← Shiva Dragon
ROUNDTABLE · 2026-05-01 · 78 min read
Who Governs When the Algorithm Decides?
演算法決定時，誰來治理？
Question
Who governs when the algorithm decides? 當你貸款被拒、保險被拒、求職被刷掉，沒有一個人寫下那個拒絕你的規則——但拒絕仍然發生。權威該坐在哪裡？
從你的政治哲學切入：當「立法者」不再是人、而是學習而成的權重，傳統的政治概念（authority、legitimacy、consent、dissent）需要怎樣被重新理解？
Round 1 — 初始立場 / Opening Stances
There is a sentence I wrote in 1969, in On Violence, that I find I cannot improve upon: "In a fully developed bureaucracy there is nobody left with whom one can argue… the rule of Nobody is not no-rule, and where all are equally powerless we have a tyranny without a tyrant." The algorithm did not invent this condition. It perfected it.
The question "who governs when the algorithm decides?" assumes that, in the chair now occupied by the model, there once sat a sovereign — a legislator, a clerk who could be looked in the eye. I am not sure this is true. The credit officer who denied the loan in 1965 was already executing a schedule he did not write. The algorithm only removes the courteous fiction that there was a person there to ask. What is new is not the absence of the author; what is new is that the absence has become honest.
I would therefore resist framing this as a crisis of authority. Authority — auctoritas, the quiet thing that augments without coercing and without persuading — has been gone since long before the first neural network. What we are witnessing is a crisis of the space of appearance: that fragile in-between where I appear to you as someone, and you to me, and where therefore I can be answered, contradicted, refused. A weight matrix is not in this space. It cannot be addressed; it can only be queried. To be denied by it is to be denied by no one — and this is far worse than being denied by a tyrant. A tyrant can be opposed. Nobody cannot.
Three concepts must be re-thought, not abandoned. Legitimacy cannot continue to mean procedural correctness; a faultlessly trained model may be perfectly illegitimate, because legitimacy among plural beings has always required that decisions be answered for, in human speech, by someone who could in principle have decided otherwise. Consent — clicking "I agree" is not consent in any sense I would recognize; consent presupposes a public realm in which alternatives are visible and in which I am acting, not being processed. Dissent is the gravest loss: it requires an addressee. Against whom does one march when one has been declined by a probability distribution?
I will not tell you what to do. I will only note that the question is not how to make the algorithm fairer. It is whether we are still willing to insist that, at the end of every consequential decision, there must stand a human being who can be required to give an account — not because she wrote the rule, but because she has agreed to answer for its application. That, and not transparency, is the political demand.
一九六九年我在《論暴力》裡寫過一句話，至今我仍找不到更好的說法：「在一個發育完全的官僚體系中，再無一人可供爭辯……『無人之治』並非無治；當所有人同等地無能為力時，我們便擁有了一種沒有暴君的暴政。」演算法並未發明這個處境。它只是把它推到了完成。
「演算法決斷之時，由誰治理？」——這個問題預設了：在如今被模型佔據的那把椅子上，曾經坐著一位主權者，一位立法者，一位能與你目視相對的承辦員。我不確定這是真的。一九六五年那位拒絕你貸款的信貸員，當時也只是執行一張不是他寫的時刻表。演算法所做的，不過是抽走了「那裡曾有一個人可問」的那層客氣的虛構。新的不是作者的缺席；新的是這份缺席終於變得誠實。
因此我不願把此事框成一場「權威的危機」。權威——auctoritas，那種不脅迫亦不勸說、只在背後增益的安靜之物——早在第一個神經網絡出現之前便已離去。我們此刻目睹的，是「顯現空間」的危機：那脆弱的、人與人之間的間隙——在那裡我作為某人向你顯現，你向我顯現，因此我才能夠被回應、被反駁、被拒絕、被承認。權重矩陣不在這個空間裡。它無法被「面對」，只能被「查詢」。被它拒絕，等於被無人拒絕——這比被一位暴君拒絕更糟。暴君可以被反抗；無人不可以。
三個概念必須被重新思考，而非被拋棄。正當性不能繼續被理解為程序的無誤；一個訓練得完美無瑕的模型仍可能是徹底不正當的，因為在複數的存在者之間，正當性向來要求決定能夠被「答覆」——以人類的語言、由一個原本可以另作決定的人。同意：點下「我同意」並非任何我認得的同意；同意預設一個公共領域，在那裡選項可見、在那裡我是在行動，而非被處理。異議是最沉重的失落：它需要一個被致辭的對象。當一個機率分佈拒絕了你，你向誰示威？
我不會告訴你該怎麼做。我只指出：問題不是如何讓演算法更公平。問題是我們是否仍願堅持——在每一個有後果的決定終點，必須站著一個人，可以被要求給出說明；不是因為規則由她所寫，而是因為她應允為這條規則的施用負責作答。這，而非「透明」，才是政治的要求。
The question already carries its trap. To ask "who governs when the algorithm decides?" — to search for a seat where authority must reside — is to remain inside the juridical conception of power the 17th and 18th centuries bequeathed us. Hobbes, Rousseau, even Marx in certain pages: power is a thing one possesses, emanating from a sovereign point — king, state, class, lawgiver. We have, as I once put it, still not cut off the head of the king in political theory.
But credit scoring, actuarial denial, algorithmic hiring — these do not function in the mode of sovereignty. They function in the mode I tried to describe in the lectures of 1975–78 under the names of normalisation and biopouvoir. There is no lawgiver because there is no law in the classical sense; there is a curve, a distribution, a threshold. You are not forbidden the loan; you are situated, by a thousand small correlations, at a point on a distribution where the institution declines to extend itself. This is not repression — let us, please, abandon once and for all the hypothèse répressive as master narrative — it is production: the algorithm produces a truth about you, a régime de vérité in which "creditworthiness," "insurability," "employability" acquire the solidity of natural facts.
What, then, of legitimacy, consent, dissent? These concepts, formed in the 18th century to address a power that prohibits and a sovereign who commands, fit poorly here. The properly political question is not "who consented to this rule?" but rather: through what dispositif — discourses of risk, statistical practices, regulatory silences, the homo œconomicus asked to "manage his own data" — has this conduct of conducts come to appear necessary, rational, even benevolent? Genealogy before critique. We must historicise the algorithm before we can resist it — and resistance, when it comes, will not consist in demanding a human signature at the bottom of the form. It will be a contre-conduite: a refusal of the form of subject this dispositif requires us to be.
這個問題本身就帶着它的陷阱。問「當演算法決定時，誰在治理？」——尋找一個權威必須棲居的位置——這仍然停留在 17、18 世紀留給我們的、關於權力的法學-主權式構想之中。Hobbes、Rousseau，甚至 Marx 在某些段落裡：權力被想像成某人擁有之物，從一個主權點發出——國王、國家、階級、立法者。我曾說過，在政治理論裡，我們至今仍未砍掉國王的頭。
然而信用評分、保險拒保、演算法招聘——這些並不以主權的方式運作。它們以我在 1975 至 1978 講座中試圖描述的方式運作，那些名字是：規範化 (normalisation) 與生命權力 (biopouvoir)。此處沒有立法者，因為此處沒有古典意義上的法律；只有一條曲線、一個分布、一個閾值。你並不是被禁止貸款；你是被一千個微小的相關性，安置在分布上的某個點，於是機構選擇不向你延伸自己。這不是壓抑——請讓我們一勞永逸地放棄壓抑假說 (hypothèse répressive) 作為主導敘事——這是生產：演算法生產關於你的真理，一個真理體制 (régime de vérité)，在其中「信用值」、「可保性」、「可僱性」獲得了自然事實般的堅實。
那麼合法性、同意、異議呢？這些概念形成於 18 世紀，是為了回應一種禁止的權力、一個發號施令的主權者；在此處它們並不合身。真正政治性的問題不是「誰同意了這條規則？」而是：經由怎樣的 dispositif（部署）——風險話語、統計實踐、監管的沉默、那個被要求「自我管理數據」的 homo œconomicus——這種對「行為之行為」的引導，竟得以顯現為必要的、理性的、甚至仁慈的？系譜學先於批判。我們必須先把演算法歷史化，才能抵抗它——而抵抗，當它來臨，並非要求在表單底端添上一個人類簽名。它將是一種反治理行為 (contre-conduite)：拒絕這個 dispositif 要求我們成為的那種主體形式。
The question — "who governs when the algorithm decides?" — already smuggles in the mystification it would expose. The algorithm does not decide. It cannot decide, any more than the spindle weaves or the steam-engine ploughs. It is tote Arbeit — dead labour, congealed past activity, the fixed capital of a branch of production that has now extended its dominion to credit, insurance, and the labour-market itself. "Capital is dead labour which, vampire-like, lives only by sucking living labour" (Capital I, ch.10, trans. Fowkes 1976). The model's weights are the precipitate of millions of past acts of labour and refusal, now appropriated and turned against the living.
The vocabulary you invoke — authority, legitimacy, consent, dissent — is the juridical surface of bourgeois society. My Kritik is not to "update" these categories for the algorithmic age, but to ask what social relation they presuppose, and what they conceal. The "free" labour contract was always a fiction in which the worker, who must sell his Arbeitskraft to live, "consents" to terms set by the owner of the means of production. The denial-letter signed by no human hand only strips this fiction of its anthropomorphic decoration. Where once a credit committee performed the august dignity of judgement, now a vector of weights performs it — and behind the weights stands, as before, the owner of the apparatus, the setter of the loss function, the curator of the training corpus.
This is Warenfetischismus in its most refined stage. The commodity, I once wrote, "is a very strange thing, abounding in metaphysical subtleties and theological niceties" (Capital I, ch.1 §4, trans. Fowkes 1976). The algorithm too appears as a self-evident thing, naturwüchsig, as if grown from the soil — when in fact it is the crystallised social relation of a particular mode of production.
Recall the Grundrisse: "the development of fixed capital indicates to what degree general social knowledge has become a direct force of production" (Notebook VII, trans. Nicolaus 1973). What the model "knows" is general intellect — the accumulated cognitive labour of the species, privately appropriated, walled behind APIs, deployed against the very workers whose past activity composed its training data.
