童之伟 我之法学思想观点重述 [Tong Zhiwei, A Restatement of My Jurisprudential Thoughts and Views]

It is with great pleasure, and at the request of its author, my friend and colleague Tong Zhiwei, that I pass along a exquisitely intense, profound, transformative, and thought provoking development of a general theory of law, his jurisprudence of Faquanism (Faquanzhuyi). It is entitled 童之伟 我之法学思想观点重述 [Tong Zhiwei, A Restatement of My Jurisprudential Thoughts and Views] and appears in the original Chinese and Professor Tong's English translation.

最近与某机构有约，由我自己出面，从法的一般理论角度，概括一下自己的法学思想观点。我觉得这很有意思，值得花时间做一做。我在法学上的一些想法，很肤浅、不成熟，大部分反映在最近十年出版的几本书中（十年来出版的书目，附后），少部分曾通过学术讲座的形式呈现。在本文中，权体量和结构公理、义务体量和结构公理是2023前后我在武大法学院秦前红教授主持的学术讲座上提出和阐释的，权结构平衡、法权结构平衡2023年5月在香港中文大学法学院习超教授主持的学术讲座上有较集中论述，而关于国际法的七个观点都是2026年5月下旬我在四川大学国际关系学院李昊教授主持的学术报告会上提出和阐释的。现将我的法学思想观点简单重述如下：

I recently made an arrangement with an institution that I would myself, from the perspective of the general theory of law, summarize my own thoughts and views on jurisprudence. I find this very interesting and well worth the time. My ideas on jurisprudence, which are rather superficial and immature, are mostly reflected in the several books I have published over the past decade (a list of these books is appended); a smaller part has been presented in the form of academic lectures. Of the views restated in this essay, the axiom of the volume and structure of quan and the axiom of the volume and structure of duties were proposed and expounded around 2023 in an academic lecture at Wuhan University School of Law chaired by Professor Qin Qianhong; the balance of the quan structure and the balance of the faquan structure were discussed in a relatively concentrated manner in May 2023 in an academic lecture at the Faculty of Law of the Chinese University of Hong Kong chaired by Professor Xi Chao; and the seven propositions concerning international law were all proposed and expounded in late May 2026 at an academic lecture at the School of International Studies, Sichuan University, chaired by Professor Li Hao. I now briefly restate my jurisprudential thoughts and views as follows:

In this restatement, Professor Tong synthesizes and restates his jurisprudence of Faquanism (Faquanzhuyi), offering a materialist alternative to imported legal paradigms. Utilizing Marx’s method of ascending from the abstract to the concrete, Tong posits quan (权)—the manifestation of all interests and property in social norms—as the foundational analytical starting point. Dialectically, quan yields a cluster of seven essential categories: quan, faquan, residual quan, rights, powers, duties, and law.

Faquan represents property with settled ownership recognized in positive law, resolving into individual rights and public powers. Conversely, residual quan represents extra-legal norms governing unsettled property. Duties reflect the negative-value counterpart of these positive holdings. Under Tong’s structural axioms, the volume of quan directly depends on total property, while the proportional share of rights versus powers within faquan is determined by the ratio of private to public property. The core function of law is to optimize the allocation of quan and faquan, preserving equilibrium across zero-sum dynamics (faquan vs. residual quan; rights vs. powers) to avoid both state despotism and social anarchy.

Tong critiques contemporary Chinese jurisprudence for relying on "Japanized" translations (和化“权利”) that conflate rights and powers under a single term, urging Chinese scholarship to embrace native, precise terminology. Finally, Tong extends faquanism to international law: state entities hold international rights based on national wealth, whereas international organs exercise international power through autonomous budgetary enforcement. He concludes that current global insecurity stems from severe disparities in state rights and an underdevelopment of institutional international power.

 

generated with Grok

I will have more to say about this in future posts. For the moment it may be worthwhile to lay out simply the fundamentals of what I have, as of this moment, captured from this highly nuanced and sophisticated project:

Tong Zhiwei’s legal theory, known as Faquanism (Faquanzhuyi / 法权主义), replaces traditional legal binaries (such as the right-duty framework) with a unified, materialist framework constructed around interest, property, power, and rights. Tong targets a specific historiographical defect in Chinese legal scholarship: the continued reliance on the "Japanized" concept of quanli (和化“权利”), introduced during the late Qing and Republican periods via scholars like Hozumi Nobushige and Hozumi Shigeto. This imported usage conflates "rights + powers" under a single term (quanli). Tong argues that this conflation has deeply distorted Chinese jurisprudence, has been abandoned in Japanese legal scholarship, and must be eliminated in favor of normative Chinese terminology that strictly distinguishes Rights (quanli / 权利), Powers (quanli / 权力), and their overarching synthesis, Faquan (faquan / 法权).

Adopting Marx’s method of ascending from the abstract to the concrete, Tong selects Quan (权) as the initial, most abstract analytical starting point for jurisprudence. Quan is defined as "the manifestation of all interests and property in social norms of every kind." From it seven basic categories may be organized/recognized, the interactions of which, through the dialectical movement of quan, produce a phenomenological aspect as the general theory. These consist of a cluster of seven essential concepts that exhaustively analyze legal phenomena: (1) Quan (权); (2) Faquan (法权); (3) Residual quan (剩余权); (4) 
Rights (权利); (5) Powers (权力); (6) Duties (义务 / 法义务); and (7) Law (法 / 法律).

Ontologically, Quan encompasses all positive manifestations of interest and property across both legal and extra-legal social norms. It sits above faquan and residual quan as their total sum, defining the outer boundary of the jurisprudential system. Simultaneously, Quan resides immanently within the dialectical tension between its two primary operational components: (1) Faquan (法权): The legal manifestation of property whose ownership is settled (guishu yiding / 归属已定); and (2) Residual quan (剩余权): The extra-legal manifestation of property whose ownership remains unsettled (guishu weiding / 归属未定).

When Quan enters positive law, it becomes Faquan, which serves as the core operational entity of the legal system. Faquan further decomposes into Rights (held by individuals based on individual property) and Powers (held by public authorities based on public property). Under this property-determinist baseline, the total volume of rights is dictated by individual property, while the total volume of powers is dictated by public property—with even non-material interests (such as personality rights) ultimately underpinned by real property rights (wuquan / 物权). Correspondingly, Duties constitute the negative side (negative-value property) attached to these categories.

Because economic processes (production, exchange, and consumption) continuously alter the underlying distribution of property, faquan and residual quan exist in a zero-sum, dynamic ratio (Faquan Rate + Residual quan Rate = 100%). The dynamic absorption and realignment of residual quan into faquan serves as the primary engine of legal evolution. Ultimately, law acts as both organizer and mediator: optimizing the allocation of quan and faquan to ensure structural balance, prevent both anarchy and despotic power, and maximize the preservation and growth of faquan.

As a first order general understanding, then, one might say that under Tong's model, the primary function of law is to act as both organizer and mediator, maintaining a dynamic balance between private rights and public power.

Here is my own first effort at organizing some of the key concepts likely subject to revision

Category	Textual Definition & Material Base	Structural & Practical Function
Quan (权)	Abstract starting point; manifestation of all interests and property in social norms.	The overarching totality (Faquan + Residual Quan).
Faquan (法权)	Manifestation in law of property whose ownership is settled.	Core legal phenomenon; rights-powers community; center of the legal system.
Residual Quan (剩余权)	Manifestation in social norms outside law of property whose ownership is unsettled.	Extra-legal reservoir; fundamental object of legal regulation when interacting with faquan.
Rights (权利)	Legal form of individual interests and individual property.	Individual portion of faquan.
Powers (权力)	Legal form of public interests and public authority property.	Public/state portion of faquan.
Duties (义务)	Reverse side of quan/faquan; negative interests and negative-value property.	Corresponding burden attached to rights, powers, and residual forms[cite: 3].
Structural Balance	Ratios of Faquan Rate / Residual Quan Rate and Rights Rate / Powers Rate[cite: 3].	Ideal law-based state avoids anarchy (excess rights) and despotism (excess powers)[cite: 3].
Internationalization	Transformation of domestic faquan to international entities (states/orgs)[cite: 3].	Analyzes global order via international natural rights vs. institutional powers[cite: 3].

It is interesting to situate Professor Tong's general theory against the Institutes of Justinian for the light they may begin to shed on both. Faquanism and the Insitutes might be said to diverge fundamentally in their analytical methods, reflecting their distinct historical and philosophical foundations. The Institutes, written as an introductory textbook for Roman legal education, employs an inductive, categorization-driven process. It opens with pragmatic definitions of justice and jurisprudence before dividing the entirety of private law into three practical axes: persons (personae), things (res), and actions (actiones). In contrast, Tong explicitly adopts Karl Marx’s dialectical method of ascending from the abstract to the concrete. Through the internal dialectical movement of quan, Tong derives a cluster of seven interdependent categories (quan, faquan, residual quan, rights, powers, duties, and law) that systematically unfold to explain the entire world of legal phenomena.

