaning without inventing or explicitly redefining them. He changed their context.
- Energy conservation looks like a higher-level law that conflicts with nothing. It looks that way only in hindsight, after caloric theory fell.
2A field gets productive once it shares one model of good work
A group starts making visible progress only when it agrees on one concrete model of good work, and until then effort scatters across rival schools.
- Without a shared model, every school rebuilds the field from scratch and collects facts almost at random. Ch. II · The Route to Normal Science
- Each school stresses the phenomena its own theory handles best and patches the rest ad hoc.
- With no model, every fact looks equally relevant, so people gather whatever data is close at hand, often from crafts.
- Each writer addresses rival schools as much as nature. The members are real investigators, but the net output falls short of a science.
- Optics from antiquity to the late seventeenth century had no single view of light. Epicurean, Aristotelian and Platonic schools treated light as particles from bodies, as a change in the medium, or as the medium interacting with an emanation from the eye.
- Early eighteenth-century electricity split three ways. One camp treated attraction as basic and repulsion as mechanical rebound. A second treated attraction and repulsion as equally basic but struggled with conduction. A third treated electricity as a flowing fluid and struggled with attraction and repulsion.
- Pliny and the seventeenth-century Baconian histories formed a morass. Heating by mixture sat beside the warmth of dung heaps. Chaff bouncing off a rubbed glass rod went almost unrecorded because it looked mechanical. Unconfirmable claims like heating by antiperistasis went in too.
- Facts never interpret themselves. If no theory is present, a metaphysic, another science or accident supplies one. Pre-paradigm work can still yield real discovery and invention, and many creative fields still work this way.
- A model wins by looking better than its rivals, not by explaining everything. Ch. II · The Route to Normal Science
- Divergence usually ends when one school's narrow focus produces a revealing result.
- Older schools fade as their members convert. Holdouts get read out of the profession and ignored.
- A clear model tells the group which experiments are worth running.
- The fluid camp's push to bottle electricity produced the Leyden jar, a device random exploration would rarely find. Franklin's theory still could not account for every known case of repulsion.
Truth emerges more readily from error than from confusion.
- Only rarely, as in ancient statics, dynamics and geometrical optics, did loosely gathered facts speak clearly enough to produce a first model on their own.
- Once a model is settled, the group professionalizes, works fast, and becomes hard for outsiders to follow. Ch. XIII · Progress through Revolutions
- Foundations move into textbooks. Research shifts from books for any reader to short papers for colleagues, with journals, societies and a place in the curriculum.
- Peers who share one set of standards judge the work, without pressure from lay demands. So members finish problems faster and pick ones they expect to solve, not merely urgent ones.
- The gap with outsiders comes from the very mechanism of advance. It is not just an unfortunate side effect.
- Cavendish, Coulomb and Volta seem further from Gray, Du Fay and even Franklin than those writers are from the sixteenth century.
- Mathematics and astronomy stopped being readable to educated laymen in antiquity. Dynamics became esoteric in the later Middle Ages and briefly regained readability in the early seventeenth century. Electricity needed translation before 1800, and most physical sciences followed in the nineteenth century.
- Mature fields then move from model to model. Optics went from Newton's corpuscles to Young and Fresnel's waves to photons. Medicine, technology and law, driven by outside social need, do not follow the pattern of professionalizing after adopting a model.
- Fields that keep debating whether they are "scientific" are really missing consensus on what they have achieved. Ch. XIII · Progress through Revolutions
- Within any single school, good work always counts as progress. Progress looks doubtful only when rival schools attack each other's foundations.
- Psychology and other social sciences argue at length about their status. Economists argue less, likely because they agree more about economics.
- Agreement on achievements settles the worry. A definition does not.
- While painting aimed at representation, it counted as cumulative. Pliny and Vasari logged foreshortening and chiaroscuro as steps forward, and Leonardo moved freely between art and science. The split came only when painting dropped representation as its goal.
- Every recorded civilization had technology, art, religion, politics and law. Only those descended from Hellenic Greece had more than rudimentary science, and most of it comes from Europe in the last four centuries. Communities that support this kind of work are rare.
- Whether parts of social science have a first model at all remains open. Even in biology, some first models, such as the one for heredity, are recent.
3The shared model is worked examples held by a specific group, not a rulebook
A group's real common ground is a set of solved problems that members extend by resemblance, and that set belongs to a specific community, not to a topic.
- Members share concrete solved problems, not an explicit set of rules. Ch. V · The Priority of Paradigms
- A group's standard examples are easy to find. They recur in its textbooks, lectures and lab exercises.
- Its rules are hard to pin down. Any wording comes out a little too strong, and some members would reject it.
- New cases get handled by likeness to accepted ones, the way people apply a word like "game".
- Scientists can agree that Newton, Lavoisier, Maxwell or Einstein settled a set of problems for good. They can still disagree, often without noticing, about what made those solutions work.
- Wittgenstein found that nothing is shared by all games and only games. People call a new activity a game because of overlapping, crisscrossing likenesses to known games. That is a family resemblance.
Lack of a standard interpretation or of an agreed reduction to rules will not prevent a paradigm from guiding research.
