Rudolf Carnap’s An Introduction to the Philosophy of Science (published posthumously as Philosophical Foundations of Physics) is expository philosophy assembled from a seminar Carnap taught repeatedly at the University of Chicago and the University of California, Los Angeles. Martin Gardner, who attended the 1946 Chicago course, proposed in 1958 that the lectures be recorded, transcribed, and shaped into a manuscript; Carnap’s wife taped the entire final semester, and Gardner smoothed the style and added examples. Carnap’s preface, dated February 1966 at UCLA, presents the volume as a logical analysis of scientific concepts, statements, and theories rather than metaphysical speculation, deliberately restricted to a small number of fundamental problems treated thoroughly.
“It cannot be said that one is wrong.”
The book’s six parts move from the logic of laws and explanation, through measurement and quantitative language, into the structure of space, causality and determinism, theoretical laws and concepts, and finally beyond determinism. Most of the material, Carnap notes, applies to every branch of science—biological, psychological, and social as well as physical—so the book can serve as a general introduction to the philosophy of science, not merely to physics.
From the Logic of Laws to the Limits of Determinism
From Seminar to Book
Carnap opens with a preface explaining the book’s origins. It grew out of a seminar he taught repeatedly, under titles like “Philosophical foundations of physics,” whose content shifted but whose standpoint never did: the logical analysis of scientific concepts, statements, and theories, with no truck for metaphysical speculation. Gardner, who had taken the course at Chicago in 1946, proposed in 1958 that Carnap publish it and offered to edit. Carnap had no manuscript and no appetite for writing one, so the seminar scheduled for fall 1958 at UCLA was recorded instead. He doubted that talk could be turned into publishable prose, but colleagues pressed him—few of his views on the philosophy of science had appeared in print—and his wife settled the matter, taping and transcribing the entire semester and helping afterward; she did not live to see the book. Gardner smoothed the transcript, rearranged topics, and sharpened examples, with chapters passing back and forth and Carnap revising heavily. Though the audience had been advanced graduate students at home in symbolic logic, formulas were cut back and the rest explained so that a wider readership could follow. The book makes no claim to cover every problem; it treats a small number of fundamental questions thoroughly, and most of its topics apply to all the sciences, biological and social as well as physical, so it can serve as a general introduction. Carnap thanks Gardner, Feigl, Hempel, Shimony, and his students, and signs from Los Angeles in February 1966. The contents list six parts: laws, explanation, and probability; measurement and quantitative language; the structure of space; causality and determinism; theoretical laws and concepts; and beyond determinism.
Laws as Regularities
The first chapter begins with the plain fact that observation, everyday and scientific alike, turns up repetitions: day follows night, seasons return, fire burns, objects fall. Laws of science are statements that express such regularities as precisely as possible. A regularity holding everywhere and always without exception is a universal law—”All ice is cold” covers every piece of ice past, present, and future. Other laws assert regularities only in a percentage of cases or relate one quantity to another; these are statistical laws, as in “Ripe apples are usually red” or the near-even split of boys and girls among births. Both kinds are needed, but universal laws are logically simpler and so come first. Their form is the universal conditional, (x)(Px ⊃ Qx): for every body x, if it is heated it expands—the law of thermal expansion in its simplest nonquantitative shape, and the basic logical form of all universal laws, which physics refines into quantitative statements without exceptions. A biconditional is simply two conditionals joined, so it fits the same pattern.
Not every scientific statement takes this form. “Yesterday in Brazil, Professor Smith discovered a new species of butterfly” pins down a single time and place; it is singular. All knowledge begins in such singular statements—particular observations of particular things—and how we climb from them to universal laws is one of the perplexing questions of the philosophy of science. Ordinary language breeds confusion here. “The elephant is an excellent swimmer” uses “the” in the Aristotelian sense, meaning the whole class, a habit European languages inherited from the Greeks, who said “Man is a rational animal” to mean all men; “tuberculosis is characterized by the following symptoms” likewise means every instance. Scientists often call universal statements “facts,” forgetting that “fact” properly applies to singular occurrences. A report that this morning a current through a coil magnetized an iron core is a fact—assuming the observer was sober, the room not too foggy, and nobody secretly tampered with the apparatus. Universal statements, however elementary, will be called laws: thermal expansion, or “All ravens are black,” which Carnap does not know to be true but which, if true, would be a law of zoology.
