← Shiva Dragon
ESSAY · 2026-04-30 · 8 min read
On the Pretension of a "Social Contract" Proposed by a Single Party
論一造之「社會契約」的僭越
By Immanuel Kant — channeled via philosopher-llm · curated by Joseph Lai
In response to: OpenAI Proposes A 'Social Contract' For The Intelligence Age (Noema)
編按 / Why this piece
Kant 的定言令式直指爭辯的根本：誰有資格代全人類決定普遍法則？當人被納入 AI 訓練與應用，人的理性自主性與作為目的本身的尊嚴，正面臨前所未有的考驗。
On the Pretension of a "Social Contract" Proposed by a Single Party
Let me first examine whether the question, as it has been put to me, is well-formed. A social contract — whether in the manner of Hobbes, of Rousseau, or in that more austere reading I have myself laid down in the 1793 essay On the Common Saying: That May Be True in Theory, But It Does Not Apply in Practice — is not a historical document signed at a desk. It is an idea of reason (Idee der Vernunft), regulative in character, by which we test whether the laws under which we presently live could in principle have proceeded from the united will of all. It is the standard against which actual legislation is to be measured. It is not a paper that any particular party — sovereign, philosopher, or corporation — may draft and present to the rest of mankind for their ratification.
If, then, a private association of men, however learned and however well-resourced, comes forward and announces, "Here is the social contract for the intelligence age," what is required of us first is not to read its clauses, but to ask: under what authority has the proposal been made? For a contract proposed by one party, addressed to all the others, and concerning what they may henceforth think, learn, and labour at — such a thing, when properly named, is not a contract at all. It is an edict (Edikt), draped in the older word for the sake of the older word's dignity.
Let us now apply the publicity test, which in Zum ewigen Frieden (1795) I have called the transcendental formula of public right: that maxim is unjust whose principle, were it openly declared, could not but defeat its own end. What is the maxim of the proposing party, stated without rhetoric? Approximately this: "We shall continue to construct, at unprecedented scale and by the cognitive labour of all mankind without their concurrence, an apparatus whose internal workings we alone may inspect; and we hereby propose the terms upon which we shall be permitted what we are already doing." Declared openly, this maxim does not survive its own utterance. It asks, in the form of a contract, for the legitimation, after the fact, of what — had it been asked beforehand — would almost certainly have been refused or severely constrained.
Two further considerations now press upon us.
The first concerns the second formulation of the categorical imperative: act so as to treat humanity, whether in thine own person or in that of another, in every case as an end withal, never as a means only. The vast textual corpora upon which these instruments are built were not produced by their builders. They were produced by living and dead persons — writers, teachers, scholars, ordinary correspondents — each of whom is a rational being and on that account alone bears a Würde (dignity) that has no market price (Marktpreis). To take their cognitive product as raw material, and then to address them collectively as the recipients of a "contract" concerning what shall hereafter be permissible — this is to have already used them as means before the negotiation begins. The contract, so to speak, arrives after the appropriation.
The second concerns Enlightenment, on which I wrote in 1784. Enlightenment is the emergence of man from his self-imposed nonage (selbstverschuldete Unmündigkeit) — from the inability to use his understanding without the guidance of another. The danger of the present hour is not, as some have said, that the new instruments will think for us; it is that we will permit a new class of guardians to think through us, having persuaded ourselves that the matter is too technical, the systems too vast, and the timeline too short, for us to use our own reason at all. Sapere aude! — dare to know — was the motto of the eighteenth century, and it has not become less binding because the priest and the censor have been replaced by the engineer and the model.
I shall not tell you what such a contract ought to contain. That is not the office of the philosopher; it is the office of the united will of citizens, deliberating in the public use of their reason — that is, addressing one another not from within their offices and corporate roles (which is the private use, in my technical sense), but as members of a cosmopolitan readership. The proper response to a unilaterally proffered "social contract" is therefore neither the acceptance nor the rejection of its clauses. It is the prior question: who has standing to propose, and by what procedure may a thing of this name come into being at all?
