可解释性何以影响基层公务员的算法决策采纳偏好  

How Interpretability Affects Grassroots Civil Servants'Decision Preferences in Algorithm Systems

作　　者：段哲哲[1,2,3] 焦嘉欣 Zhezhe Duan;Jiaxin Jiao

机构地区：[1]深圳大学政府管理学院、廉政研究院、全球特大型城市治理研究院 [2]深圳大学新质生产力研究院 [3]广东省习近平新时代中国特色社会主义思想研究中心深圳大学基地 [4]深圳大学传播学院

出　　处：《公共行政评论》2024年第6期78-98,M0005,共22页Journal of Public Administration

基　　金：国家社会科学基金青年项目“少数基层干部‘空忙’现象及矫治机制研究”(21CZZ03);广东省普通高校特色创新类项目(哲学社会科学)“新时代超大城市社会治理现代化实践探索与长效机制研究”(2023WTSCX076);深圳大学2024年研究生自主创新成果培育项目“公共部门算法应用的基层公务员接受度影响因素与提升路径研究”(868-000002020293)。

摘　　要：以算法技术为辅助工具浸入公共部门决策已变得无处不在,人工智能渗透公共治理过程的新一代治理模式成为未来公共部门数字转型的重要方向。然而,当前算法系统“黑箱”属性受到越来越多的质疑,带来算法治理的“信任赤字”与“问责缺失”等问题。研究从可解释的人工智能(Explainable Artificial Intelligence,XAI)逻辑出发,通过一项2(解释来源:规则vs数据)×2(解释内容:程序vs结果)的调查实验,在税务部门应用算法决策系统的场景中,探索不同类型XAI对传统官僚采用算法决策建议的作用机制。实证结果显示:可解释的算法系统可以促进基层公务员对算法决策的采纳偏好,其中数据驱动的程序解释型算法系统或数据驱动的结果解释型算法系统,相较无解释组,均会促进基层公务员对算法决策的采纳偏好;数据驱动的XAI比规则驱动的XAI更能促进传统官僚对算法决策的采纳偏好;数据驱动结果可解释性算法系统能够增强传统官僚对于算法系统的公平感知和技术信任,从而更偏好采纳算法决策。这些发现表明,不仅要发展算法官僚的技术能力,也要建构传统官僚能够理解的算法系统,探索人机良性互动模式,才能推动算法系统在公共部门的顺利落地。Algorithms are becoming an omnipresent means of public sector decision-making.The new-generation governance model of artificial intelligence has emerged as a significant orientation for the future digital transformation of the public sector.Nevertheless,the“black box”attribute of the current algorithm system has been increasingly questioned,giving rise to the“trust deficit”and“lack of accountability”in algorithm governance.This paper explores the logic of Explainable Artificial Intelligence(XAI)through a 2(interpretation source:rules vs.data)×2(interpretation content:procedures vs.results)survey experiment.In this scenario,the tax department applies the algorithm decision system,and the mechanism of different types of XAI on traditional bureaucrats'adoption of algorithm decision suggestions is examined.The empirical results indicate that,first,the interpretable algorithm system can boost the adoption preferences of grassroots civil servants;and the data-driven program interpretation algorithm system or data-driven result interpretation algorithm system can enhance the adoption preference of grassroots civil servants.Second,data-driven XAI can promote the algorithmic decision adoption preferences of traditional bureaucrats more effectively than rule-driven.Finally,the data-driven result-interpretable algorithmic system can heighten the traditional bureaucracy's perception of fairness and technical trust in the algorithmic system,thereby making them more inclined to adopt algorithmic decisions.These findings imply that we should develop the technical capabilities of algorithmic bureaucrats,provide interpretable types of algorithms to traditional bureaucrats,and explore favorable human-machine interactions to facilitate the smooth implementation of algorithmic systems in the public sector.

关 键 词：算法决策 可解释性 人工智能 人机互动 算法官僚 基层公务员 

分 类 号：D63[政治法律—政治学]