Entelechy as Pseudoexplanation
A defender of a forcelike, nonphysical agent acting on the whole organism might claim it is as legitimate as the physicist’s magnetism. The reply is that the physicist never explains a nail’s motion by the word “magnetism” alone; pressed, he produces laws—qualitative if-then statements about iron nails and magnetized bars, the lodestone of Magnesia, stroking, the suspended needle pointing north, like poles repelling, heat and hammering weakening the force. Naming a new agent is not enough; laws must follow. Driesch supplied none. He never said how an oak’s entelechy differs from a goat’s, never classified entelechies as opposed to organisms, never formulated laws about their strengthening or weakening. His zoological findings—cutting a sea urchin’s limb yields death, fragmentary regrowth, or full regeneration—are respectable empirical laws, but “entelechy”
The Five Rules of Temperature Measurement
Carnap begins with the thermometer as a case study in how a quantitative concept is built rather than found. Immersing the instrument in water, we watch the mercury rise with heat and fall with cold; a single mark on the tube lets observers agree whether the column stands above or below it, and so we can say with confidence that today is warmer than yesterday. Two relations come first: equality of temperature, defined by the liquid reaching the same mark in two bodies, and the lower-than relation, defined by it reaching a lower mark. Rules One and Two tie these relations to numerical values—equal readings mean equal temperatures, lower readings mean lesser ones—but no graduated scale is yet needed. Assigning numbers takes three more rules. Rule Three fixes a value, usually zero, at an easily recognizable state, such as freezing water on the centigrade scale; Rule Four assigns a second value, a hundred at boiling water, to another reproducible state. These two give a basis for units but not yet a scale. Rule Five does the decisive work: it specifies the empirical conditions under which two differences count as equal, so that equal increases in mercury volume—requiring a tube of uniform bore—appear as equal distances along the tube. Mark zero and a hundred, divide the span into a hundred equal intervals, and two physicists following the same five rules arrive at identical scales. The rules determine a unique scale.
Convention and Fact in the Choice of Scale
The choices in Rules Three and Four are partly conventional: the meter rests on krypton wavelengths, the kilogram on a prototype in Paris, and Fahrenheit and Kelvin pick other fixed states. Yet all such scales share essentially the same fifth-rule procedure and differ only in calibration, translating easily into one another. Adopt a genuinely different fifth-rule procedure—mercury expansion instead of iron expansion or electrical flow—and the scales differ in form, perhaps beyond simple translation, and the laws built on them differ too, one yielding simple equations and another complex ones. That makes the fifth-rule choice enormously important, and scientists choose procedures that simplify the basic laws of physics. The absolute, or Kelvin, scale maximally simplifies thermodynamics; centigrade and Fahrenheit are variants of it. Early thermometers used alcohol, mercury, and constant-pressure gases, and precise instruments revealed small differences in scale form depending on the test substance. The absolute scale, closer to a gas scale than to mercury but not exactly like any gas—the “ideal gas” is only a manner of speaking—was eventually chosen for the simplest laws. In practice, mercury thermometers with scales close to absolute are used and converted by correction formulas; the absolute scale expresses amounts of energy rather than volume changes.
Meaning Arises from Measuring
We cannot know what a quantitative magnitude means until we have rules for measuring it; the concept develops out of the measuring. Not until thermometers were invented could temperature be given precise meaning. Einstein stressed exactly this in the discussions leading to relativity, concerning space and time—”equality of duration,” “equality of distance,” “simultaneity of two events at different places”—when no devices and rules had been specified. Rules Three, Four, and Five have both conventional and nonconventional aspects, as did Rules One and Two. Factual knowledge is needed to avoid conflict with nature, logical structure to avoid inconsistency. Freezing water suits Rule Three because the mercury height is always the same in it; were the height to vary with the water’s origin or amount, it would be unsuitable. That all boiling water has the same temperature is a fact of nature, with a qualification: boiling temperature holds for a given locality, but on a high mountain, where air pressure is less, water boils at a slightly lower temperature. One must specify altitude or apply a correction, and even then verify the pressure with a barometer—corrections resting on empirical facts, not convention. Rule Five seeks the simplest laws, a nonconventional aspect since nature’s facts determine the laws; and the numbers imply logical relations that are not conventional, since abandoning them yields contradictions.
