Abstract:

The rapid spread of generative artificial intelligence is reshaping how content is produced and disseminated on social media platforms, intensifying challenges such as cross-platform cascading diffusion of misinformation, rising moderation costs, and blurred accountability chains. Current practice largely relies on linear “human + algorithm” governance and minimal compliance. Addressing gaps in existing research—namely the lack of mapping between technical architectures and governance tiers, overly static collaboration mechanisms, and the difficulty of parameterizing and transparently assessing the safety-efficiency-fairness trade-offs—this paper proposes an integrated, four-tier collaborative governance framework based on the co-construction of “technical architecture-governance level.” At its core, the framework establishes a clear correspondence of “who is responsible—what is governed—how it is governed—how it is evaluated,” linking technical layers (data, models, content distribution) with governance tiers (institutional/regulatory, industry, platform, societal). It is supplemented by a bidirectional governance loop (top-down rules and bottom-up risk signals) and cross-platform coordination mechanisms, translating abstract governance principles and values into measurable, adjustable, and auditable metrics and weighting schemes. This enables multi-stakeholder collaboration to be operationalized as executable and assessable meso-level designs, bridging gaps between regulations, standards, and platform operations; supporting agile transitions between routine and crisis contexts; shortening response latency; curbing cascading spread; and improving exposure fairness and external accountability at a manageable efficiency cost. The framework provides a theoretical foundation and a systematic analytical toolkit for platform governance in the generative AI era.

Key words: generative artificial intelligence, social media, multi-stakeholder collaborative governance, technical architecture, platform governance

摘要：

生成式人工智能的迅速普及重塑了社交媒体平台的内容生产与传播方式,使虚假信息跨平台级联扩散、审核成本攀升和责任链模糊等问题愈发突出。现有实践主要依赖“人工+算法”的线性治理与最低合规执行。针对现有研究中技术架构与治理层级映射缺位、协同机制偏静态、安全—效率—公平的价值权衡难以参数化和公开评估难等空白,本文提出“技术架构—治理层级”互构整合的四级协同治理框架,本框架的核心是建立起一套清晰的“谁负责—管什么—如何管—怎样评”的对应关系,把技术架构(如数据、模型、内容分发)与治理层级(制度、行业、平台、社会)打通;在此基础上辅之以双向治理闭环(规则下行/风险上行)和跨平台协同的治理机制,将抽象的治理理念和价值观转译成为可测量、可调整与可审计的评价指标与权重设置。进而,能够将多元协同治理落地为可执行、可评估的中层设计,弥合法规、标准与平台运营之间的落差,在常态与危机情境中实现敏捷切换,缩短响应时滞,抑制级联扩散,以可控的效率代价改善曝光公平与外部问责。为生成式AI时代社交平台治理提供了学理基础和系统化分析工具。

关键词: 生成式人工智能, 社交媒体, 多元协同治理, 技术架构, 平台治理