- Practitioners state the hypotheses behind current work well. On their field's foundations and legitimate methods, they do little better than laypeople. Their skill shows in the doing.
- Arguments over rules flare up only when the shared model feels shaky. Ch. V · The Priority of Paradigms
- While solutions are accepted without question, work runs without rule talk.
- Deep debates over methods and standards cluster in pre-paradigm periods and just before and during revolutions.
- Quiet about first principles is not proof of agreement. The question may just feel irrelevant.
- Pre-paradigm optics, electricity, seventeenth-century chemistry and early nineteenth-century geology were full of such debates. The debates defined rival schools and produced no agreement.
- Rule debates returned when Aristotelians, Cartesians and Leibnizians contested Galileo's and Newton's mechanics. They came back with Maxwell's theory, statistical mechanics and the move to quantum mechanics, where some continue.
- People learn a framework's terms by solving its problems and seeing new ones as like old ones. Postscript, 1969
- A student grasps "force" and "mass" more from applying them than from incomplete textbook definitions.
- Students often understand a chapter and still fail its end problems. They succeed once a problem looks like one already solved.
- A general form like f = ma is a law-sketch. It gets rewritten for each case, such as free fall or the pendulum.
- Galileo saw a ball rolling down one incline and up another as like a point-mass pendulum. Huyghens solved a physical pendulum's center of oscillation by treating it as Galilean point-pendula freed mid-swing. Daniel Bernoulli treated water leaving an orifice like Huyghens' pendulum and derived its speed.
- All three used one law, vis viva: "Actual descent equals potential ascent." Stated alone, the law is nearly useless without those problems.
- A theory is never even a candidate for acceptance without concrete applications.
- This knowledge is tacit but not private. The group shares it, tests it and passes it on through training.
- A framework belongs to a group, so find the group before you look for its framework. Postscript, 1969
- Large groups show up in degrees, society memberships and journals read.
- Small groups show up in special conferences, shared drafts and galley proofs, correspondence and citations.
- A revolution can happen inside fewer than 25 people without outsiders noticing.
- The phage group could not be found through field or journals. It had to be traced through its conferences, drafts and citation links.
- There was no physics community before the mid-nineteenth century. It formed by merging parts of mathematics and natural philosophy.
- Asked whether a helium atom is a molecule, a chemist said yes, because it acts like one in the kinetic theory of gases. A physicist said no, because it shows no molecular spectrum. Each answered from their own training problems.
- Quantum mechanics means different things to those who learned only its principles and those who learned its chemistry or solid state uses. X-rays were a revolution only for Kelvin, Crookes and Roentgen. Maxwell's equations were revolutionary for a small group of specialists.
- The broad shared framework has four parts, and worked examples are the deepest. Postscript, 1969
- The disciplinary matrix holds symbolic generalizations, models, values and exemplars.
- Exemplars also mark the lines between subgroups.
- One reader indexed at least twenty-two uses of "paradigm". Most were stylistic. Two genuinely different senses remained.
- To map a group's matrix, list its undisputed generalizations and note which act partly as definitions. List its models, from heuristic analogies to beliefs about what exists. List its values. Then compare its exemplars to find subgroup borders.
- Members need not share every model. Early nineteenth-century chemists shared constant proportion, multiple proportion and combining weights, yet argued sharply over whether atoms existed.
4Normal work is puzzle-solving: the answer is expected, the route is not
Most of a field's effort goes into making an early model's promise good, and a problem earns that effort when the model guarantees it has an answer.
- A new model is a ruling to extend, not a template to copy, and most careers go to extending it. Ch. III · The Nature of Normal Science
- A Latin verb pattern is copied onto other verbs. A common-law ruling is sharpened under new and harder conditions. A paradigm works like the ruling.
- At first it is narrow and imprecise. It wins because it handles a few problems the field finds pressing better than its rivals do.
- Its early success is mostly a promise, shown in a few incomplete examples. The follow-up work fills most working lives.
- Newton derived Kepler's Laws, some lunar deviations, and pendulum and tide data. With ad hoc assumptions he also derived Boyle's Law and the speed of sound. He treated pendulum bobs as mass points, mostly ignored air resistance, and ignored attraction between planets, so agreement was only approximate.
- Euler, Lagrange, Laplace and Gauss did some of their best work closing those gaps. They built the mathematics for many-body orbits, hydrodynamics and vibrating strings.
Mopping-up operations are what engage most scientists throughout their careers.
- The narrow focus hides phenomena that do not fit. It is also what forces study in great depth. When the model stops working, the restrictions loosen.
- Normal work falls into three classes: key facts, contact points, and articulation. Ch. III · The Nature of Normal Science
- Measure the facts the model marks as revealing, more precisely and in more situations. Precision measurers from Tycho Brahe to E. O. Lawrence built reputations this way without finding anything new.
- Build places where predictions meet nature directly, as Atwood's machine did for Newton's second law.
- Articulate the model by measuring constants, finding laws, and choosing among possible applications.
- Atwood's machine came almost a century after the Principia. Other demonstration devices include telescopes to show Copernican parallax, Foucault's apparatus showing light moves faster in air than in water, and a giant scintillation counter for the neutrino. Even now, general relativity has no more than three areas where it can be checked against nature.
