学报首页 >> 期刊浏览 >> 正文  标    题： 社会治理中的深度学习算法公平性（创刊号）（2022 第1卷 第1期 3） 作       者： 郁建兴，刘宇轩 文章栏目： 智慧城市 摘       要： 在人工智能高速发展的今天，深度学习在提高决策的准确性与稳定性方面已取得突破性进展，在社会治理中扮演着越来越重要的角色。然而，深度学习常用算法的公平性在近年来广受争议。缺少公平性的算法在公共政策与社会治理领域可能带来巨大危害。本文基于社会治理的现实场景，尝试归纳算法不公平出现的制度诱因与技术源头，并分别从社会治理和深度学习算法两个角度分析双方需要向对方领域提出的诉求与挑战。在社会治理中，技术专家和算法模型都需要转变自身角色，融入社会治理体系，特别是需要将社会治理作为特殊且具体的场景，研发与社会治理目标相匹配的深度学习算法。政府需要打破传统技术外包模式，保证与技术专家领域互通，让深度学习算法更好地融入社会治理生态。在此基础上，深度学习算法还需要从公平性指标、数据和模型等三个维度回应社会治理场景下的公平问题。 关  键  词： 深度学习算法，算法公平性，社会治理 Abstract： With the rapid development of artificial intelligence , deep learning has made breakthrough progress on the accuracy and stability of decision making, which is now playing a more and more important role in social governance. However, the fairness of deep learning algorithms is widely debated. Prior research shows that if an algorithm lacks fairness, it may harm public policy and social governance. This paper uses apractical, social governance scenario to investigate the institutional inducement of and technical causes for algorithmic discrimination. It analyzes , from the perspective of social governance and deep learning, respectively , the costs and the benefits of using a deep learning algorithm in social governance for both the social governance goals and the proper functioning of the deep learning algorithm. During social governance , technologists should alter their role together with algorithmic models to better integrate the technology into the social governance system. Particularly, there is a need to develop deep learning algorithms that match social governance goals by considering social governance as a special and specific technical scenario. The government ought to break from the traditional technology outsourcing approach and ensure that effective communication exists between technical experts , so that deep learning algorithms can be better applied to social governance. Ideally, a deep learning algorithm should respond to the naturally arising fairness problems in socialgovernance within the preset dimensions of fairness as measured by indicators , data , and models. Keywords： Deep Learning Algorithms : Fairness; Social Governance 作者简介： 郁建兴，浙江工商大学党委书记、校长，浙江大学公共管理学院教授、社会治理研究院院长，E-mail: yujianxing@zju.edu.cn。 刘宇轩（通信作者），浙江大学公共管理学院博士生, E-mail: yuxuanliu@zju.edu.cn。 链       接： 阅读原文 分享新闻: 0

	友情链接