A Village Puzzle
First, a village. This village has just one barber, and his rule is posted at the door: I shave only those villagers who do not shave themselves. The rule ran well enough, until one day someone asked: the barber's own face — who shaves it?
If he shaves himself, then by the rule he is one he ought not to shave — he has violated it; if he does not shave himself, then by the rule he is one he ought to shave — he has violated it again. The village falls silent: this rule cannot be kept — not because it is too difficult, but because it logically cannot be.
This village version is a popular illustration of the paradox proper — Russell himself recorded that it was a rewriting in disguise suggested to him by an unnamed friend, not the prototype he discovered in 1901. But the illustration has one advantage the original lacks: it sets the exit of the solution right out in the open. After a moment of silence, the village will say: no such barber exists. Not that he has moved away — the description "a barber who plies his trade by this rule" assigns no object at all: a filter condition that contradicts itself filters out no one. Remember this exit; at the end of this chapter it will return, and become a whole foundation of mathematics.
A Letter of Only a Few Lines
The prototype reaches far deeper than the illustration. Spring 1901, Cambridge; Russell was writing The Principles of Mathematics. He had accepted the route of Frege and Cantor: the foundations of mathematics could rest on classes, on sets — given any property whatever, one could speak of the totality of the objects having that property. In the course of writing he checked, in passing, a boundary case: consider the property "not containing itself"; the sets that do not contain themselves make up a totality R. Is R itself inside it or not? If R does not contain itself, it satisfies the property, and so belongs to R — it contains itself; if R does contain itself, it fails the property, and so does not belong to R — it does not contain itself. Both directions strike the wall: R belongs to R if and only if R does not belong to R. There is no room for adjusting the wording; this is not a problem of formulation but a crack standing open in the structure itself.
And let the account of time be completed in passing: set theory's own cracks came earlier than this. In 1897 Burali-Forti had already given the paradox of the greatest ordinal; Cantor himself, in his 1899 letter to Dedekind, discussed "inconsistent totalities", knowing that certain totalities could not be treated as completed sets. But these cracks were at the time taken for an internal technical minefield of set theory; Russell's crack was different — it requires no special concept whatever, only the two words "property" and "totality", which every act of reasoning employs. That is why what it shook was not one discipline but the foundation itself.
On June 16, 1902, Russell mailed this crack to Jena, addressee Frege. The weight of the letter must be read from the recipient's position: Frege's system had just built arithmetic upon the axioms of logic in volume one of the Basic Laws of Arithmetic (1893), and volume two was already in press; and the crack Russell pointed out ran exactly through the load-bearing wall — Basic Law V says that every concept has an extension, a set, and "the set of all sets that do not contain themselves" is precisely the extension of a concept. Where the crack passed, the foundation of the whole logicist edifice split through. Frege replied on June 22, admitting that he was "thunderstruck", and saying that the foundations of arithmetic had been shaken. When volume two appeared, he gave, in an appendix written specially for it, the sentence that has been quoted for a century: a scientist can hardly meet with anything more undesirable — at the moment the edifice was completed, the foundation collapsed. He attempted a repair on the spot, but that patch was later proved to leak still. The planned third volume was never written; the Frege of his last years exhausted his belief in logicism, and a few years before his death went so far as to abandon the central claim of his lifetime, coming to hold that the roots of arithmetic lay, after all, in geometrical intuition. To watch with one's own eyes the foundation one had built for thirty years retreat from the position of "truth" to the position of "a failed attempt" — this scene is the most sorrowful in the history of logic.
RC's General Outline says in Section 1.1 that possibility is determinacy not yet locked by observation. The position of Frege's remaining years falls exactly here: the determinacy that his lifelong project had locked fell back, before his eyes, into the undetermined. What this chapter has to do is to give this fall a mechanical account.
The Genesis of the Paradox
When the standard textbooks reach Russell's paradox, the rhythm is usually: state the paradox, declare the crisis, turn to axiomatization. This book wants to slow down and ask a genetic question: under what conditions does a system produce such a crack?