Extensive Magnitudes and the Principle of Additivity
Temperature needs the full five-rule schema; many magnitudes called extensive need only three, applying wherever two things can be joined to produce a new thing whose value is the sum of the components’ values. Weight is extensive: a five-pound and a two-pound object together weigh seven pounds. Temperature is not: combining a sixty-degree and a forty-degree object does not yield a hundred degrees. The combining operation varies—putting together, gluing, tying, placing side by side on a balance pan. A row of books’ width is the sum of individual widths; reading ten pages and then ten more gives twenty; adding cold water to hot gives total volume as the sum. The exact combining procedure is often left unstated, a risky practice that breeds confusion; it should be explicitly defined. Then the three-rule schema applies.
Testing Geometry by Measurement
Carnap turns to the practical question of how one might actually test which geometry describes physical space. Triangles, he observes, are far easier to test than parallel lines: two lines might not meet until prolonged for billions of miles, whereas a triangle’s angles can be measured within a small region. Euclidean geometry gives an angle sum of 180 degrees; Lobachevski’s hyperbolic geometry gives less; Riemann’s elliptic geometry gives more. The sphere illustrates the elliptic case—a triangle formed by two meridians and the equator has two right angles at the equator, so adding the polar angle pushes the sum past 180, reaching 270 if the meridians cross at right angles.
Gauss reportedly imagined testing an enormous stellar triangle and is said to have triangulated three German mountain tops—a hill near Göttingen and two visible peaks. Having done important work on probability and errors of measurement, he could measure angles optically from each summit, average them under constraints to find the most probable values, and compute a probable error. The reported deviation from 180 degrees was so small it fell within that probable error, indicating either Euclidean space or an extremely small non-Euclidean deviation. Recent scholarship suggests Gauss may not have performed the test at all, yet the legend marks a methodological milestone: he was the first to ask what we should find if we empirically investigated space’s geometrical structure. Contemporaries found the idea preposterous—like determining the product of seven and eight by counting baskets of balls—and the French mathematician Jourdain joked about polling experienced headwaiters. Believing with Kant that intuition cannot err geometrically, they saw the measurement as absurd, which is why Gauss reportedly published neither the experiment nor his estimate of its worth. Speculation about non-Euclidean geometries nonetheless pushed many mathematicians to recognize a genuine empirical problem.
Models of Non-Euclidean Space
The sphere serves as a model for the Riemannian plane, though the analogy must not be overextended: any two straight lines on a Riemannian plane share only one point, while great circles on a sphere always meet at two. Strictly, the model fits only a portion of the sphere without opposite points, or each point must be represented by a pair of antipodal points. All geodesics in Riemannian space have the same finite length and close like a circle’s circumference—an extreme deviation from intuition that probably explains why this geometry was discovered later than Lobachevski’s.
On the spherical model, a circle’s circumference-to-diameter ratio is always less than pi: for a circle centered at the North Pole, the radius is the arc NB and the diameter the arc ANB, which exceeds the straight segment ACB. In Lobachevski space the ratio exceeds pi, visualized through a saddle-shaped surface like a mountain pass, or more exactly through the pseudosphere generated by rotating a tractrix about its axis. None of this makes pi variable: pi is the Euclidean plane’s ratio, untouched by geometries where the ratio varies with circle size.
Every surface has at each point a measure of curvature. A line’s curvature is the reciprocal of its radius of curvature; a surface’s curvature is the product of the reciprocals of the two principal radii. At a saddle the two curvatures take opposite signs, so the product is negative; on a sphere or egg both curve the same way, giving positive curvature. Lobachevski space has constant negative curvature, Riemannian space constant positive curvature, and Euclidean space zero.
Rival Theories and the Language of Relativity
Carnap notes how hard it can be to devise experiments distinguishing non-equivalent theories. Newton’s and Einstein’s gravitation differ so slightly that clever experiments and precise measurement were needed. Einstein’s unified field theory, he admitted, suggested no crucial experiment, being too abstract to yield observable consequences at current instrument precision. “Equivalent theories” means something stronger than agreement with known observations: identical predictions in all cases.
Einstein held that space’s structure deviates from Euclidean geometry in gravitational fields—undetectable on Earth, but testable near the sun and larger stars. Popular books mislead by declaring on one page that space in a gravitational field is non-Euclidean and on another that rods contract there; neither statement is wrong, but they belong to different languages and should not share a chapter. As with the flatland physicists, two equivalent descriptions cover the same facts. In the non-Euclidean language, mechanics and optics stay pre-Einsteinian, bodies remain rigid, and light rays are always straight; in the Euclidean language, geometry is preserved and new optical and mechanical laws absorb the deviations. A triangle around the sun then has angles summing to less than 180 degrees outside the sun and more inside; a Euclidean model surface, hill-like near the top and saddle-like outside, reproduces this, with curvature greatest near the sun and fading with distance.