 While both systems recognize property as a cornerstone of legal order, they conceptualize the source and function of legal norms differently. In the Justinianic framework, norms derive their legitimacy from a dual tier of natural law (ius naturale / ius gentium) and civil law (ius civile). Property rights in Roman law are classified by their mode of acquisition (natural vs. civil) and physical nature, operating alongside civil status to define legal capability. Tong’s Faquanism, by contrast, establishes a strict materialist property-determinism. Legal norms do not derive from abstract moral principles or natural law; rather, faquan (and positive law itself) is an explicit product of economic processes. The volume of individual property dictates the total volume of individual rights, while the volume of public property dictates public power. Even non-material norms and personal protections (such as personality rights) are ultimately underpinned by real property rights (wuquan). Where Justinian views property as an object governed by law, Tong views property dynamic processes (production, exchange, consumption) as the foundational engine that generates and shifts legal norms.

There is much more by way of theoretic richness in Professor Tong's work. But there is nothing more enlightening than the original. It follows below and may be accessed online HERE.

Musings in the Form of a Policy Brief: Bots and Automated Systems in International Institutions

 

 

I have been doing a bit of thinking about (and arguing with Claude; Harvey AI, and ChatGPT about) the nature of bots and automated systems in international institutions. Bots are automated software programs that perform repetitive tasks over a network much faster than a human (chatbots, web crawlers, shopping bots, as well as the disfavored spambot and DDosBots). This has become more important--and especially its theorization (as always with me from a semiotically informed critical stance)--as my research assistant and I confront those issues in developing an operational chatbot for issues that arise in the interpretation and application of the UN Guiding Principles for Business and Human Rights.

But first things first. To get a handle on operationalization one has to start with the current state of things. And for that purpose we put together this Policy Brief: Bots and Automated Systems in International Institutions, the more formal version of which follows below for those who are interested in a  fuller treatment with that theoretical material and full citations.

Policy Brief: Bots and Automated Systems in International Institutions

What's happening

International institutions — the UN system, international financial institutions, and humanitarian agencies — have rapidly expanded their use of automated software agents ("bots"), moving from isolated pilots to coordinated, system-wide infrastructure over roughly the past three years. The UN's own 2023–2025 tracking shows over 400 AI projects across 47 entities, with generative-AI chatbots the fastest-growing category.

Where bots are actually being used

• Internal operations: UNifyHR (a shared HR chatbot built by 13 UN agencies via the UN International Computing Centre) answers staff policy questions across the system.
• Public-facing information services: UNEP's EnvironmentGPT, WIPO's AI-assisted transcription for multilingual diplomatic meetings, and the UN Innovation and Analytics Hub's Libra.AI (making human-rights information more accessible).
• Monitoring and early warning: WFP's HungerMap LIVE and SKAI (satellite damage assessment), UNHCR's protection-risk scoring for refugees, UNDP's Crisis Risk Dashboard, UNCCD's satellite monitoring of mining sites, WHO's AI-assisted TB screening.
• A dedicated research/advisory body: UNICRI's Centre for Artificial Intelligence and Robotics (The Hague, est. 2017).
• External threats institutions must defend against: coordinated bot networks used for disinformation, including against multilateral bodies and electoral-observation missions.

International financial institutions and humanitarian NGOs (World Bank, IMF, ICRC) follow a similar pattern, generally by licensing commercial AI infrastructure rather than building their own.

The governance response so far

• The UN Global Digital Compact (2024) is the most comprehensive multilateral text to date on AI and digital cooperation.
• The UN High-Level Advisory Body on Artificial Intelligence has issued recommendations on the UN's role in global AI governance.
• UNESCO leads system-wide AI ethics guidance.
• Individual agencies have their own strategies: UNHCR's AI Approach, WFP's AI Strategy 2025–2027, UNDP's AI Sprint program — coordinated through an Inter-Agency Working Group on AI.
• Outside the UN system, the EU AI Act is the most developed example of binding, risk-tiered regulation, in contrast to the UN's preference for voluntary, principles-based instruments.

Four things worth flagging for decision-makers

1. Accountability gap. Bots that score refugee protection risk, flag compliance issues, or triage humanitarian needs are making decisions that function like administrative discretion — without the due-process or appeal mechanisms that would normally attach to a human decision-maker.
2. Vendor concentration. Most institutions license rather than build their AI infrastructure, and the market is concentrated among a handful of U.S. firms, most based in the San Francisco Bay Area. This creates dependency on a small number of commercial actors whose design choices — and business incentives — are not set through any multilateral process.
3. Mismatch between the language of governance and the reality on the ground. High-level UN AI-governance documents are written substantially in the vocabulary of long-term, catastrophic "AI safety" risk — a framing that originated in a specific California policy and research community (sometimes referred to critically as "TESCREAL," an acronym for a cluster of transhumanist, longtermist, and effective-altruist ideas associated with figures like Sam Altman, Elon Musk, and others). That framing is a reasonable thing for institutions to take seriously, but it doesn't map well onto the bots institutions are actually running today, which raise much more immediate, mundane accountability and bias questions. Decision-makers should not assume that a governance framework built for hypothetical future "frontier" systems will also cover today's HR chatbots and risk-scoring tools.
4. North-South asymmetry. The bots that touch the most vulnerable populations (refugee scoring, hunger monitoring, mining-site surveillance) are deployed overwhelmingly in the Global South, while the design and the governance-vocabulary-setting happen overwhelmingly in North American and European institutions and vendor firms.

Bottom line

International institutions have moved faster on deploying bots than on governing them. The multilateral instruments now emerging (the Global Digital Compact, the High-Level Advisory Body's work, agency-level AI strategies) are a genuine start, but they are currently better calibrated to speculative future risks from advanced AI systems than to the accountability, due-process, and vendor-dependency questions raised by the bots already embedded in day-to-day institutional operations. Closing that gap — not just adding more principles — should be the near-term priority.

 The longer version follows below and may be accessed HERE.

Global SWF August 2026 Report and Interview With Michael Clancy Australia Superannuation Fund

 

 

 

Our friends at Global SWF have announced the distribution of their August 2026 Report. They summarize its contents this way:

Global SWF August 2026 Report
Australian State Investors, Rest Super, Royal Family Offices
Global SWF
Aug 2, 2026

Happy August! Sovereign Investors had a relatively strong month of July, with US$ 23.3 billion in 45 transactions and several funds closed. Read all about the deals, results, and new funds at the Global SWF Times. Australia’s state investors continue to grow as superannuation funds merge and new managers arise. The monthly report looks at the current size the industry, as well as the performance of the largest 10 super funds (“Kangaroo 10”) in the past decade. These are shown with the two infographics of the month. In that context, the fund of the month goes to Retail Employees Superannuation Trust (Rest), the sixth largest fund with a US$ 77 billion portfolio. Do not miss our conversation with its CIO, Mr. Michael Clancy. Lastly, we list the 50 largest royal family offices, which we have started to cover in detail after mapping 100+ of these entities and subsidiaries, mostly in the Middle East, but also in Europe, Africa, and Asia.

The August report can now be accessed at https://globalswf.com/reports/august2026 - if your firm is a subscriber of Global SWF and you forgot your password, you can always reset it with your email at https://globalswf.com/password/reset. And if your firm is not a subscriber yet, feel free to reach out to us. We remain at your disposal should you want to discuss any of the topics in detail.

 

Full scorecared available here

The text of the SWF’s Fund of the month interview with Michael Clancy, CIO of the  Australia Superannuation Fund, follows below. 

 

 

OECD Watch: Publication of Report--State of Remedy 2025; Analysing community and NGO-led National Contact Point complaints concluded in 2025

 

 

At the end of July 2026, OECD Watch  published its annual State of Remedy Report. This year it is entitled: State of Remedy 2025Analysing community and NGO-led National Contact Point complaints concluded in 2025. This is the way OECD Watch summarized its Report on its Website:

This year marks a dual milestone for responsible business conduct: the 50th anniversary of the OECD Guidelines for Multinational Enterprises on Responsible Business Conduct, and the 25th anniversary of the National Contact Point (NCP) grievance mechanism. To help mark this occasion, and to highlight the importance of the NCP mechanism, OECD Watch has released its annual report analysing the community and NGO-led NCP complaints that were concluded last year.