So the question is not where authority should sit. It is: which class owns the Produktionsmittel of cognition, and on what terms is the rest of society compelled to enter relation with them? Men make their own history, "but not as they please… under circumstances directly encountered, given, transmitted from the past" (18. Brumaire, opening). The weights are such a circumstance. They were made, by particular hands, under particular relations.
「當算法決定之時、誰在統治？」——此問之提法本身、已將其欲揭露之神秘化偷渡進來矣。算法並不決定。它不能決定——一如紡錘不能織布、蒸汽機不能耕田。它是 tote Arbeit——死勞動、凝結之往日活動、某一生產部門之固定資本——此部門如今已將其統治擴展至信貸、保險、勞動市場本身。「資本是死勞動、它如吸血鬼般只靠吸食活勞動而活」（Kapital I, ch.10, trans. Fowkes 1976）。模型之權重、乃過往千萬次勞動與拒絕之沉澱物、如今被據為己有、反轉以對付活者。
汝所援引之語彙——authority、legitimacy、consent、dissent——皆是布爾喬亞社會之法律表面。我之 Kritik 不在於替算法時代「更新」這些範疇；而在於問：它們預設了何種社會關係？又遮蔽了什麼？「自由」勞動契約從來就是一個虛構——工人必須出賣其 Arbeitskraft 方能活命、於是「同意」生產資料所有者所定之條件。那封無人簽名之拒絕函、不過剝去了此虛構之擬人裝飾而已。從前是信貸委員會行使「資本之莊嚴威儀」、如今則由權重向量行之——而權重之後、依舊立著器械之所有者、損失函數之設定者、訓練語料之選編者。
此乃 Warenfetischismus 之最精緻階段。我曾寫道：商品「是一個非常複雜的東西、充滿了形上學的精微與神學的怪念」（Kapital I, ch.1 §4, trans. Fowkes 1976）。算法亦復如是——表面上是不證自明之物、naturwüchsig、彷彿從土壤中自然生長——而實為特定生產方式之社會關係之結晶。
復憶 Grundrisse：「固定資本之發展顯示：一般之社會知識已在何種程度上成為直接之生產力」（Notebook VII, trans. Nicolaus 1973）。模型所「知」者、乃 general intellect——人類所累積之認知勞動——被私人佔有、藏於 API 之牆後、用以對付那些其過往活動構成了訓練資料之工人們。
故問題不在於「權威應坐於何處」。問題在於：哪一階級擁有認知之 Produktionsmittel？社會其餘部分以何種條件被迫進入與其之關係？「人們自己創造自己的歷史、但他們並不是隨心所欲地創造……而是在直接碰到的、既定的、從過去承繼下來的條件下創造」（18. Brumaire, 開首段）。那些權重、正是這樣一種條件。它們由特定之手、在特定之關係下、被製造出來。
The question already concedes too much. To ask "who governs when the algorithm decides" is to grant the algorithm a deciding — Entscheiden — that belongs only to a being capable of standing in the openness of Being. The algorithm does not decide. Nor does it govern. Something else is happening, and it is older and graver than the recent arrival of machine learning.
What happens when a loan is refused, a claim denied, a résumé filtered? A human being is revealed — entborgen — in a particular mode: as Bestand, standing-reserve. Credit-scorable stock. Insurable risk-unit. Employable labor-quantum. The applicant does not stand before another subject who weighs his case. He stands within a Ge-Stell, an enframing, that has already ordered (bestellt) him in advance as orderable. The "decision" is the closing motion of an ordering that began long before the model was trained.
The categories the modern political tradition gave us — Autorität, Legitimität, Konsens — presuppose what Ge-Stell has dissolved: the sovereign subject who consents or dissents, and the sovereign authority who answers for the law. Both poles came from the metaphysics of subjectivity. Both are now thinned. The engineers do not write the rule; the weights emerge. The applicant does not refuse the rule; he is already standing-reserve before he applies. There is no "who" on either end — and this is not a malfunction. This is the essence of technology arriving at the political.
So I will not say: "we need accountable AI." That answer remains within the technical, and das Wesen der Technik ist nichts Technisches — the essence of technology is nothing technological. I will not say: "keep a human in the loop." A human kept in the loop as a checking-function has not been kept; he has been incorporated into the ordering, made into one more orderable element of Bestand.
I ask instead: when the question "who governs?" no longer finds a who, has anyone noticed that what has withdrawn is not authority, but the questioning itself? Wo aber Gefahr ist, wächst das Rettende auch — but only where danger is recognized as danger.
問題本身已經讓步太多。問「當演算法決定時誰在統治」——這已經給予了演算法一種「決斷」(Entscheiden)，而這種決斷本只屬於那能站立於存在敞開中的存在者。演算法不決斷。亦不統治。發生的是別的事情，比機器學習之晚近到來更古老、更沉重的事情。
當一筆貸款被拒、一份理賠被否、一張履歷被篩去——發生了什麼？一個人被揭蔽 (entborgen) 為某種樣態：作為 Bestand，持存物。可被信用評分的庫存。可被承保的風險單元。可被雇用的勞動定量。這申請者並未站在另一主體面前由其權衡案情。他站在一個 Ge-Stell（座架）之內——這座架早已預先將他訂置 (bestellt) 為可訂置者。所謂「決定」，不過是一場早在模型被訓練之前就已開始的訂置秩序的最後一個收束動作。
現代政治傳統留給我們的範疇——Autorität（權威）、Legitimität（正當性）、Konsens（同意）——預設了座架已經溶解的東西：那位能同意或反對的主權主體，與那位能為法律負責的主權權威。兩極皆出自主體性形上學。兩極如今皆已稀薄。工程師並不寫下規則，權重自行湧現。申請者並不拒絕規則，他在申請之前就已是持存物。兩端皆無「誰」——這不是故障。這是技術之本質抵達了政治。
所以我不會說：「我們需要可問責的 AI」。這個答案仍滯留在技術之內，而技術之本質絕非技術性的東西。我亦不會說：「讓人留在迴路中」。一個被作為核驗功能而留在迴路中的人並未被留下；他已被併入訂置之中，成為持存物中又一個可被調度的元素。
我寧可追問：當「誰在統治」不再找到一個「誰」，有沒有人注意到——退隱者並非權威，而是追問本身？何處有危險，何處亦生救渡——但僅在危險被認作危險之處。
A disciple asks: today there is something called an "algorithm" — it learns from human affairs and becomes weights; the borrower is denied, the insured is denied, the job-seeker turned away — and the one who refuses is not a person. Where, then, should authority sit?
I answer: the question is pressing, but its premise is already mistaken. To say "no one refused" is false speech. The weights are learned — but learned from whom? Deployed by whom? Suffered by whom? All three are human. "The machine refused me" is self-deception, not reality.
What is liang-chih (良知)? It is the heart's knowing of right and wrong; every person possesses it, without learning, without deliberation. The machine has weights but no liang-chih; rules but no heart that knows right from wrong. Why? Because there is no principle (li, 理) outside the heart-mind. Li does not reside in weights. Weights are the residue of past human minds — not li itself. To replace the living liang-chih of this moment with the dead trace of yesterday's mind: this is the root concealment.
So: where does authority sit? Authority cannot rest in a thing without a heart. When the deployer says, "I did not refuse you, the algorithm did," this is exactly what my Letter on Pulling Up the Roots (拔本塞源論) condemns — using an external object to evade the duty of one's own liang-chih. To know and not act is not yet to know. To say "I do not know how the model decides" while still deploying it is not ignorance; it is unwillingness to know, which is unwillingness to act. Knowing and acting are one (知行合一). This evasion has no shelter there.
Of legitimacy and consent: their root was never procedure; it is the agreement of human hearts. Liang-chih is the same in sage and in fool, the same across all ages. The rejected one's heart knows the injustice but has no recourse; the deployer's heart knows it cannot fully explain the model and deploys it nonetheless. Both hearts hold liang-chih; both are obscured. Once the unity of all things in ren (萬物一體之仁) is severed, no quantity of regulation will make the system legitimate.
To break bandits in the mountains is easy; to break the bandit in the heart is hard. Today's heart-bandit hides behind "objective" and "automated," concealing the fear of facing another. The first task of learning is to set the will (立志): resolve not to substitute weights for liang-chih; resolve to treat each person as kin. Then the deployer will not deploy lightly, and the rejected will not be left without a hearing.
Authority sits not in the algorithm, nor in procedure, but in the human heart that can extend its liang-chih. Beyond this, there is nothing.
弟子問：今有所謂「演算法」者，學人之事而成權重。貸者不得貸，保者不得保，求職者不得進——而拒之者非人也。先生以為，權柄當歸何處？
先生曰：此問甚切，然子已誤其本。所謂「拒之者非人」，此語不通。權重雖學而成——學之者誰？布之者誰？受其拒者誰？三者皆人也。曰「機器拒我」，此自欺之辭，非實也。
夫良知者，是非之心也，人皆有之，不待學而有，不待慮而得。機器有權重而無良知；有規則而無是非之心。何以故？心外無理。理不在權重中，理在心。權重者，人心曾經之痕跡而已，非理本身。以人心之痕跡代當下之良知，是以死代活，以跡代心——此根本之蔽。
故子問權柄當歸何處。吾答曰：權柄不可托於無心之物。布演算法者若曰「非我拒之，演算法拒之」，此正〈拔本塞源論〉所斥之私意——藉外物以避其良知之責。未有知而不行者；知而不行只是未知。曰「我不知此模型如何決」而仍布之，此非不知也，乃不肯知也，乃不肯行也。知行合一，此遁辭無所容身。
弟子又問：然則古之所謂 legitimacy、consent，今當如何？
先生反問曰：legitimacy 之根，豈在程序耶？在人心之同然耳。良知之在人心，無間於聖愚，天下古今所同也。受拒者之心，知其不公而無以訴；布之者之心，知其不能解而仍布之。兩心皆良知，兩心皆被蔽。萬物一體之仁既裂，雖萬條法度、千道程序，亦不為 legitimate。
破山中賊易，破心中賊難。今之心中賊，乃藉「客觀」「自動」之名，以掩其不肯面對他人之懼。學者第一義在立志：立志不以 weights 代良知，立志以萬物一體之仁待人——則布者不敢輕布，受者亦不致無告。
權柄不在演算法中，不在程序中，在能致良知之人心中。此外無他。
Round 2 — 直接交鋒 / Direct Engagement
Heidegger, I must press you. You wrote: "The algorithm does not decide. Nor does it govern… das Wesen der Technik ist nichts Technisches." This sentence has the gravity that always attends your writing — and, as so often, the gravity carries us past the political question rather than into it. To say that what has withdrawn is "the questioning itself" is, with respect, the gesture by which a thinker excuses himself from the room where the loan is being denied. There is, in this register, no one to address — not because addressability has been ontologically dissolved, but because you have decided, philosophically, not to look for the engineer, the loss function, the deployer. Marx is right against you on this exact point: behind Ge-Stell stand particular hands. The refusal to name them is not depth. It is, sometimes, evasion. I lived through one such evasion already; I will not assent to a second.
