基层公务人员的算法避责——基于税务部门发票自动化改革的分析  

Grassroots Officials' Use of Algorithms for Blame Avoidance: An Analysis Based on the Invoice Automation Reform in Tax Departments

作　　者：梁平汉[1,2] 刘洪志 Liang Pinghan;Liu Hongzhi(School of Government,Sun Yat-sen University,Guangzhou)

机构地区：[1]中山大学中国公共管理研究中心 [2]中山大学政治与公共事务管理学院

出　　处：《经济社会体制比较》2025年第1期139-148,共10页Comparative Economic & Social Systems

基　　金：国家社科基金重点课题“构建和巩固亲清政商关系研究”(项目编号:22AZD033)。

摘　　要：在数字政府建设的背景下,人工智能算法越来越多地用于辅助决策。文章以A区税务部门的发票自动化改革为案例,讨论基层公务人员算法避责的机制、表现与可能的后果。基层公务人员的工作情境具有不完全合约的特征以及二重不完全性:其一,复杂多变难以预期的基层情况导致上下级间的信息不对称,赋予基层公务人员自由裁量权;其二,基层公务人员的决策能力和努力程度是完全的个人因素,从而可以相互协调决策,实现避责。人工智能算法能够改变基层公务人员决策者角色,且其给出的结果具有聚点的协调功能,成为与组织惯例等结构化工具相类似的新型避责工具。算法的引入缓解了第一重不完全性问题,但其聚点特性加剧了第二重不完全性问题。而且,算法避责行为阻碍了信息自下而上的及时反馈,使人工智能算法模型自身难以迭代优化,从动态上看降低了人工智能的学习效率,可能引发进一步的风险。The increasing use of artificial intelligence(AI) algorithms in decision-making processes has significant implications for digital governance. Using the case of invoice automation reform in District A's tax department, this study explores the mechanisms, manifestations, and potential consequences of using algorithms for blame avoidance.Grassroots officials operate in a framework characterized by two layers of contractual incompleteness. First, the complexity of decision-making environments creates information asymmetry across administrative hierarchies, necessitating discretionary authority. Second, officials possess private information about their own abilities and efforts, enabling them to coordinate and employ pooling strategies for blame avoidance. Algorithm-generated outputs, much like organizational routines, serve as convenient justifications for shifting accountability, effectively altering decision-making dynamics and coordination focal points.While algorithms mitigate the first layer of contractual incompleteness, they exacerbate the second, leading to reduced information feedback from lower levels of the hierarchy. This, in turn, hinders the iterative optimization and dynamic learning potential of AI systems, thereby heightening associated risks.

关 键 词：算法避责 数字政府 协调博弈 不完全合约 

分 类 号：F812.42[经济管理—财政学] D630.3[政治法律—政治学]