- Newton's work implied a universal gravitational constant. His own problems did not need its value, and no one measured it for a century. Cavendish did in the 1790's. Similar work continues on the astronomical unit, Avogadro's number, Joule's coefficient and the electronic charge.
- Caloric theory could explain compression heating several ways. Gas might mix with a void that held heat, or specific heat might change with pressure. Experiments tried to pick between them, often run by the people who proposed them.
- On the theory side, computing useful facts such as ephemerides counts as hack work. Reformulation does not. Mathematical physicists from Euler and Lagrange to Hamilton, Jacobi and Hertz recast mechanics into clearer, more uniform equivalent forms.
- Useful measurement needs a qualitative model first. Ch. III · The Nature of Normal Science
- Boyle's experiments only made sense once air was seen as an elastic fluid that hydrostatics could describe.
- Pan-balance measurements of electrical force produced no consistent pattern. Coulomb's purpose-built apparatus did.
- Coulomb's design assumed every particle of electric fluid acts on every other at a distance. That force was the one most likely to be a simple function of distance. His results then refined the theory.
- Because the field already expected his result, it surprised no one. Several contemporaries had predicted it. The pan-balance results stayed isolated facts that drew little attention.
- Electron-scattering maxima were recorded before they meant anything. They became measures of electron wave length only after a theory predicted wave-like matter. The apparatus then had to be redesigned so readings tied cleanly to that theory.
- A problem is a puzzle when a solution is assured and rules bound it. Its importance is not the test. Ch. IV · Normal Science as Puzzle-solving
- Experts avoid repetitive work like computing ephemerides. They chase problems where the outcome is known but the route is not.
- Curing cancer or securing lasting peace may have no solution, so they are not puzzles.
- A jigsaw built from two random boxes probably cannot be solved, so it cannot test skill.
- A real jigsaw has rules. Every piece is used, face up, and interlocked without forcing. Scientific rules limit both acceptable answers and allowed steps.
- A result outside the expected range usually counts as the researcher's failure, not a fact about nature.
- The same test screens out problems the model cannot state. They get called metaphysical, assigned to another field, or set aside as too troublesome. Chasing pressing but unsolvable problems distracted parts of seventeenth-century Baconianism and some social sciences.
- On the moon's motion, swapping the inverse square law would have changed the paradigm. That would have set a new puzzle, not solved the old one.
- The rules come in four levels, but the paradigm runs deeper than any of them. Ch. IV · Normal Science as Puzzle-solving
- Explicit laws: Newton's Laws made quantity of matter a basic category. Fixed and definite proportions posed the atomic-weight problem.
- Instrument commitments: views on fire in chemical analysis, and the history of chromatography.
- Assumptions about what exists, plus commitments all scientists share.
- After about 1630, and especially after Descartes, most physical scientists assumed the world was microscopic corpuscles. Boyle therefore studied reactions that looked like transmutations, since they best showed corpuscles rearranging.
- Helmholtz met strong resistance from physiologists to the idea that physical experiments could illuminate their field.
- The shared general commitments are to understand the world, extend its order in precision and scope, examine nature in detail, and treat apparent disorder as a call to refine techniques or theories.
- Rules come from paradigms, not the reverse. A paradigm can guide work where no explicit rule exists.
5Precise expectations are what make surprises visible
Narrow, rule-bound work is the best source of real surprises, because a deviation only stands out against a sharp prediction, and turning it into a discovery takes a stretch of joint conceptual and empirical work.
- A discovery is a process that ends only when the surprise has become expected. Ch. VI · Anomaly and the Emergence of Scientific Discoveries
- It has three stages. Someone notices nature breaking an expectation, explores the area, then adjusts theory and procedures until the result is predictable.
- Knowing that something exists and knowing what it is are separate recognitions. Both take time.
- So "who found it, and when" often has no proper answer. Teams should budget for a period of reframing and measuring after the first odd signal, not wait for a single eureka.
- Oxygen had at least three claimants. Scheele, a Swedish apothecary, prepared it first but published only after others announced it. Priestley, heating red oxide of mercury, called it nitrous oxide in 1774 and air with less phlogiston in 1775.
- Lavoisier in early 1775 called it purer, more respirable common air. By 1777 he concluded it was a distinct constituent of the atmosphere. Yet until his death he held oxygen to be a 'principle of acidity' combined with caloric. Other chemists in the early 1770's likely had enriched air without knowing it.
- Long priority fights point to a flawed picture of discovery as an instant act, like seeing. The credit a discovery earns grows with how hard it broke expectations. Finds the model predicts, like new elements filling the periodic table, are normal work and add no new kind of fact.
- The split between discovering a fact and inventing a theory is artificial. A new fact is not fully a scientific fact until the theory has been adjusted to hold it.
- Expectations are built into instruments and standard tests, not only into theory. Ch. VI · Anomaly and the Emergence of Scientific Discoveries
- Choosing an apparatus assumes that only certain circumstances will arise.
- A finding that no theory forbids can still overturn practice when it shows the standard tools were misleading.
- Teams should audit what their metrics, dashboards and test procedures assume can occur. When a new finding implies the tools were producing it unnoticed, check which past results need redoing.