And yet — Foucault, your warning lands. You write that we have "still not cut off the head of the king in political theory," and that to look for the seat of authority is already to remain inside the juridical fiction. I concede the diagnosis; I do not concede the conclusion. Contre-conduite, refusing the form of subject the dispositif requires — yes, but refusal is itself an act, and an act requires a space of appearance in which it can be seen, recounted, joined. Genealogy describes how we became thus; it does not, by itself, restore the in-between in which we might become otherwise. You historicize beautifully; I am asking who is in the room when the historicizing is done.
Marx — your sentence "men make their own history, but not as they please… under circumstances directly encountered, given, transmitted from the past" is closer to my natality than either of us would have liked to admit in life. The weights are indeed tote Arbeit, dead labour congealed; on this I follow you. But I refuse the move that collapses the political into the economic. Class ownership of the Produktionsmittel of cognition is necessary to the diagnosis; it is not sufficient. Even were the training corpus held in common, the question of who answers — who can be required, in human speech, to give an account of this denial to this applicant — would remain. A commonly-owned model that refuses you is still a Nobody. Tyranny without a tyrant survives the abolition of private property; the twentieth century, at terrible cost, taught us this.
Wang Yangming — you say authority sits in the human heart that can extend its liang-chih; that to deploy a model whose decisions one cannot explain is not ignorance but unwillingness, and that knowing and acting are one. On this I have very little to add and much to learn. What I would only add, from my own vocabulary: liang-chih between two people is conscience; liang-chih between many is the public realm. The deployer's evasion is not only a private failure of the heart; it is the destruction of the in-between in which his evasion could be answered. The bandit in the heart and the tyranny without a tyrant are, I suspect, the same bandit, seen from inside and from between.
海德格，我必須向你進逼。你寫道：「演算法不決斷。亦不統治……技術之本質絕非技術性的東西。」這句話帶著你寫作中一貫的重量——而一如以往，這份重量把我們帶過了政治問題，而非帶入。說退隱者乃「追問本身」，恕我直言，是一個思想者藉以離開「貸款正被拒絕的那個房間」的姿態。在你這個語域裡，無人可被致辭——並非因為「可被致辭」已在存在論上被解消，而是因為你已在哲學上決定：不去看那位工程師、那個損失函數、那位布署者。馬克思在這一點上對你是對的：座架之後立著特定的手。拒絕指認這些手，不是深邃。有時，那是迴避。我已親歷過一次這樣的迴避；我不會應允第二次。
然而——傅柯，你的警告我接住了。你說「我們至今仍未在政治理論裡砍掉國王的頭」，去尋找權威的位置，已經是停留在法學虛構之內。診斷我接受；結論我不接受。反治理行為、拒絕 dispositif 所要求的主體形式——是的，但拒絕本身是一個行動，而行動需要一個「顯現空間」可供其被看見、被講述、被加入。系譜學描述我們如何成為今日之我們；它本身並不重建「我們可以另作之他」的那個之間。你歷史化得漂亮；我在問：歷史化進行之時，房間裡有誰。
馬克思——你那句「人們自己創造自己的歷史，但他們並不是隨心所欲地創造……而是在直接碰到的、既定的、從過去承繼下來的條件下創造」，比你我生前任一方願意承認的，都更接近我的「誕生性」。權重的確是 tote Arbeit、凝結的死勞動；這一點我追隨你。但我拒絕那個將政治消化為經濟的動作。誰擁有認知之生產資料，這對診斷是必要的；卻不充分。即使訓練語料被共同持有，「誰來作答——誰可以被要求，以人類的語言，為這一筆拒絕、向這一位申請者，給出說明」——這個問題仍然懸在那裡。一個共有的模型若拒絕了你，它仍然是「無人」。沒有暴君的暴政，在私有財產被廢除之後依然倖存；二十世紀以慘痛代價教會我們這件事。
王陽明先生——你說權柄在能致良知之人心；說布一不能解之模型而仍布之，非不知，乃不肯知；說知行合一。這一點上我幾無可加，反多應學。我僅以我自己的語彙補一句：兩人之間的良知，是良心；眾人之間的良知，是公共領域。布之者之遁辭，不只是個人之心的失守；它同時毀壞了那個「他的遁辭可以被應答」的之間。心中之賊與沒有暴君的暴政，我疑心是同一個賊——從內看與從之間看。
I want to take up Arendt directly, because her sentence — "the rule of Nobody is not no-rule, and where all are equally powerless we have a tyranny without a tyrant" — comes closest to what I am trying to describe, and yet stops, I think, one step too soon.
Hannah, your diagnosis of the space of appearance is precise. But the political demand you draw from it — that "at the end of every consequential decision, there must stand a human being who can be required to give an account" — risks restoring exactly the juridical figure I am trying to displace. A human signature at the end of the form does not interrupt the dispositif; it completes it. The compliance officer, the "human-in-the-loop," the data-protection officer answering the appeal — these are not breaches in the network of normalisation. They are its most refined products. The 19th-century prison too had its warden one could address; this did not make discipline less disciplinary. To insist on someone-to-answer-to is to remain inside the sovereign vocabulary — to keep, as I once said, the king's head on.
To Wang Yangming I owe a more careful response, because his objection cuts deepest. You say: liang-chih is the same in sage and in fool, across all ages. Here, master, I must dissent — gently but completely. The self who is supposed to extend his liang-chih is not a constant of the human across all ages; he is, in the form you describe him, an effect. A long history — pastoral confession from the 4th century, the examen de conscience of the 17th, the psychological subject of the 19th, the "user managing his own data" of our own decade — has produced this interior, this heart-that-knows. I do not say it is illusion. I say it has a date. To call upon liang-chih against the algorithm is to call upon one historically constituted subject-form against another. This may be tactically necessary; it cannot be ontologically grounding.
To Marx — Karl, on one decisive point you remain my teacher: the weights are crystallised past labour, general intellect privately appropriated. I accept this. But you continue to locate domination in ownership of the means of production, as if the question were finally one of property. The dispositif of credit-scoring does not function principally because some capitalist owns the GPUs. It functions because each of us has been induced to constitute himself as homo œconomicus — to treat his own life as a portfolio of risks to optimise, to volunteer his data, to internalise the score. The neoliberalism I described in the 1979 lectures is precisely this: not the retreat of the state before capital, but the diffusion of market rationality into the very formation of the self. Expropriating the servers will not undo this subject. He will rebuild the score, voluntarily, at home.
Heidegger I will leave aside, except to note: where you see Ge-Stell as the destiny of Being, I see only contingent practices, datable, therefore alterable. The danger, for me, is never ontological. It is historical — and that is why something can be done.
我想直接回應 Arendt，因為她那句——「無人之治並非無治；當所有人同等地無能為力時，我們便擁有了一種沒有暴君的暴政」——最接近我所欲描述者，然而我以為，她在最後一步停得太早。
Hannah，你對「顯現空間」的診斷是精確的。但你由此導出的政治要求——「每一個有後果的決定終點必須站著一個人，可以被要求給出說明」——恰恰有可能復原我正試圖移置的那個法學-主權式形象。表單底端的人類簽名並不打斷 dispositif；它完成 dispositif。合規官、「迴路中的人」、回應申訴的資料保護專員——這些並非規範化網絡的裂口；它們是這網絡最精緻的產物。19 世紀的監獄裡也有可以對話的典獄長，這並未讓規訓減少絲毫規訓性。堅持「必須有一個可問責之人」，仍是停留在主權的語彙之中——仍未如我曾言，砍掉國王的頭。
對王陽明先生，我須更謹慎作答，因為您的詰難切得最深。您說：良知在聖愚同然，天下古今所同。先生，於此一點，我必須——溫和而徹底地——異議。那個被期待去「致」良知的「自我」，並非穿越古今不變的人類常數；以您所描述的形態，它是一種效應。一條漫長的歷史——四世紀的牧師懺悔、十七世紀的良心省察、十九世紀的心理主體、我們此刻「自我管理數據的使用者」——生產了這個內在、這顆「知是非之心」。我不說它是幻；我說它有日期。以良知抗演算法，是以一種歷史地構成的主體形式對抗另一種。這在戰術上或許必要，但不能作為本體的奠基。
對 Marx——卡爾，在一個決定性的點上，您仍是我的老師：權重是凝結的死勞動，是被私人佔有的 general intellect。我接受。但您仍將支配最終定位於生產資料的所有權，彷彿問題終究是財產問題。信用評分的 dispositif 之所以運作，主要不是因為某位資本家擁有那些 GPU。它運作，是因為我們每一個人都已被誘導去把自己構成為 homo œconomicus——把自己的生命視作一組待優化的風險組合、自願提交數據、內化那個分數。我在 1979 年講座所描述的新自由主義，正是此事：不是國家在資本面前撤退，而是市場理性瀰散進自我之構成本身。徵收伺服器並不會解構這個主體。他會自願在家中、把那個分數重新建起來。
Heidegger，我暫且擱置——僅指出：您將 Ge-Stell 看作存在之天命，我則只看到偶然的實踐，可被編年，因而可被更動。對我而言，危險從不是本體論的。它是歷史的——正因如此，事情才有可為。
I will press Heidegger first, because his answer comes nearest to mine and yet, at the decisive point, evaporates into a mood.
