生成式人工智能的法律定位與分層治理

The Legal Positioning and Hierarchical Governance of Generative AI

生成式人工智能改變了數字社會的生產架構，向既有的技術治理體系提出挑戰。一是生成式人工智能呈現“基礎模型—專業模型—服務應用” 的分層業態，無法在現有的“技術支持者—服務提供者—內容生產者” 監管框架中找到適配的法律定位;二是其傳播模式和技術指標使得原有規制工具難以適用。生成式人工智能治理應符合其功能業態的技術邏輯，也應基於其在數字社會生產的地位，重新認識和調整生成式人工智能的法律定位，將模型作為新型的法律治理對象。生成式人工智能的基礎模型因可接入千行百業而具有通用性，因同時提供信息內容與機器所需合成數據而具有強大賦能性，是人工智能時代的新型數字基礎設施。應構建“ 基礎模型—專業模型—服務應用” 的分層治理體系，在不同的層次適配不同的規制思路與工具。在基礎模型層以發展為導向，將其作為數字社會新型基礎設施設置法律制度;在專業模型層以審慎包容為理念，進行分級分類並設置合理避風港規則;在服務應用層實施敏捷治理，建立合理容錯制度。由此，我國得以從單一場景的算法治理發展為適應不同治理目標的複合型系統性治理。

Generative AI has changed the architecture of digital social production and challenged the established technology governance system. The first reason is that generative AI presents a hierarchical business model of “base model-professional model-service application”, which cannot find a suitable legal position in the existing regulatory framework of “technology supporter-service provider-content producer”. And the second reason is that its dissemination mode and technical indicators make it difficult to apply the original regulatory tools. The governance of generative AI should be in line with the technical logic of its functional mode of operation and based on its status in the production of digital society, the legal positioning of generative AI should be reconceptualized and adjusted, and the model should be treated as a new type of legal governance object. The base model of generative AI is a new type of digital infrastructure in the era of AI, based on the fact it empowers industries and provides both information content and synthetic data. A hierarchical governance system of “base model- professional model-service application” should be established, and different levels should be matched with different regulatory ideas and tools. In the base model layer, the development-oriented legal system should be set up as a new type of infrastructure for digital society; in the professional model layer, the concept of prudence and inclusion should be adopted to classify and set up reasonable safe harbor rules; in the service application layer, agile governance should be implemented, and a reasonable fault tolerance system should be established, which can help evolve our country from algorithmic governance of a relatively single scenario to a composite systemic governance of generative AI adapted to different governance objectives.