AI倫理的兩面性初探──人類研發AI的倫理與AI倫理

A Preliminary Study of AI Ethical Duality: AI Ethics and Ethical Ais

目前AI國際倫理準則在規範方向上偏向AI研發及使用者，比較缺乏針對AI系統本身，但AI的自主學習能力，使得AI的倫理規範應該涵蓋這兩個面向，本文提出較為含括的哲學倫理觀來說明。這項倫理觀包含了AI的自主道德決策能力，如依據其目前及可預期的技術基礎，如與人類相較雖屬於知性與低限度的，但在決策的責任承擔上仍具備分散式行為主體性。最後針對AI兩面向的倫理規範，我們提出底線式道德義務作為所有規範的起點。並對於落實AI規範的議題，提出個案研發的倫理評估機制設立方向，以為回應。

AI (artificial intelligence) is characterized by autonomous learning and reactive capabilities. Arguably, these characteristics mean AIs can be held liable for failing to adhere to ethical, legal and social regulations when undertaking autonomous acts. In this paper, we consider the du-ality of AI ethics: that AI ethics should work both for humans and Arti-ficial Intelligences. We lay out a more inclusive ethical view which would accommodate the idea of AIs characterized by distributive agency, which posits that AIs possess only a minimal degree of autonomy or intelligent autonomy in virtue of which AIs can be given a certain agency status. The agency status of AI will differ from that of human beings, governments, and corporations, nonetheless, the inclusive view also suggests that AIs ought to be held accountable for their acts: shared responsibility. We also suggest an idea of a bottom-line view of moral obligations as the starting point for the duality of AI ethics. Finally, we suggest an implementation mechanism of AI ethics based on case eval-uations.