祝融说。

The Eclipse a Seventeen-Year-Old Remembered

1919, Vienna. A seventeen-year-old boy was making his way through the crowded marketplace of ideas in the capital of a defeated empire. The year stayed in his memory all his life: Eddington's expedition had brought the eclipse plates back to London, and the newspapers proclaimed Newton's ideas overthrown — the very ritual Chapter Fourteen told of. What impressed him most was not the content of relativity but the shape of the event: a theory staked itself in public, was adjudicated by the observations of the whole world, and, losing, accepted the verdict. The same marketplace was crowded with other doctrines besides — the Marxist circles he had at one time and another moved in, the Adlerian psychoanalysis professed by acquaintances of his — and these too were coherent, these too explained everything, yet never once did they stake themselves on any photographic plate.

The boy's name was Karl Popper (1902–1994). He said afterward that it was precisely this contrast that grew into his philosophical question: what, after all, is the difference between Einstein and those doctrines? The question later acquired its learned name — the problem of demarcation: where the line falls between science and non-science (pseudoscience, metaphysics, ideology). It is sharper than the "problem of truth", for it does not ask which theories are right; it asks only which theories are working in the mode of "being capable of being proved wrong".

Chapter Fourteen has already given the RC reading of 1919: the re-observation ritual of re-convergence. This chapter changes the focal length — looking not at the content of the accounts that ritual changed, but at the discipline it exposed: the channel between theory and observation must be kept open. This is precisely what the chapter has to discuss: two examiners of the twentieth century, one of whom legislated for this channel with a single criterion, the other of whom pointed out that the law itself stood in need of correction. And after the two of them, RC will give its own verdict: the health of reason is judged not by how many true theorems it holds, but by its capacity to keep converging.

The Predicament of Verification

Popper's opponent must first be set squarely on the stage. The Vienna Circle of the 1920s — Schlick, Carnap, and the others — exchanged philosophy's old questions for a new standard: the verification principle. For a proposition to have cognitive meaning, it must be verifiable by experience — it must be possible, in principle, to indicate which observations would support it. The standard was aimed at metaphysics: the grand assertions of the Hegelian kind, admitting of no supporting observation and no opposing observation either, should simply not count as knowledge. It was a movement of clearance, its motive of one lineage with Bacon's clearing away of the idols (Chapter Twelve).

The predicament was hidden in the standard itself, and its root is the introduction's second puzzle. Hume had long since proved: from "it has been so every time in the past" one cannot infer "it will also be so in the future" — no finite number of confirmations can, in logic, establish a universal law. Chapter Thirteen turned this crack into an institution: the scientific ledger differs from the oracle in that it keeps the difference between "so far" and "forever" forever open and forever audited. The verification principle demands, in logic, precisely that an unlimited check be paid out of a finite number of deposits — the condition of acceptance it lays down for science is one that science, on Hume's accounting, simply cannot meet. More awkward still is the asymmetry in the other direction: a hundred white swans found to confirm "all swans are white" do not balance the falsification of a single black one. The innate inequality of the logical weights of confirmation and refutation was seen through at a glance by Popper, and turned into the whole of his starting point.

Falsifiability

In 1934 Popper published Logik der Forschung; in 1959 the greatly expanded English edition appeared under the title The Logic of Scientific Discovery. Here the standard was inverted: the territory of science is to be staked out not by verification but by falsification — a theory is scientific if and only if it runs the risk of being overruled by observation: it makes definite predictions, the predictions may conflict with observation, and the theory accepts that possibility. The scientific character of a theory lies not in how often it has been confirmed — astrology's "verifications" come by the cartload — but in what it forbids: a theory that permits no conceivable observation to refute it forbids nothing, and therefore says nothing.

On this book's reading, the criterion can be put straight into the vocabulary of Part One and translated, and the translation fits astonishingly well. Falsifiability = keeping an open channel for observational divergence. That a theory is falsifiable means that it keeps, for its dialogue with the world, the possibility of observational divergence arising — that there exists a class of observations which, should they occur, would constitute a questioning of the consensus (General Outline 1.4: "observational consensus, too, can be questioned or negated by different modes of observation, producing observational divergence"). That a theory is not falsifiable means that it has welded the channel shut: whatever observation comes in is absorbed by the framework as a reconfirmation of itself — the turkey of Chapter Two lived in just such a framework.