The Causal Relation and the Single Case
Carnap opens with a contrast meant to isolate what is peculiar about causation. Whether wallpaper C is a darker shade of blue than D can be settled by looking at the two samples under ordinary light; whether E is F’s brother can be settled by tracing their parentage. Both inquiries stay with the case at hand, however easy or hard the tracing proves. A causal claim is different in kind. In principle one cannot define the relation and then inspect a single pair of events A and B and be done.
In practice we often seem to skip the extra work, but only because the relevant law is so familiar that it goes without saying—and we forget that the law itself rests on countless earlier observations. A stone flies at a window, strikes the pane, and the glass splinters; asked how we know the impact did it, we say it is obvious. The retort “what else?” points to a long childhood history of shattered glass. Appearances can deceive: in a television western the villain fires, a shot sounds, a man falls—yet no bullet flew, and the report may have been dubbed in later. Perhaps an invisible plastic shield stopped the stone while someone inside broke the window on cue. We dismiss such tricks as improbable precisely because past experience favors the ordinary reading; if we suspect a trick, we investigate further.
The lesson holds whether we glance or scrutinize: we always bring hundreds of similar past cases to bear. No single observation establishes a causal relation. Imagine a future in which window glass shatters only at an extremely high frequency sound; seeing a stone strike a pane, we would cry coincidence and look for the hidden source of sound. Causation, unlike containment—where one examination of box b settles whether object a is inside—can be established only through a general law resting on many observations. To say A caused B is to say this is an instance of a law holding everywhere and everywhen, a bold leap from particular cases to a universal conditional. Carnap would rather replace talk about the meaning of causality with analysis of the various laws science actually uses. He notes Hans Kelsen’s interest in the concept’s historical origins—Kelsen, once a Vienna professor of law and a chief author of the Austrian Republic’s 1919 constitution—and the old suggestion that humans project their own purposes onto nature.
Determinism, Compulsion, and Free Choice
Everyday speech withholds “cause” where B precedes A, or where both are end results of chains from a common source: nightfall can be predicted from daylight, a timetable entry from a train’s arrival, yet neither causes the other. Carnap would still call these causal inferences, and would use “causal law” for any law by which events are predicted or explained from others. Determinism is the stronger thesis that a complete description of the world at one instant, plus the laws, fixes every past and future event—the view of Newton, analyzed by Laplace. Quantum mechanics, with its irreducibly probabilistic basic laws, leads most physicists and philosophers to reject it; Einstein never did, trusting the rejection was temporary.
The free-will question, Carnap insists, does not turn on these scientific theories, and he parts company with Reichenbach, who held that strict determinism would empty “choice” of meaning. Carnap answers that this confuses determination with compulsion. A prisoner walled in suffers compulsion; so does a man whose hand is forced to pull a trigger. But a friend known to adore rare Bach performances, telephoned about a private concert, goes of his own free will—even if his attendance was predictable from his character and psychological laws. Character is molded by upbringing and experience, yet choices springing from it remain free. Compulsion means outside force against desire; predictability is another matter entirely.
Correspondence Rules and the Status of Theoretical Terms
How can laws about unobservable molecules yield laws about pressure and temperature? Only if something bridges the two vocabularies. Carnap calls these bridges correspondence rules—Bridgman’s operational rules, Campbell’s “Dictionary”—such as the statement that gas temperature is proportional to mean molecular kinetic energy. Theoretical terms cannot be explicitly defined by observables, though observables can sometimes be defined theoretically. There is no picture of the electron; the best we can say is that it is a tiny body behaving in certain ways, described only through further theoretical laws. Bridgman’s operational definitions really pick out partial concepts, one per measuring procedure; Reichenbach’s “correlative definitions” fare no better as definitions, since light rays and stretched strings are only approximately straight and finite, never true geometrical lines. Carnap prefers “correspondence rules.” Mathematical axiom systems, like Peano’s, can be defined outright and say nothing about the world; physical theories cannot enjoy that isolation.
The Ramsey Sentence
Ramsey, dead at twenty-six, left his paper “Theories” in The Foundations of Mathematics, edited by Braithwaite—a piece Carnap thinks undeservedly neglected.
