文章摘要
算法透明层次论
Hierarchy of Algorithmic Transparency
  
DOI:
中文关键词:  算法;透明原则;可解释人工智能;算法黑箱
英文关键词:  algorithm; the principle of transparency; explainable artificial intelligence; algorithmic black box
基金项目:
作者单位
安晋城  
摘要点击次数: 566
全文下载次数: 53
中文摘要:
      算法透明的功能价值、技术基础与规范体系问题值得反思讨论。价值层面,除改进与证明两项工具价值外,算法透明还可增进算法社会的交往、理性与信任,监督算法权力,从而促进人格自由、保障人性尊严,具有内生价值。技术层面,可解释人工智能近年来飞速发展,形成了“模型为中心”的透明与“用户为中心”的透明、“内在透明”与“事后透明”、“可观察、可分解与可模拟的透明”等技术层次,为打开黑箱、实现算法透明提供了技术基础。规范层面,现行算法透明制度在规范性质上左右摇摆,应将前述价值目标、技术基础与规范体系相互耦合,建构由柔性规范、中性规范和刚性规范组成的多层次体系。
英文摘要:
      The functional value, technical foundation and normative system of algorithmic transparency are worthy of discussion. At the value level, in addition to the two instrumental values, namely the value of improvement and the value of proof, algorithmic transparency also has such endogenous values as enhancing communication, rationality and trust in the algorithmic society and supervising the power of algorithms, which promote personal freedom and safeguard human dignity. At the technical level, explainable artificial intelligence has developed rapidly in recent years, forming technical levels such as model-centric transparency and subject-centered transparency, intrinsic transparency and post-event transparency, and observable, decomposable and simulated transparency, which provide a technical basis for opening the algorithmic black box. At the normative level, the current algorithmic transparency system swings left and right in the nature of norms, and the aforementioned value objectives, technical foundation and normative system should be coupled with each other to build a multi-level system composed of flexible norms, neutral norms and rigid norms.
查看全文    下载PDF阅读器
关闭