- Theory in 1895 did not forbid X-rays. It already accepted visible, infrared and ultraviolet radiation. Roentgen found the agent traveled in straight lines, cast shadows and was not deflected by a magnet. Yet Lord Kelvin first called X-rays a hoax.
- The shock came because cathode ray equipment in wide use must have been producing X-rays unnoticed. Apparatus now needed lead shielding, and earlier normal work had to be redone. X-rays also fed the crisis that led to twentieth-century physics.
- In the goodness-of-air test, experience said ordinary air leaves one volume of residue and other gases leave more. A result close to one volume led both chemists to misidentify their gas. Their commitment to the procedure amounted to a commitment that no gas like oxygen existed.
- Fission was identified late partly because researchers chose chemical tests aimed mainly at elements at the upper end of the periodic table. Dropping standard tests is no fix. Procedures are as necessary as laws, even though they limit what can be seen.
- An earlier suspicion that something is wrong lets one person see what peers miss. Ch. VI · Anomaly and the Emergence of Scientific Discoveries
- A surprising result does not change theory by itself. Awareness that the model is off is what prepares the observer.
- Teams should keep a written record of nagging doubts about their current models.
- In 1772, before any work on the gas, Lavoisier deposited a sealed note with the Secretary of the French Academy. It recorded that phlogiston theory was flawed and that burning bodies absorb part of the air.
- The oxygen work gave that suspicion its shape. It let Lavoisier read in Priestley's experiments what Priestley could not. His 1777 papers announced the oxygen theory of combustion, the keystone of the chemical revolution.
- In pre-paradigm or crisis periods, speculative theories can guide discovery too. The result usually differs from the prediction, and theory and experiment are refined together until a paradigm emerges.
- Resistance to novelty is useful, because the anomalies that break through reach the core. Ch. VI · Anomaly and the Emergence of Scientific Discoveries
- People first fit an anomaly into an existing category. Only extended exposure forces the category to change.
- The more precise and far-reaching the model, the more sensitive it is as an anomaly detector:
Anomaly appears only against the background provided by the paradigm.
- Resistance keeps people from being distracted by every blip. To get past your own resistance, look at the anomaly again and again.
- In the Bruner and Postman experiment, subjects identified playing cards at gradually longer exposures. Some cards were anomalous, such as a red six of spades or a black four of hearts. A run ended after two successive correct identifications.
- The steps were consistent. Anomalous cards were first confidently named as normal ones. Then came hesitation, a sense that something was wrong, and growing confusion. Categories then often shifted suddenly. After two or three such shifts, later anomalous cards caused little trouble.
- A minority never adjusted. Even at forty times the exposure needed for normal cards, more than 10 per cent of anomalous cards went unidentified, and some failing subjects showed acute distress. Whether this is a metaphor or a fact about the mind is left open.
- Professionalization narrows vision as the price of precision. Simultaneous discovery in several laboratories shows both how traditional normal work is and how completely it prepares its own overthrow.
6Retool only when the model fails at its own long-solved problems
A working model should be kept through ordinary anomalies and dropped only when its old, settled problems keep breaking and a replacement is ready to stand in its place.
- Anomalies are normal, so log them and keep working. Ch. VIII · The Response to Crisis
- Every model leaves problems unsolved. Those leftovers are the everyday puzzles of the work.
- Most discrepancies yield to ordinary practice in time. Waiting is often the right call while other problems remain open.
- A puzzle and a counterinstance are the same problem seen two ways. Only crisis plus a rival turns one into the other.
- Newtonian predictions long disagreed with the measured speed of sound and with Mercury's motion. Nobody seriously questioned the theory. The sound gap closed through heat experiments done for other reasons. Mercury's gap vanished with general relativity, after a crisis it played no part in creating.
- A model with nothing left unsolved stops generating research. Geometric optics seemed complete and became an engineering tool.
There is no such thing as research without counterinstances.
- An anomaly deserves escalation when it threatens core generalizations, blocks an important application, gains weight from new tools, or resists solution for a long time. Usually several of these combine. No general rule settles which anomalies matter.
- Crisis means old, supposedly solved problems keep failing, and it has visible symptoms. Ch. VII · Crisis and the Emergence of Scientific Theories
- Fixes move the error elsewhere instead of removing it.
- Complexity grows faster than accuracy.
- Versions of the model multiply until no one can say exactly what it is.
- Failure on a new kind of problem only disappoints. Failure on a settled one shocks, and that shock is the signal.
- The Ptolemaic system, built over the last two centuries before Christ and the first two after, predicted planets as well as Copernicus later did. Astronomers patched each mismatch with compounded circles. Alfonso X in the thirteenth century and Domenico da Novara in the sixteenth registered the failure. Copernicus called the inherited tradition a monster.
- Chemistry hit crisis in the 1770s. Joseph Black showed in 1756 that fixed air differed from normal air. Cavendish, Priestley and Scheele, all phlogiston believers, then found so many gases that there were nearly as many versions of the theory as pneumatic chemists. Weight gain on roasting drew patches such as phlogiston with negative weight.
- Once the wave theory was accepted around 1815, ether drift became a normal problem. Fresnel, Stokes and others built ether-drag versions that explained null results, Michelson and Morley's included. Maxwell's theory could not easily absorb ether drag, so those old null results turned anomalous.