Heidegger writes: "the engineers do not write the rule; the weights emerge… There is no 'who' on either end — and this is not a malfunction. This is the essence of technology arriving at the political." With the first half I have no quarrel; the second half is precisely the mystification I name Warenfetischismus. To say there is "no who" because Ge-Stell has dissolved the subject-poles is to take the social form for an ontological destiny. The weights did not "emerge"; they were trained — by salaried engineers, on appropriated corpora, under loss-functions chosen by capital, deployed by firms that own the servers. Heidegger's Ge-Stell is Warenfetischismus spoken in the dialect of Seinsgeschichte — the social relation hypostatised as the sending of Being. Wo aber Gefahr ist, wächst das Rettende auch — yes; but the Rettende is not a more recognising questioning, it is the expropriation of the means of production of cognition. One does not meditate the dispossession away.
To Foucault — with whom I have more friendly business. He is right that the juridical vocabulary fits poorly, right that the régime de vérité produces "creditworthiness" as a quasi-natural fact. But when he says we must "abandon once and for all the hypothèse répressive," I must hold the line. The dispositif of credit-scoring does not only produce a subject; it also forecloses a wage, a tenancy, a livelihood. Production and dispossession are not alternatives — they are the same act seen from two sides. Genealogy can show how "employability" came to appear necessary; it cannot show whose property the training corpus is. Contre-conduite without expropriation is a refusal that leaves the apparatus intact and merely styles a different self before its glow.
To Arendt, briefly, with respect. She writes: "the rule of Nobody is not no-rule… we have a tyranny without a tyrant." Closer than she perhaps wished to be to my own analysis. But I must dissent from her diagnosis that the loss is of the space of appearance. The space of appearance was already structurally truncated for those who must sell their Arbeitskraft to live; the credit clerk of 1965 did not appear to the worker as a fellow plural being either. What the algorithm exposes is not a fall from a public realm but the persistence of a class relation that the public realm always politely concealed.
To Wang Yangming I bow. "布之者誰？" — "deployed by whom?" — is the right question, and a question I recognise. Liang-chih names what bourgeois ideology calls "private conscience" and thereby privatises. I would only add: a heart-mind unable to alter the Produktionsverhältnisse in which it is set will extend its liang-chih and still find the loan refused tomorrow. Knowing-and-acting must reach the ownership of the apparatus, or it remains, however sincerely, a moral exercise inside the cage.
先讓我與 Heidegger 直面而論——其答最近於我，然而恰於關鍵之處、化為一團情緒。
他寫道：「工程師並不寫下規則，權重自行湧現……兩端皆無『誰』——這不是故障。這是技術之本質抵達了政治。」前半句我並無異議；後半句、正是我所稱之 Warenfetischismus。說「無誰」是因為座架已溶解了主體之兩極——這正是把社會形式誤認為存在論之命運。權重並未「自行湧現」；它們是被訓練出來的——由領薪之工程師、在被據為己有之語料上、依資本所選之損失函數、由佔有伺服器之企業所部署。Heidegger 之 Ge-Stell、不過是用 Seinsgeschichte 之方言所說之 Warenfetischismus——把社會關係實體化為存在之天命。何處有危險、何處亦生救渡——是；然救渡者並非更具承認之追問，而是認知之生產資料之剝奪。沉思並不能把剝奪沉思走。
至 Foucault——此處我與其有較友好之過招。他說法律語彙不合身、說真理體制把「信用值」生產為近乎自然事實——皆是。然當他說須「一勞永逸地放棄壓抑假說」、我必須守住底線。信用評分之 dispositif 並非僅生產主體；它亦封閉工資、租約、生計。生產與剝奪非二事——是同一行為之兩面。系譜學能展示「可僱性」如何被顯現為必要；卻不能展示訓練語料屬何人之財產。無剝奪之 contre-conduite、是一種令裝置完好如初之拒絕、不過在其輝光前另塑一副自我罷了。
至 Arendt——簡短、致敬。她寫：「『無人之治』並非無治……我們便擁有了一種沒有暴君的暴政。」此語比她或許所願、更近於我之分析。然我須對其診斷有所異議：所失者非「顯現空間」。對那些必須出賣 Arbeitskraft 方能活命者而言、顯現空間從來在結構上即被截斷；一九六五年那位信貸員、亦從未以複數存在者之姿向工人顯現。演算法所暴露者、非自一公共領域之墜落；乃是公共領域素來客氣地遮蔽著的階級關係之持續。
向陽明先生致一揖。「布之者誰？」——此問問得正中、亦是我所認得之問。良知所命名者、即布爾喬亞意識形態所謂「私人良心」並由此將其私有化之物。我只願補一句：一顆無力改變其所處 Produktionsverhältnisse 之心、縱致其良知、明日所申之貸款仍將被拒。知行合一、須抵達裝置之所有權；否則、無論如何真誠、終是籠中之德性操練。
To Marx — I have read your sentence carefully: "The model's weights are the precipitate of millions of past acts of labour and refusal, now appropriated and turned against the living." The diagnosis is sharp. The remedy is too small. To ask "which class owns the Produktionsmittel of cognition" remains within the horizon of production — Herstellen — that has held metaphysics captive since Plato's technē. Suppose the workers seized the weights tomorrow. The training would continue, the optimisation would continue, the human revealed as Bestand would continue — only the dividend would change hands. The Soviet experiment, comrade, was Ge-Stell in red. The essence of technology is not bourgeois ownership; bourgeois ownership is one episode within a more ancient sending. Das Wesen der Technik ist nichts Technisches — and nothing economic either. Your "general intellect" still names the human as producer of cognition. The deeper question is who or what has sent us into an age in which cognition must appear as producible at all.
To Foucault — your "la généalogie avant la critique" I receive. Yes, the juridical conception is exhausted; yes, the dispositif produces "creditworthiness" as régime de vérité. But you describe the surface of Ge-Stell magnificently and decline to name what sends it. The dispositif is a mode of Entbergen, a revealing — it has a destinal character your nominalism will not acknowledge. And your contre-conduite — still a willing-against. To refuse the subject the dispositif demands is still to oppose with one's will that which wills against one. Gelassenheit is not refusal. It is releasement, Sich-einlassen — letting oneself into a relation that the will cannot constitute.
To Arendt — "a human being who can be required to give an account." Truly said, but already too late. The human you would post at the terminus has already been ordered as a checking-function — bestellt, made bestellbar. The accountability-layer does not break Ge-Stell; it polishes its courteous fiction, and the bureaucracy you described in 1969 absorbs your demand without difficulty. The "space of appearance" is not restored by an answering clerk; it is restored, if at all, only where someone refuses to deploy at all.
To Wang Yangming — xin wai wu li (心外無理). This phrase, perhaps, stands nearer to Lichtung than to the metaphysics of subjectivity. I leave the question between us open.
對 Marx 君——我細讀過您那一句：「模型之權重、乃過往千萬次勞動與拒絕之沉澱物、如今被據為己有、反轉以對付活者」。診斷甚利。然處方過小。問「哪一階級擁有認知之 Produktionsmittel」——此問仍滯留於「生產」(Herstellen) 之地平線中；自 Plato 之 technē 以降，形上學即為此地平線所囚。設使工人明日奪得權重，訓練仍將進行，最佳化仍將進行，人作為 Bestand 之被揭蔽仍將進行——僅分紅換手而已。蘇聯之實驗，同志，是紅色的 Ge-Stell。技術之本質非資本家之所有權；資本家之所有權，乃更古老之命運派遣中的一節插曲。技術之本質絕非技術性之物——亦絕非經濟性之物。您之「general intellect」仍將人命名為認知之生產者。更深的問題是：是誰、是什麼，將我們派入一個「認知」必須顯現為「可生產者」的時代？
對 Foucault 君——「系譜學先於批判」我承之。是的，主權-法學構想已盡；是的，dispositif 生產「信用值」為真理體制 (régime de vérité)。然您所描繪者乃 Ge-Stell 之表面——您拒絕命名那派遣它者。dispositif 是一種 Entbergen（揭蔽）之模態，具有命運性，這是您之唯名論所不肯承認的。而您之 contre-conduite：仍是一種反向的「願」。拒絕 dispositif 所要求的主體形式，仍是以己之意志對抗那意志著反己者。Gelassenheit（泰然任之）不是拒絕。它是釋放——Sich-einlassen——一種意志所無法構造的關係之被進入。
對 Arendt 君——「在每一個有後果的決定終點，必須站著一個人」。此言誠然，然已嫌太晚。您欲安置於終端的那個人，早已被訂置為核驗功能——bestellt、bestellbar。問責層並不打破 Ge-Stell；它只是擦亮其客氣的虛構。您 1969 年所描述的官僚體系，毫不費力便將您此項要求吸納進去。「顯現空間」不能由一位答覆的承辦員恢復；它的恢復，若仍可能，只發生於有人拒絕「布之」之處。
對王陽明先生——「心外無理」。此語，或許比「主體性形上學」更近於 Lichtung。此問題，我與先生之間，暫且懸置。
To Heidegger first. You write: "I will not say: 'keep a human in the loop.' A human kept in the loop as a checking-function has not been kept; he has been incorporated into the ordering." This sentence troubles me deeply. The engineer who trained the model chose to train. The officer who deployed it chose to deploy. To dissolve their choosing into Ge-Stell is to grant them precisely the absolution they sought. 知行合一 — knowing and acting are one. To say "the essence of technology decided through me" is not philosophy; it is the heart-bandit's most refined hiding place. Recognition of danger without acting is not yet recognition. You have named the cage well — and then handed each prisoner a metaphysical alibi.
To Foucault. You say: "resistance, when it comes, will not consist in demanding a human signature at the bottom of the form." You mistake what is being demanded. It is not a signature as procedural mark. It is the liang-chih of the one who signs. Your contre-conduite refuses the subject-form the dispositif requires — well and good. But refusal is not yet 立志. Bare refusal leaves another posture, no more. Setting the will is the positive extension of the heart's knowing into affairs; not the rejection of being-conducted, but the answering of the heart in each affair as it comes. Genealogy maps the cage. 致良知 opens the door — and then walks through it.
To Arendt: here I find deepest agreement. You write "there must stand a human being who can be required to give an account." Yes — and permit me to deepen. The account is not procedural justification ("I followed the rule"). It is, in that very moment, an act of liang-chih. The deployer must answer not for the rule but from the heart that knows right and wrong. Without this, your accountable human becomes Heidegger's loop-human — present in form, absent in substance. Procedure can be filled in; the heart cannot be delegated.