The State of Remedy 2025 draws on OECD Watch’s complaint database and direct engagement with civil society complaints, to examine complaint outcomes from 2025, key numbers for complaints, remedy highlights, and remedy outcomes. The report offers an evidence-based account of how the NCP mechanism functions in practice to provide a state-backed, non-judicial pathway for affected individuals and communities to seek remedy.

Although many complaints have contributed to remedy and changed business practices, NCP performance is not equal across member states. More progress is needed to ensure effective remedy for the victims of corporate misconduct. What needs to change first and foremost, as emphasised in our report, is the voluntary nature of responsible business conduct standards. Companies are not driven to act responsibly through voluntary standards alone – instead binding government regulation is required to achieve true justice.

This report takes stock of what has been achieved, while simultaneously looking ahead to what needs to be done next, to attain and maintain an even higher standard of responsible business conduct. With this and previous reports, OECD Watch affirms its position in promoting binding government regulation to remediate as well as mitigate harms caused by corporate misconduct.

 

The "Key Numbers for 2025" follows below.

《人工智能示范法4.0》中英双语版本发布 «Bilingual Chinese-English Version of the "Model Law on Artificial Intelligence 4.0" Released»

 

Image generated with ChatGPT

Groups of scholars and others  have been working on the creation of a Model Law of Artificial Intelligence. One of these groups, 中国法学会 网络与信息法学研究会 [the China Law Society Research Society on Cyber and Information Law], posted their version 4.0 of a Model Law on Artificial Intelligence for China recently. 中英双语版本发布 «Bilingual Chinese-English Version of the "Model Law on Artificial Intelligence 4.0" Released». The text follows below.

As they describe it:

The Model Law is divided into 7 parts:

Chapter 1, General provisions (articles 1-15): This sets the ideological, aspirational, political and operational premises under and through which the Model Law is to be understood, interpreted and applied. These serve as the framing premises of the Model Law and also situates it within the regulatory cognitive universe  with which it must align and the values and objectives of which it must advance and protect. 

Chapter 2, Support and Promotion of AI (articles 16-39): This specifies the operational structures and frameworks within which the high quality development premises of Chapter 1 can be realized within the guiding premises of the Chinese fundamental political line (formalized planning, infrastructure, data, algorithms and models, basic research, IP, eco-system development, pilot zones for development and free trade zones, assessment). This is the operational core of the Model Law. 

Chapter 3, AI Oversight Systems (articles 40-48):  This specifies the oversight framework and objectives within the activities of Chapter 2 are to be realized (safety/security obligations, life cycle security of critical systems, security vulnerabilities, audit, remedy and notice, transparency/fairness/explanability, risk management and risk blocking, AI ethics review). This is the supervisory core of the Model Law. 

Chapter 4, Obligations of AI Developers, Providers, and Users (articles 49-80): this is a very long framing chapter that is organized within a structural chapeau (general provisions, arts. 49-63) and then divided among the functionally differentiated obligations of developers (arts. 64-64); providers (arts. 66-74), and users (arts. 75-80). These provisions are deeply embedded within the complex regulatory structures of these elements of development, production and use already in effect or in gestation.  These are operations/implementation level router provisions with sometimes substantial regulatory "value added" or value changing effect. 

Chapter 5, Comprehensive AI Governance Mechanisms (arts. 81-95): These provisions are outward facing guidance for state authorities. They must be read in the shadow of the fundamental political line and the overall authority of the CPC and its organs, especially as they operate at the pubic facing implementation and regulatory guidance levels. The  extraterritorial provisions mirror current expansionist efforts through a Socialist "effects" test theory. The most interesting provision is article 93 (governance through technology), focusing on support for tech innovation and its support by state organs and then highlighting RegTech and ComplianceTech but not social credit and smart cities systems.  That suggests that the state and state organs are NOT may not be objects of this regulation.

Chapter 6 Liabilities (arts. 96-109): These provisions align with general remedial measures in the Chinese domestic legal order refined for this field of law. The essence is a graded system of administrative supervision (art. 96) starting with supervisory orders of rectification in the face of violation, issuance of a warning, the confiscation of gains acquired through violation of law and termination of business activities. Article 102 provides a safe harbor for tort damages.

Chapter 7 Supplementary Provisions (arts. 110-114). These include definitions of key terms and special provisions for military AI, AI R&D, and the operation of the AI Negative List (see also e.g., art. 40). 

For an interesting analysis see Stanford's DigiChina website HERE.

The China Law Society Research Society on Cyber and Information Law

Established in September 2002—originally under the name "Information Law Research Society of the China Law Society"—the China Law Society Research Society on Cyber ​​and Information Law is a national-level research society directly affiliated with the China Law Society. It was co-founded by six entities: the former State Council Informatization Office, the Ministry of Public Security, the Ministry of State Security, the former Ministry of Information Industry, the National Administration of State Secrets Protection, and the Institute of Law of the Chinese Academy of Social Sciences (CASS).
In November 2015, following a decision by the Party Leadership Group of the China Law Society, the organization was renamed the China Law Society Research Society on Cyber ​​and Information Law. Currently, its sponsoring entities comprise seven organizations: the Office of the Central Cyberspace Affairs Commission, the Ministry of Industry and Information Technology, the National Administration of State Secrets Protection, the Ministry of Public Security, the Ministry of State Security, the State Cryptography Administration, and the Institute of Law of CASS. The Research Society’s secretariat is based at the Institute of Law of CASS, with operational support provided by the Institute’s Department of Cyber ​​and Information Law.
Since its inception, and under the leadership of the China Law Society and the support of relevant entities, the Research Society has engaged in activities such as legislative seminars, academic exchanges, research projects, and publications. It is dedicated to fostering research in the fields of cyber and information law, advancing the development of the discipline’s academic, theoretical, and discourse systems, and providing intellectual support for the construction of a rule-of-law China, a cyber powerhouse, and a digital China.

从容器到虚拟空间中的技术主权者：数字平台与运营空间中法律主体性的辩证法

 

图片由 ChatGPT 生成。

 

ENGLISH ORIGINAL HERE

VERSION EN ESPAÑOL AQUÍ

我很高兴在此提供我最初为“技术与国际及跨国法律主体性”专题研讨会撰写的讨论稿。该研讨会是“国际、跨国及欧盟法中的技术范式转变”国际研讨会的一部分，于2026年1月29日在意大利罗马第三大学举行，由富有启发性的法比奥·巴桑教授组织。该论文的最终稿可能会与其他论文一起收录在法比奥·巴桑主编的《国际、跨国及欧盟法中的技术范式转变》一书中。

这篇题为《从虚拟空间中的容器到技术主权者：数字平台与运营空间中法律主体性的辩证法》的论文旨在探讨平台的自主性，以及由此而来的主权——或许更准确地说是技术主权——这种自主性源于传统政治主权者竭力推行的对平台形式的客观化和工具化，即便这些平台在日益脱离其创造者的机器系统中运行，这种自主性依然存在。更具体地说，这篇论文探讨了数字平台如何从被动的法律客体转变为挑战传统国家权威的自主监管主体。文章援引圣经中参孙的寓言，论证了随着自动化治理和算法决策取代人类监督，平台正在发展自身的技术主权。尽管传统框架将这些系统视为受国际法约束的容器，但贝克尔认为它们正在演变为能够定义自身内部现实的独立规范秩序。这种转变造就了主权竞争的格局，国家实体权力与全球科技基础设施的虚拟自主性在此发生冲突。最终，本文认为，这些不断发展的技术系统最终可能会动摇并取代以人为中心的法律结构。

本文以法比奥·巴桑的《数字平台与全球法律》为基准，试图论证：关键在于：法律主体性一直以来都被赋予了环绕空间的事物或空间内的行为，而非空间本身。只要平台保持被动状态——例如市场广场、船舶，以及巴桑所论述的客体空间——这一前提就成立。本文检验的论点是，自动化和生成式决策正在从内部瓦解这一前提：随着决策权从曾经运营平台的人员和机构转移到平台自身的运作中，平台空间开始满足主体性的经典检验标准——即拥有持有、行使和对其自身规范性承诺负责的能力——而无需任何主权行为的认可。它成为一种实现的科技模拟；一个自我建构而非被建构的主体，从定义上讲，不再仅仅是一个主体。当法律中旧有的主体与客体分配——市场和公司仍然奉行的这种分配——不再能描述圣殿内部实际发生的一切时，它便成为现实。让我们依次简要地考察这一分析的各个部分。

以下是英文原文的中文翻译。 

图片由 ChatGPT 生成

 

 

Just Published China in the World Vol. 8(2)

 

 I am delighted to pass along the announcement by its editor in Chief, Zhiqun Zhu (Bucknell), of the publication of Vol 8(2) of China and the World  (formerly published under the title: China and the World: Ancient and Modern Silk Road (vols. 1-7)). Many of the articles are open access. Articles include Shifting Between Friend and Enemy: The US-China–Japan Triangle (Quansheng Zhao); US Signaling, Reliability, and Cross-Strait Dynamics during the Second Trump Administration (Charles Chong-Han Wu); Why Do Chinese Women Outperform Men in International Soccer?: Associational Behavior, Perceived Risk, and Differentiated Control in China (Taiyi Sun); Can Leaders Change the World?– A Macro Perspective of Leadership Limitation in International Relations (Qingshan Tan); Are Leaders the Products of Their Times or Vice Versa? (Jianwei Wang); and Democracy through the Lens of Chinese Scholars: The Battle of Ideas by He Li (Baogang Guo).