- Never drop a model without a replacement, and expect alternatives to be ignored until crisis. Ch. VIII · The Response to Crisis
- Rejecting a model with nothing in its place means leaving the field, because there is no work without one.
- Rival ideas are easy to invent. Retooling is costly, so it waits until crisis shows it is needed.
- Pitch a new idea against a recognized, painful failure. Ideas with no trouble spot to fix get ignored.
- Rey, Hooke and Mayow proposed combustion-by-air theories in the seventeenth century. They were ignored because they touched no recognized trouble spot.
- Leibniz and others attacked Newton's absolute space in the late seventeenth century. The critique was purely logical, tied to no observational problem, and it died out early in the eighteenth.
- Copernicus' system was neither simpler nor more accurate than Ptolemy's. The recognized crisis, not the system's merits, drew astronomers to it.
- Rejection is always a two-way comparison: both models against nature and against each other. No historical case looks like falsification by direct comparison with nature alone.
- Run crisis as extraordinary research, and look to newcomers for the new model. Ch. VIII · The Response to Crisis
- Crisis ends one of three ways. The old model solves the problem after all, the problem is shelved for later tools, or a battle over a new candidate begins.
- Isolate the anomaly, push old rules to their limit, magnify the breakdown, run experiments just to see, and float cheap, disposable hypotheses.
- New models usually come from the very young or people new to the field.
- The same problems read differently. Copernicus saw counterinstances where Ptolemy's successors saw puzzles. Lavoisier saw one where Priestley saw a solved puzzle. Einstein did the same against his peers in physics.
- Months before Heisenberg's matrix mechanics paper, Pauli wished he had been a movie comedian. Less than five months later he said it gave him hope and joy in life again.
- Crisis clusters discoveries. Malus found polarization by reflection while starting work on the Academy's prize essay on double refraction, a known trouble area. The neutrino hunt and the parity hypothesis were attempts to localize diffuse anomalies.
- Timing varies. Thomas Young described the wave theory at a very early stage of an optics crisis. Einstein saw how black-body radiation, the photoelectric effect and specific heats were related before he had any new theory. More often no such structure appears in advance.
7A new framework changes what you see, measure and count as a problem
A revolution replaces the old model with an incompatible one. The new model decides which data get collected, how things are grouped, and which problems are worth solving.
- A successor model does not contain its predecessor, and walling off the old one to keep it safe stops research. Ch. IX · The Nature and Necessity of Scientific Revolutions
- Newton is an approximation of relativity, not a special case of it. Laws derived from Einstein still use Einsteinian space, time and mass.
- Newtonian mass is conserved. Einsteinian mass converts with energy. The same word names different things.
- Phlogiston theory explained why bodies burn, why metals share more properties than their ores, and why air shrinks during confined burning. Had its claims stopped there, nothing could have challenged it.
- The test runs in steps. Write the new laws as E1 to En. Add a limit such as (v/c)2 << 1. Derive N1 to Nm, which look like the old laws. Then check whether the variables still carry the new meanings. If they do, you have not recovered the old theory. If you redefine them, you have not derived it.
- The derivation still has a use. It explains why Newton seemed to work, and it justifies a driver or a surveyor acting as if the old laws hold.
- Restricting a theory to the evidence it already covers is logically sound. The cost is that a model with no reach beyond its precedents produces no puzzles, no surprises and no crises.
- Each model sets its own legitimate problems and standards, so rival camps argue in circles. Ch. IX · The Nature and Necessity of Scientific Revolutions
- The new science of the seventeenth century treated "nature" explanations as word-play. Molière mocked a doctor who said opium causes sleep through its dormitive potency.
- Lavoisier gave up explaining what makes a substance metallic or acidic. For much of the nineteenth century, failing to explain the qualities of compounds was not counted against a chemical theory.
- Once Maxwell's theory took hold, building mechanical ethers stopped being a standard problem.
- Newton's innate gravity looked as occult as the scholastic tendency to fall. Neither its critics nor its defenders won the argument. By the mid-eighteenth century scientists accepted innate attractions almost universally, because they could not work without the Principia. That was a reversion, not a step backward.
- Standards can reverse. Quantum mechanics brought back explanations of color and state of aggregation. Einstein explained gravitational attraction. A model is
not only with a map but also with some of the directions essential for map-making.
- After a switch, people see new things with old instruments in old places. Ch. X · Revolutions as Changes of World View
- Western astronomers first saw change in the heavens within half a century of Copernicus. The Chinese, whose cosmology allowed change, had recorded new stars and sunspots centuries before Galileo.
- Ready to find planets, astronomers identified twenty asteroids with standard instruments in the first fifty years of the nineteenth century.
- Once Hauksbee's large apparatus magnified electrical repulsion, repulsion became the basic sign of electrification. Attraction became the thing that needed explaining.
- Subjects wearing inverting lenses at first see the world upside down and become badly disoriented. After a period of confusion, the whole visual field flips back. Absorbing the anomaly changed the perception itself.
- Late sixteenth-century astronomers, some using tools as simple as a piece of thread, kept finding comets moving through space once reserved for planets and stars.
