人工智慧演算法之法規與監理議題──自歐盟透明性要求之規範內涵談起

Regulatory Issues of Artificial Intelligence Algorithms － Starting with the Transparency Require-ments of EU Normative Implications

在大量利用資料分析作為決策實證基礎的今日，越來越多的決定來自於各種人通智慧(AI)與機器學習演算法運作結果。 AI演算法的透明性(transparency)就是基於此需求而產生的呼聲，要求將AI演算法的黑盒子(black box)予以揭開，以避免AI決策所涉及的公眾或個人利益可能遭受損害。然而，AI演算法設計可能各有不同，即可能以各種形式之智慧財產進行保護，其中包括營業秘密的保護。在維護資料被使用者或受決策影響者的利益的透明性要求，與演算法設計者以營業秘密作為智慧財產保護之保密性要求，如何衡平是未來可能遇到的問題；而演算法揭露是否真能達到揭開AI黑盒子的目的，也是對要求演算法揭露的質疑聲音。本文將整理歐盟對於AI相關法規要求，嘗試釐清規範上對於透明性要求的價值判斷方式，並提出可能之政策判斷與監理建議，期能提供未來政策與法規範制定方向之思考。

Nowadays, where data analysis is extensively used as an evidence basis for decision-making, more and more decisions come from the results of various artificial intelligence (AI) and machine learning algorithms. The demand for transparency in AI algorithm has emerged from this context, aiming to uncover the “black box” of AI algorithm to prevent potential harm to the public or individual interests associated with AI decision-making. However, different AI algorithms designs may be protected by various forms of intellectual property, including the protection of trade secrets. The balance between the transparency requirements of safeguarding the interests of data owners or those affected by AI decision-making, and the confidentiality requirements of algorithm designers’ trade secrets is a critical problem that may be encountered in the future. Moreover, there are doubts regarding whether algorithm disclosure can truly achieve the goal of unveiling the AI black box. This article will summarize the AI-related regulations of European Union and attempt to clarify the value judgment approach to transparency requirements. Finally, this article will provide possible policy and regulatory suggestions or potential decision-making and supervision, aiming to contribute to the future consideration of policy and regulatory frameworks.