歐盟人工智慧法中可信任AI原則的衝突與權衡

The Conflicts and Trade-offs of Trustworthy AI Principles in the EU Artificial Intelligence Act

本文探討了歐盟人工智慧法（AI Act）中可信任AI原則的衝突與權衡，特別關注這些原則在實踐中面臨的挑戰與調整。文章首先介紹了可信任AI的七大原則及其在歐盟AI治理中的重要性，並深入分析原則間的衝突，包括技術上的矛盾與目標上的衝突。通過比例原則設置的「先限制再開放豁免」規範，歐盟在可信任AI與國防安全、執法等公共利益間進行權衡，並針對可信任AI原則內部衝突在法條中羅列了特殊情境。針對尚未明確列舉的衝突情境，本文利用實證實驗結果探討依應用場景直接指定所有原則優先排序的可行性，並進一步建議通過標準化規範促進法律明確性，從而幫助企業降低法規遵循風險。最後，本文強調可信任AI是具備高度相關性的概念，歐盟的立法框架並不代表完全的真理，我國應透過跨領域合作和定期更新法規，制定符合我國發展方向的彈性框架。

The paper explores the conflicts and trade-offs of the Trustworthy AI principles in the EU Artificial Intelligence Act (AI Act), mainly focusing on the challenges and adjustments of these principles in practice. The article first introduces the seven key principles of Trustworthy AI and their importance in EU AI governance, then elaborates on the conflicts among these principles, including technical contradictions and goal-related conflicts. By applying the principle of proportionality to establish the“default restriction with conditional exemptions”regulation, the EU balances Trustworthy AI with public interests such as national security and law enforcement and then lists specific scenarios within the legislation to address internal conflicts among the principles. Regarding potential conflicts not yet explicitly addressed, this article uses empirical, experimental findings to examine the feasibility of directly prioritizing principles based on application scenarios, further suggesting that standardization could enhance legal clarity and provide companies with possible approaches for compliant implementation of Trustworthy AI. Finally, the article points out that, with a highly relative nature, Trustworthy AI principles require flexible trade-offs in different application scenarios. While the EU’s legislative framework offers valuable insights, Taiwan should independently develop flexible regulations aligned with its values, prioritizing interdisciplinary collaboration and regular legislation updates to keep up with the development of AI technology.