- A child who narrows "mama" from all humans to all females and then to one person learns more than a word. Expectations and the perceived world change together.
- The model decides what gets measured, so the numbers shift with it. Ch. X · Revolutions as Changes of World View
- Aristotelians saw a swinging body as constrained fall. They would track its weight, the height it was raised, the time to reach rest, and the resistance of the medium.
- Galileo saw a pendulum. He measured weight, radius, angular displacement and time per swing, the exact data that yield the pendulum laws.
- Galileo's view came out of the impetus paradigm. In the fourteenth century Buridan described a vibrating string as impetus implanted, used up, and restored. Later that century Oresme gave the first known account of a pendulum. Before then there were only swinging stones.
- Galileo's vision was not more accurate. He claimed the period stays the same for amplitudes as great as 90°, more regularity than anyone finds today.
- Aristotle measured total distance and total time, which gives what is now called average speed. Impetus theory later split speed into average and instantaneous.
- When camps classify the same things differently, more evidence cannot settle the dispute. Ch. X · Revolutions as Changes of World View Postscript, 1969
- Proust said all reactions occur in fixed proportions. Berthollet said they do not. Where Berthollet saw a variable compound, Proust saw a mixture.
- Dalton made fixed proportions true by definition. Any reaction without them simply did not count as chemical.
- Ohm's law met resistance because it redefined "current" and "resistance." A shift can also move whole classes, as when metals went from compounds to elements.
- Affinity theory had classed salt in water, alloys, glass and oxygen in the atmosphere as compounds. Dalton treated air as a mixture, which created a new anomaly: he could never explain why the heavier oxygen did not settle out.
- Proust measured an oxygen ratio of 1.47:1 for the two copper oxides, not the 2:1 the atomic theory required. Chemists adopted Dalton anyway. Berthollet was never convinced.
- It took almost another generation to bring nature into line. Afterward even the percentage composition of well-known compounds had changed.
8New approaches win by persuasion, and progress has no fixed destination
A new approach is judged only against a real rival, wins converts gradually on promise, and moves the field forward without moving it toward any preset truth.
- An approach is never tested alone against reality, only against an actual competitor. Ch. XII · The Resolution of Revolutions
- During normal work, a failed attempt tests the attempt, not the framework. It is like a chess player trying a move.
- Falsification fails as a rule because every theory has misfits. If any misfit forced rejection, every theory would always be rejected.
- "Does it fit the facts?" has no precise answer. "Which of these two fits better?" does.
- Neither Priestley's theory nor Lavoisier's matched the observations exactly. Even so, few contemporaries needed more than a decade to judge Lavoisier's the better fit.
Verification is like natural selection: it picks out the most viable among the actual alternatives in a particular historical situation.
Nobody can tell whether a better option existed among ideas that were never available.- Rejection really involves two steps. First an anomaly calls up a competitor. Later, rejecting the old view and verifying the new one happen as a single step.
- Requiring only "severe" misfit does not rescue falsification. It needs a measure of degrees of falsification, and that measure runs into the same problems as probability-based theories of verification.
- No neutral procedure picks the winner, and that spreads risk across the group. Postscript, 1969
- Shared values such as accuracy, simplicity, consistency and plausibility are weighed differently by each person.
- Because of that, a few people bet on the new idea while most keep doing productive work. In effect the community makes the choice, not any single member.
- Einstein saw an inconsistency in the old quantum theory as intolerable, something that blocked normal work. Bohr and others saw the same issue as a difficulty that ordinary work would eventually clear up.
- Reasons outside science count too. Sun worship helped make Kepler a Copernican. Personality, nationality and the innovator's reputation also sway who converts.
- Debates over theory choice cannot take the form of proof, because the disagreement is over premises. Good reasons still exist and still decide the matter for the group.
- Early adopters back future promise on faith, and their development work produces the arguments that convert everyone else. Ch. XII · The Resolution of Revolutions
- The strongest arguments come later: solving the crisis problem, clearly better quantitative precision, and predictions nobody expected.
- Early on, the incumbent can be patched and still wins outside the crisis area.
- Kepler's Rudolphine tables beat any tables computed from Ptolemy's theory, and they converted many astronomers. Planck's radiation law and the Bohr atom persuaded through quantitative success, even though both raised more problems than they solved.
- Copernicus's theory was no more accurate than Ptolemy's at first. About sixty years after his death, the telescope revealed mountains on the moon, the phases of Venus and countless unknown stars. That evidence won converts, especially among non-astronomers.
- Einstein apparently did not expect general relativity to account exactly for the long-known anomaly in Mercury's perihelion. Old data can still surprise.
- Old frameworks fight back. Tycho Brahe's earth-centered system answered the challenge from Copernicus with real success. Before the composition of water was known, hydrogen combustion favored phlogiston over Lavoisier.
- Allegiance shifts a few people at a time, and holdouts are not irrational. Ch. XII · The Resolution of Revolutions
- Lord Kelvin never accepted the electromagnetic theory.
- Confidence that the old model will solve its problems is the same confidence that makes normal work possible.
- In a deadlock, translate between the camps instead of trying to prove a point.
