| 關 鍵 詞： |
歐盟人工智慧法；AI 系統；提供者；通用 AI 模型；系統風險；布魯塞爾效應 |
| 中文摘要： |
2024 年 5 月 21 日歐盟理事會通過「人工智慧法」的最終文本，全球首部直接規範人工智慧系統與模型的法律於焉誕生。規範的要點包括：寬鬆定義「AI 系統」並強調系統的自主性與推論能力；適用的行為主體廣泛；按風險程度進行分級管理，包括應予禁止的 AI 系統實施方式、嚴密管制高風險 AI 系統、有限度風險 AI 系統與透明性義務；新增監管「通用 AI 模型」專章，以及在歐盟建立層級化監理與執行機制等。歐盟人工智慧法對未來全球經貿秩序的影響力，也就是發生布魯塞爾效應的程度，雖然受到美中競爭、G7 或聯合國合作倡議，以及國際標準制定組織等外在因素的影響，但預期歐盟規則仍持續會在與含內高風險 AI 產品的貿易、應用 AI 系統提供跨區服務的平臺、有限度風險 AI 系統的透明性義務規範，以及應用通用 AI 模型的服務貿易方面，持續發揮其法規領導力。我國應歡迎在歐盟立法倡議下，促使各國積極研議如何適當規範人工智慧的發展應用及相關標準的建立。我國目前不宜制定可能影響或限制 AI 相關產品或服務進出口的法規，但由於是以半導體相關產品占出口主要部分，故宜全面盤點最可能適用該法中有關「產品內 AI」規定的產品，以利廠商因應準備，且於立法時（無論是專法形式或以部分條文納入其他法律）有較為明確的方向與目的，避免條文流於空洞的政策宣示。
|
| 英文關鍵詞： |
EU Artificial Intelligence Act (AI Act)；AI System；Provider,；General-Purpose AI Model；Systemic Risk；Brussels Effect |
| 英文摘要： |
On 21 May 2024, the Council of the European Union approved the final text of the EU Artificial Intelligence Act (AI Act) and the first horizontal and comprehensive rules for artificial intelligence (AI) systems in the world has come to exist. Several takeaways of the newest version of the AI Act include: AI systems are broadly defined with a focus on their autonomy and inference abilities; the scope of subjects under compliance obligations encloses the participants across the whole AI value chain; a risk-based approach is adopted and AI is classified into different categories, prominently depending on their use cases, namely: a closed list of prohibited AI practices, dually defined high-risk AI systems, certain limited-risk AI systems and transparency obligations. A new chapter regulating general-purpose AI (GPAI) model (not “system”) is also added. Lastly, a layered compliance and enforcement structure is architected. In terms of the possible implications of the AI Act for global trade in the future, in other words, to what extent the so-called “Brussels effect” continues to be effective, it is expected that the exterritorial impacts of the AI Act could be seen at least in the cross-border trade relating to high-risk AI products, international platform services with built-in high-risk AI systems, as well as the transparency requirements, such as adding a disclosure note in those limited-risk AI services. We also project that the AI Act would have strong influences on the digital trade based on the GPAI model, including services provided by large language models (LLM). However, the EU law’s extraterritorial impact could be constrained due to some factors such as: strong global competition from U.S. and China, international cooperation initiatives on international forum, e.g., G7 and UN, and the mediation and compromises in the international standard setting organizations. Taiwan’s economy is characterized as export-oriented and the export of semiconductor-related products accounts for a significant percentage of our annual trade values, so we should welcome the EU’s engagement in regulating AI and facilitating the global governance in AI development and application. At this moment, we need not pursue a domestic AI regulation which might undermine the import or export of AI-related products and service. However, identifying our AI built-in products which are most relevance when applying AI Act is imperative. That would help our manufactures well prepared for adjustment and could set clearer purposes or goals when the procedure for AI legislation, either in the form of separate statute or articles incorporated into existing laws, has actually taken place.
|
| 目 次： |
壹、前言－新法誕生
貳、2024 歐盟「人工智慧法」要義及其初步研析
一、限縮本規則適用的範圍，緩和法規對事業與產業的限制
二、寬鬆定義「AI 系統」並強調系統的自主性與推論能力
三、適用本規則的行為主體範圍甚廣
四、按使用 AI 系統的風險程度進行分級管理，包括：列舉應予禁止的 AI 系統實施方式；嚴密對高風險 AI 系統的管制；對有限度風險的特定 AI 系統提供者與部署者課以透明性義務
（一）應予禁止的 AI 系統實施方式（Prohibited Artificial Intelligence Practices）
（二）嚴密對高風險 AI 系統的管制
（三）有限度風險的特定 AI 系統與透明性義務
（四）其他低（無）風險的 AI 系統
五、新增監管「通用 AI 模型」專章
六、建立層級化監理與執行機制
七、罰則
八、生效與適用
參、歐盟「人工智慧法」對全球經貿秩序的影響初探
一、產品內的高風險 AI
二、各項人類服務應用的高風險 AI
三、對特定 AI 系統的透明性要求
四、通用 AI 模型專章的規定
肆、我國人工智慧法制政策的省思－代結論
|
| 相關法條： |
 |
| 相關判解： |
|
| 相關函釋： |
|
| 相關論著： |
|