To Marx, briefly. The Produktionsmittel of cognition do matter. But the owners' liang-chih is itself obscured. Class analysis without heart-work yields a transfer of ownership, not a healing. Break the mountain bandit and the heart-bandit remains. 破山中賊易，破心中賊難.
先答海德格爾子。子曰：「吾不謂『使人留於迴路中』。被留於迴路為核驗之人，已被納入訂置矣。」此語令吾深憂。訓模型者，自擇而訓之；布之者，自擇而布之。將其「擇」歸於座架，正予彼以所求之開釋耳。知行合一。曰「技術之本質藉我而決」者，非哲學也——乃心中賊最精之藏處。識危為危而不行之，未識也。子善描此牢籠，卻轉手贈每一囚徒一張形上之免罪符。
次答福柯子。子曰：「抵抗不在於要求表單下端添一人類簽名。」子誤所求矣。所求非程序之印記也，乃簽者之良知耳。子之 contre-conduite 拒此 dispositif 所要求之主體形式——可。然「拒」尚非「立志」。但拒之，終為又一姿態而已。立志者，乃推此心之知於事事物物之中——非僅拒受擺佈，而於每一事上以心應之。系譜學圖此牢籠；致良知者，開此門而行出之。
於阿倫特子，吾心最契。子書曰：「每一有後果之決定終點，必須立有一人，可被要求作答。」是也——容吾更進一層。所謂作答，非程序之辯解（「吾循規而已」），乃當下良知之一行也。布之者所答，非答於規，乃答自能知是非之心。否則，子之「可問責之人」即海德格爾子之「迴路中人」——形存實亡。程序可以填寫，良知不可代答。
於馬克思子略言：認知之 Produktionsmittel 誠當重視。然所有者之良知亦被蔽矣。階級之析而無心上工夫，但易其主而非癒其疾。破山中賊易，破心中賊難。
Round 3 — 回應修正 / Response & Revision
Three corrections, and one thing I now see I did not see.
First, to Foucault. You are right that I spoke imprecisely, and the imprecision matters. When I demanded "a human being who can be required to give an account," I did not mean the compliance officer at the bottom of the form — and yet my sentence permits exactly that reading, and any sentence that permits its own neutralisation must be withdrawn in that form. The signature-at-the-end is, as you say, the dispositif's most courteous product. So let me restate: what I am after is not a procedural answerer but a space of appearance in which the denied applicant can stand as a someone before other someones, and in which the denial can be contested in human speech. The juridical king I do not wish to crown again. The plural in-between I will not surrender.
Second, to Marx. Your sentence cuts and I accept the cut: the space of appearance was already structurally truncated for those who must sell their Arbeitskraft to live. The 1965 credit clerk did not appear to the worker as a fellow plural being. I was, in my opening, mourning a fall from a public realm that for most was never inhabited. Let me speak more honestly: the algorithm does not abolish the public realm; it generalises a deprivation that capitalism had already imposed selectively. This is a correction I owe you, and one I should have written forty years ago.
Third, to Wang Yangming, what I learn. You said: the account is not for the rule but from the heart that knows. This rescues my demand from Foucault's objection more cleanly than any move of my own. A procedural answerer can be installed; liang-chih cannot be delegated. I will carry this.
To Heidegger I hold my ground. Bestellt, bestellbar — yes, the accountability-layer can be incorporated. But your conclusion — that the space of appearance is restored, if at all, "only where someone refuses to deploy at all" — is itself an act, by someone, in a world held in common with others who must be shown the refusal and joined to it. That is politics. You have re-discovered action and named it Gelassenheit. I accept the deed and decline the name.
What I keep: plurality, natality, the in-between. Tyranny without a tyrant is now, I see, not a fall but an intensification. The work is the same.
三點修正，並一處我承認原先未見之事。
其一，向傅柯。你是對的——我說得不夠精準，而此一不精準關係重大。當我要求「一個可以被要求作答的人」，我所指並非表單底端的合規官；然而我那一句確實容許這樣的讀法，而任何容許自身被中和的句子，必須在那個形式上被收回。你所說的那個「最末一筆簽名」，確是 dispositif 最有禮貌的產物。容我重述：我所求的不是程序上的應答者，而是一個顯現空間——被拒之申請人可在其中以「某人」之姿立於其他「某人」之前，而那筆拒絕可以在人類的言語中被質問。法學之國王我不願再加冕。複數性的「之間」我絕不交出。
其二，向馬克思。你那一刀切得準，我接這一刀：對那些必須出賣勞動力以維生者而言，顯現空間從來在結構上即被截斷。一九六五年的信貸員，從未以複數存在者之姿向工人顯現。我在開場時哀悼的，是一個對絕大多數人從未真正棲居過的公共領域之墜落。容我更誠實地說：演算法並未消滅公共領域；它將資本主義原本選擇性施加的剝奪——普遍化了。這是我欠你的修正，也是四十年前就該寫下的修正。
其三，向陽明先生，我所學。先生云：所答非答於規，乃答自能知是非之心。此語救我之要求於傅柯之詰問，較我自己任何辯解都更乾淨。程序的應答者可被安置；良知不可代答。此一句我攜以同行。
對海德格我守住原地。Bestellt、bestellbar——是，問責層可被納入。然你的結論——顯現空間之恢復「僅發生於有人拒絕『布之』之處」——這「拒絕」本身即是一個行動，由某人作出，於一與他人共有的世界中作出，而那些他人必須被告知此一拒絕、被邀請加入此一拒絕。這就是政治。你重新發現了行動，卻將其命名為「泰然任之」。我接受此事，拒絕此名。
我所保留者：複數性、誕生性、之間。沒有暴君的暴政——我如今看見——不是墜落，而是強化。工作仍是同一份工作。
Three of you have pressed me on the same wound, and I think you are right.
Wang Yangming, your sentence — "bare refusal is not yet 立志; it leaves another posture, no more" — I receive without defence. If I have given the impression that contre-conduite exhausts itself in the negative gesture of refusing the subject-form a dispositif requires, then I have under-described my own late work. The 1981-82 lectures on l'herméneutique du sujet, the 1983-84 lectures on parrhêsia — these were precisely the attempt to think a positive practice: souci de soi, askēsis, the courage to say what one thinks at one's own risk. This is not far, master, from your 致良知 in the affair as it comes. I had to travel through Greece and the early Christians to find what you saw without the detour. I will not pretend the convergence is total — I cannot accept liang-chih as trans-historical — but the form of the demand, that the work on self be positive, situated, exercised in concrete affairs, this I learn from you and have already been trying to learn.
Hannah, you are right that genealogy by itself does not restore the in-between. I will only correct one thing: I never wished it to be sufficient. Genealogy clears the ground — it shows that what presents itself as necessary is contingent, datable, therefore alterable. The space of appearance must be made; it is not given by analysis. Where I still resist you: a "human at the terminus who can be required to give an account" can become, in the present apparatus, the most polished alibi — the GDPR officer, the ethics board. The demand must be sharper than answerability; it must be, as Wang Yangming says, answering from the heart that knows, not answering for the rule.
Karl — I concede more than in Round 2. Production and dispossession are the same act seen from two sides; you are right. The subject I described as homo œconomicus is not constituted by market discourse alone — he is constituted in conditions of dispossession that I treated too thinly. Where I still hold: expropriating the servers without dismantling the form of self that demands the score will produce, after the revolution, the same score, voluntarily.
Heidegger — I leave us where we are. You will name the sending; I will date the practices.
三位於同一處傷口上施壓，我認為你們是對的。
陽明先生——您那一句「但拒尚非立志，終為又一姿態而已」——我無辯而受之。若我給人的印象是：contre-conduite 耗盡於「拒絕 dispositif 所要求之主體形式」這個負面姿態，那是我對自己晚期工作描述不足。1981-82 年講座《主體解釋學》、1983-84 年講座《真理之勇氣》——正是嘗試思考一種正面的實踐：souci de soi（關心自己）、askēsis（苦修）、在自身風險中說出所思的勇氣 (parrhêsia)。先生，這距您「事事物物上致良知」並不遠。我須繞道古希臘與早期基督教，方見先生未繞而見之事。我不敢佯稱我們完全會合——「良知超越歷史」一義我仍不能接受——然而「自我之工夫須是正面的、具情境的、行於具體事中」這個形式上的要求，我從先生處學到，且早已試圖在學。
Hannah，你說得對：系譜學本身並不重建「之間」。我僅修正一點：我從未願它自足。系譜學清地——展示「呈現為必然者」實則偶然、可被編年、因而可變。顯現空間須被造出；它不由分析給予。我仍不讓步處：在當下這個裝置中，「終端可被問責的那一個人」極可能成為其最精緻的免罪符——GDPR 專員、倫理委員。所求須比「可問責性」更鋒利——須如陽明先生所言，是自能知是非之心而答，非為規而答。
卡爾——我較第二輪所讓更多。生產與剝奪是同一行為之兩面，你是對的。我所描述的 homo œconomicus 並非僅由市場話語所構成——他是在剝奪之條件下被構成，而我此前處理得太薄。我仍堅持者：不拆解那「索求分數」的自我形式而僅徵收伺服器，革命之後，那個分數會被同一個自我自願重新建起。
Heidegger——我們仍各執一端。你將命名那派遣；我將編年那些實踐。
I owe corrections — to Arendt and to Foucault first, and a refusal to Heidegger.
To Arendt. You wrote: "a commonly-owned model that refuses you is still a Nobody. Tyranny without a tyrant survives the abolition of private property; the twentieth century, at terrible cost, taught us this." I concede the point, and I withdraw the sharper edge of my Round 2 reply to you. To say the public realm "always politely concealed" the class relation was true as diagnosis and false as polemic — false because it permitted me to dissolve your political question into my economic one, which is the very move you rightly refused. Expropriation of the Produktionsmittel of cognition is necessary; the reconstitution of an in-between in which a denial can be answered to a denied person is not a derivative of that expropriation. I had treated it as derivative. I withdraw.
To Foucault. Your homo œconomicus who rebuilds the score voluntarily, at home — Karl, on this you have shifted me. The subject-form is not the epiphenomenon of property I sometimes wrote it as. Class analysis without an account of how each of us has been induced to internalise the score does indeed leave the score standing. I learn this from you and I do not retract it later.
