AI與量刑——以妨害性自主犯罪為例

AI and Sentencing: Sex Crime as an Example

本文為司法院委託研究計畫案「妨害性自主量刑資訊系統之AI標註與優化發展」之研究計畫成果改寫而成1。在刑法妨害性自主罪章的21個條文（刑法第221條第1項至第332條第2項第2款等）中，以判決數量較多的八個罪名，即刑法第221條第1項、第222條第1項、第224條、第224條之1、第225條第1項、第225條第2項、第227條第1項、第227條第3項，將1,240篇判決書中法官呈現出量刑上的考量文字進行人工標註，供作AI學習訓練的資料。繼而開發出一個與性侵害犯罪相關案件的AI自動標註系統，使未來不需使用人工來標註量刑審酌事項。此外，也更新了量刑資訊系統的使用者頁面，使之成為更為直覺且對使用者友善的系統。

This article depicts the research result of a project commissioned by the Judicial Yuan, entitled“AI-Assisted Annotation and Optimization for the Sentencing Information System concerning Sexual Offenses”. Among the 21 provisions related to sexual offenses in the Criminal Code (from Article 221(1) to Article 332(2)(2)), eight offenses that have 1240 judgments, specifically in Article 221(1), Article 222(1), Article 224, Article 224(1), Article 225(1), Article 225(2), Article 227(1), and Article 227(3) were examined. The sentencing considerations presented by judges in the judgments for these offenses were manually annotated for the training data for AI learning and training process. Consequently, an AI automated annotation system related to sexual offense cases was developed which eliminates the future need of manual annotation of sentencing factors. This article also updated the user-interface of current sentencing system and made it more intuitive and user-friendly.