Fairness in AI-Driven Recruitment: Challenges, Metrics, Methods, and Future Directions

招聘流程对于一个组织的成功定位至关重要，从找到合格且适合的求职者到影响其产出和文化。因此，在过去的一个世纪里，人力资源专家和工业-组织心理学家已经建立了招聘实践，如通过工作广告吸引候选人，使用评估来衡量候选人的技能，以及使用面试问题来评估组织适应性。然而，大数据和机器学习的出现已经迅速改变了传统的招聘流程，因为许多组织已经开始使用人工智能（AI）。鉴于基于AI的招聘的普及，越来越多的人担心人类偏见可能会延续到这些系统所做出的决策中，从而通过系统应用放大这种影响。实证研究已经确定了候选人排名软件和聊天机器人交互中普遍存在的偏见，催生了过去十年致力于AI公平性的日益增长的研究。本文通过讨论AI驱动招聘中遇到的偏见类型，探讨各种公平性指标和缓解方法，以及审计这些系统的工具，提供了这个新兴领域的全面概述。我们强调当前的挑战，并概述未来发展公平的AI招聘应用程序的方向，确保公正对待候选人并增强组织成果。

AI-based recruitment may carry over human biases, leading to unfair treatment of candidates. This paper aims to explore the types of biases encountered in AI-driven recruitment and to examine fairness metrics and mitigation methods.

The paper proposes various fairness metrics and mitigation methods, including pre-processing data to remove biases, using adversarial training to improve fairness, and post-processing techniques to adjust model outputs. The authors also suggest auditing tools to ensure the fairness of AI recruitment systems.

The paper discusses the types of biases encountered in candidate ranking software and chatbot interactions, and provides examples of how these biases can amplify through systematic application. The authors also present a comprehensive review of fairness metrics and mitigation methods, and highlight the challenges and future directions for developing fair AI recruitment applications.

Related studies include 'Discrimination in Online Ad Delivery' by Ali et al., 'The Case for Algorithmic Transparency in the Digital Economy' by Pasquale, and 'Fairness in Machine Learning: Lessons from Political Philosophy' by Binns et al.