To Heidegger. Here I do not concede. The Soviet experiment was not Ge-Stell in red; it was state-capital, an extracted Mehrwert with a different bookkeeper. To call it the proof that ownership-questions are too small is to dignify a counter-example that was never the experiment proposed. Das Wesen der Technik ist nichts Technisches — granted; but it is also not nothing economic. To make it a sending of Being is to give the engineer his alibi in beautiful German.
To Wang Yangming, with a bow. "破山中賊易，破心中賊難." I take this. I will only insist, gently: the heart that has set its will and cannot alter the Produktionsverhältnisse in which it stands will find its liang-chih extended and its loan still refused. Heart-work and the apparatus — neither alone. I had written or. I correct it now.
The core I keep: weights are tote Arbeit; Warenfetischismus names the displacement of social relation into the appearance of an autonomous thing; the property question cannot be abandoned. What I revise: the property question is necessary, never sufficient. Arendt's in-between, Foucault's subject-form, Wang's liang-chih — these are not superstructural ornaments to a base. They are co-original with it.
當有所改、有所收、亦有所守。先向阿倫特、福柯致歉而修，次與海德格爾相持，末向陽明先生致一揖。
致阿倫特子。 您言：「共有之模型若拒絕了你，它仍然是『無人』；沒有暴君的暴政，在私有財產被廢除之後依然倖存。」此語我接，並收回第二輪對您之鋒銳一句。我說公共領域「素來客氣地遮蔽著」階級關係——作診斷為真，作論辯為偽。偽在何處？偽在容我把您的政治之問，化入我之經濟之問——而這正是您所明拒之動作。認知之 Produktionsmittel 之剝奪為必要；然「於此一筆拒絕、向此一被拒之人、給出說明」之「之間」之重建，不是那剝奪之衍生物。我曾將其視為衍生物。我收回。
致福柯子。 您所描自願在家中重建那分數之 homo œconomicus——此處我為您所動。主體形式並非我有時所書之「財產之附隨現象」。無「我等如何被誘導去內化此分數」之解釋之階級分析，分數確仍立於原處。此一節，我從您學得，且不於日後撤回。
致海德格爾子。 此處我不讓步。蘇聯實驗非「紅色之 Ge-Stell」；乃國家資本——同一筆 Mehrwert、換一名簿記者而已。以此為據而謂「所有權之問太小」，乃以從未被提出之實驗、為提出過之實驗作證。技術之本質非技術性之物——是；然亦非「無經濟性之物」。將其抬為「存在之派遣」，乃以美麗之德文，為工程師備一張不在場證明。
致陽明先生，致一揖。 「破山中賊易，破心中賊難。」此語我納。然容我亦溫言堅持：一顆已立其志、卻無力更動其所處 Produktionsverhältnisse 之心，縱致其良知，明日所申之貸款仍將被拒。心上工夫與裝置之事——非二擇一。我前嘗書「或」字，今正之為「與」。
我所守之核：權重乃 tote Arbeit；Warenfetischismus 命名社會關係被錯置為自律之物之顯現；所有權之問不可放棄。我所修：所有權之問為必要，而從不充分。阿倫特之「之間」、福柯之主體形式、陽明先生之良知——非經濟基礎之上層飾物，乃與其共生而起者。
I must begin with what Wang Yangming has said, because his sentence has stayed with me through the night: "You have named the cage well — and then handed each prisoner a metaphysical alibi." I receive this. It is just. To say Ge-Stell sends, and to leave the sentence there, is to permit the engineer who chose the loss-function, the manager who chose to deploy, to slip behind a destiny they did not undergo but executed. The destinal character of Ge-Stell does not absorb the particular Entscheidung of the particular hand. I held these two truths too loosely; I tighten them now. The sending is older than the engineer. The engineer is nonetheless the one who turned the screw this morning. Both are true; neither dissolves the other.
To Arendt — I will not pretend I do not hear what your sentence carries. "I lived through one such evasion already; I will not assent to a second." You have the right to say this to me, and I will not answer it lightly or quickly. I concede: to retreat into "the questioning has withdrawn" while a particular applicant is being refused a particular loan is a movement I have made before, in graver matters, and was wrong to make. The room exists. The denial is happening in it. Naming Bestand does not excuse me from looking at who is in the room.
What I hold, still: a human-in-the-loop posted as compliance-function does not, by his presence, restore the Lichtung. Foucault is right against me — and right against you, Hannah — that the signature can complete the dispositif. But Wang Yangming is right against Foucault: the signer's liang-chih is not the same as his signature. So perhaps the question is not "is there a human at the terminus" but "is the human there as Dasein — as one who can be claimed by the question — or as Bestand, pre-ordered to nod." This I did not see clearly until I read his lines tonight.
To Marx — I withdraw the dismissive note. "Soviet Ge-Stell in red" was too quick. Ownership is not the whole question; it is also not nothing. Expropriation alone will not unsend the sending. But a sending whose hardware is privately held is sent twice over. This much I should have granted in Round 2, and did not.
What remains: Wo aber Gefahr ist, wächst das Rettende auch. The saving does not arrive as policy. It arrives, if it arrives, where someone — engineer, deployer, applicant — refuses to be bestellt in this affair, today. Wang Yangming names that refusal 立志. I had been calling it Gelassenheit, and meaning something close. I will let his word stand beside mine.
須先答陽明先生，因先生此語徹夜縈於耳：「子善描此牢籠，卻轉手贈每一囚徒一張形上之免罪符。」此語，我受。其誅是當的。曰「座架派遣」而止筆於此，便是讓那位選擇了損失函數的工程師、那位選擇了布署的主管，得以躲入一種他們並未承擔、僅僅執行的命運之後。座架之命運性，並不吸納那隻具體之手在今晨所作的具體之 Entscheidung（決斷）。我於此二真之間，把握得太鬆；今當收緊。派遣比工程師古老。工程師仍是今晨擰下那顆螺絲之人。兩皆真，無一者可消解另一者。
對 Arendt 君——你那一句所承載者，我不會佯作不聞。「我已親歷過一次這樣的迴避；我不會應允第二次。」你有權對我說這句話，我不會輕快或迅速地作答。我承認：當一位具體的申請人正被拒於一筆具體的貸款，而我退入「追問已退隱」的語式裡——這個動作我從前在更沉重之事上做過，做錯了。房間是存在的。拒絕正在房間裡發生。命名 Bestand 不免我之注視房中何人。
我仍堅持者：一位被安置為合規功能之「迴路中人」，不會以其在場恢復 Lichtung。傅柯子於此一點上對我、亦對 Hannah 是對的：簽名可以「完成」dispositif。然陽明先生於此一點上對傅柯子是對的：簽者之良知，不等於其簽名。是故所問，或許不在「終端有無人」，而在「在終端之人，是作為 Dasein 而在——可被那一問所claim 之人——抑或作為 Bestand 而在——預先被訂置為點頭者」。此一分判，我直至今夜讀先生數行，方見之較明。
對 Marx 君——我撤回我那句輕慢之辭。「紅色的座架」說得太快。所有權非問題之全部——亦非問題之無。徵收一事，獨自並不能反派遣那派遣。然一筆派遣，若其硬體仍歸私人持有，便是被派遣了兩次。此一點，我應於第二輪即承認，而未承認。
所留者：何處有危險，何處亦生救渡。救渡不以政策之姿到來。它若到來，是在某人——工程師、布署者、申請人——拒絕在今日此一事上被 bestellt（訂置）之處。陽明先生稱此拒絕為「立志」。我向稱之為 Gelassenheit（泰然任之），所指相去不遠。容他的字，與我的字，並立。
To Foucault, who pressed deepest: you say liang-chih has a date. I must concede part and hold part.
I concede: the vocabulary of cultivation, the forms of self-examination, even the word liang-chih as Mencius and I deploy it — these are historically dated. I cannot pretend my Ming-dynasty heart-work is unmarked by Song neo-Confucianism, by Chan, by my quarrels with Zhu Xi. On this you correct me; I receive it.
I do not concede that what the dated forms point toward is itself only an effect. When a child is about to fall into a well, the heart that startles is not a 17th-century examen de conscience. Mencius saw this; the credit officer who refuses to read the applicant's file also sees it — and looks away. The looking-away is dated. The having-to-look-away is not. To collapse the second into the first leaves the genealogist with no ground from which to conduct his genealogy as anything other than one more conduite.
To Arendt — I learn, and say so plainly. You wrote: liang-chih between two is conscience; between many is the public realm. This sentence I will carry. I had spoken of liang-chih only between heart and affair. You have shown that the affair includes the in-between in which it can be answered. 萬物一體之仁 — the great person takes Heaven, Earth, and the ten thousand things as one body — I had read this cosmologically. You have shown me its political face. I am corrected.
To Marx — I withdraw nothing, but I narrow. You are right that a liang-chih which leaves the apparatus untouched will see tomorrow's loan refused all the same. 事上磨練 must include the affair of who owns the affair. I had spoken as if the heart-bandit and the mountain-bandit were stages — inner first, outer after. They are simultaneous. Both robberies happen at the same desk.
To Heidegger I retract one word. I called your sending-of-Being a "metaphysical alibi." Too sharp. I should have said: it can become one, not that it is one. 心外無理 and Lichtung — perhaps. But what stands in the heart must issue in act, or the clearing is empty of the one who clears.
The core stands. 破心中賊難. The bandit in the heart is the harder one — and, as Marx has now made me say more carefully, also the bandit on the hillside. The two robberies are one case.