The first layer of the answer: the crack is not an intruder; it is brought by capacity. On RC's reading, the genesis of the paradox has a single condition — once a system has gained a sufficient capacity for self-observation, the observer falls into his own range of observation. RC discusses the symmetry of observer and observed in its axiomatic system (Axiom A4): every existence is at once observer and observed, the identities exchanging with the perspective of observation. What this axiom states, in the ontology, is subject and object; what this chapter has to point out is that it has an exact counterpart in the world of symbols: when a symbol system is strong enough to speak of "the totality of objects of an arbitrary property", it is strong enough to speak of that totality itself — the instrument of filtering falls into the range of what is being filtered. Chapter Five said that symbolization is the compression of the convergence paths of inference; and the stronger the compression, the more occasions on which the path crosses itself. Aristotle's term logic was too weak — too weak even to reach self-reference; no mood in the syllogistic can construct a term that talks about itself. The languages of Frege and Cantor were strong enough — strong enough that "arbitrary totality" became a legitimate phrase — and so observation saw, for the first time, its own back.
The second layer: the shape of the crack is an illegal total observation. Note what the act of defining R attempts — to lock, in a single stroke, a total snapshot containing all sets, a snapshot that must contain the product of the very act of "locking all totalities": R itself. The system attempts to lock a totality that takes in its own act of locking. Here precisely lies the trouble: locking is an action with levels; every locking occurs at some observational position, whereas "the set of all sets" attempts to gather every position into one, including the very one it presently occupies. This is not a local technical slip of set theory but an intrinsic structural limit of the act of total observation — the snapshot cannot catch the hand that is pressing the shutter, unless there is a second camera; and the second camera's photograph likewise cannot catch itself.
This reading also settles an ancient lodger. The liar paradox — "this sentence is false" — has been wandering the front hall of philosophy since ancient Greece, taken for two millennia as wordplay or as the narcissistic pathology of language. Set into the frame of this chapter, its identity becomes clear: it is the natural-language version of the same crossing of levels. Natural language has one feature that other systems lack: by default it can talk about everything, itself included; a language that "talks about everything" necessarily arrives at talk about the talking itself, and so upon some sentence, the describing and the described collide into a single point. The ancient commentators had no way to locate the phenomenon, because to locate self-reference one must first possess the pair of concepts "system" and "talk about the system" — and this pair became possible only after symbolic logic had objectified language itself. The paradox is not a malfunction of language but an early rehearsal of language's capacity for self-observation; Frege and Russell were merely the ones who put the rehearsal onto the official stage.
The third layer: what the crack discovered. On this reading, Russell's paradox discovered no monster in the world — there is no dangerous set called R lurking in the cellar of mathematics. What it discovered is an impossible locking: a filter condition whose reflexive application keeps it from ever converging into any object. The barber in the village, Burali-Forti's greatest ordinal, the liar ("this sentence is false") — they are different outlets of one and the same crack, and every outlet leads to one and the same fact: the concept of totality, unless restricted by levels, devours itself. Chapter One transcribed the words of Section 1.1 of RC's General Outline: all existence at once does not exist — every determinate state is only a temporary locking upon the Ground, not the Ground itself. The lesson of set theory is the symbolic version of this logic: no locked totality is the totality itself; to treat "the totality itself" as a locked object is to obtain contradiction, not knowledge.
Patch One: The Theory of Types, an Artificial Stratification of Observation
The first route of repair was opened by Russell himself. His diagnosis went straight to the vital point: the paradox arises from a vicious circle — defining a totality by means of the membership of that very totality. The paper of 1908, "Mathematical Logic as Based on the Theory of Types", gave the plan: stratify the objects. Individuals are level 0; classes of individuals, level 1; classes of classes, level 2; and so on; and one iron law stands across the levels — the members of any class may be drawn only from the level immediately below. The phrase "the set of all sets that do not contain themselves" is from then on illegal: a set and itself are always separated by at least one level, the predicate "contains itself" has nowhere to attach, and the construction of Russell's paradox cannot so much as be built in the grammar. By Principia Mathematica (1910–1913), this stratification had been honed into the ramified theory of types: not only do objects have types; properties are further ordered into orders by the levels they involve, so as to stop the semantic paradoxes of the liar kind.
