从算法霸权到智能公域：全球人工智能治理的范式重构与中国方案

From Algorithm Hegemony to the Intelligent Commons: Paradigm Reconstruction of Global AI Governance and the China Approach

“技术是中立的，但技术的使用从来不是中立的。”

“Technology is neutral, but the use of technology never is.”

引言：AI作为“新型全球公共事务”

Introduction: AI as a “New Global Public Affair”

人工智能（Artificial Intelligence, AI）已从单纯的算法工具演变为重塑国际权力结构、经济分工与社会伦理的颠覆性通用技术（Disruptive General-Purpose Technology）。与气候变化、公共卫生一样，AI的跨境流动性、系统性风险与指数级演进特征，使其成为典型的全球公共事务（Global Public Affair）。然而，当前的全球AI治理正处于“原则过剩、规则割裂、执行乏力”的阵痛期：技术领先国谋求“算法霸权（Algorithm Hegemony）”与标准输出，发展中国家面临“智能鸿沟（Intelligence Divide）”的进一步拉大，而联合国主导的多边协调机制仍在艰难成型。

Artificial Intelligence has evolved from a mere algorithmic tool into a disruptive general-purpose technology​ reshaping international power structures, economic divisions, and social ethics. Like climate change and public health, AI’s cross-border mobility, systemic risks, and exponential evolution make it a quintessential global public affair. Yet, current global AI governance is trapped in the painful phase of “principle surplus, rule fragmentation, and weak enforcement”: technologically leading states pursue “algorithm hegemony” and standard export; developing countries face a widening “intelligence divide”; and UN-led multilateral coordination mechanisms are still struggling to take shape.

本文立足于2025–2026年的最新地缘科技（Geotech）动态——包括联合国“人工智能独立国际科学小组”的设立、欧盟《人工智能法案》的落地调整、中国《全球人工智能治理倡议》的深化以及“世界人工智能合作组织（WAICO）”的成立——从比较政治与全球治理的智库视角，剖析中美欧三大治理模式的底层逻辑，探讨全球AI治理从“碎片化博弈”走向“智能公域（Intelligent Commons）”的范式重构路径。

Based on the latest 2025–2026​ geotech dynamics—including the establishment of the UN Independent International Scientific Panel on AI, the implementation adjustments of the EU AI Act, the deepening of China’s Global AI Governance Initiative, and the founding of the World Artificial Intelligence Cooperation Organization (WAICO)—this paper analyzes the underlying logics of the three major governance models (China, US, EU) from the perspective of comparative politics and global governance, exploring the paradigm shift from “fragmented gaming” to the “Intelligent Commons.”

一、全球AI治理的三大模式：价值取向与制度分野

I. Three Models of Global AI Governance: Value Orientation & Institutional Bifurcation

学术界通常将当前全球AI治理格局概括为“三极博弈”。尽管各国均在谈论“以人为本（Human-centric）”与“可信AI（Trustworthy AI）”，但其战略优先级与规制工具存在根本性差异。我们可以借用比较框架将其归纳为三类理想类型：

The global AI governance landscape is often summarized as a “tripolar game.”​ Despite widespread rhetoric on “Human-centric”​ and “Trustworthy AI,”​ there are fundamental differences in strategic priorities and regulatory tools. We categorize them into three ideal types:

1. 美国：守夜人看护与安全本位（US: Guardian Supervision & Security-First）

美国的AI治理呈现“内松外紧”的二元结构。在国内，联邦层面回避系统性立法，依赖行业自律（Industry Self-Regulation）、事后诉讼与州级碎片化立法，以维持硅谷的创新活力与全球竞争力，体现“守夜人看护”的轻触式监管（Light-touch Regulation）；在对外，则以国家安全（National Security）为绝对优先，通过《芯片与科学法案》、出口管制与“小院高墙”策略，谋求技术脱钩与联盟封锁，将AI治理工具化服务于地缘霸权。

The US exhibits a “loose inside, tight outside”​ dual structure. Domestically, it avoids systemic federal legislation, relying on industry self-regulation, ex-post litigation, and fragmented state laws to maintain Silicon Valley’s innovation vitality—embodying “light-touch regulation”​ as a guardian. Externally, it prioritizes national security, using the CHIPS and Science Act, export controls, and “small yard, high fence” strategies to decouple technology and build alliances, instrumentalizing AI governance for geopolitical hegemony.