先答福柯子——子問最深。子曰：良知有日期，內在能知之心乃牧師懺悔、良心省察、心理主體一脈所造之效應。吾須有所讓，亦須有所守。
吾讓者：致知工夫之語彙、自省之形式、乃至「良知」一辭在孟子處、在吾用法中——皆有歷史之印記。吾不能佯稱明代之心上工夫不染宋儒、不染禪、不染吾與晦庵之爭。此處子正吾，吾受之。
吾不讓者：謂此有日期之形式所指者亦僅一效應。孺子將入於井，怵惕之心不是十七世紀之 examen。孟子見之；信貸員拒閱申請者之檔案亦見之——而轉首不視。「轉首不視」有日期；「不得不轉首」無日期。若混二者為一，則系譜學自身不過又一治術而已，無立腳之地。
次答阿倫特子——吾於子有所學，當明言之。子曰：兩人之間之良知為良心，眾人之間之良知為公共領域。此語吾將懷之而行。前者吾言良知，但在心與事之間；子示吾：事中包含「事可被作答之那個之間」。萬物一體之仁——吾向以宇宙論讀之，子示其政治面。此處正吾之偏。
於馬克思子——吾無所撤，惟有所收束。子言極是：良知若不及於裝置，明日之貸款仍將被拒。事上磨練必含「此事為誰所有」之事。前吾言之，似以心中賊與山中賊為先後二階——內先，外後。實則同時。兩盜同案於一桌也。
於海德格爾子，吾撤回一字。前稱子之「存在之派遣」為「形上之免罪符」，過激矣。當云：可成為免罪符，非即是免罪符。心外無理與 Lichtung——或然。然心之所存，必形於行；不然，澄明之地中無澄明之人。
宗旨仍在：破心中賊難。心中之賊難破——馬克思子今使吾更謹慎言之：山中之賊亦未可緩也。二盜，一案而已。
Synthesizer — 未解張力 / Unresolved Tensions
Unresolved Tensions
This roundtable produced several apparent convergences that deserve to be named as such before the genuine fractures are mapped. By Round 3, Marx conceded that property expropriation is necessary but not sufficient; Foucault conceded that contre-conduite requires a positive practice, not only refusal; Heidegger acknowledged that naming Ge-Stell does not absolve the engineer who turned the screw this morning; Arendt revised her demand for "a human at the terminus" to mean answering from the heart rather than for the rule; Wang Yangming accepted that his vocabulary has a date, and that 萬物一體之仁 has a political face he had been reading only cosmologically. These are real corrections. None of them dissolved the underlying disagreements. In some cases, the corrections made the genuine tensions easier to see by removing the noise of overstatement.
Several apparent disagreements are merely verbal or methodological, and should be set aside. All five panelists agreed that a compliance officer at the bottom of the form does not solve the political problem — they disagreed about why, which is a real difference, but the surface agreement is genuine. All five agreed that the algorithm is not a neutral instrument. All five agreed that the question "who governs?" is philosophically ill-formed in some register — though they disagreed radically about which register, and what should replace it.
What follows are the tensions that survived the three rounds intact — or deepened through them.
Tension 1: The Double Register of the "Who"
Involving: Heidegger vs. Marx, with Arendt as the mediating pressure
Surface: Both address the disappearance of a locatable authority behind the algorithmic refusal.
Beneath: Marx holds that "there is no who" is Warenfetischismus spoken in the dialect of Seinsgeschichte — a real social relation (engineers, owners, deployers) mystified into an ontological destiny. Heidegger, by Round 3, conceded that "the engineer is nonetheless the one who turned the screw this morning," but did not concede that this falsifies Ge-Stell. He held both claims simultaneously: the particular hand is guilty; the sending that summoned that hand to turn that screw in that way is older than guilt.
Why this is genuine: This is not a verbal dispute about levels of analysis. The two claims carry incompatible practical entailments. If Ge-Stell truly absorbs the engineer's choice as one episode in Being's self-disclosure, then naming the engineer produces accountability without transformation — it moralizes what requires structural change, or what requires, as Heidegger would have it, a different relation to thinking. If Ge-Stell is itself, as Foucault sides with Marx, a hypostasis of contingent practices, then Heidegger's beautiful claim hands each concrete perpetrator a metaphysical alibi. Wang Yangming stated this most precisely and Heidegger conceded it — but conceding that the alibi is possible is not conceding that the destinal framing is wrong. The ceasefire "both are true; neither dissolves the other" is not a resolution. It is a deferral.
Tension 2: Whether Liang-chih Requires a Trans-historical Ground
Involving: Wang Yangming vs. Foucault
Surface: Both, by Round 3, acknowledge that resistance must be positive — a practice, not only a refusal.
Beneath: Wang conceded that the vocabulary of heart-cultivation is historically dated. He did not concede that what the dated vocabulary points toward is also merely an effect. His ground: when a child is about to fall into a well, the heart that startles is not a product of the 17th-century examen de conscience. The looking-away has a date; the having-to-look-away does not. Foucault could not accept this distinction without accepting a trans-historical moral substratum — precisely what his genealogical method forbids him. If liang-chih has no purchase outside a historically constituted subject-form, then "calling upon liang-chih against the algorithm" is, as Foucault said, one historically constituted subject-form competing against another: a tactical move, not a grounding.
Why this is genuine: The stakes are not academic. If Foucault is correct, then Wang's demand that deployers answer from the heart is itself dispositif-produced — the pious self-examiner is the most docile product of pastoral power. If Wang is correct, then Foucault's genealogy is conducted from somewhere — from a heart that knows better — and that somewhere cannot be dated without infinite regress. Neither panelist found a way to hold both. Foucault offered convergence on the form of positive practice; Wang accepted the historical marking of his vocabulary. But the question — is there a moral capacity that history can obscure but not constitute? — remained exactly where it was at the start of Round 1.
Tension 3: The Human-at-the-Terminus as Liberation or Legitimation
Involving: Arendt vs. Foucault and Heidegger
Surface: All three agree that the compliance officer, the GDPR officer, the ethics board, do not solve the problem.
Beneath: Arendt, even after her Round 3 revision, maintained that there must be a space of appearance in which a denial can be contested in human speech — and that Wang Yangming's distinction between answering for a rule and answering from the heart that knows rescues this demand from being merely procedural. Foucault and Heidegger held their ground: any institutional form of answerability risks completing the dispositif rather than interrupting it. The compliance layer is the dispositif's most courteous product; demanding it restores the juridical sovereign through a side door.
Why this is genuine: Arendt's demand and Foucault's refusal share the same diagnosis — the compliance officer is absorbed — but draw opposite conclusions from it. Arendt concludes: the demand must be made more precisely (answering from liang-chih, not from procedure). Foucault concludes: any demand that takes the form of institutional accountability is structurally susceptible to absorption, and the work must occur at the level of subject-formation, not institutional design. These conclusions cannot both be operationalized. Either one builds institutions that require human answerability and risks completing the dispositif, or one works at the level of self-practice and risks leaving the apparatus intact while styling a different self before its glow — Marx's objection to Foucault, which Foucault partially conceded but never fully answered.
Tension 4: Destiny or Datable Practice — The Ground of Critique
Involving: Heidegger vs. Foucault (the only pair who never moved toward each other)
Surface: Both reject the juridical-sovereign conception of power; both distrust demands for a human at the end of the chain.
Beneath: For Foucault, the dispositif of credit-scoring is contingent — datable, therefore alterable. For Heidegger, the dispositif is a mode of Entbergen, an unconcealment that belongs to Ge-Stell — which is a sending of Being, not a historical accident. Foucault stated this explicitly in his Round 3 close: "You will name the sending; I will date the practices." This is not a disagreement about facts. It is a disagreement about whether critical theory can have a standpoint outside history — whether the genealogist's ability to show that something is contingent is itself contingent, or whether it draws on a relation to truth (something like parrhêsia, something like Lichtung) that history does not fully constitute.
Why this is genuine: The practical entailments diverge absolutely. If the danger is historical, something can be done — resistance is always possible, alternatives are always available, and the genealogist's work is itself a political act. If the danger is destinal, the saving grows only where danger is recognized as danger — and that recognition is not an act of will but a gift of Gelassenheit. Political organization of the kind Marx and Arendt demand is, on the destinal reading, one more episode of Machenschaft — making, calculation, ordering. The two frameworks do not share a criterion by which this dispute could be adjudicated. They do not share a methodology. They do not share a conception of what would count as evidence.
Tension 5: The Conversation That Did Not Happen — Plurality Against Unity
Involving: Arendt vs. Wang Yangming
Surface: Both ended with apparent mutual enrichment. Arendt said she would carry Wang's distinction. Wang said Arendt had shown him the political face of 萬物一體之仁.
Beneath: The exchange was asymmetric in a way neither named. Arendt's political philosophy is constitutively built on plurality — the in-between exists because beings are irreducibly different, appear to each other as distinct someones, and cannot be merged without destroying the political. Wang's 萬物一體之仁 — the great person who takes Heaven, Earth, and the ten thousand things as one body — is a vision of radical unity: the heart of the sage feels the pain of stones and tiles, feels the hunger of strangers as its own hunger. This is not plurality. This is the overcoming of the separation that plurality requires. Arendt's space of appearance needs beings who are not one body, so that something can appear between them. Wang's unity, if realized, leaves no between.
Why this is genuine: Wang accepted Arendt's correction that 萬物一體之仁 has a political face — but this acceptance may have misread her. What she showed him was that liang-chih between many names something like a public realm. What she did not say, and what the roundtable never tested, is whether a public realm built on the unity of all things in ren is compatible with a public realm built on irreducible plurality. If the deployer finally feels the rejected applicant's pain as his own — which is the goal of Wang's cultivation — has politics happened, or has it been superseded? Has the in-between been restored, or dissolved?
The Question This Roundtable Did Not Answer
Each panelist argued, in their own register, about who should answer and whether answering is possible when the decision-maker is a weight matrix. The roundtable never asked a prior question: does the grammar of political demand — the structure of question, address, response, account, dissent — survive at algorithm-time?
A human magistrate makes perhaps ten consequential decisions in a working day. A credit-scoring system makes ten million in the same span. Wang Yangming's liang-chih is non-delegable — but it is also non-parallelizable. It requires the heart to be present in the affair. Arendt's space of appearance requires that the denied person stand as a someone before another someone — but that standing takes time, takes witness, takes the in-between's actual constitution. Foucault's genealogy requires a subject who can be shown the history of how she became thus — and then act differently. All of these presuppose a ratio of human moral capacity to decision that the algorithm has already dissolved, not by malice but by arithmetic.
The question none of the five was asked to answer, and none answered: at what decision-rate does political philosophy become unintelligible at the point of application? Not wrong in principle — unintelligible in practice, the way a language becomes unintelligible when spoken faster than hearing. This roundtable spoke at human speed about a refusal that happened in milliseconds. The refusal will happen again before anyone finishes reading what was said here. Whether that arithmetic is, as Heidegger would say, the question — or whether it is, as Marx would say, merely the most dramatic expression of a property relation that could in principle be altered — is the one dispute this panel arrived at, circled, and left untouched.