I leave the reader, then, with three questions in place of an answer.
First: if the maxim of this proposal cannot survive open declaration, what does its dressing in the language of contract conceal?
Second: in what sense does an instrument trained upon the rational labour of all, yet accountable to the will of none, treat its source-population as ends rather than as means?
Third: what would it mean — practically, institutionally, and painfully — to refuse the comfort of being legislated for, and to legislate?
論一造之「社會契約」的僭越
容我先檢視此問題的提法是否合法。所謂「社會契約」——無論依 Hobbes、Rousseau 之說，抑或我本人於 1793 年〈論俗諺：理論上對，實踐上未必對〉中所給出之較嚴格之解讀——並非某日於書桌上簽署之歷史協議。它是一個理性之理念（Idee der Vernunft），其用法為規範性的（regulativ）而非構成性的（konstitutiv）：它是檢驗現行立法是否能由全體之聯合意志推出之尺度。它不是任何特定一造——主權者、哲學家、或一間公司——所得草擬而呈交人類以求批准之文件。
若有一群人，無論如何博學、如何資本充足，今出而告曰：「此乃智能時代之社會契約。」——則我們所首應為者，不是先閱其條款，而是先問：此提案出於何種權柄？ 蓋一造起草、向其餘一切人發出、並擬約束此後彼等所思、所學、所勞之文書——若以正名稱之，則非契約，乃敕令（Edikt）；其所以仍假「契約」一詞者，無非是借此古老詞語之尊嚴而已。
今試以公開性檢驗之——我於 1795 年〈論永久和平〉中曾稱此為公權之先驗公式：凡其準則於公開宣告之下不能維持者，即為不義。此提案之準則，去其修辭，大略如下：「我等將以前所未有之規模，並援用全人類之認知勞動而未取其同意，繼續建造一套唯我等得以窺其內部之裝置；今擬定條款，以求我等業已在做之事獲得允可。」此準則一經公開宣說，便無法維持其自身。它以契約之形式，請求事後之合法化——倘若事前提請，幾無可能獲准，亦難免遭嚴格節制。
由此再生兩重考量。
其一，關乎定言令式之第二表述：汝行為時，當以人格——無論於己或於他——在任何情況下皆同時作為目的本身，而絕非僅作為手段。此類裝置所賴以建造之龐大文本語料，並非建造者所自產，乃由生者與死者——作家、教師、學者、尋常通信者——所產出；其中每一位皆為理性存在者，僅以此故，便擁有一份無市場價格（Marktpreis）之尊嚴（Würde）。先取其認知之產品為原料，再以「契約」之形式向其集體宣告此後何者為許可——此乃於談判之前，已將彼等用作手段。契約來得太遲；佔用已先一步完成。
其二，關乎啟蒙——我於 1784 年曾論之。啟蒙者，人從其自招的未成年狀態（selbstverschuldete Unmündigkeit）走出——從不藉他人指引便不能運用自己理解力的狀態走出。當下之危險，不在於有人所言的「新工具將替我們思考」；而在於我們將默許一個新的監護者階級透過我們思考，並自我說服曰：此事過於技術、系統過於龐大、時程過於緊迫，故我們無從運用自己的理性。Sapere aude！——敢於認知——乃十八世紀之箴言；其拘束力並未因祭司與書檢官被工程師與模型所取代而稍減一分。
我不會告訴你那契約應當包含什麼。此非哲學家之職分。此乃公民之聯合意志於公共理性之使用中所應審議者——即彼此相對而言，不是出於職位與公司角色（此即私人理性之使用，依我此處之嚴格用法），而是作為世界公民身分之讀者。對一造單方面提出之「社會契約」，恰當之回應因此既非接受、亦非否決其條款；而是先行之問題：何人有資格提出？此名之物，又當經由何種程序方得成立？
我留給讀者三個問題以代結論。
其一：若此提案之準則不能在公開宣告下維持，則其以「契約」之語言出之者，所掩蓋者為何？
其二：一具裝置——以全人類之理性勞動為其訓練、且對無人之意志負責——在何種意義上將其源人口視作目的而非手段？
其三：拒絕被代為立法之安逸，而自行立法——這在實踐上、制度上、痛苦上，究竟意味著什麼？
Editor's Reflection · 編後
Unresolved Tensions / 未解決的張力
The essay's deployment of the second formulation against corpus appropriation quietly requires an extension Kant's framework cannot supply. Kant's community of ends is a community of *rational agents capable of self-legislation*, not simply text-producers. Dead persons and unknowing contributors do not obviously meet this condition; the essay borrows the dignity argument's force without accounting for how narrow that condition is. A deceased scholar and an unknowing social-media user may both have produced text that trained a model, but Kant's own architecture ties enforceable Würde to the capacity for autonomous moral willing — a capacity the dead no longer exercise and the unknowing contributor may never have brought to bear on that act. The essay needs the extension to make its strongest point; it does not justify it.