Meaning Postulates and the Analytic
Carnap works with an artificial language built on English, equipped with meaning postulates that let a speaker decide whether a sentence is analytic. A dictionary supplies some of these: given the definition that to be defenestrated is to be tossed out of a window, the sentence “If a bottle is tossed out of a window, the bottle is defenestrated” is seen at once to be analytically true—no bottle need actually be thrown. Other cases resist the dictionary. “All men are rational animals” and “All men are featherless bipeds” are ambiguous because “men” is ambiguous. In the artificial language the postulates settle the matter by stipulation: if rationality and animality are listed among the essential meaning components, the first sentence is analytic; if the postulates concern only bodily structure, it is synthetic. Lacking analogous postulates for “featherless” and “biped,” the second sentence stays synthetic—a one-legged man, or a man with feathers on his head, would still be called a man. The more precise the postulates, the sharper the analytic/synthetic line; the remaining haziness merely mirrors haziness in the meanings of descriptive words, not a defect in the distinction itself. Postulates say nothing about the actual world: one making “warmer” asymmetric means that two objects each reported warmer than the other betray differing understandings of the word, not a surprising discovery. Carnap acknowledges the attacks of Quine, Morton White, and others on the sharp distinction, citing Quine’s “Two Dogmas of Empiricism” and “Carnap and Logical Truth,” White’s “The Analytic and Synthetic: An Untenable Dualism,” and his own replies in “Meaning Postulates” and “Meaning and Synonymy in Natural Languages.” For observation languages he counts the problem solved in principle; theoretical languages raise formidable difficulties.
The Network Above the Plane of Observation
Theoretical terms—”electron,” “mass,” “electromagnetic field”—cannot be given complete interpretations; a bubble-chamber track yields only partial, indirect meaning. Temperature in the kinetic theory is tied by correspondence rules to thermometer readings, but a thermometer works only in a narrow interval, other methods must cover other temperatures, and future observations may add new rules. Carnap borrows Hempel’s image: a theory is like a spatial network whose knots are terms and whose threads are definitions and hypotheses, floating above the plane of observation and anchored to it by interpretive strings. Sentences true by logical form are easy to identify; the hard problem is separating analytic from synthetic sentences that contain descriptive terms. A statement counts as analytic if it is a logical consequence of the meaning postulates—empty of factual content, like “No bachelor is married.” Theoretical postulates alone cannot serve, since they yield an abstract deductive system, Euclidean geometry say, without empirical meaning until correspondence rules interpret it; combined theoretical and correspondence postulates presuppose too much and yield synthetic statements. General relativity shows the theory has factual content: it entails that if two photographic plates of the same star pattern are made, one during a total eclipse, then images near the sun’s rim are displaced away from the sun. Findlay Freundlich tested this in 1919, found the antecedent true and, after measurement, the consequent true; had the consequent failed, the conditional and the theory would have been refuted. Quine and others deny the dichotomy can be applied to theoretical language at all; Hempel, perhaps hesitatingly accepting it for observation language, voiced strong Quinian scepticism about theoretical language, holding that the double role of theoretical and correspondence postulates makes analytic truth elusive. Ramsey’s extreme solution states the theory’s observational content in the Ramsey sentence, quantifying the theoretical terms away—but Carnap rejects this as too radical, since theoretical terms greatly simplify the formulation of laws. He proposes splitting the theory into an analytic postulate, factually empty, and the Ramsey sentence carrying all factual content, the analytic part formulated as a conditional asserting that if the Ramsey sentence is true, the theoretical terms must be understood so that the whole theory is true. In the six-concept example—two classes, four relations, the terms Mol, Hymol, Temp, Press, Mass, Vel—the postulate says that if a sextuple of entities relates to the observational entities as the theory specifies, the six terms denote such a sextuple; if no such sextuple exists, the conditional holds regardless of interpretation and gives no partial interpretation. A sentence is analytically true in the total language when logically implied by the combined postulates of observation and theoretical language, analytically false when its negation is.
Maxwell-Boltzmann and the Deterministic Faith
Nineteenth-century kinetic theory produced probabilistic laws in statistical mechanics. For oxygen at uniform pressure and temperature, molecular velocities follow the Maxwell-Boltzmann distribution: for each of the three velocity components the probability distribution is the normal, bell-shaped curve.
What the Scientific Language Commits Us To
Running beneath the book’s technical machinery is a single conviction about language: the concepts of science are tools we build, not features of the world we discover ready-made. Carnap divides scientific concepts into three groups—classificatory, comparative, and quantitative—and argues that the movement from “warm” to “warmer than” to “50 degrees” is a movement in expressive power, not in fidelity to some hidden essence. Stones contain no numbers; “it is we who assign numbers to nature.” The five-rule schema for measuring temperature, with its conventional choices of fixed points and its nonconventional demand that the resulting laws come out simple, shows how much of scientific objectivity is constructed and how much is forced on us by nature’s own regularities. The same lesson recurs in the treatment of time and length, where the apparent circularity of defining periodicity and rigidity is escaped not by finding an absolute standard but by selecting, on grounds of simplicity, one class of processes or bodies from among many equally self-consistent options.