And the act of "keeping an open channel for observational divergence" has a name ready and waiting in RC's theory of practice. The General Outline, discussing sustainable decision-making in Section 4.3, says: there is no such thing as an optimal decision; what passes for optimal is in essence the sustainable decision; sustainable decision-making is keeping optionality present at all times, keeping the elasticity to offset the solidification of rules. Falsifiability is precisely the cognitive edition of this principle: a theory that refuses to weld itself into irrefutability is, along the dimension of time, preserving a margin for trial and error — it concedes to "the next observation" the right to change the account. The two vocabularies meet head-on at this point: with a criterion of scientific methodology, Popper said what RC says with Axiom A7 (conservation of margin) — in locking determinacy, observation does not exhaust the Ground; space is left for re-convergence. Chapter Sixteen has just gathered the three forms of reason into "cognitive relay stations", and Popper's criterion can now supply a footnote: the difference between a relay station and a terminus is written in falsifiability — a terminus sells no tickets for the next leg.

The criterion also lights up two accounts of Chapter Thirteen. The notarization of prediction — locking the path publicly beforehand, with signature and deadline stated, handing it over to the world for verdict — is precisely the institutionalized form of falsifiability; the "risky predictions" Popper admired are, structurally, one and the same thing as the date of Halley's comet's return. And ad hoc hypotheses — the corrections Popper expressly forbade, those made only to square the old account while staking nothing new — are Chapter Thirteen's paper money: each note printed shaves one degree off the framework's persuasive power. How much life a framework has left is seen not in the positive assets on its books but in whether it still dares to open new bets — that verdict of Chapter Thirteen is very nearly the accounting translation of Popper's criterion.

Lakatos's Correction

The standard was stood up, but the materials of twentieth-century history of science refused to fall wholly into line. The pressure came from two sides. From logic: Duhem at the end of the nineteenth century and Quine at mid-century argued that theory never faces observation alone — what a crucial experiment overrules is always a bundle of hypotheses (the conjunction of theory plus auxiliary hypotheses plus instrument theory plus data processing); so long as the auxiliary hypotheses are willing to yield, the theory proper can always be spared. From history: Chapter Fourteen's Lorentz contraction is the ready specimen — a hypothesis exactly canceling the observation, not independently measurable in principle; by naive falsificationism it should long since have shot ether theory dead, yet the physics community went on doing business with that paper money for more than a decade, and the accounts, by and large, balanced.

The corrector was Imre Lakatos (1922–1974). In 1970, in the paper "Falsification and the Methodology of Scientific Research Programmes", he gave his scheme — a paper that later became the title piece of his posthumous collection The Methodology of Scientific Research Programmes (published in 1978). His picture: the basic unit of science is not the single theory but the research programme. Each programme has a two-layer structure: the hard core — the irrefutable-by-fiat basic assumptions of the programme (the hard core of the Newtonian programme is the three laws of motion plus universal gravitation); to give up the hard core is to give up the whole programme; and the protective belt — the auxiliary hypotheses, initial conditions, and data-processing conventions surrounding the hard core, which receives the impacts of anomalies and can be continuously adjusted. The death of a theory comes not from a single counterexample — people can live with anomalies — but from the degeneration of the whole programme: when every adjustment of the protective belt merely swallows old accounts and issues no new successful predictions, the community begins to abandon the programme; conversely, a programme that keeps producing predictions in excess of expectation is progressive. The specimens Lakatos himself cited are precisely the ones Parts Three and Four of this book have told: the Newtonian programme in its best years was the model of progress — every adjustment of the protective belt (booking one more source of perturbation into the account) carried with it a new cash check, and Chapter Thirteen's Neptune was the summit redemption of that check; while Vulcan, on Mercury's account, is the specimen of the degenerating phase — the protective belt was still moving, yet it could no longer issue a single certificate that dared stake coordinates and a time. One and the same structure, two account ages: the dividing line between progress and degeneration is exactly Chapter Thirteen's verdict — whether it still dares to open new bets.