• 核心逻辑：创新优先（Innovation First），安全通过技术主导权实现。

• Core Logic: Innovation first; security achieved via technological dominance.

• 典型举措：NIST AI Risk Management Framework（非强制）；组建“广岛AI进程（Hiroshima AI Process）”等技术联盟。

• Typical Measures: Voluntary NIST AI RMF; “Hiroshima AI Process” tech alliances.

2. 欧盟：家长式防控与规范型权力（EU: Paternalistic Prevention & Normative Power）

欧盟依托《人工智能法案（AI Act）》建立了全球首个横向综合性AI立法，采用风险分级（Risk-based Classification）逻辑（不可接受风险、高风险、有限风险、最小风险），对生成式AI与大模型施加透明度义务。这被称为“家长式防控（Paternal Prevention）”——通过强规制确立“布鲁塞尔效应（Brussels Effect）”，以统一标准外溢至全球，保障基本权利与民主价值，但客观上增加了合规成本，引发“过度监管抑制创新”的担忧。2025年欧盟理事会已提议简化部分规则以减轻企业负担，显示其在权利与产业间摇摆。

The EU’s AI Act​ establishes the world’s first horizontal comprehensive AI legislation, adopting a risk-based classification​ (unacceptable, high, limited, minimal risk) and imposing transparency obligations on GPAI models. This “paternalistic prevention”​ leverages the “Brussels Effect”​ to project normative power globally, safeguarding rights at the potential cost of stifling innovation via compliance burdens. In 2025, the EU Council proposed simplifying some rules to ease industry burden, signaling a wobble between rights and competitiveness.

• 核心逻辑：风险规避（Risk Aversion），权利保障优于技术扩张。

• Core Logic: Risk aversion; rights protection prioritized over tech expansion.

• 典型举措：GDPR延伸；AI法案禁令清单（如社会评分、实时远程生物识别限制）。

• Typical Measures: GDPR extension; prohibited AI practices list (social scoring, real-time remote biometric ID limits).

3. 中国：伴侣型监管与发展安全平衡（China: Companion Regulation & Balanced Development-Security）

中国走的是“有序创新（Ordered Innovation）”之路，被学界定义为“伴侣型监管（Companion Regulation）”——政府在战略引导、顶层设计（如《新一代人工智能发展规划》）与底线管控（算法推荐管理规定、生成式AI服务管理办法）间寻求动态平衡，强调“伦理先行、软硬结合”。不同于纯粹的市场放任或立法前置，中国通过监管沙盒（Regulatory Sandbox）、标准体系建设与央地协同，实现政府与企业的深度协作。

China pursues “ordered innovation,”​ defined as “companion regulation”—seeking dynamic balance between strategic guidance, top-level design (Next Gen AI Development Plan), and bottom-line control (Algorithm Recommendation Provisions, Generative AI Service Measures). It emphasizes “ethics-first, combining soft and hard law”​ via regulatory sandboxes, standard systems, and state-local synergy for deep government-enterprise collaboration.

• 核心逻辑：统筹发展与安全（Balance Development & Security），普惠包容。

• Core Logic: Balancing development and security; inclusive and beneficial to all.

• 典型举措：《全球人工智能治理倡议》（2023）、《人工智能全球治理行动计划》（2025）、成立世界人工智能合作组织（WAICO）（2026）。

• Typical Measures: Global AI Governance Initiative (2023); Action Plan for Global AI Governance (2025); founding WAICO​ (2026).