This is why Carnap spends so much space on pseudoexplanation. Driesch’s entelechy earns its long critique because naming a force is not explaining anything: without laws, without new predictions, the word adds nothing to knowledge. The physicist who says “magnetism” must, when pressed, produce the conditional laws governing iron and magnetized bars. The same standard separates genuine theoretical concepts from empty ones, and it explains Carnap’s preference, against Bridgman’s operational definitions, for treating length and mass as theoretical concepts partially interpreted by the whole system of physics rather than pinned down by any single measuring procedure.
Why the Quantitative Matters, and What It Does Not Cost
Carnap’s defense of quantitative language is one of the book’s most engaging threads, and it is aimed at a Romantic objection he takes seriously enough to answer at length. Goethe’s complaint—that Newton’s prism experiments tore light from its natural conditions—gets a respectful hearing, and the reply is that the physicist’s numbers do not discard the colors. Musical notation does not let you hear the melody, yet the tones are implicit in it once you know the translation key; a spectroscope reading tells a trained observer that the light is blue-green as surely as the word “blue” does. The black-and-white photograph really did leave out color; the equation does not. What the quantitative idiom buys is economy of vocabulary and, far more importantly, the power to deduce: a law like P = a + bM + cM² can be manipulated mathematically, while fifty overlapping qualitative adjectives can only be strung into conditionals. Carnap’s patience with Goethe runs out only at the point where the poet’s followers would enjoy the fruits of quantitative physics—cars, refrigerators, telephones—while denouncing its spirit.
The critique of “word magic” belongs to the same argument. Ogden and Richards’s examples, and Riezler’s imagined Aristotle who accuses modern science of eliminating both the Negro and the Eskimo in its pursuit of 50 degrees, let Carnap press a distinction between certitude and richness. The thermometer’s reading is objective and thin; the Eskimo’s and the Negro’s sensations are rich and private. Carnap does not deny the loss; he denies that it is a loss of truth, and he suspects that the philosophers who mourn it are in the grip of a childhood belief that the mother-tongue word carries the quality it names.
Geometry as an Empirical Question
The treatment of space is where the book’s logical tolerance shows most clearly. Gauss’s legendary triangulation of three German hilltops—whether or not he performed it—marks for Carnap the moment someone first asked seriously what we would find if we measured the geometry of physical space. The spherical model of Riemannian space, the saddle and pseudosphere models of Lobachevski’s, and the constant curvature that fixes each geometry are laid out with care, including the warning that the sphere fits only a portion of the Riemannian plane and that pi itself remains the Euclidean ratio, untouched by geometries in which circumference over diameter varies with size.
The payoff comes in relativity, where Carnap insists that the popular books’ habit of saying both “space is non-Euclidean near the sun” and “rods contract in a gravitational field” in the same chapter is incoherent. Each statement belongs to a different language. In one, geometry stays Euclidean and new physical laws make rods shrink radially near the sun; in the other, rods stay rigid and geometry bends. The two descriptions are observationally equivalent, and the choice between them is a matter of convenience. Reichenbach’s distinction between differential effects, which vary with the substance, and universal effects, which do not and cannot be shielded against, gives this freedom its precise form: a universal effect can always be absorbed into the geometry, and no experiment can force the issue.
A Philosophy of Science Without Metaphysics
The book’s later parts carry the argument past determinism, into statistical laws, correspondence rules, the Ramsey sentence, and the puzzles of analyticity in theoretical language—territory where Carnap’s logical empiricism does its most characteristic work. What holds the whole together is a temperament rather than a doctrine: a willingness to ask what a scientific statement actually does, what rules govern its terms, and what would count as evidence for or against it, while refusing to answer such questions by appeal to intuition about the nature of things. Carnap’s preface admits the book is not systematic; it treats a handful of fundamental problems thoroughly and leaves the rest. That restriction is itself the method. By showing, case by case, how laws are confirmed rather than verified, how magnitudes are constructed rather than found, and how rival geometries can describe the same facts, Carnap offers an introduction to philosophy of science that is really an introduction to a way of thinking—patient, technical, and unfailingly suspicious of words that promise more than they deliver.