Set this structure back into this book's ledger, and one sees that it is very nearly the history-of-science transcription of a principle in Section 3.4 of RC's General Outline. The General Outline, discussing dynamic evolution in Section 3.4, says: "In the early stages of divergence, formal stability at the level of surface institutions can usually be maintained, while the deeper rules undergo continuous adaptive adjustment." Chapter Fourteen has already redeemed this sentence with the Lorentz contraction — the ether was the surface institution, the textbooks going on being written, the professors going on being appointed, while the deep rules — length, simultaneity, the transformation properties — were being rewritten one by one in the dark. Lakatos's protective belt is the methodologization of this phase law: the hard core is the formal stability of the surface, the protective belt is the deep adaptive adjustment, and "degeneration" — paper money printed in ever greater volume, cash never again issuable — is precisely the phase where the deep adjustments surface to the top and the divergence enters its last stage. Chapter Fourteen's play in three acts (divergence on suspense, patches maintaining, re-convergence) and Lakatos's narrative (anomaly, adjustment of the protective belt, succession of programmes) develop the same batch of historical materials twice, and this is no coincidence: the two describe one and the same mechanism, one in RC's vocabulary, the other in the vocabulary of philosophy of science.

Lakatos's correction of Popper can, on this book's reading, be said in one sentence: falsifiability is not wrong; what was wrong was fixing the unit of testing too small. The channel of divergence is to be kept open — this Lakatos accepts whole; but the channel is not a straight passage for every single observational divergence; it is a bookkeeping channel with the programme as its unit. In Chapter Thirteen's notation: accounts are to be settled by programme, not entry by entry — Neptune-style cash and Vulcan-style paper money must accumulate through a full accounting period before the balance is known. This is also exactly the position of General Outline 1.4: the accumulation of observational divergence and the occurrence of re-convergence both take place at the level of the "rule framework", not at the level of single laws. Of the two examiners, one established the principle of the channel; the other measured the channel's width.

The Demarcation Problem: Channel, Not Content

The question of that seventeen-year-old can now be answered head-on: where, after all, does the line between science and ideology fall?

Popper's answer (draw it by falsifiability) needs, on this book's reading, one correction and one preservation. The correction: the "content" of single propositions cannot draw this line. One and the same proposition can appear in science and in pseudoscience — "a certain psychological disposition exists" can be a psychoanalytic dogma or a hypothesis under experimental test; "the heavenly bodies influence human affairs" can be astrology's all-purpose patter or the theory of the tides. The line, then, lies not in what is said, but in how it is said — and in what is done after saying it.

What is preserved is Popper's genuine insight: the line is in the channel. Science is a way of organizing that institutionalizes the open channel of divergence — the two meta-norms of Chapter Twelve are exactly it: results public (every result, with its method, its data, and its errors, enters the public record) and method reproducible (anyone, anywhere, by the published steps, should be able to walk the same path again); add peer review — which Chapter Three described as making "the mutual verification of observation across subjects" into procedure. The function of these three institutions is one and the same thing: to welcome and amplify the signal of divergence. The more contested a result, the more surely it is handed to those who took no part in it for re-examination; the more unexpected an observation, the more surely it is redone by other instruments, other groups, other continents — divergence is not treated as a threat; it is the formal entry by which the account is updated. The joint session of Chapter Fourteen and the Bell experiments of Chapter Fifteen, tightening the loopholes generation after generation, are this channel running at the highest specification.

Ideology — meaning here its pathological form, not every system of belief — installs the channel in reverse. Its three instruments, RC's General Outline has already listed in Section 2.2, discussing the bidirectional representation of language, and the list fits seam for seam: information overload — blurring the key parameters and piling up redundant noise to weaken the subject's capacity of judgment; semantic polarization — with extreme positions, abused original meanings, and boundless extension, rewriting the words' original space of expression and dissolving the basis of discussion among subjects; emotional override — replacing long-persistent solutions with short-term release. The General Outline gives the collective name of these three operations to contextual pollution — language used to accomplish discipline rather than cognition. Set the two side by side and one sees: not one of these three instruments "gets the content wrong"; each of them closes a channel — keeping divergence from arising, from being transmitted, or, once it has appeared, from being heard above the noise. So the answer to the demarcation problem within the RC framework can stand in one sentence: the difference between science and ideology lies not in what is written in the ledger but in whether the ledger still accepts overruling. The one makes observational divergence the entry by which the account is updated; the other makes observational divergence a threat to the security of the account — in Chapter One's vocabulary, the one keeps the interface for re-convergence; the other is welding it shut.