二、治理困境：碎片化、智能鸿沟与伦理异化

II. Governance Dilemmas: Fragmentation, Intelligence Divide & Ethical Alienation

尽管多边机制频出，全球AI治理仍面临三重结构性张力：

Despite frequent multilateral initiatives, global AI governance faces three structural tensions:

1. 规则碎片化与地缘割裂（Regulatory Fragmentation & Geopolitical Fission）

中美欧的路径差异导致合规孤岛（Compliance Islands）。企业需同时满足欧盟的严苛透明度、美国的出口管制与中国的数据安全审查，极大抬升了全球AI供应链的交易成本。技术联盟（如G7 Hiroshima AI Process）与主权治理并行，加剧了“数字铁幕（Digital Iron Curtain）”的风险。

Divergent paths of China, US, and EU create “compliance islands.”​ Enterprises must navigate EU’s transparency, US export controls, and China’s data security reviews, raising supply chain costs. Tech alliances (e.g., G7 Hiroshima AI Process) vs. sovereign governance heighten “digital iron curtain”​ risks.

2. 智能鸿沟的代际放大（Intergenerational Amplification of the Intelligence Divide）

全球南方（Global South）在算力、数据、人才上的先天劣势，使其难以参与规则制定。联合国数据显示，绝大多数AI研发集中在少数发达国家。若治理话语权被垄断，“AI红利”将转化为“AI剥削”，复制并加剧南北不平等。中国在联合国推动的“人工智能能力建设国际合作决议”（140+国联署）正是试图填补此缺口。

The Global South​ lacks compute, data, and talent, excluding them from rule-making. UN data shows most R&D concentrates in few developed states. Without intervention, “AI dividends” turn into “AI exploitation,”​ replicating North-South inequality. China’s UN resolution on International Cooperation on AI Capacity-Building (140+ co-sponsors) aims to bridge this gap.

3. 科技价值异化与主体性危机（Alienation of Tech Values & Subjectivity Crisis）

从深度伪造（Deepfake）操控选举到自主武器（LAWS）的伦理红线模糊，AI正在挑战人类主体性（Human Agency）与最终决策权。西方叙事下的“伦理普世化”往往掩盖了特定意识形态的植入，而开源大模型的失控风险则使“算法黑箱（Algorithmic Black Box）”成为全球安全的达摩克利斯之剑。

From Deepfakes manipulating elections to blurred red lines on LAWS, AI challenges human agency​ and final decision-making authority. “Universal ethics” in Western narratives often mask ideological imprinting, while uncontrolled open-source models make the “algorithmic black box”​ a global security sword of Damocles.

三、多边主义进阶：联合国机制与“中国方案”的供给

III. Multilateralism Upgraded: UN Mechanisms & the Supply of “China’s Approach”

面对治理赤字，2025–2026年成为全球AI治理从“软倡议”转向“硬机制”的关键窗口：

Amid governance deficits, 2025–2026 is a critical window for global AI governance shifting from “soft initiatives”​ to “hard mechanisms”:

1. 联合国双轨机制的确立（UN Dual-Track Mechanisms）

2025年8月，联合国大会一致通过决议，设立“人工智能独立国际科学小组（Independent International Scientific Panel on AI）”与“人工智能治理全球对话（Global Dialogue on AI Governance）”。前者提供跨学科的风险与影响评估（科学背书），后者为193个成员国及利益相关方提供平等议事平台（政治磋商）。首届全球对话定于2026年7月在日内瓦举行，标志着全球AI治理在最具普遍性的多边框架下落地。

In Aug 2025, the UNGA unanimously established the Independent International Scientific Panel on AI​ (scientific assessment) and the Global Dialogue on AI Governance​ (political consultation for 193 members). The inaugural Dialogue in Geneva, July 2026​ marks the operationalization of AI governance under the universal multilateral framework.

2. 全球人工智能治理倡议与中国行动（Global AI Governance Initiative & China’s Actions）

中国于2023年提出的《全球人工智能治理倡议》确立了“以人为本、智能向善（Human-centric, AI for Good）”的宗旨。在此基础上，2025年WAIC发布的《人工智能全球治理行动计划》提出13项务实举措（风险测评、开源共享、绿色算力等）。2026年7月，世界人工智能合作组织（WAICO）在上海正式成立，29国签署协定，成为首个聚焦AI国际合作与政府间共治的常设机构，体现了中国推动治理从“西方主导”转向“共商共建共享”的努力。

China’s Global AI Governance Initiative​ (2023) set the “human-centric, AI for good” tenet. The Action Plan for Global AI Governance​ (2025 WAIC) proposed 13 measures (risk assessment, open-source sharing, green compute). In July 2026, the World Artificial Intelligence Cooperation Organization (WAICO)​ was founded in Shanghai with 29 founding members, marking a shift from “Western-led” to “co-constructed” governance.