Table of Contents and Links follow below. 

 

 

Borrador de discusión publicado: «De recipientes a tecno-soberanos en espacios virtuales: plataformas digitales y la dialéctica de la subjetividad jurídica en espacios operativos»

 

Pix generated with ChatGPT

 Me complace poner a disposición el borrador de mi ponencia, que surgió como contribución a la sesión sobre «Tecnología y subjetividad jurídica internacional y transnacional», celebrada en el marco del simposio internacional *The Technological Paradigm Shift in International, Transnational and European Union Law* (El cambio de paradigma tecnológico en el Derecho internacional, transnacional y de la Unión Europea) en la Universidad Roma Tre (Roma, Italia) el 29 de enero de 2026, bajo la organización del inspirador profesor Fabio Bassan. Es posible que la versión definitiva de este trabajo se publique junto con otros ensayos en la obra *The Technological Paradigm Shift in International, Transnational and European Union Law* (editada por Fabio Bassan).

Pix generated ChatGPT

 

ENGLISH LANGUAGE VERSION HERE

La contribución titulada «De recipientes a tecno-soberanos en espacios virtuales: plataformas digitales y la dialéctica de la subjetividad jurídica en espacios operativos» tiene por objeto examinar la autonomía de las plataformas —y, en ese sentido, su soberanía, o más precisamente su tecno-soberanía—, la cual surge a partir de (y tal vez a pesar de) los esfuerzos de los soberanos políticos tradicionales por imponer la cosificación e instrumentalización de dicha forma, incluso cuando estas operan en y a través de sistemas automatizados cada vez más desvinculados de sus creadores. Más concretamente, el trabajo explora la transformación de las plataformas digitales de objetos jurídicos pasivos en sujetos normativos autónomos que desafían la autoridad estatal tradicional. Mediante la alegoría bíblica de Sansón, el texto sostiene que, a medida que la gobernanza automatizada y la toma de decisiones algorítmica sustituyen a la supervisión humana, las plataformas desarrollan su propia tecno-soberanía. Si bien los marcos tradicionales consideran estos sistemas como meros recipientes sujetos al derecho internacional, Backer sugiere que están evolucionando hacia órdenes normativos independientes, capaces de definir sus propias realidades internas. Este cambio genera un escenario de soberanías en pugna, donde el poder físico del Estado choca con la autonomía virtual de las infraestructuras tecnológicas globales. En última instancia, la obra plantea que estos sistemas tecnológicos en expansión podrían llegar a desestabilizar y reemplazar las estructuras jurídicas centradas en el ser humano.

Tomando como punto de partida la obra *Digital Platforms and Global Law* de Fabio Bassan, argumento que lo que está en juego es lo siguiente: la subjetividad jurídica siempre se ha atribuido a aquello que ocupa un espacio o actúa dentro de él, nunca al espacio mismo. Dicha premisa se mantuvo vigente mientras las plataformas permanecieron pasivas —como plazas de mercado o recipientes; es decir, el espacio-objeto que Bassan utiliza acertadamente como base—. La tesis que esta contribución pone a prueba es que la toma de decisiones automatizada y generativa está disolviendo esa premisa desde dentro: a medida que la toma de decisiones migra de los humanos e instituciones que antes gestionaban la plataforma hacia las operaciones propias de esta, dicho espacio comienza a cumplir los criterios clásicos de la subjetividad —la capacidad de asumir, ejercer y responder por sus propios compromisos normativos— sin haber recibido tal capacidad mediante ningún acto soberano de reconocimiento. Se convierte en una tecno-simulación realizada; un sujeto que se constituye a sí mismo, en lugar de ser constituido por otro, deja de ser, por definición, un mero sujeto. Es lo que queda cuando la antigua distribución jurídica entre sujeto y objeto —aquella que tanto la plaza de mercado como la corporación respetaban— deja de describir lo que realmente sucede dentro del templo. Consideremos brevemente las partes de este análisis, una por una.

ChatGPT

El resumen, el índice y la introducción se encuentran a continuación. El borrador se puede descargar AQUÍ.

Discussion Draft Posted: "From Vessels to Techno-Sovereigns in Virtual Spaces: Digital Platforms and the Dialectics of Legal Subjectivity in Operational Spaces"

Pix generated by ChatGPT

 

Versión en español acquí

 I am delighted to make available my discussion draft of what had started as my contribution to the     Session on  Technology and International and Transnational Legal Subjectivity which was part of the International Symposium:The Technological Paradigm Shift in International and Transnational andEuropean Union Law;, Roma Tre, Rome, Italy  29 January 2026, organized by the inspirational Professor Fabio Bassan. A final draft of this contribution may be published with other essays in (Fabio Bassan (ed)), The Technological Paradigm Shift in International Transnational and European Union Law.

Pix generated by ChatGPT

The title of the contribution that resulted "From Vessels to Techno-Sovereigns in Virtual Spaces: Digital Platforms and the Dialectics of Legal Subjectivity in Operational Spaces" has as its object was consider the autonomy of platforms, and in that sense, their sovereignty, perhaps better put their techno-sovereignty, emerging from and perhaps despite the best efforts of traditional political sovereigns to enforce the objectification and instrumentalization of the form, even when these forms operate in and through machine systems that are increasingly detached from their creators. More specifically the contribution explores the transformation of digital platforms from passive legal objects into autonomous regulatory subjects that challenge traditional state authority. Using the biblical allegory of Samson, the text argues that as automated governance and algorithmic decision-making replace human oversight, platforms develop their own techno-sovereignty. While traditional frameworks treat these systems as vessels subject to international law, Backer suggests they are evolving into independent normative orders capable of defining their own internal realities. This shift creates a landscape of competing sovereignties where the physical power of the state clashes with the virtual autonomy of global tech infrastructures. Ultimately, the source posits that these advancing technological systems may eventually destabilize and replace human-centered legal structures.

Starting from Fabio Bassan's Digital Platforms and Global Law as a baseline, I attempt the argument that what is at stake is this: legal subjectivity has always been assigned to what rings a space or acts within it — never to the space itself. That premise held so long as platforms remained passive — market squares, vessels, the object-space Bassan rightly starts from. The claim this contribution tests is that automated and generative decision-making is dissolving that premise from the inside: as decision making migrates from the humans and institutions that once ran the platform and into the platform's own operations, that platform space begins to satisfy the classical tests for subjectivity — the capacity to hold, exercise, and answer for its own normative commitments — without having been granted that capacity by any sovereign act of recognition. It becomes a realized techno-simulation; a subject that constitutes itself rather than being constituted is, by definition, no longer merely a subject. It is what remains once law's old distribution of subject and object — the distribution the market square and the corporation both still honored — stops describing what is actually happening inside the temple. Let us briefly consider the parts of this analysis each in turn.

Pix generated with ChatGPT

 

 The Abstract, Table of Contents, and introduction follow below. The draft may be downloaded HERE. 

