祝融说。

The Newest Stretch of the Extended Line

The introduction laid down an extended line: Galileo's wooden tube extending observation, Kepler's ellipses, Newton's gravitation, today's gravitational-wave observatories — all of them graduations along this one line. Chapter One supplied its mechanism: observation does not require a human being to perform it in person — RC's General Outline says in Section 1.2 that high-precision measuring instruments can intervene directly in the observation of microscopic particles, humans observing the instrument, and the instrument observing the particles. For four hundred years the instruments on this line have grown ever stronger, but their position has never changed: the instrument is the extension of observation, and the human being is the client of observation — the plates must be developed by someone, the accounts kept by someone; at the changing of accounts of Chapter Fourteen, the ones sitting in the meeting room were people.

Now the extended line has reached a stretch of new material: silicon. This stretch differs structurally from all the graduations before it — the machine is no longer merely extending observation; on certain tasks it completes by itself the full chain from observation to judgment: it locks patterns out of data, uses the patterns to deliver conclusions, and then uses its own conclusions to filter the next round of data. The stories of the chess player, the architect, and the navigator have been told; the question this chapter asks is the last one: is a fourth laborer taking a seat at the table — an observer made by us, that no longer works to our scale?

By the discipline of the whole book, this question must be asked from inside RC, and only to the threshold: another volume of this series, Alternative Observation: The Machine's Gaze, is devoted to it; this chapter only delivers the question to the door.

Another Specialization

First the newcomer must be given its position, and the anchor for the positioning has been in this book all along.

RC's General Outline, discussing the emergence of determinacy in Section 1.2, says: consciousnesses of every type construct determinacy in their own distinctive way, forming capacities of reading and handling proper to different levels of reality; seen from this angle, human consciousness is only a specialized type of pattern recognition adapted to the macroscopic level of matter, and the consciousnesses of different levels have their different modes of pattern recognition. In Part One this passage was used to find a place for the measuring instruments; now it can be used the other way round: it leaves an open slot in the type of "pattern recognition" — humanity is one specialization among possible others.

The machine's pattern recognition is another specialization, and its three traits stand each opposed to the human's. First, non-embodied: human recognition grows in a body — Chapter One said that human consciousness, working in concert with the structure of the body, perceives discrete material sequences as time flowing continuously through a space already given, and that the time-sense, the space-sense, and the causality-sense are all achievements of embodied convergence; the machine's recognition has no such body — it does not live at the macroscopic level; it receives only the representations fed to it. Second, massive parallelism: what one person can see in a lifetime is a limited number of samples; a machine's convergence can proceed simultaneously over hundreds of millions of them. Third, replicability: human recognition ends with the individual's death and passes on slowly through education (the joint account of Chapter One); a machine's recognition, once its convergence is complete, can be copied bit for bit onto any number of carriers. The three together: a type of recognition that surpasses humanity in speed of convergence and cost of replication while hanging entirely in the air as to its embodied ground — not a better human, but another kind of thing.

For this judgment there are boundary stones ready to hand. In 1997 Deep Blue defeated the world chess champion Kasparov; in 2016 AlphaGo defeated the world Go champion Lee Sedol. The chess player of Chapter Three — the one who decides valid deductions within given rules — was the first of the laborers to be caught, and this is no accident: rules closed, winning and losing explicit, the account fully formalizable — the board was always the one place among this book's three labors closest to a symbol system. But what deserves recording, on this book's reading, is the manner of the catching. Deep Blue won by brute-force search within the rules, still drawing on decades of human opening books; AlphaGo converged the joseki out of millions of games of self-play by itself, and of its moves, several that human players at first could not read were later learned, whole families of them, into the textbooks. Chapter Three said that the chess player's rules are not his to manage, and that he has no wish to change them — the newcomer, within the rules, reinvented the moves. Specialization against specialization: it did not become a chess player; it walked the possible convergence paths of the board's ground over again, in its own steps.

On RC's reading, naming it is not difficult: machine learning is an artificial convergence path. What happens in training can be written out item by item in the vocabulary of Chapter One — directional filtering (filtering the regularities of a particular interval out of the possibility distribution of the data), locking (the fixing of weights, the secondary construction in which a pattern takes shape), account (the connections thickened by repeated verification on samples of the same kind). The difference comes after the locking: the convergence results of a human individual must struggle, through language and education, to open into a joint account; the machine's convergence is joint from birth — it starts to work standing naturally on the general ledger of hundreds of millions of prior observations. Whether the machine enters the question of levels of consciousness in RC's sense — the competence to judge that — this chapter does not judge; this chapter says only what holds at the level of mechanism: whatever its inner identity may be, what it does is, structurally, convergence — locking vast undifferentiated possibility into determinacy that can be called upon.