This reading also settles, in passing, an old pending case of the demarcation problem: is the falsifiability criterion itself falsifiable? Argued the old way, the criterion is not an empirical proposition and cannot be overruled by any observation, so it seems unscientific by its own standard. On RC's reading, the question itself asks the wrong category: falsifiability is not a theory about the world; it is an institutional design concerning the channel — just as "the accounts shall be public" is not a theorem of accounting but the charter that makes accounting possible. It does not accept the verdict of observation; it accepts a verdict of another kind: whether the cognitive systems organized by it really do converge more strongly, more quickly, more transferably. The evidence that four hundred years of the history of science have delivered on verdicts of this kind is precisely the whole content of Part Four of this book.

The Health of Reason

After the two examiners, RC can give its own verdict — the one that is this chapter's title.

The health of reason is measured by the capacity to keep converging, not by the stock of true theorems held. This is not rhetoric; the two are entirely different measures. By the latter, reason is an inventory of assets: how many proved theorems, how many verified laws, it holds — the longer the list, the healthier. By the former, reason is a metabolic capacity: can the signal of divergence still come in; once in, can it still be converted into updates of the account; after the update, is the channel still open. The General Outline, discussing processual completeness in Section 2.1, laid down the principle long ago: the completeness of cognition should ultimately and only inhere in the capacity of the process to iterate, not in the degree of completion of its conclusions. The oracle cannot be wrong, and so cannot learn — that verdict of Chapter Thirteen is the shortest statement of the standard of health.

The difference between the two measures is invisible in times of stability and a matter of life and death in times of disturbance. A system holding true theorems, meeting an anomaly, reacts first by defending the inventory — because the inventory is all it has; a system keeping the capacity to converge, meeting an anomaly, reacts first by booking it — because the anomaly is proof that the channel is open. The ether physics of Chapter Fourteen and the history of quantum mechanics of Chapter Fifteen acted out these two reactions each in its turn: the former kept the zero result off the books for eighteen years with the contraction hypothesis; the latter wrote "theory and the single observation never agree" directly into the theory's grammar. Which was the more modest, which lived the longer — history has settled the account.

And it is precisely here that this chapter must say one word in defense of "the modesty of reason", lest it be read as a posture. On this book's reading, modesty is not a rhetorical virtue but a structural property: it names the fact that a system keeps margin for what is not yet locked (A7), together with all the channels designed to maintain that fact — publicity, reproducibility, falsifiability, exchangeable foundations, re-observation. Popper legislated for one of these channels; Lakatos measured its width; and RC points out that the channels share one and the same root system: the root system of sustainable decision-making, keeping optionality present. The modesty of reason is the structure of reason, not its facial expression.

A landslide in the opposite direction must also be held back in passing. The twentieth century did indeed see people slide from "science also errs, also edits its history, also depends on community" all the way to "science is no different from myth — anything goes" — Feyerabend's Against Method (1975) is often cited as the source of that slogan. Chapters Seven and Nine have already refuted the Gödel edition and the geometry edition of this slide; here only the science-edition verdict need be added: it has confused the channel with the content once again. That the history of science is many-headed, confused, and contingent this book's four parts have never denied — convergence has never been a straight line; but from "the process is confused" it does not follow that "there is no difference"; the difference is written in the accounts: a system that leaves a channel for divergence has, in four hundred years, exchanged its ledger for ever thicker ones; a system that welds the channel shut can keep its balance page smooth only by contextual pollution. The health of the channel is observable: the rate at which divergence enters, the honesty of the bookkeeping, the degree to which the instruments are not burned when the account is changed — these are all quantities weighable within the RC framework; and the other face of anything goes is, precisely, the refusal to weigh.

The Handing On

The task of this chapter is hereby complete: the demarcation problem, rewritten within the RC framework as the problem of the channel; falsifiability and the research programme transcribed, respectively, as optionality kept present and as continuous adjustment of the deep rules; the health of reason given its measure. But one last layer remains unlifted. The channels this chapter has discussed — publicity, re-verification, falsification, the changing of accounts — all take the observer's end as given: humans observe, humans re-verify, humans decide whether to change the account. Yet along this four-hundred-year extended line, the position of the observer is itself being handed over to something new — a pattern-recognition device made by us that no longer works to our scale. When the observer itself becomes a manufactured device of convergence, what will become of the channel? That is the question of Chapter Eighteen — and the only reason this book dares to name this part "looking ahead".