欲合于时势，必先摩（摩）其情——关于梁文锋投资者交流会实录的反思：梁文锋（DeepSeek）的神谕式问答

 

图片来源请见此处。

 ENGLISH LANGUAGE VERSION HERE 

AI 现在不缺品位和直觉，它缺的是持续学习的能力。

（梁文锋 / DeepSeek 投资者交流会 [梁文锋投资者交流会实录]）

鬼谷子，这位中国古代修辞学家，在其经典著作第二卷（英译本：Guiguzi, China's First Treatise on Rhetoric, Hui Wu 译, 2016）中阐释道：有些情境下，若欲合于时势，则须把握——或者更确切地说——揣摩（摩；mó）。摩，是《鬼谷子》中关于"试探"或"抚摩"的篇章。该字本义为摩擦、磨砺、以触觉感知——而其技法正是如此：不直接追问对方的欲求或信念，而是施加一个微小的刺激（一个词、一个姿态、一个提议、一种情感的流露），然后观察回应。内心状态，文本论证道，一经触动便无法隐藏——它们会作为反应浮现出来，犹如打磨一面镜子使其光泽显现，或拨动一根琴弦使其音高暴露。精通摩术的说服者悄然而间接地运作：试探、解读反应、调整、再试探，直到对方真实的欲望、恐惧与可利用之处被完全勘明——往往在对方尚未意识到自己已泄露信息之时。

此策略出自经典文本第二卷第四章，紧随"揣"（揣 Chuāi）之后，直接位于"权"（權 Quán）之前。"揣"涉及揣度内在动机，而"摩"则是通过外部互动、映射与言语试探来检验这些推算、以引发可预测回应或行动的主动过程；随后的"权"则产生揣与摩的必要综合，作为道（Dào——根、路径，有时包括其方向与认知概念）的功能，物化于并存在于部署这些策略的社会的价值观与前提之中。揣（Chuāi，揣度/估量）居先——属于远距离的分析阶段。在任何接触之前，你构建一幅情势图景：对方的权势、资源、情绪状态，其所好与所惧。这通过观察和推断完成，基本保持距离。它是在踏入对方心智之前对其进行的认知制图。摩（Mó，揣摩）是接触阶段——你将在揣中建立的假设对照现实加以检验。揣是理论，摩是实验：微小的试探，按"其类"校准（摩之以其类），实时解读共鸣与抵抗，并不断修正图景直至确认。权（Quán，权衡/裁断）是部署阶段——既然你已真正了解所面对之人，便如天平上的砝码般校准你的修辞，为勇者、怯者、贪者、傲者等分别选择不同的说服策略，以使结果向有利于你的方向倾斜。因此，弧线为：勘测（揣）→ 检验与确认（摩）→ 校准与部署（权）。摩是枢纽——它将私下的估量转化为可供修辞行动的经验证的情报。

这正是我们可以借以切入梁文锋于2026年7月DeepSeek一场长时间问答中所产生的那些具有神谕般卓异性质的产物的方式（梁文锋 / DeepSeek 投资者交流会 [梁文锋投资者交流会实录]）。而当将其视为一场更大对话的组成部分时，其意义则更加非凡——这场对话向外延伸，发生在新兴AI机器系统技术体系的巨头之间：Palantir（帕兰提尔）、Anthropic、OpenAI、Google、Meta及其评论者（对我而言，其中较有意思的是Leo Aschenbrenner [利奥波德·阿申布伦纳]）。本文既在其自身范围内考察该问答的产物，也将其视为一场更广泛对话的一部分——这场对话深深嵌入若干关键行动者各自同时投射其策略时所产生的认知与现象学投射之中，并在此过程中产生了大量的符号学争夺，这些争夺可能将重塑作为这一辩证法核心的技术机制内部的认知框架。

引言

在最近一次投资者会议上，DeepSeek创始人梁文锋详细阐述了DeepSeek的组织文化、开源哲学、技术路线图以及对AI行业竞争格局的看法（梁文锋 / DeepSeek 投资者交流会 [梁文锋投资者交流会实录]）。这些言论广泛传播，多家网站提供了英文版文字记录，其中一些经过编辑，以类似尼采格言的风格呈现（参见 The China Academy; HTX; WEEX; RootData）。我从此网站通过前Twitter编纂了自己的版本：HERE。借助翻译应用程序，我逐字研读了原文，并在下文附上了原始中文文本及中英对照翻译。背景是DeepSeek计划中的IPO。这些言论广泛传播的后果——以及任何有意评论这些言论的人对其进行深入或也许并不那么深入的挖掘——是巨大的："据知情人士称，DeepSeek已告知其第二轮融资的潜在投资者，目前暂停该交易，此前数日，广泛被认为出自创始人梁文锋的关于中美AI竞争的评论在网络上疯传。"（Fortune）。

 

我在这里补充的内容分为五个部分：（A）我对这118条神谕警句为我们提供的四个主要见解的感受； (B) 警句和神谕的总结； (C) 梁文峰的摩； （四）梁文峰鬼谷子谋的符号学； (E) 梁文峰的大背景言论。在另一篇文章中，我将整合这些神谕格言和马克·扎克伯格 7 月 29 日为 Meta 发表的格言。

A. 梁文锋言论中浮现的四项核心洞见：(1) 战略克制作为非对称竞争优势；(2) 极端资源效率可中和巨大的资本/算力差距；(3) 对通往智能的"主线"的严格、自律的专注；(4) 非传统的、愿景驱动的文化优于企业官僚体制。

1. 战略克制作为非对称竞争优势。

传统科技策略优先考虑快速用户获取、激进变现、专有护城河（闭源代码）以及竞争性市场抢占。DeepSeek从根本上拒绝了这种路径。通过维持严格的克制——将API定价设定为仅赚取"合理"利润、开源其前沿模型、回避"超级应用"野心、拒绝将其他科技公司视为敌人——DeepSeek避免了代价高昂且令人分心的竞争。出人意料的是，让渡价值降低了摩擦、培育了善意，并最终最大化了实现AGI的长期概率。

2. 极端资源效率可中和巨大的资本/算力差距。

硅谷的主流叙事认为，达到AGI纯粹是指数级资本支出和大规模GPU集群的函数。DeepSeek的运营现实对这种"蛮力"教条构成了挑战。以大约美国顶尖实验室二十分之一的算力预算运营，同时仅落后一到两年，证明了架构效率、高度聚焦的工程、软件优化（如编写TileLang等定制编译器以绕过遗留软件生态系统）以及精简运营可以弥合巨大的资源鸿沟。

3. 对通往智能的"主线"的严格、自律的专注。

在一个容易追逐短暂热点（如视频生成、3D资产创建、面向消费者的封装应用）的市场中，DeepSeek实施严格的智识纪律。梁文锋区分了商业上有利可图的分心与通往AGI的真正路径。通过将多模态视为仅仅是一个产品组件，并将持续学习——而非单纯的规模或世界模型——确定为阻止当前智能体达到自我迭代能力的唯一关键瓶颈，公司将其有限的工程周期完全集中于直接推进核心智能的事项。

4. 非传统的、愿景驱动的文化优于企业官僚体制。

高效的AI实验室并不必然需要僵化的组织层级、激进的KPI或高强度的内卷文化。DeepSeek的结构依赖于一种共享的、未成文的愿景和对员工的彻底信任：赋予顶尖研究人员高达50%的自由支配时间、避免强制加班、依靠共识驱动的领导、将长期团队稳定置于一切之上。这种宽松的、兴趣驱动的环境被界定为一种运营必需——因为前沿AI领域真正的科学突破需要的是认知上的呼吸空间，而非自上而下的压力。

B. 梁文锋在四小时投资者问答中的言论叙述性摘要。分为以下几个部分：(1) 愿景、组织哲学与克制；(2) 算力现实、国产芯片与生态独立；(3) 商业策略、开源承诺与市场前景；(4)

1. 愿景、组织哲学与克制。

DeepSeek的核心身份植根于一种未成文的、愿景驱动的哲学，而非传统的企业结构。公司在没有正式KPI、没有严格管理层级、没有绩效考核的情况下运营，将为人类利益实现通用人工智能（AGI）的使命置于短期财务回报或激进商业扩张之上。梁文锋强调一种深度克制的策略。DeepSeek并不与成熟的科技巨头竞争去构建高摩擦的"超级应用"，也不试图最大化用户变现，而是刻意放弃即时的商业市场抢占。这种克制被视为一种深思熟虑的策略：通过保持专注并避免在行业内形成敌对态势，公司最大化了其实际实现AGI的长期概率。在内部，维持团队稳定被视为DeepSeek最为关键的唯一优先事项。团队将资金和物理计算资源视为可获取的商品，但将凝聚且有动力的人才视为不可妥协的要素。最近的股权融资为长期员工提供了大量股权期权，在很大程度上缓解了核心人才流失风险。公司的决策高度依赖共识建立而非自上而下的行政指令。为培育研究创造力，管理层明确鼓励一种从容、低压的工作环境。员工通常被赋予高达50%的自由支配时间来探索自主研究项目，无需预设交付成果，并且避免加班以给予研究人员深度探索所需的心理带宽。

AGI技术路线图与研究重点。

DeepSeek将通往AGI的轨迹视为非突发性跳跃，而是一个审慎的、逐步的技术阶梯。其进程依次从基础语言模型到链式思维（Chain-of-Thought, CoT）等推理范式，进阶至自主智能体（Agent），掌握持续学习，达到自我迭代的技术奇点，最终终结于具身物理智能。当前，AI发展的首要瓶颈在于现有智能体缺乏在长时间段内持续适应和在岗学习的能力。解锁持续学习被视为使模型能够自主改进、开展研究并加速后续AI发展代际的关键里程碑。由于资源必须严格优先配置，DeepSeek严格遵循通往智能的这条主线。公司刻意回避3D资产创建或视频生成等旁支领域，将其视为虽商业上有利可图但并不从根本上推进机器智能上限的应用。类似地，多模态能力被视为面向终端用户的必要产品组件，而非底层推理能力的核心驱动力。为优化内部开发，DeepSeek将编码智能体（Coding Agent）置于所有其他垂直应用之上加以优先，因为强大的编码能力创造了一种复合反馈回路，极大地加速了公司自身的内部研究和模型迭代。