- At the close of the Origin of Species, Darwin set his hopes on young and rising naturalists. Planck said new truths win because their opponents eventually die out.
- Seemingly decisive proofs arrive after the switch. The Foucault pendulum showed the earth's rotation, and the Fizeau experiment showed light is faster in air than in water. Both came long after the new views were accepted, so they serve later textbooks, not the original debate.
- Translation steps: treat the other side as a different language community. Find the terms that cause trouble only between groups. Use shared everyday words to clarify them. Ask what the other person would see and say in a given situation. Never explain their view as error or madness.
- Being persuaded is not the same as going native. Many people who met relativity or quantum mechanics in midcareer accepted them but used them like foreigners, relying on practitioners trained in them.
- Progress is real as more solved problems, but it moves away from what was known, not toward a fixed truth. Ch. XIII · Progress through Revolutions
- The unit of achievement is the solved problem. The number of solved problems and their precision keep growing, though the field's breadth may shrink.
- Winners rewrite history as a straight line and do not see what was lost.
- Before adopting a new approach, a community needs three things. Nature must first make past results look shaky. The candidate must solve a recognized problem nothing else can solve. It must also keep much of the old problem-solving ability.
- Copernicus destroyed an old explanation of terrestrial motion and offered no replacement. Science has no equivalent of an art museum, so discarded work simply disappears from view.
- Each theory solves more puzzles than the one before, yet the picture of reality shows no steady direction. In some respects Einstein's general relativity is closer to Aristotle than either is to Newton.
- Darwin faced the same obstacle. By 1859 evidence for evolution had built up for decades, and Lamarck, Chambers and Spencer had all described evolution as following a plan. What many readers found hardest to accept was that Darwin removed the goal entirely.
9Put it to work
Name the shared model your team works from, write down what it predicts before each launch, and treat stubborn surprises as signals to investigate rather than noise.
- Agree on one concrete model of good work Pick a framework the team will build on, even if it leaves some facts unexplained. A shared model ends the re-arguing of basics and frees people for narrow, hard problems. Ch. II · The Route to Normal Science
- The trap is waiting for a model that explains every observation. No winning model ever does. It only has to beat its rivals.
- Electricity had nearly as many theories as major experimenters until Franklin explained the Leyden jar. That explanation became the common model, and work shifted to articles written for other electricians.
- Teach the model through worked examples Onboard people with solved past projects and practice problems, not only principles. People learn what the terms mean by applying them. Ch. V · The Priority of Paradigms
- The same words can mean different things to different specialists. Asked whether a single helium atom is a molecule, a chemist said yes and a physicist said no, both at once. Each answered from the problems they had been trained on.
- State what you expect before every test or launch A real surprise only shows up against a precise prediction. Also check what your standard metrics and procedures quietly assume can happen. Ch. VI · Anomaly and the Emergence of Scientific Discoveries
- Priestley and Lavoisier both tested their gas with two volumes to one of nitric oxide, and both misidentified it. The fixed procedure hid the answer until Priestley later used four times the nitric oxide and almost no residue remained.
- Log anomalies, but keep working the model Every model has misfits, and most of them yield to ordinary work in time. Escalate an anomaly only when it hits core assumptions, blocks an important use, or has resisted for a long time. Ch. VIII · The Response to Crisis
- For sixty years the predicted motion of the moon's perigee was only half the observed value. In 1750 Clairaut showed the error lay in the math, and Newton's theory stood.
- Do not drop a model until a ready alternative exists to compare it against. Rejecting one model without replacing it leaves no way to do the work at all.
- Watch for the signs of real crisis Count how many competing versions of your core model the team now holds. Check whether patches add complexity faster than accuracy, and whether long-solved problems have started failing. Ch. VII · Crisis and the Emergence of Scientific Theories
- Ptolemaic astronomers fixed each mismatch by adjusting their circles. Each fix caused errors somewhere else.
- By the early 1770s there were nearly as many versions of phlogiston theory as pneumatic chemists. That spread of versions was the symptom.
- When you switch models, re-check what you measure A new framing changes which numbers matter and how they get recorded. Re-express old data in the new form, and audit which old tests no longer apply. Ch. X · Revolutions as Changes of World View
- Chemists stopped recording two carbon oxides as 56 per cent and 72 per cent oxygen. They wrote one weight of carbon combining with 1.3 or 2.6 weights of oxygen, and a 2:1 ratio stood out.
- Win adoption by persuasion, not proof Recruit a few champions, often newcomers, and let their work produce the stronger arguments. Lead with the incumbent's known failures and with predictions nobody saw coming. Ch. XII · The Resolution of Revolutions
- Poisson, an opponent, showed that Fresnel's wave theory implied a white spot at the center of a disk's shadow. When Fresnel demonstrated the spot, French resistance collapsed suddenly and fairly completely.
- Plan for slow conversion. Copernicanism won few converts for almost a century after Copernicus died, and Priestley never accepted the oxygen theory.
10The book's words
Nine terms cover how a working model rises, does its job, breaks and gets replaced.
- Paradigm A concrete achievement a group accepts as its model of how the work should be done.
- Franklin's account of the Leyden jar played this role for eighteenth-century electricians.
- Normal science Everyday work inside one paradigm that extends its reach and tightens its fit with the facts.
