以人工智慧輔佐法院之刑罰裁量：破譯「黑盒子」的可能與顧慮

Using Artificial Intelligence to Assist Sentencing in Courts? Perspectives on Statistical Evidence and Big Data

大眾文化中常見以人工智慧或其他科技手段，取代人類擔任審判者的描寫，反映出社會對於司法系統的期待。然而，今日人工智慧的發展進程，是否已經使這種期待有實現的可能？本文從最有可能引入人工智慧和大數據分析等新興科技，輔助刑事法院判斷的證據類型：「統計證據」之觀點切入，說明當代的行為科學研究已經充分證明「人類智慧」在判斷上難以避免的種種缺陷，因此有求助於「人工智慧」的必要性。繼而說明人工智慧的發展歷程，探討「以人工智慧取代或輔助人類審判者」的可能性。
本文分別以使用於美國多個法域的刑事司法系統輔助軟體COMPAS，及我國司法院建置的第一代與第二代量刑輔助系統為例，說明以人工智慧和大數據分析技術輔助法院進行量刑決策時，其原理、具體方式、優缺點以及可能的憲法爭議。除了分別說明2種不同的科技手段如何輔助刑事法院進行量刑決策外，本文也比較了人類社會對於這2種科技手段產出判斷結果的接受程度，以及「人類智慧」與「人工智慧」間的互動與信任關係。
本文主張，至少以今日的技術發展而論，仍不可能以人工智慧「取代」人類法院對於量刑的判斷。一方面，司法當局已經意識到：若強制要求審判者以科技的判斷結果決定量刑，固然更可能達成一致的量刑，卻會有侵害審判權核心的疑慮；另一方面，社會大眾也不易逕行接受人工智慧判斷的結果，因為人工智慧的技術發展至今，仍難提供人類願意接受司法裁判的必備要素：充分的可解釋性與透明性。

In popular culture, the depiction of artificial intelligence or other technological means replacing human judges reflects society's expectations of the judicial system. However, has the current development of artificial intelligence made these expectations potentially, or practically, attainable? This article approaches the question from the standpoint of ''statistical evidence,'' a type of evidence that could most possibly be introduced with emerging technologies like artificial intelligence and data analysis to aid criminal court judgments. By quoting literature, this article explains that contemporary behavioral science research has adequately demonstrated the unavoidable limitations of ''human intelligence'' in judgment, thus necessitating the involvement of ''artificial intelligence'' to aid humans in decision-making, in particular in court decision-making.
This article further traces the three waves of development of artificial intelligence and explores the possibility of ''replacing human judges with artificial intelligence'' in terms of the technology development today. Using the widely employed U.S. criminal justice system assistance software, the COMPAS, as well as the first and the second generation of ''Sentencing Assistance System'', established by the Judicial Yuan (Taiwan’s highest judicial authority), this article elaborates on the specific theories, methodologies, advantages, disadvantages, and potential constitutional disputes when using artificial intelligence and big data analysis technology to assist courts in sentencing decisions. In addition to detailing how these two different technological approaches aid criminal courts in decision-making, the article also compares society's acceptance levels of judgment results produced by these technologies and delves into the interaction and trust relationship between human intelligence and artificial intelligence.
This article holds that artificial intelligence is not able to“replace”the human decision-making in court sentencing, because of two reasons. First, the judicial authority would not allow this practice, because it seems to inevitably infringe the core of the judicature. Second, the public would not allow that the sentencing decisions are made by AI machines which lacks transparency and explainability.