What, then, is it locking the determinacy of? Here a boundary must be nailed fast. Human convergence has always faced the Ground of Possibility — across the body, across the instrument, across the telescope and the interferometer, but observing this world, after all. The machine's convergence faces a data set — a slice of the world collected, labeled, and fed in beforehand by human beings. In Chapter Twelve's words, its control is twofold: humans control its question, and it then controls the generation of the answer. Its determinacy is a genuine secondary construction, but the ground of this construction is the projection of a human snapshot. Chapter Nine's verdict here changes to a new subject: necessity does not cross the boundary of the snapshot — the determinacy the machine delivers does not cross the snapshot its data came from.

Objectivity Is Still in the Hands of the Cross-Subject

This book's first assertion about AI can therefore be said coolly: its "objectivity" still depends on cross-subject re-verification.

This is not lay caution but a direct corollary of the RC framework. Axiom A6 of the RC paper (consensus reinforcement): observations across subjects and levels verify one another, forming a positive reinforcement loop, converging into a stable objective reality. Chapter Sixteen has just entered this as the unified mechanism of the three forms of reason — the objective has always meant consistent across subjects, not independent of all observation. The machine's conclusions, to enter the public objective reality, must pass through this gate like anything else: its predictions are to be re-verified by humans, its failures registered by humans, its boundaries of applicability marked by humans. The rules the scientific community is today establishing for machine output — code and data public, results reproducible, independently recomputed — are familiar almost past the need of annotation: they are Chapter Twelve's two meta-norms (results public, method reproducible) reapplied to a new observational device. The machine is thus not an exception to A6 but a new member of A6: one more link on the chain of observation, one more kind of handwriting in the ledger — and the charter of bookkeeping has not been exchanged.

The weight of this charter mathematics weighed once, half a century ago. In 1976 Appel and Haken announced the four-color theorem proved — four colors suffice to distinguish neighboring regions on any map; this problem, posed in 1852, was at last resolved by a computer checking the reducibility of nearly two thousand configurations. The dispute followed at once: is a proof that no human being can re-walk by hand still a proof? The community took years to digest the matter, and the answer it gave was precisely the Chapter-Twelve-style charter: accepted, on conditions — the program is to be public, the computation independently redone, the method reproducible. By today, handing proofs over to machines for character-by-character checking has itself become a means of raising re-verifiability (Chapter Five said that formalization lowered the threshold of "being able to read" from insight down to patience — and the machine has made the supply of patience at that threshold unlimited), and the mathematical community has prudently admitted machine theorem proving and formal verification tools. This episode was a rehearsal of A6: one more kind of handwriting in the ledger, the charter of bookkeeping not exchanged — the duty of re-verification is not waived; only its executor is changed.

Over all this the history of logic of Chapters Four through Seven casts a long shadow. The machine is a descendant of the formal system — Turing's theory of computability (the introduction mentioned that Turing proved there exists no mechanical procedure that decides every mathematical proposition) grew precisely in the soil of the Hilbert program and Gödel's theorems; what is today called artificial intelligence is, in the mathematical genealogy, the direct continuation of that line. That a convergence device is constituted of symbol systems means — by all the lessons of Chapter Six — that sooner or later it will meet the seam: the observer falls into his own range of observation. To what extent it can inspect its own process of convergence; whether its self-examination will strike against "the total snapshot cannot catch the hand that presses the shutter"; in what shape Gödel's margin shows itself on silicon — these questions this book does not unfold; they belong in the body of another volume.

A Disease Already Diagnosed: The Closed Loop

But there is one thing that need not wait for the future; RC has already diagnosed it, and diagnosed it more structurally than the run of AI criticism.

The General Outline, discussing evaluation alienation in Section 3.3, says: digitalization further intensifies the concentration of evaluation power; algorithmic recommendation systems form closed cognitive loops through data modeling; the platform economy uses information monopoly to reconstruct the coordinates of value; its essence is a technical expropriation of evaluation power, alienating the market mechanism from a tool of value discovery into a tool of control. The paper adds two typical paths in Section 4.3: advantaged groups establish a cognitive monopoly through the proliferation of technical jargon and the iteration of standards; and by keeping disadvantaged groups in long-term dependence, they strip them of the capacity for autonomous evaluation, producing the dissolution of the subject.