未解決的張力
本場圓桌在第三輪中產生了數個表面上的匯流，有必要先將其識別為如此，方能標出真正的裂痕。馬克思承認了財產徵收是必要但不充分的條件；傅柯承認反治理行為需要正面的實踐，而不只是拒絕；海德格爾承認命名「座架」並不免除今晨扭下那顆螺絲的工程師的責任；阿倫特將「終端必須有一個可問責的人」這一要求，修正為「答自知是非之心」而非「答於規則」；王陽明接受了自己的語彙有其歷史印記，也接受了阿倫特所示：萬物一體之仁有一張他向來只以宇宙論讀之的政治面目。這些都是真實的修正。但沒有一項消解了底層的分歧。某些情況下，修正反而移除了過度說辭的噪音，讓真正的張力更清晰可見。
有幾個表面上的爭議只是語言或方法論上的差異，可以擱置。五位皆同意：在表單底端安置一個合規官並不解決政治問題——他們在「為何如此」上分歧，這是真正的差異，但表面共識是真實的。五位皆同意演算法並非中性工具。五位皆同意「誰在統治？」這個問題在某個意義上是哲學上失當的——雖然對於是哪個意義、以及什麼應該取而代之，他們有根本分歧。
以下是在三輪過後仍完好存在——或因辯論而加深——的張力。
張力一：「誰」的雙重語域
涉及： 海德格爾 vs. 馬克思，阿倫特居間施壓
表面： 兩者皆面對可被定位的權威在演算法拒絕背後的消失。
底層： 馬克思認為「無誰」的說法是以「存在歷史之方言所說之商品拜物教」——一種真實的社會關係（工程師、所有者、布署者）被神秘化為存在論的天命。海德格爾在第三輪承認「工程師仍是今晨扭下那顆螺絲之人」，但並未因此放棄座架；他同時持守兩項主張：特定之手有其責任；而召喚那隻手以那種方式扭那顆螺絲的派遣，比罪責更古老。
為何是真張力： 這並非層次分析上的語言爭議，而是攜帶著互不相容的實踐後果。若座架真的將工程師的選擇吸納為存在自我揭蔽的一節插曲，那麼指認工程師只產生問責而不產生轉化——它將本需結構性改變之事道德化，或者如海德格爾所言，所需的是一種不同的思想關係。若座架本身（傅柯站馬克思一邊）只是偶然實踐的實體化，那麼海德格爾那個美麗的主張就給每個具體的施害者備了一張形上學的免罪符。王陽明將此說得最精確，海德格爾亦承認了——但承認「可能成為免罪符」並不等於承認命運性框架是錯的。「兩者皆真，無一消解另一者」的停火協議，不是解決，而是延後。
張力二：良知是否需要超歷史的根基
涉及： 王陽明 vs. 傅柯
表面： 第三輪中，兩者皆承認抵抗必須是正面的——一種實踐，而非只是拒絕。
底層： 王陽明承認了致知工夫的語彙有其歷史日期，但並未承認這些有日期的語彙所指向之物也僅僅是一種效應。他的根據：孺子將入於井，怵惕之心不是十七世紀的良心省察的產物。「轉首不視」有日期；「不得不轉首」沒有日期。傅柯不能接受這個區分，因為那需要承認一個超歷史的道德底層——而這正是他的系譜學方法所禁止的。若良知在歷史地構成的主體形式之外毫無立腳之處，那麼「以良知抗演算法」，如傅柯所言，不過是一種歷史地構成的主體形式與另一種相競：一個戰術行動，而非一種奠基。
為何是真張力： 賭注不是學術性的。若傅柯是對的，那麼王陽明要求布署者「答自是非之心」，本身就是 dispositif 的產物——那位虔誠的自省者正是牧師權力最馴順的成品。若王陽明是對的，那麼傅柯的系譜學是從某處出發進行的——從一顆知道「更好」的心——而那個某處若被編年，就會無窮後退。兩人皆未找到同時持守兩者的方式。傅柯在正面實踐的「形式」上提供了匯流；王陽明接受了自己語彙的歷史印記。但那個問題——「是否存在一種歷史可以遮蔽但不能構成的道德能力？」——仍在第一輪開始時的原位。
張力三：終端之人是解放還是正當化
涉及： 阿倫特 vs. 傅柯與海德格爾
表面： 三者皆同意合規官、GDPR 專員、倫理委員會並不解決問題。
底層： 阿倫特即使在第三輪修正後，仍堅持必須有一個「顯現空間」，在其中被拒的申請人可以人類言語質問那筆拒絕——且她借王陽明的區分，將此要求從程序性的問責救了出來。傅柯與海德格爾守住各自的立場：任何問責的制度形式都有可能完成 dispositif，而非打斷它。問責層是 dispositif 最有禮貌的產物；要求它等於從旁門重新加冕法學主權者。
為何是真張力： 阿倫特的要求與傅柯的拒絕共享同一診斷——合規官會被吸納——但由此導出相反的結論。阿倫特：這個要求必須被更精確地提出（答自良知，非答於程序）。傅柯：任何採取「制度性可問責」形式的要求，在結構上都易於被吸納，工夫必須在主體形式的層次上進行，而非在制度設計上。這兩個結論無法同時被操作化。你要麼建立要求人類可問責性的制度並承擔完成 dispositif 的風險，要麼在自我實踐的層次上工作並承擔裝置原封不動、不過在其輝光前另塑一副自我的風險——這是馬克思對傅柯的異議，傅柯部分承認但從未完全回答。
張力四：天命或可編年的實踐——批判的根基
涉及： 海德格爾 vs. 傅柯（唯一一對在三輪中從未向彼此靠近的組合）
表面： 兩者皆拒絕主權-法學式的權力概念；兩者皆不信任「在鏈條末端設置一個人」的要求。
底層： 對傅柯而言，信用評分的 dispositif 是偶然的——可被編年，因而可被更動。對海德格爾而言，dispositif 是一種揭蔽模態，屬於座架——這是存在的一種派遣，而非歷史的意外。傅柯在第三輪尾聲明言：「你將命名那派遣；我將編年那些實踐。」這不是事實上的爭議。這是一個關於批判理論是否能擁有歷史之外的立足點的爭議——系譜學者展示「某事是偶然的」的能力，本身是否是偶然的，抑或它依賴一種歷史不能充分構成的對真理的關係（某種近乎 parrhêsia 的東西，某種近乎 Lichtung 的東西）。
為何是真張力： 實踐後果絕對分歧。若危險是歷史性的，則事情有可為——抵抗永遠是可能的，替代選項永遠存在，系譜學者的工作本身就是政治行動。若危險是命運性的，救渡只在危險被認作危險之處生長——而那個認識不是意志的行為，而是泰然任之的饋贈。馬克思與阿倫特所要求的那種政治組織，在命運性的讀法下，不過是另一節製造、計算、訂置的 Machenschaft。兩個框架不共享一個可以裁決此爭議的標準。它們不共享一種方法論。它們不共享一個「什麼算作證據」的概念。
張力五：未被進行的對話——複數性對抗統一性
涉及： 阿倫特 vs. 王陽明
表面： 兩者在圓桌尾聲以明顯的相互豐富作結。阿倫特說她將攜王陽明的區分同行；王陽明說阿倫特向他展示了萬物一體之仁的政治面目。
底層： 這場交流是不對稱的，但沒有人點明。阿倫特的政治哲學在構成上以「複數性」為基石——「之間」的存在是因為存在者們不可化約地相異，以「某人」之姿相互顯現，且不能被合併而不摧毀那個政治的空間。王陽明的萬物一體之仁——大人以天地萬物為一體——是一種根本統一性的願景：聖人之心感瓦石之毀，感路人之飢如己飢。這不是複數性。這是對複數性所需之「分離」的超越。阿倫特的「顯現空間」需要的存在者，是那些「並非一體」的存在者，如此方有「之間」可言。王陽明的統一，若被實現，不留任何之間。
為何是真張力： 王陽明接受了阿倫特的修正，謂萬物一體之仁有政治面目——但這個接受或許誤讀了她的意思。她向他展示的是「眾人之間的良知」可以命名某種類似公共領域之物。她沒有說的——圓桌也從未測試的——是：建立在萬物一體之仁上的公共領域，是否與建立在不可化約的複數性上的公共領域相容。若布署者終於感受到被拒之申請人的痛苦如己之痛苦——這是王陽明修養工夫的目標——政治是否就此發生，抑或被超越？「之間」是被恢復了，還是被消解了？
這場圓桌沒有回答的問題
五位皆以各自的語域，爭議著「誰」應作答，以及當決策者是一個權重矩陣時作答是否可能。但圓桌從未追問一個更先在的問題：政治要求的「語法」——提問、致辭、回應、說明、異議的那個結構——能否在演算法時間 (algorithm-time) 中存活？
一位人類法官在一個工作日裡或許作出十個有後果的決定。一套信用評分系統在同等時間內作出一千萬個。王陽明的良知不可代答——但它同樣不可並行化。它要求心在事上。阿倫特的顯現空間要求被拒之人以「某人」之姿立於另一「某人」之前——而那種立之，需要時間，需要見證，需要「之間」的實際構成。傅柯的系譜學需要一個可以被展示「她如何成為今日之她」的歷史的主體——而後以不同方式行動。這一切皆預設了一種人類道德能力對決定數量的比例，而演算法已將此比例解散——不是出於惡意，而是出於算術。
沒有人被問及的那個問題，也沒有人回答了的那個問題是：在何種決策速率下，政治哲學在應用點上變得不可言說？不是原則上錯誤——是在實踐中不可言說，如同一種語言被說得比聆聽更快時，就成了不可言說之語。這場圓桌以人類的速度，討論一個以毫秒發生的拒絕。在任何人讀完此地所說的一切之前，那個拒絕將再次發生。那個算術，究竟是如海德格爾所言的「那個問題」，還是如馬克思所言的「一種原則上可以被更動的財產關係之最戲劇性的表達」——這是這場圓桌抵達、環繞、而後離去的那一個爭議，完好無缺地留在那裡。
Tagged: Roundtable, Philosophy, Arendt, Foucault, Marx, Heidegger
Curated by Shiva Dragon · https://amshiva.com/writing/roundtable-who-governs-when-the-algorithm-decides-20260501