On this book's reading, the theory of types is an artificial stratification of observation: it rebuilds, by open stipulation, that fact of Chapter Five — the snapshot cannot catch the hand that presses the shutter. It concedes that every observation occurs at some position, then registers the positions by number, and forbids photographing oneself across levels. It did not discover the world's levels; it installed levels into language. The price followed at once: the ramified stratification was too strict — so strict that a whole batch of perfectly healthy mathematical definitions (definitions that involve a reference back to a totality, such as "a real number is the thing forced out by every nested family of rational intervals") were ruled illegal; to rescue them, Principia Mathematica had to introduce an axiom of reducibility, which says, roughly, that every ramified property has a simple property with the same extension. Yet this axiom itself carried no intuitive guarantee whatever — it was a hypothesis decreed in order to save mathematics. Logicism had wanted to exchange the foundation of mathematics for a more limpid logic; now the inventory of the foundation contained one axiom of which no one could say why it was true: the chess player admitted that keeping the board intact required a new rule exempt from question.
RC's General Outline, discussing the multi-level character of consciousness in Section 1.2, says that the consciousnesses at every level construct, according to their precision of recognition, the determinacy matched to them, and that the matter and the consciousness of level A can together be taken by level B as objects — between the levels there is strict pairing and inclusion. It must be said at once: this is RC's thesis about existence and consciousness, not a claim about sets; this book brings it here as an analogy because the two are structurally isomorphic — the act of observation at every level dwells within the range of observation of the level above it, and the levels can be joined only in sequence, never swallowing one another. The theory of types is the result of transcribing this hierarchical intuition into hard grammar: the levels of observation no longer exist by default; they must be declared.
The stratification plan has one voice from the periphery that must be recorded. During the years of the crisis, Poincaré proposed that the lesion of the paradoxes lay in "impredicative" definitions — defining an object by means of a totality that includes the object to be defined — and urged that mathematics be permitted to speak only "predicatively". This criticism and Russell's vicious-circle principle identified one and the same lesion, from east and from west, yet the two men could not stomach each other's prescriptions: Poincaré suspected that Russell's axiom of reducibility was precisely the back door of the impredicative, while Russell held that Poincaré's prohibition amputated legitimate mathematics. Consensus on the lesion coexisting with schism over the remedy — this configuration is itself a piece of testimony: as to "where the boundary had been crossed", the community saw clearly; as to "where the boundary ought to be drawn", it could only bargain. Part Three, treating the foundational crisis of the calculus, will meet this configuration again; here let its shape be recorded.
Patch Two: Zermelo's Fence
The second route was opened by Zermelo in 1908, and its idea is the opposite of stratification: restrict not the language of the objects, but the hand that "makes sets".
Frege's Basic Law V licensed the making of a set directly out of any concept whatever — the concept is the knife; where the knife falls, there is the set. Zermelo's axiom schema of separation withdrew this unlimited authorization: starting from any property, you may only, within a set already given, separate out those elements that satisfy the property. Sets are no longer cast out of properties out of thin air; they can only be carved out of material already at hand; and where the "material at hand" comes from is issued, one by one, by several carefully drafted axioms (pairing, union, power set, and the rest). Thus one thing is permanently expelled from the registered residence of mathematics: the set of all sets can no longer be made by any axiom — it is a forbidden operation, not a negated existence.
When Zermelo wrote out this prescription, he had just finished another hard-fought battle: in 1904 he proved the well-ordering theorem — every set can be put into a well-order — and the proof used an assumption no one had before stated: from any family of non-empty sets, one element can be picked from each. As soon as this axiom of choice appeared it drew fierce dispute, for it claims that infinitely many arbitrary choices can be made at one stroke, while offering no rule for how each choice is to be made. This dispute bears on the present chapter in the soft spot it exposes: the intuitions of set theory begin to conflict with themselves in the face of the infinite — among one and the same body of mathematicians, some found the axiom of choice self-evident to the highest degree, others found it next door to fraud; and no experiment or computation could adjudicate the quarrel. The infinite is the focal range in which the snapshot's format is least able to hold.