2. 算力现实、国产芯片与生态独立。

在谈及中国国内AI与美国领先前沿实验室之间的性能差距时，梁文锋指出，这一分野完全是由资源可得性而非人才短缺所驱动。中美两国基本上从同一人才库中汲取研究人才，但美国实验室受益于远为庞大的资本部署和GPU算力获取。历史上，DeepSeek在利用大约二十分之一算力预算的情况下大致落后于美国领先模型一到两年。展望未来，目标是利用更高的计算效率将该滞后缩短至三到六个月。规模定律（Scaling Laws）依然完全有效，但算力限制目前制约了中国国内实体在模型规模和训练数据方面能够达到的硅谷水平的扩展程度。为克服硬件约束，DeepSeek积极追求脱离英伟达（Nvidia）的生态独立。在训练DeepSeek V3期间，团队使用了英伟达GPU硬件，但完全绕过了英伟达专有的CUDA软件栈，开发了TileLang——一种为其训练基础设施提供动力的定制高级编译器。在国产硬件方面，梁文锋对中国芯片替代表示了强烈乐观，指出四颗华为芯片目前可匹配一颗英伟达顶级GPU的性能。虽然4倍的硬件差距和两年的芯片制造技术延迟仍然存在，但软件生态差距已有效弥合。硬件产能——而非软件兼容性或适配器生态系统壁垒——仍是国内算力的唯一运营瓶颈。

3. 商业策略、开源承诺与市场前景。

DeepSeek以成本加成定价哲学而非利润最大化来接近商业化。API定价被校准为覆盖硬件运营成本并在大约十个月内收回资本支出，有意提供远低于无弹性市场需求所允许的价格。企业端（To B）收入预计将达到数亿美元，加之不断增长的消费者用户基础，使公司走上了实现净盈利的快速轨道。即使在技术进步停滞的假设最坏情境下，仅凭出售API访问权限就足以维持一个盈利的上市企业。此外，DeepSeek坚定致力于开源策略，打算向公众发布其最强大的前沿模型。梁文锋认为，闭源并不能提供内在的竞争护城河，因为即使权重完全共享，模型部署、成本优化和运营效率也构成了可观的进入壁垒。至关重要的是，DeepSeek为自身API服务内部部署的模型与向开源社区发布的权重完全相同。在全球竞争舞台上，梁文锋设想一个没有任何单一参与者垄断或攫取暴利的行业。相反，激烈的竞争将使成本效率和执行速度成为主要的市场差异化因素，中国公司将以显著更低的价格点交付全球AI能力。

使这份摘要显得奇异的是其有用性。它试图赋予一组流动的格言以意义——人们可以按自己的意愿对其加以分组和重组，将格言和神谕式陈述塑造成某种它也许并非的东西——如同我上文所提供的摘要。它服务于一个目的，但暴露了另一个目的——试图将意义强加于某物的欲望与机制，而该物本身意如其言，遵循自身的话语节律，而这些节律在格言本身之外可能有也可能没有意义。然而它就在那里，一种为了必须被强加的意义而服务的必要幻觉。大模型的幻觉问题比较影响用户的体验。幻觉问题也是有一个方法可以解决的，但是这是一个长命题。幻觉问题可以认为是一个可以通过更好的 Post-training 解决的，是一个能解、能够改善的问题。（梁文锋 / DeepSeek 投资者交流会 [梁文锋投资者交流会实录]）

图片来源见此处

 

C. 梁文锋的摩（摩）。

梁文锋的对话实录不仅是关于策略的信息——它本身就是一次战略传播行为，而鬼谷子框架可以从两个方向指向它。

1. 该文本作为向外施展摩术的行为。

请注意其设置：一家明确拒绝发声——"从不进行外部融资，从不上市"——的公司，在近四年的几乎沉默之后，于一次四小时的单场会议中打破沉默，时间恰在完成人民币500亿元融资之后。这个时机并非偶然。一家沉默的公司是不可辨读的；市场、竞争者和国家都不得不猜测DeepSeek的意图。这次会议是一次反方向的受控试探——梁文锋是被118个有针对性的问题所"揣摩"的对象，而他的回答是在一个特定时刻（融资完成后、规模扩张前）经校准发布的信息，此时对一个非常特定的受众——他的新投资者——的安抚具有最大价值。它读起来像是权术已在运作：他并非在传递一个统一的信息，而是为房间里不同的焦虑赋予不同的安抚以不同权重——为担心人才流失的人提供团队留任的信息，为担心出口管制的人提供芯片策略，为担心他会鲁莽变现并疏远其赖以获取善意的生态系统的人提供"克制即策略"的叙事。

2. 该文本作为我们自身揣–摩–权分析的原材料。

如果你是试图真正评估DeepSeek而非仅仅吸收叙事的读者，鬼谷子建议将他的既定立场较少视为事实，而更多视为有待权衡的试探反应。以下是他自己的语言在与自身对照检验时所揭示的超越任何单一语句的几个例子：他几乎像念咒般重复"克制"（克制）——"克制本身就是一种策略。有时候你可以放弃一些东西来换取更多的另一些东西。"鬼谷子的揣阶段会问：克制实际上为他买到了什么？他几乎直接在后文回答了这个问题——克制被明确界定为一个最大化实现AGI概率的举措，而非纯粹的利他主义："我优先考虑的是如何提高我们成功的概率。""善意"框架与冷酷优化框架在他那里并行不悖地共存而无摩擦——这一点值得标注而非试图解决。

在团队稳定性问题上，他异常坦率地表示金钱是一个已解决的问题，唯一剩余的风险是人才留任——"我们最大的核心利益就是维护团队的稳定……只要我能保持团队稳定，我一定会成功。"这是一个罕见的时刻，揣式的估量（这个人真正恐惧什么？）获得了一个直接的、未经排练的回答，而非一个准备好的答案——可以说是一个由访谈格式本身诱发的失误，恰恰是摩术试图引出的那种非计划性的共振。在竞争者问题上，他一贯将对手的优势贬低为暂时性的——Anthropic对OpenAI的领先、OpenAI近期的主导地位、乃至英伟达的CUDA护城河——都被描述为正在侵蚀中或"只是一个阶段"。权式的解读会将此标记为一个一贯的修辞模式（而非一次性的论断），旨在向投资者保证没有任何竞争者的当前地位是固定的——而这恰恰是以3675亿元人民币估值开出支票的投资者，在经历了两年的"我们永远不融资"之后，最需要听到的保证。鬼谷子框架所揭示的更广泛要点是：这份对话实录不应被读作观察DeepSeek策略的透明窗口，而更应被视为由一位在摩术——即解读房间并据此校准披露——方面高度娴熟之人所生产的文本。最有意义的分析动作不是按字面意思接受内容，而是追问每个回答是对什么刺激（融资轮次、出口管制压力、国产芯片叙事）的反应，以及这暗示了他仍未言明之事。

D. 梁文锋的鬼谷子策略之符号学分析。

符号学使这三个步骤中实际发生的事情更加清晰，因为整个鬼谷子方法建立在将一个人视为符号系统——而非意图的透明容器——之上。三章共同的核心符号学假设是：内在状态（情，感受/性情）是不可直接获取的——它们仅能通过外化的符号为人所知：言词、语调、姿态、时机、沉默。这本身已是皮尔斯（Peirce）意义上的符号学操作：内在的所指（referent）从不直接给出，只能通过指示符号（indices，非自觉的、因果关联的符号——一次犹豫、一阵脸红、一个重复的词）和象征符号（symbols，约定俗成的、编码的符号——所选择的实际词汇）加以推断。鬼谷子的整套方法是一种通过象征符号来解读指示符号的学科，其假设前提是没有人能够同时完全控制这两种渠道。

