镜像与花园之辩:算法歧视争议下的价值目标与伦理实践——基于工程师的访谈

Mirror world or bias-free garden: Exploring ethical goals and practices in the context of algorithmic discrimination

算法歧视现象构成个人数字化生存的伦理挑战。通过访谈22位国内算法工程师,本研究从技术的社会建构视角出发,探讨了他们的价值目标和伦理实践如何作用于有偏算法系统之产生、发展与应用。研究发现,算法歧视与工程师对真实和准确的理解、追求有关——他们试图打造“镜像世界”时,以数据、社会、统计因果等为“求真”的依归,最终导致了歧视的加固与强化;同时,面向算法公平和正义的算法“向善”,被“求真求准”的价值目标和特定业务目标不断挤压;工程师所受教育和所处组织环境,让他们认为“求善”并非他们的职责。加上公平等伦理概念本身难以量化,更为建设“数字花园”制造了技术难度。这些都让“求善”沦为工程师的一种弹性选择,影响了他们“去偏”和“消歧”的伦理实践方案。这一研究为理解算法歧视提供了来自工程师伦理层面的实证依据,为理解有偏算法的社会建构提供了基于中国语境的业内见解。

Algorithmic discrimination poses a danger to the individual's digital survival. Based upon semi-structural interviews with 22 algorithm engineers, this study explores how engineers' ethical values and judgments are embedded in the discriminatory algorithm from the perspective of the social construction of technology. The study finds that algorithmic discrimination can be attributed to engineers' understanding and pursuit of reality and accuracy—for building the mirror world, they rely on data records, social reality and statistical causality, eventually leading to the reinforcement of discrimination. Besides, for engineers, their goal of algorithmic fairness is constantly squeezed by their ethic goal of seeking reality and accuracy with certain business goals. And engineers are always educated and work in the orientation that making a bias-free digital garden isn't an obligatory target for them. What's more, quantifying ethical concepts like fairness remains a technical obstacle for algorithmic workers. All these make “algorithm for good” a flexible choice for engineers, finally defining their practice to debias. This study provides an empirical window for understanding algorithmic discrimination, as well as detailed and professional insights into the construction of biased technologies in Chinese context.