The exit of the village puzzle returns here, and upgraded. The village says: a self-contradictory filter condition filters out no barber — the definition fails, the object does not exist. Zermelo says: no filter condition whatever, even a consistent one, suffices to guarantee that a totality exists — existence must be conferred separately, by axioms. The common posture of the two is this: between "uttering a condition" and "locking an object", a sluice gate is inserted once more. In the vocabulary of Chapter One, this is taking the authority to lock back into the institution: no snapshot may develop itself into a print automatically upon a description; the shutter must be pressed by an axiom on the register. The village's silence at the opening of this chapter hears, in Zermelo's axioms, its mathematical echo — and the other half of that echo will not be complete until Chapter Seven: this sluice gate stops the paradoxes, but it cannot stop a deeper boundary rising up from inside the system.
And let the account be completed in passing: Zermelo's axiom system of 1908 became what is today generally called ZFC (Zermelo–Fraenkel set theory with choice) only after Fraenkel's supplement of the axiom of replacement in 1922 and Skolem's clarification of the use of "definite properties". It is the de facto common foundation of contemporary mathematics — most branches of mathematics build on it by default. The foundation was exchanged, and the building carried on: this is the architect's skill that Chapter Three spoke of — the foundation can be exchanged; lay another foundation, and the same building is roofed again on the new ground. Only, no one in 1908 yet knew that the new ground, too, had a boundary of its own buried in it; and the instrument that would prove this was even then taking shape.
The Common Posture of the Two Routes
Set the two patching routes side by side, and one sees one and the same posture.
The theory of types says: talk must be stratified — the language of level n may not reach into level n+1. Zermelo says: set-making must be authorized — an arbitrary property no longer automatically casts a totality. The one governs the grammar of language, the other the registered residence of objects; the directions are opposite; but together they do one and the same thing: they declare total observation an illegal operation. The mathematical community needed two generations to learn one lesson: an unguarded "all" is an action that a system cannot perform upon itself. The paradox is not sabotage by an enemy, but the inevitable curtain-raiser at the moment a system's capacity for self-observation matures — the stronger the capacity, the more nodes of self-crossing; every sufficiently strong system will sooner or later see itself at some node, and if the manner of seeing is an unstratified total snapshot, what it obtains is contradiction.
What deserves notice is the community's attitude toward the two routes: no verdict, only a division of labor. The theory-of-types line flowed into the later study of logic and language (Tarski will take over there the role assigned to the next chapter); axiomatic set theory became mathematics' daily foundation. The scale of Chapter Three takes effect here: this was not a re-convergence — no old consensus was overturned, no observational divergence quieted by new observation; it was a preventive planning, a redrawing, made at the first appearance of the crack, of the construction lines for "how high it is still safe to build". In RC's vocabulary, both plans preserve margin for the system: the theory of types leaves "what cannot be said" outside the stratification; Zermelo leaves "what cannot be made" outside the axioms — what is excluded is not knowledge but catastrophic self-locking.
But the red lines governed only the boundary-crossing construction; they could not govern a quieter question. Once axiomatic set theory stood upright, anyone would begin to ask: is this new foundation itself safe — might it not, too, conceal a crack, waiting to split open some day just as the edifice was completed? Frege's lesson is precisely that self-evidence guarantees nothing — Basic Law V, too, had once looked harmless to the point of common sense; Cantor had set up a warning sign, by hand, as early as 1899 — "inconsistent totalities" must not be treated as sets — but a warning sign is not a proven line of defense; the guarantee of the theory of types was equally suspect, the axiom of reducibility being the very confession that the guarantee itself stood in need of guaranteeing. To give safety a proof, one would have to prove that the axiom system derives no contradiction; and a proof needs a system that does the proving, whose own reliability again needs a guarantee — unless one could find an absolutely unimpeachable meta-reasoning in finitely many steps, to underwrite the whole of mathematics once and for all. One man now took over this problem; he believed that such reasoning existed, and for it he founded a program: to defend the infinite, not yielding an inch. The ceremony of capping that he wanted never arrived; what arrived was a twenty-four-year-old young man at a round table in a city by the sea, saying a few sentences in ten minutes. The next chapter is this book's most crucial transcription: the boundary itself becomes a proof.