2023年第10期(第42卷总第558期)揭开“算法的面纱”：关于构建算法解释框架的思考刘烨(同济大学法学院,上海200092)摘要:如何解决算法可解释性问题是算法治理中的一个重要法律议题,囿于算法使用者与受众之间不断扩大的“解释鸿沟”,现阶段算法解释困境存在于算法运行、决策形成到应用的过程之中,具体体现在数据识别的失衡、证明依据的不足和损害结果的泛化三方面。考虑到不同算法运用场域之下解释需求的差异,借助体系思维构建起算法解释框架,或将成为解决可解释性问题的突破口。以解释对象为逻辑起点将解释方法划分为定向告知、公开披露和行政报备三种模式,并基于“场景公正”理念将其应用于医疗、信息推荐、金融等领域,针对不同业务和场景区分可解释性程度和标准,以期实现算法可解释。关键词:算法治理;自动化决策;可解释性;解释标准中图分类号:D922.14文献标识码:ADOI:10.19358/j.issn.2097-1788.2023.10.012引用格式:刘烨.揭开“算法的面纱”:关于构建算法解释框架的思考[J].网络安全与数据治理,2023,42(10):72-78.Unveilingthe"algorithmicveil":reflectiononbuildinganalgorithmicinterpretationframeworkLiuYe(SchoolofLaw,TongjiUniversity,Shanghai200092,China)Abstract:Apropersolutionoftheproblemontheexplainabilityofalgorithmisanimportantlegalissueinalgorithmgovernance,limitedbytheexpanding"explanationgap"betweenalgorithmusersandaudiences.However,atthisstage,thedilemmaoftheexplainabilityofalgorithmexistsintheprocessofalgorithmoperation,decision-makingformationandapplication,whichisem-bodiedinthreeaspects:imbalanceofdataidentification,insufficientproofbasisandgeneralizationofdamageresults.Combinedwiththedifferencesininterpretationrequirements,explanationmethodsandcriteriaofdifferentapplicationfields,buildinganal-gorithminterpretationframeworkwiththehelpofsystematicthinkingmaybecomeabreakthroughtosolvetheproblemontheex-plainabilityofalgorithm.Takingtheobjectasthelogicaljumping-offpoint,theexplainabilitymethodisdividedintothreemodes:directionalnotification,publicdisclosureandadministrativereporting.Thismethodcanbeappliedtomedical,informationrecom-mendation,financeandotherfieldsbasedontheconceptof"scenejustice",andthedegreeofexplainabilityandcriteria...