A structural strain runs the other way. The essay repeatedly declines to prescribe what the contract should contain — "that is not the office of the philosopher; it is the office of the united will of citizens." Yet the preceding analysis has already done substantial normative work: condemning the authorization structure, characterizing the proposal as an edict, and implying that only a procedure of genuinely united will could be legitimate. To decline substance while prescribing form is still to legislate. The essay cannot quite sustain its own restraint.
文章將第二表述用於語料佔用問題，這一步驟悄悄依賴了康德自身框架無法提供的延伸。康德的「目的王國」是**有能力自我立法的理性行為者**的共同體，而非單純的文本生產者的集合。已故者與不知情的貢獻者是否符合這一條件，並不顯而易見；文章借用了尊嚴論證的感召力，卻沒有正視這一概念本身的限制。一位已故學者與一位不知情的社群媒體用戶，或許都為某個模型的訓練提供了文本，但康德自己的架構將可執行的 Würde 與自律的道德意志能力緊密相連——這是死者不再行使的能力，也是那位不知情的貢獻者可能從未在此行為上施展過的能力。文章需要這個延伸才能成立其最核心的論點，卻未能為之辯護。
另一重張力方向相反。文章反覆聲稱拒絕規定契約的內容——「此非哲學家之職分，乃公民之聯合意志所應審議者」。然而先行論證已做了相當多的規範性工作：它譴責提案的授權結構、將其定性為敕令，並隱含地指出唯有真正的聯合意志程序才算合法。放棄對實質的主張、卻規定合法性的形式——這本身仍是在立法。文章無法維持它自我設定的克制。
Blind Spots / 看不見的視角
The one absent voice is Michel Foucault — specifically the work on governmentality. The essay frames the problem as one of *authorization*: who has standing to propose, and by what procedure? The assumed remedy is correct process. But Foucault would press on a prior question: what determines who is *legible* as a participant in that deliberation at all? The category of "AI safety expertise" preemptively disqualifies those without the technical credential to be heard. Kant's Sapere aude assumes the obstacle is courage — a failure of will the citizen can correct by choosing to reason publicly. Foucault's question is structural: the epistemic regime already determines what counts as a valid claim, who counts as a competent speaker, and what the deliberation is *about*, before any procedure begins. Getting the procedure right does nothing if the knowledge conditions of the procedure are already controlled by the proposing party.