- Newton's work implied a universal gravitational constant. Cavendish measured it only in the 1790's, a century after the Principia.
- Puzzle A problem whose solution is assured but whose route nobody yet knows.
- A jigsaw built from pieces of two different boxes is not a puzzle, because no solution is guaranteed.
- Anomaly A result that keeps refusing to match what the paradigm says should happen.
- Roentgen saw a shielded screen glow during routine work. Another investigator had seen the glow and discovered nothing.
- Crisis A stretch when a paradigm keeps failing at its own core problems and its rules start to loosen.
- After 1890, ether-drift experiments kept failing and theories multiplied before special relativity emerged in 1905.
- Scientific revolution The replacement of a paradigm by an incompatible new one, at any scale.
- A revolution can happen inside a group of fewer than twenty-five people and look like nothing to outsiders.
- Incommensurable Rival paradigms differ in their problems, standards and word meanings, so no single scale can measure both.
- Critics called Copernicus mad because part of what "earth" meant was a fixed position.
- Exemplar A concrete solved problem that teaches members to see a new case as like an old one.
- Galileo's inclined plane, Huyghens' pendulum and Daniel Bernoulli's flowing water all rest on one law. The law is nearly useless without the problems that show how to apply it.
- Disciplinary matrix The full set of a group's shared commitments: symbolic generalizations, models, values and exemplars.
- The term replaced the broad sense of "paradigm" after a reader found the word used in at least twenty-two ways.
11Chapter by chapter
Where each idea is in the book.
- I Introduction: A Role for History Science looks cumulative only because its image comes from textbooks. Its actual history shows long stretches of rule-bound normal science broken by revolutions that replace a community's basic commitments and change the world scientists work in.
- II The Route to Normal Science A field becomes a mature science when its community accepts a shared paradigm, a concrete achievement that ends debate over fundamentals and directs research toward precise, esoteric problems. Before that, rival schools gather facts almost at random and each rebuilds the field from scratch.
- III The Nature of Normal Science Normal science is the detailed mop-up work of turning a paradigm's early, partial promise into actual success. Its narrow focus, which comes from trust in the paradigm, is what lets science study nature in depth.
- IV Normal Science as Puzzle-solving Normal science does not try to produce major novelty. It solves puzzles: problems whose outcome is largely expected and whose solution is assured by the paradigm, but whose route to that outcome is unknown. Those puzzles are bounded by rules that flow from the paradigm, though a paradigm can guide work even where no explicit rule exists.
- V The Priority of Paradigms Normal science is guided first by shared concrete achievements, called paradigms, and not by an explicit set of rules. Scientists learn and extend these models by resemblance, and rules become important only when the models feel insecure.
- VI Anomaly and the Emergence of Scientific Discoveries Fundamental discoveries are not single moments but extended processes that begin with noticing an anomaly against paradigm expectations, continue through exploration, and end only when theory and procedures are adjusted so the anomaly becomes expected. Normal science, though not aimed at novelty, is the most effective way to produce it, because its precision makes anomalies visible.
- VII Crisis and the Emergence of Scientific Theories New scientific theories appear only after the old paradigm visibly fails at its own long-standing problems. That crisis signals that it is time to retool, and without it even correct alternatives are ignored.
- VIII The Response to Crisis Scientists never abandon a paradigm just because it faces anomalies. Rejecting one paradigm happens only when another is ready to take its place, and crisis matters because it loosens the rules of normal research, multiplies competing versions of the paradigm, and prepares the ground for a new one to emerge.
- IX The Nature and Necessity of Scientific Revolutions Scientific revolutions are non-cumulative episodes in which an older paradigm is replaced, wholly or partly, by an incompatible new one. They are necessary because real novelty only appears where existing expectations fail. Paradigm choice cannot be settled by normal-science logic alone, since paradigms supply the very standards, problems and concepts used to judge them.
- X Revolutions as Changes of World View When a paradigm changes, scientists do not just reinterpret fixed observations. They come to see, measure and even record different things, so in a real sense they work in a different world afterward, even though the world itself has not changed.
- XI The Invisibility of Revolutions Scientific revolutions look invisible because textbooks, popularizations and textbook-style philosophy are rewritten after each revolution and recast the past as a steady, cumulative march toward today's paradigm. That rewritten history hides the real engine of science, which is the reformulation of questions, concepts and facts together.
- XII The Resolution of Revolutions A new paradigm does not win by proof or by a single falsifying test. It wins as a growing share of a community shifts its allegiance, on faith in future promise, after crisis, and through persuasion that builds as early adopters develop the new view.
- XIII Progress through Revolutions Science looks uniquely progressive because its communities share one paradigm, answer only to peers, and train through textbooks, which makes them extremely efficient puzzle solvers and ensures each revolution is seen and made as progress. That progress is evolution away from what was known, not movement toward a fixed truth set in advance.
- Postscript, 1969 Scientific knowledge belongs to specific communities of specialists. The deepest part of what such a group shares is not rules but concrete worked examples, called exemplars, that teach members to see situations as alike. That makes theory choice a matter of shared values applied by a community and of translation across viewpoints, not a fixed algorithm. It is not irrational and not mere relativism.