核心贡献在于：中国方案强调发展权（Right to Development）与主权平等（Sovereign Equality），反对将AI作为遏制他国发展的工具，倡导将AI打造为全球公共品（Global Public Good），而非少数巨头的专利。

Its core contribution: The China approach emphasizes the Right to Development​ and Sovereign Equality, opposing AI as a tool for containment, and advocates AI as a Global Public Good, not a monopoly of a few tech giants.

四、范式重构：走向“智能公域”的未来图景

IV. Paradigm Reconstruction: Toward the “Intelligent Commons”

未来的全球AI治理不应是霸权的延伸，而应构建“智能公域（Intelligent Commons）”——即在承认国家主权与数据边界的前提下，对算法模型、算力基础设施与安全基准进行开放性共享与协同治理。

Future global AI governance should not extend hegemony but build an “Intelligent Commons”—open sharing and collaborative governance of algorithms, compute infrastructure, and safety benchmarks, while respecting national sovereignty and data boundaries.

• 技术层：推动开源大模型（Open-source LLMs）的安全互信与互操作（Interoperability），避免生态割裂。

• At the technical level: Promote security, trust, and interoperability​ in open-source LLMs​ to avoid ecosystem fragmentation.

• 制度层：在联合国框架下融合“硬法（Hard Law）”（如致命自主武器禁令）与“软法（Soft Law）”（伦理准则、标准互认），形成梯度化约束。

• At the institutional level: Under the UN, blend hard law​ (e.g., LAWS ban) and soft law​ (ethical codes, standard mutual recognition) for graded constraints.

• 价值层：超越西方中心主义的“权利话语”，纳入全球南方的发展诉求与人类命运共同体的共生逻辑。

• At the value level: Move beyond Western-centric “rights discourse” to include Global South’s development demands​ and the community-with-shared-future symbiosis.

结语：治理即未来

Conclusion: Governance is the Future

人工智能的终点不是算法的无敌，而是人类文明的有序延展。在2026年这个节点回望，从赵云“一身都是胆”的宿卫之责，到今日各国对AI“安全可控”的底线坚守，本质都是对力量与责任关系的拷问。中国以“伴侣型监管”平衡创新与安全，以“世界人工智能合作组织”搭建共商平台，正是在回答这一千古之问：技术何以载道？答案在于——唯有将算力置于人类共同价值的驾驭之下，算法才不会沦为利维坦，而是化作照亮普惠未来的银鞍白马。

The endpoint of AI is not algorithmic invincibility but the orderly extension of human civilization. Looking back from 2026, from Zhao Yun’s “courage in every fiber” as a loyal guardian to today’s global insistence on “safe and controllable AI,”​ the essence remains the interrogation of power versus responsibility. China’s “companion regulation”​ balancing innovation and security, and the WAICO​ platform for shared deliberation, answer an eternal question: How can technology carry the Dao (Way)? The answer lies in this—only when compute is harnessed by shared human values will algorithms cease to be a Leviathan and instead become the silver saddle and white steed illuminating an inclusive future.

主要参考与数据来源 Primary Sources：

• 陈寿《三国志》（引喻宿卫与责任）；

• 联合国大会决议 A/RES/79/325 (2025)；

• 欧盟理事会《人工智能法案》修订提案 (2025)；

• 中国《全球人工智能治理倡议》(2023)、《人工智能全球治理行动计划》(2025)；

• 世界人工智能合作组织（WAICO）成立协定 (2026)；

• 中国国际友好联络会、CIIS、清华AI国际治理研究院相关智库报告；

• UN Global Dialogue on AI Governance 官方文件。

继续阅读：算法霸权   全球   人工智能治理   中国   AI治理   AI   生成式AI   大模型

此文由 怡心湖 编辑，若您觉得有益，欢迎分享转发！：首页 > 观·世界 » 从算法霸权到智能公域：全球人工智能治理的范式重构与中国方案