［美］安德里娅·罗斯  著 董帅，张弛，阳平 译 汪诸豪 校

【摘 要】本文探讨了刑事裁判中“机器”的兴起。如今，人证常常让位于机器装置和解析软件，陪审团对道德归责性的复杂判断也让位于犯罪构成的机械化指标，而法官的复杂裁判则让位于量刑指南和精算工具。尽管机器化具有提高刑事司法客观性和准确性的光明前景，但由于当前机器化司法的开发和应用水平不均衡、不成体系，该前景仍未实现。目前，司法证明、责任认定和量刑方面的机器化现状存在着可预见但却未受到认真对待的自动化缺陷：“黑箱”程序中潜藏着主观性和错误；对归责问题设置的指标过于简化（而且往往非常不准确）造成决策失真；人类安全阀所保护的价值（例如，尊严、公平和仁慈）受到减损；在某些方面，机器本可以成为强大的消除偏见的工具，但由于公共政策层面的动因不足，其开发和应用水平却甚显不足。例如，政府一方面在提高通常会使匹配数据更具有罪倾向的DNA解析软件的客观性，另一方面却又信奉指印和工具痕迹分析人员所作的主观判断；而在上述软件面前，这些分析人员对身份同一性的主张可能会显得苍白无力。同样，在庭审中，当测谎结果通常仅由被告提供时，政府指责测谎仪是“不可靠”的设备；但在撤销缓刑听证会上 , 政府却又经常使用测谎仪，彰显其作为“真相机器”的传统地位。本文最后提出一种系统化的解决方案，即“人机交互审判”，用以防范机器自动化的病理缺陷；同时，通过“衡平监控”和其他手段来对机器化中的明显缺失进行盘查。

【关键词】机器；算法；法庭证据；DNA

【中图分类号】D915.13

【文献标识码】A

【文章编号】1674-1226（2021）01-0087-42

Trial by machine. Andrea Roth. Professor, UC Berkeley School of Law. Translated by Dong Shuai (Institute of Evidence Law and Forensic Science, CUPL), Zhang Chi (Leiden University, Advanced LLM in Law and Digital Technologies), Yang Ping(Assistant Researcher, China Academy of Discipline Inspection and Supervision), Wang Zhuhao (Institute of Evidence Law and Forensic Science, China University of Political Science and Law, Beijing,China, 100088).

【Abstract】This Article explores the rise of "machines" in criminal adjudication.Human witnesses now often give way to gadgets and interpretive software, juries' complex judgments about moral blameworthiness give way to mechanical proxies for criminality, and judges' complex judgments give way to sentencing guidelines and actuarialinstruments.Although mechanization holds much promise for enhancing objectivity and ccuracy in criminal justice, that promise remains unrealized because of the uneven, unsystematic manner in which mechanized justice has been developed and deployed. The current landscape of mechanized proof, liability, and punishment suffers from predictable but underscrutinized automation pathologies: hidden subjectivities and errors in "black box" processes; distorted decision making through oversimplified--and often dramatically inaccurate--proxies for blameworthiness; the compromise of values protected by human safety valves, such as dignity, equity, and mercy; and even too little mechanization where machines might be a powerful debiasing tool but where little political incentive exists for their development or deployment. For example, the state promotes the objectivity of interpretive DNA software that typically renders match statistics more inculpatory, but lionizes the subjective human judgment of its fingerprint and toolmark analysts, whose grandiose claims of identity might be diluted by such software. Likewise, the state attacks the polygraph as an unreliable lie detector at trial, where results are typically offered only by defendants,but routinely wields them in probation revocation hearings, capitalizing in that context on their cultural status as "truth machines. " The Article ultimately proposes a systems approach--"trial by cyborg" --that safeguards against automation pathologies while interrogating conspicuous absences in mechanization through "equitable surveillance" and other means.

【Key Words】machine, algorithm, forensic evidence, DNA