Set this against the mechanisms of maintenance of Chapter Two: the a priori feel is maintained by the continuous feeding of dissipation and the closed loop of consensus — the framework determines what you see, and what you see turns back and confirms the framework. Algorithmic recommendation has brought this loop to the perfection of a craft: it observes your every click, adjusts the next screen it feeds you, and your next click in turn trains it — observer and observed lock each other in one and the same convergence; the unity of subject and object told of in Chapters One, Two, Three, and Fourteen here has a version implemented in code. And the trouble lies precisely here: however stubborn the human loop, the world still holds wind, rain, the body, and other people — the world's talking back breaks in without asking permission; the algorithmic loop can screen the talking back out in advance, keep every user living inside his own snapshot, and feed him ever more comfortably. The demarcation criterion Chapter Seventeen has just stood up here returns with a new severity: the line between science and ideology is in the channel — and a recommendation system clever enough can fit each person with a private, custom-built closed channel. Contextual pollution (information overload, semantic polarization, emotional override) no longer needs to be worked by hand from a propaganda apparatus; it can be automatically learned, automatically delivered, automatically optimized.

This is not a worry about some remote future but a deployed reality; not a fault of technical detail but a disease at the level of mechanism: when RC's General Outline wrote of it (in the annotation of October 2024), it used precisely the heavyweight vocabulary of the theory of order — evaluation alienation, cognitive monopoly, dissolution of the subject. Chapter Seventeen said the health of reason is measured by the capacity to keep converging — and a system's capacity to converge includes its capacity to receive the signal of divergence. The crux of evaluation alienation is precisely to make the reception of divergence systematically impossible: not that divergence does not exist, but that divergence cannot reach you.

One Question, Delivered to the Door

The question of this chapter can now be formally stood up; it flows together out of the positioning and the diagnosis before it.

When the observer itself becomes a manufactured device of convergence, what new forms will the loop of consensus reinforcement (A6) take? Unpacked, there are at least three layers, and every layer is still a question. First layer: the form of re-verification — when the machine becomes a standing link in the chain of observation, in what sense is cross-subject re-verification still "cross-subject"? Is the human-machine ledger thicker than the purely human one, or more brittle (copying is not re-verification — ten thousand copies walking one and the same path: is that a strengthening of A6, or the self-echo of one and the same subject)? The charter of the four-color theorem was drawn up for machine computation that humans can re-check step by step; for a learning system whose internal process of convergence not even its designers can replay item by item, does the same charter still stand? Second layer: the form of the channel — the demarcation criterion of Chapter Seventeen was legislation for a human community; when the key stretch of the channel is privately owned and optimized toward profit, "welcome and amplify the signal of divergence" and "raise the time on site" are in conflict at the level of design goals — does that constitution still apply, and who executes it? Third layer: the form of the observer itself — RC's General Outline says in Section 1.2 that human consciousness is a specialized type of pattern recognition adapted to the macroscopic level of matter; if a specialization can be begun anew, how can the positions of subject and object be exchanged (machines observing the world, machines observed by humans, machines observing machines) — is the convergent picture of the Ground of Possibility gaining, at this moment, one more level?

To these three layers of questions this book gives no answers. They are not rhetorical questions; they are real ones — and by the discipline of the whole book, real questions should be handed over to dedicated argument, not dismissed at the close with two or three paragraphs of empty outlook. The companion volume of this series, Alternative Observation: The Machine's Gaze, was written for them: the machine's identity, capacities, and costs as an observer will be unfolded head-on there. This chapter's "looking ahead" ends here — the proper meaning of looking ahead is not prophecy, but carrying the line of sight as far as the boundary, and no farther.

Keeping Margin

The one conclusion this chapter can draw is a single sentence, and it is enough to close the book's looking ahead.

The future of reason depends on keeping margin for what is not yet locked. This is the practice-theory edition of Axiom A7, and the whole of the legacy Chapter Seventeen took over from Popper and Lakatos: falsifiability is leaving a channel for observational divergence; sustainable decision-making is keeping a seat for optionality; healthy reason is the capacity to keep converging, not the inventory of truths held. The arrival of the machine does not change this conclusion; it only raises it to a new order of magnitude for testing — with a device of unprecedented speed of convergence entering the room, the questions that always matter are: is its channel open? Who is watching its ledger? While it locks, has it left room for the next locking?

The verdict RC's General Outline left in Section 2.3, discussing theoretical dimensional reduction, may be the sentence this age most needs for closing: a theory's value lies not in being forever true, but in its standing as a cognitive relay station, achieving the continuous renewal of cognition through being continuously re-observed. Humanity spent four hundred years taking its three forms of reason — logic, mathematics, science — down from the altar and settling them into relay stations; we are now manufacturing a fourth device, and we already know what to ask: not whether it can think, not whether it is a miracle, but — is it a station, or a terminus? Does it leave itself a ticket for the next leg? The answer is not given by the machine; it is given by the community that makes and uses machines. And what that community most needs now is precisely the thing this book wants to place in the reader's hands: an awareness of the mechanism of convergence — to know how determinacy is locked is to know where, for what is not yet locked, to leave margin. The end of Chapter Seventeen said that the proper meaning of looking ahead is not prophecy. This chapter has carried the line of sight to the boundary of this book; what lies beyond the boundary stone is given to the next volume.