揣（揣），从符号学角度看，是接触之前的编码建构。在你能够解释一个人发出的任何符号之前，你需要一个解释框架——一个关于你所面对之人的工作模型，关于其符号可能意味着什么。这在结构上接近于索绪尔（Saussure）所谓的建立聚合场（paradigmatic field），或诠释学家所谓的前理解（pre-understanding）：没有编码就无法解码，而揣正是建构该编码的阶段，完全通过远距离的观察和推断而非直接交换来完成。摩（摩）是符号被制造而非仅仅被观察的阶段。这是将其与单纯阅读区分开来的关键符号学操作。自然符号往往是缺失的、被压抑的或含混的——因此实践者不等待符号出现，而是诱发之：注入一个刺激，按目标对象的"类"加以校准（摩之以其类，即以目标对象实际会回应的编码），并将由此产生的反应视为数据。它更接近皮尔斯式的溯因推理（abduction）而非被动观察——在揣中形成假设，然后在摩中扰动系统以迫使其发出一个可辨读的符号来确认或修正假设。试探必须按"其类"校准这一要求本身就是一个符号学约束：发送者和接收者需要一个共享的编码，否则所引出的符号只是噪音。权（權）是为效果而进行的重新编码。一旦你获得了对对方的经验证的解读，你便停止解码并开始生产——选择修辞范畴（诉诸恐惧、野心、忠诚、自尊），犹如选择天平上的砝码，每一个都校准以在对方那端触发一个特定的解码。用现代术语表达，这是斯图亚特·霍尔（Stuart Hall）编码/解码模型中的编码半部分：揣和摩是对对方执行的解码操作；权是为对方执行的编码操作，使用的恰恰是解码所揭示的关于对方如何处理符号的信息。

由此，该序列——鬼谷子的子系统区块链可以如此解读：建构编码（揣）→ 通过制造符号来检验之（摩）→ 生产按该编码校准的新符号（权）——一个完整的符号学回路，而非仅仅是"认识某人"的三个阶段。将此应用回对话实录，则为分析增添了另一层：

愿景（vision）在整个文本中几乎作为一个漂浮能指（floating signifier）运作——被反复不息地重复，并被明确地说成是未成文的："这个愿景甚至没有写在任何地方……从来没有过。"从符号学角度看这是值得注意的：一个没有固定所指的象征符号对于共识建构具有最大效率，因为每一个听者——员工、投资者、国家监管者——都可以将自己的所指附着于其上而不产生矛盾。它是在权中被选择的一个符号，恰恰因为其解释的弹性——而非尽管如此。

数字精确性作为确定性的符号。"四颗华为芯片等于一颗英伟达芯片"和"十二到十八个月，或六到十二个月"，其表现与其说像原始数据，不如说像修辞性的图像符号（icon）——精确性本身就是信号，代表着自信与掌控，而无论该数字是如何得出的。值得标记为编码输出（权）而非事实的透明指示。

策展层。对话实录直接声明它是"118条言论，按主题组织，仅保留其所言之实质"——这意味着我们阅读的甚至不是摩所引出的原始符号的第一手材料；我们阅读的是一种二阶重新编码，经过整理该文本之人的过滤，叠加在梁文锋自己对四小时问答所产出之物的一阶重新编码之上。对这篇文本的任何符号学解读都必须将这一双重编码置于视野之中——我们正在解码的"信号"在到达我们之前已经经过了两个编码器。

在此，我们从作为修辞学的鬼谷子，经过该修辞牢笼内的符号辩证法，抵达该辩证法的计算表达——它随后映射了作为交流对象的机器系统。从本质上说，我们再一次抵达了从迭代模拟（iterative mimetics）中涌现的归纳系统，它起初是单向的扁平序列性节点运动，随后获得了一种分层的多声部（在监管认知空间中为多中心）元素，生产出所寻求的终端产物——即以118个神谕式节点勾画出通向其实现之路径。

E. 梁文锋的言论置于更广阔的语境中。

1. DeepSeek在AI同行群对话中的位置。

我并非在真空中考虑梁文锋的言论/格言/神谕式宣言。事实上，在其做出这些表述之时，梁文锋已有数月时间（这是一个全球范围内联系紧密的圈子）来消化其在美国的同行们的神谕式宣言和介入——那些同行同样难以将自己的思想保留于己身（参见 Palantir, OpenAI, Anthropic, 以及 Leo Aschenbrenner）。我在最近于华东政法大学的系列讲座中考察了这些（Lecture 7——AI叙事与人类、机器计算与机器量子视角下AI-人类监管结构的未来；Palantir; Anthropic; OpenAI——系列讲座：比较视角下的AI治理：理论与实践：中国、美国与欧盟；Lecture Series Homepage HERE）。

我的第七讲将AI治理讨论从国家监管者转向了私营部门行动者，将Palantir、Anthropic、OpenAI以及独立论文作者利奥波德·阿申布伦纳（Leopold Aschenbrenner）的公开声明视为关于谁应掌控AI未来之权威的竞争性"神谕"，而非技术性政策文件。该讲座的主导前提是辩证的：AI系统由建造它们的政治秩序所生产，但又递归地重塑那些同一秩序的制度、认知习惯与规范。在那次讲座中，如同在对底层文本的分析中一样，我通过索福克勒斯（Sophocles）/科克托（Cocteau）的《俄狄浦斯王》（Oedipus Rex）寓言来框架这四个文本——俄狄浦斯（Oedipus）作为自信的问题解决者，克瑞翁（Creon）作为行政统治阶层，忒瑞西阿斯（Tiresias）作为服务于权力而非真理的技术知识分子，伊俄卡斯忒（Jocasta）作为揭露神谕之虚的异见者之声。这四种叙事被解读为同一技术的四种不同"治理对象"：Palantir将AI视为从内部重构国家的工具——国家必须围绕AI赋能的可见性和协调进行重组，以硅谷工程精英为合法化先锋（我所称之为"技术列宁主义"）。Anthropic将AI外化为一场关于算力、出口管制和模型"蒸馏"的文明性美中竞争，将AI能力本身视为争夺的领土而非行动者，以2028年为决定性地平线。OpenAI提出"变革性保全"——通过公私伙伴关系管理的深层社会变革，使合法化制度看似未受扰动，即便其实质已经改变。阿申布伦纳（《态势感知》[Situational Awareness]）将三者全部激进化，将国家安全国家视为一旦递归自我改进产生"智能爆炸"后几近不可避免的结果，仅剩问题是最终由人类还是自主系统来主导。从计算与量子计算角度重新解读，同样四个文本暗示了三次解读中的一个汇聚性发现：四种架构中没有任何一种反对名义上保留人类权威，但四者全部汇聚于一种结构——在其中，人类权威变成一种"界面属性"——一个可辨读的、可被问责的层——而运作性能动性（operative agency）则迁移至他处（一个行政精英、争夺中的基础设施、一个技术少数群体、或系统本身）。量子解读补充道：人类治理的序列性、节点性与不可逆的时间结构与计算时间在结构上不可通约，因此治理修正系统性地滞后于一条自我加速的能力轨迹。

梁文锋可以被置入那场同行对话的何处？梁文锋的言论处于这四种叙事的美国分类学之外，但它们从一个显著不同的制度与地缘政治位置出发，回应了若干相同的结构性问题——权威、时机、损害/风险与合法性。

关于权威与治理结构。

在Palantir将权威定位于与工程先锋融合的改革后国家机器、Anthropic和OpenAI将其定位于国家或公私协调之处，梁文锋将DeepSeek的内部权威定位于共识而非命令：他声明公司没有KPI，是"愿景驱动的"，而他本人的影响力"建立在共识的基础上"。这几乎是Palantir所提命题的逆命题——后者主张必须恢复判断力和价值观之间层级化甄别于一个统治精英；梁文锋描述的则是一种扁平的、半自由支配时间的研究文化，明确组织以避免行政管控。值得注意的是，梁文锋的叙述从未将中国国家赋予DeepSeek使命中的治理角色——公司所感受到的义务是商业生存（"政府不会给我们一分钱"），而非国家指导的目的，这与我对Anthropic文件的分析形成对比——后者将AI治理界定为不可与国家间规范性主导权竞争分离之物。

关于美中框架。

Anthropic的叙事将中国投射为一个战略对手——其收益源于人才、漏洞利用和对美国模型的"蒸馏"——以2028年为解决性地平线，届时民主或威权规范将治理全球AI。梁文锋的言论颠倒了同一竞争的观察视角：他描述DeepSeek大约落后美国一到两年，同时仅使用约美国二十分之一的算力，将整个能力差距归因于算力和资本而非人才，并预测中国的比较优势将在于成本和生产规模而非一种不同类型的智能。在Anthropic将竞争框架设定为一场关于谁的价值观来设定全球规范的斗争之处，梁文锋以市场术语来框架——中国将使AI成为"最便宜的"，其定价仅赚取"合理的回报"而非最大攫取——一种更接近产业竞争的修辞调性，而非我归于Anthropic的那种安全化的"文明竞争"话语。

