数字时代刑事证据运用的风险及其规制——以算法证据为分析视角  

The Risks and Regulation of the Use of Criminal Evidence in the Digital Era

作　　者：潘金贵[1,2] PAN Jin'gui

机构地区：[1]西南政法大学法学院 [2]西南政法大学证据法学研究中心

出　　处：《法治研究》2024年第6期26-42,共17页Research on Rule of Law

基　　金：2024年度最高人民检察院检察应用理论研究课题“刑事财产性判项执行检察监督研究”阶段性成果。

摘　　要：数字信息技术对刑事司法的深度介入产生了算法证据。刑事算法证据应限于“人辅机主”型的机器学习算法证据,可归属于专门性问题报告范畴。算法通过数据结构化的过程,揭示大数据与待证事实之间的因果关系,算法证据是在司法证明整体主义模式下对大数据评价的结果,包括预测类算法证据、识别类算法证据和分析类算法证据。算法证据的司法适用存在三层风险:对无罪推定和自由心证的冲击;“概率近似正确(PAC)”理论下的事实认定错误性或歧视性问题;权力行使的隐化和异化趋势。同时存在双维困境:在实体困境维度,算法透明度和可解释性、个人数据赋权与反算法身份歧视等手段难以实现对算法的有效规制和监管,影响算法证据合理适用;在程序困境维度,数据选择与算法设计监督程序、结果告知与解释程序以及异议或质证程序的规则阙如,制约算法证据适用实效。对此,应当以事实认定的辅助性、技术性正当程序、比例原则作为算法证据司法适用的理念指引;依据算法证据生成流程,从数据可靠性、算法可靠性角度明确算法证据可靠性的具体要素;构建取证、开示、质证与认证规则,强化算法证据适用的程序规制。The deep involvement of digital information technology in criminal justice has led to the emergence of algorithmic evidence.Criminal algorithmic evidence should be limited to the category of machine-led with human assistance machine learning algorithmic evidence,which can be classified as a specialized issue report.Through the process of data structuring,algorithms reveal the causal relationship between big data and the facts to be proved.Algorithmic evidence is the result of evaluating big data under the holistic model of judicial proof,including predictive algorithmic evidence,identification algorithmic evidence,and analytical algorithmic evidence.There are three levels of risks and two-dimensional dilemmas in the judicial application of algorithmic evidence.The three levels of risks are:first,the impact on the presumption of innocence and free evaluation of evidence;second,the issue of factual determination errors or discrimination under the'probably approximately correct(PAC)'theory;third,the trend of implicit and alienated exercise of power.The two-dimensional dilemmas are:in terms of substantive dilemmas,it is difficult to achieve effective regulation and supervision of algorithms through means such as algorithmic transparency and interpretability,personal data empowerment,and anti-algorithmic identity discrimination,which affects the reasonable application of algorithmic evidence.In terms of procedural dilemmas,the lack of rules in the supervision procedures of data selection and algorithm design,result notification and interpretation procedures,and objection or cross-examination procedures restricts the effectiveness of the application of algorithmic evidence.To address these issues,first,the auxiliary nature of factual determination,technical due process,and the principle of proportionality should be used as the conceptual guidelines for the judicial application of algorithmic evidence.Second,based on the algorithmic evidence generation process,the specific elements of the reliability of algorithmic evid

关 键 词：算法证据 数字司法 证据可靠性 正当程序 专门性问题报告 

分 类 号：D925.2[政治法律—诉讼法学]