从算法分析看人工智能的价值非中立性及其应对  

The Value Non-neutrality of Artificial Intelligence and Countermeasure from the Perspective of Algorithm Analysis

作　　者：朴毅 叶斌[1] 徐飞[1] Piao Yi;Ye Bin;Xu Fei(Department of Philosophy of Science and Technology,University of Science and Technology of China,Hefei 230026,China)

机构地区：[1]中国科学技术大学科技哲学部,安徽合肥230026

出　　处：《科技管理研究》2020年第24期245-251,共7页Science and Technology Management Research

基　　金：国家社会科学基金青年项目“笛卡尔的表征主义问题研究”(19CZX043);中国科学技术大学青年创新基金项目“人工智能认识论中的表征问题”(WK2111050003)。

摘　　要：随着人工智能技术的广泛应用,“大数据杀熟”“人工智能侵权”以及“算法歧视”等负面事件频发,原先人们以为的人工智能价值中立性也因此而遭到质疑。目前国内对人工智能价值中立性的相关研究仍较多集中于概念层面。在已有研究基础上,进一步从机器学习的算法层面,从数据的社会性、算法的价值偏好以及决策的偏见强化3个角度分析人工智能的价值非中立性实质,并对其所引发的诸如“信息茧房”等伦理问题和智慧系统中的决策风险进行剖析,提出实现安全、公平、透明、道德、智慧的人工智能系统的有效途径是通过宏观层面与微观层面的有效联动,一方面推进算法透明,另一方面为人工智能的决策设置“人为”的边界。frequently,such as"big data price discrimination","artificial intelligence infringement","algorithm discrimination"and so on,the value neutrality of artificial intelligence,which was deemed by people before,has been questioned a lot.The existing relevant domestic researches are mainly discussions at the conceptual level.This article,based on the existing researches,illustrates the value non-neutrality essence of artificial intelligence at the algorithm level of machine learning,from the aspects of the sociality of data,the value preference of algorithm and the reinforcement of decision bias.It also analyzes the ethical problems,such as"information cocoon room",and decision risk in artificial intelligence systems.It is proposed that the effective way to realize safe,fair,transparent,moral and intelligent artificial intelligence systems is to promote algorithm transparency and set boundaries for decision-making of artificial intelligence through effective joint effort from macro level and micro level.

关 键 词：人工智能 机器学习 价值中立 算法歧视 决策风险 

分 类 号：NO31[文化科学] G301