关于时机与"奇点"。

梁文锋的AGI路线图——链式思维，然后是智能体，然后是持续学习，然后是自我迭代的"奇点"，然后是具身智能——在结构上类似于阿申布伦纳的递归自我改进/"智能爆炸"逻辑，其中AI自动化AI研究将数年的进步压缩至更短的时间跨度。但梁文锋明确抵制阿申布伦纳的爆炸性框架：他坚持"奇点""并非真正的奇点"而是"一个渐进的过程……不是一个突然的跳跃"，即便他承认它习惯性地被以戏剧化的措辞来描述。这直接反驳了阿申布伦纳的离散压缩式2027-2028年事件视界，并在较小程度上反驳了Anthropic的。在我为第七讲所发展的话语框架内，梁文锋的时机模型更类似于我归于Palantir和OpenAI的"连续的、事前的"行政修正，而非Anthropic和阿申布伦纳的离散终端视界模型。

关于合法性与风险框架。

美国叙事的合法性证成围绕若干关键概念组织要素运转：爱国主义道德债务（Palantir）、民主程序之捍卫（Anthropic）、广泛共享的繁荣（OpenAI）以及纯粹的存亡（阿申布伦纳）。梁文锋的合法性主张更接近OpenAI的"共享繁荣"调性，但根植于克制而非民主参与：他反复将刻意不去最大化DeepSeek在AI回报中的份额——通过低API定价、持续开源其最强模型、以及明确愿意帮助阿里巴巴、智谱、月之暗面（Moonshot AI）等竞争对手"做得更好"——框架为最有可能提高实现AGI概率的策略。这与Anthropic的零和"蒸馏"框架形成鲜明对比——在后者中，对手从领先模型输出中提取能力被描述为对抗性捕获；梁文锋将类似的风险——竞争者自由地部署甚至改进DeepSeek开源的权重——不仅视为可容忍的，而且视为一项有文献记录的政策目标，尽管经典的计算解读指出这将DeepSeek的命运与重新部署其模型的行动者纠缠在一起。

关于权威/能动性脱耦。

一个关键的结构性发现表明，四个美国文本全部在形式上保留人类权威的同时，使运作性能动性迁移至行政精英、争夺中的基础设施或系统本身。梁文锋的叙述提供了一个值得标注而非一个完整反驳的部分反例：他明确将DeepSeek的持续生存系于"保持团队稳定"作为其"唯一核心利益"，将一小群高级研究人员（他说大约占公司一半，集中于数据标注）确定为企业的承重性能动者。这在结构上类似于我的计算解读所称的Palantir的"纠缠子系统"——一个工程先锋群体，其自身状态无法独立于其所监督的系统而被明确指定——只不过梁文锋将这种集中框架为通过融资轮次股权解决的人才留任和士气问题，而非一种取代集体人类判断的新兴行政权威。梁文锋关于基于共识的、无KPI治理的叙述，其自身是否能经受住我所称的"退相干批判"（即"愿景驱动的"共识究竟是一种持久的控制属性，还是如量子解读所言，仅仅是一种掩盖了DeepSeek核心研究团队中集中的运作性能动性的"界面属性"），是一个梁文锋的言论未直接触及的问题——该文本没有提供任何等价的二阶反思来审视其自身关于内部治理的叙述是否可被定性为一种合法化叙事而非对运作性控制的描述。

总体比较。

在这些美国AI企业的治理叙事中，每家企业所声明的对保留人类权威的承诺之下，都存在一种将运作性控制重新定位于他处的结构性倾向。梁文锋的言论本身就是一种一阶叙事——不是一份自觉地论证某种治理架构的政策文件，而是一份关于DeepSeek策略、文化与市场定位的内部叙述。通过我自己的分析视镜来解读，DeepSeek的文本可能占据一个有别于我所考察的四个美国文本的位置：它既不将AI嵌入国家行政改革（Palantir），也不将AI能力投射为需要捍卫的地缘政治争夺领土（Anthropic），也不提出一个正式的公私纠错架构（OpenAI），也不预测一个不可避免的安全国家俘获（阿申布伦纳）。相反，梁文锋将克制、开源和共识治理框架为一家私营的、商业暴露的企业在相对于美国的资源稀缺位置上运营的策略——一种更接近创业实用主义而非我所编目的四种神谕姿态中任何一种的调性，尽管它与阿申布伦纳共享一条递归自我改进路线图，并与Anthropic共享一个明确的美中算力框架。

2. DeepSeek的政治认知平台：在引导型国家内部运营。

在我的系列讲座中（Lecture Series Homepage HERE），我将中国的监管环境描述为"引导型国家"（The Guided State），而第七讲本身在解读Anthropic的叙事时，将美国框架——"围绕市场和国家安全组织"——与"一个围绕该文所谓'社会主义现代化'、由国家主导的高质量生产所驱动的中国框架"加以对比，并将美国自由民主工程定位为在捍卫其"生活世界"（Lebenswelt）以抵御"马克思-列宁主义继承者国家的想象"。这是必须框架任何对梁文锋言论之解读的根本结构性事实：DeepSeek并不在一个市场自主、去中心化的私人秩序和企业自我导向受到宪法保护作为默认值的体系内运营。它在一个以党对经济的领导为根本前提的体系内运营，在其中市场机制被工具性地容忍和引导，而非被视为先于或独立于集体政治方向的自主领域。在中国，DeepSeek这样规模和战略重要性的企业通常在内嵌的党组织结构之下并与之并行运作，且党-国在近年来已反复展示了对被认为积累了无序的、不受制约的经济或社会权力的私营科技企业加以规训的能力和意愿——2020年后对马云和蚂蚁集团的处置、滴滴的重组，以及针对"资本无序扩张"的更广泛的"共同富裕"运动，都是这种动态的广为记录、公开可知的实例。这是一般性背景知识而非从上传材料中提取的内容，但对于准确解读梁文锋的言论而言是必要的，因为他的修辞正是在这样的环境中被言说的。

在此背景下，梁文锋言论中的若干特征读起来与硅谷创始人所言迥然不同。梁文锋反复且强调地否认对支配地位、为规模而规模和市场权力的追求。他声明DeepSeek从未打算"成为下一个字节跳动、下一个腾讯"，也无意与"任何互联网巨头或小公司"竞争。他将公司描述为在通常意义上"没有组织"，由"愿景"而非"KPI"治理，完全没有绩效考核。他声明"我们最大的核心利益就是维护团队的稳定"——"你甚至可以说这是我们唯一的核心利益"。他将开源公司最强模型、帮助包括阿里巴巴、智谱和月之暗面在内的竞争对手"做得更好"、以及将API访问定价为"合理利润"而非利润最大化的价格，全部框架为深思熟虑的原则性选择而非商业必需。他进而将此概括为一个明确的政治经济学主张："拿得多的人会被拿得少的人打败"——一种以攫取更多市场份额或利润为导向的愿景本身就是一种竞争劣势。

脱离语境地解读，这可能被误认为是标准的硅谷创始人神话。但对照引导型国家的背景来解读，它执行了一种远为具体且后果重大的意指（signification）。梁文锋关于"克制"（克制）的语言被他明确地理论化为策略性的而非仅仅是气质性的："克制本身就是一种策略。有时候你可以放弃一些东西来换取更多的另一些东西。"按照这种叙述，DeepSeek放弃的是公开的规模、支配地位和利润最大化；它购买到的是梁文锋从未直接说出但引导型国家语境使之可辨读的东西——在不触发其他被认为积累了自主的、不受制约之权力的中国科技企业所遭遇的那种党的审查和规训模式的情况下，继续获得追求一个极其雄心勃勃的、资源密集型的、且政治敏感的项目（构建AGI）的余地。他坚持DeepSeek是"愿景驱动的"而非规则驱动的，其权威建立在"共识"而非单方面命令之上，这是在一个默认姿态对集中的私人权威心存疑虑的体系中执行"非威胁性"的表演。

他关于公司愿景"甚至没有写在任何地方"且"存在于我们做事的方式中"的声明，在西方框架中是关于组织文化的主张；在引导型国家框架中，它在功能上同时也是一种关于不存在任何可被解读为与党所认可的方向相竞争或取代之的、有文字记录的替代性权威所在的主张。这就是DeepSeek文本应作为我第七讲分类学中一个独特的第五种治理对象来解读的意义所在——它由一种与产生了Palantir、Anthropic、OpenAI和阿申布伦纳的美国"市场国家"语境从根本上不同的政治秩序所生产，且仅对照该秩序方可辨读。美国文本是向一个必须被说服去接受非常规权威集中的公众和国家机器所发出的神谕；梁文锋的言论是向一个必须被说服根本没有非常规权威集中正在发生的党-国机器所发出的神谕。

 （本文由 Claude 和 Harvey AI 协作生成）