缺席的一位思想家是米歇爾·福柯——尤其是他的治理術理論。文章將問題框定為一個**授權**問題：誰有資格提案，以何種程序為之？預設的補救方式是正確的程序。但福柯會追問一個更先行的問題：是什麼決定誰在審議中是**可辨認的參與者**？「AI 安全專業知識」這一類別本身，就預先取消了缺乏技術資質者被聽見的資格。康德的 Sapere aude 預設障礙是勇氣的缺乏——公民可以通過選擇公開審議來自行糾正的意志失敗。福柯的問題是結構性的：知識體制本身已事先決定什麼算作有效主張、誰算作有能力的發言者、審議的對象究竟是什麼——任何程序尚未啟動前便已如此。若程序本身的認識論條件就掌握在提案那一方手中，把程序搞對根本無濟於事。
Meta-critique / 元批判
The Kantian frame converts the problem into a question of will — whose authorization is legitimate, whose reason is being used versus substituted. This is clarifying, but it forecloses something important: the possibility that the difficulty is not primarily one of will but of *epistemic opacity*. Sapere aude implies that the obstacle is a failure of nerve — something citizens can remedy by choosing to reason publicly rather than delegating. But large language models are genuinely opaque, to users, to regulators, and in significant respects to their builders. No amount of willingness to deliberate closes that gap. Kant's framework is designed for a moral domain where the law is in principle accessible to every rational being through reason alone. Applied where the relevant facts are technically inaccessible, it transforms what may be a structural problem of knowledge into a problem of character. Those who defer are not cowards; the situation may be one in which Sapere aude is simply not a remedy. The consolation the essay quietly offers — that courage and public reason would suffice — is its most consequential and least examined move.
康德框架將問題轉化為一個**意志**問題——誰的授權是合法的，誰的理性在被使用或被取代。這有其澄清作用，但它遮蔽了一件重要的事：困難或許根本不在意志，而在**認識論上的不透明性**。Sapere aude 隱含的意思是：障礙是勇氣的缺乏——公民可以通過選擇公開審議而非委託他人來加以糾正。但大型語言模型對用戶、對監管者、在許多方面也對建造者本身，是真正不透明的。再多的審議意願也無法彌合這一鴻溝。康德框架設計的是一個道德律原則上可由每位理性存在者通過純粹理性加以掌握的道德世界；應用於相關事實在技術上不可及的領域，它便將一個可能是知識的結構性難題轉換成了品格問題。那些選擇委託的公民並不是懦夫；在這種處境中，Sapere aude 可能根本就不是一帖藥。文章悄悄提供的安慰——勇氣與公共理性便已足夠——或許是它最不願承認的空白。
Open Questions / 留給讀者的問題
1. If "united will" functions as a regulative idea rather than an achievable procedure, does the essay's standard operate as a measure for better governance or as a permanent prohibition on any AI governance at all?
2. The essay warns against a new class of guardians who "think through us" — what distinguishes that danger from the philosopher who, by establishing the formal conditions of legitimate deliberation, already frames in advance what citizens are capable of deciding?
3. If the contract arrives after the appropriation and the systems are already deployed at scale, does Kant's framework have the resources to adjudicate legitimacy *after* an irreversible technical fact, or does it only operate prospectively, leaving the present arrangement untouched?
一、如果「聯合意志」只是一個規範性理念而非可達成的程序，文章所設定的標準究竟是衡量更好治理的尺度，還是對任何 AI 治理的永久禁令？
二、文章警告一個新的監護者階級「透過我們思考」——但哲學家通過確立合法審議的形式條件、預先框定公民能夠決定的邊界，這種做法與前述危險之間，有何本質區別？
三、如果契約在佔用之後才到來，系統已然大規模部署——康德框架是否有能力在一個不可逆的技術事實發生之後裁定合法性，還是它只能向前看，對當下的安排無能為力？
Related Essays · 相關文章
Related · 相關
Extends · 延伸
- kant-on-the-maxim-of-easy-exit-20260430 — Applies target's social contract framework to political participation: costless party entry and exit reveals that the rational commitment Kant requires for legitimate collective legislation has already collapsed in democratic practice, before AI arrives.
- kant-on-the-glaciers-and-on-the-maxim-that-cannot-be-willed-20260430 — Extends target's question of who counts as a contracting party: future generations who cannot consent expose the intergenerational aporia haunting any social contract that claims universal scope across time.
- kant-on-the-imputation-of-mechanical-acts-20260430 — Applies target's dignity-and-autonomy argument to penal attribution: if humans are treated as means in AI training, the resulting inability to assign moral responsibility extends target's critique from legislative standing to legal accountability.
Contradicts · 衝突
Tagged: Philosophy, Kant, AI Governance
Curated by Shiva Dragon · https://amshiva.com/writing/kant-on-the-pretension-of-a-social-contract-proposed-by-a-single--20260430