Toward a Fairer Future:Understanding and Addressing Algorithmic Bias in Hiring

AI-driven hiring systems have gradually made a difference instead of traditional hiring patterns. However, such hiring algorithms with a mask of neutrality may lead to algorithmic bias, damaging equal employment rights, which is defined as the outputs of AI algorithms benefit or disadvantage certain individuals or groups more than others without justifications for such unequal impacts. The bias can stem from two stages of algorithm design including data collection and modelling. In the stage of data collection, historical data collected containing human bias and reflecting societal inequity will perpetuate the bias, and non-representative data collected violating diversity or completeness of data will also result in the bias. In the stage of modelling, irrelevant or redundant features selected like sensitive attributes and proxy attributes closely related to sensitive attributes will introduce the bias, and opaque black-box system with less explainability may contain unknown bias. Technically, the bias can be mitigated through debiasing datasets involved in diversifying data and reweighting and resampling in preprocess, introducing fairness metrics in modelling, and increasing transparency. Regarding with legal governance, the framework of AI fairness reporting and auditing should be constructed to identify bias and achieve transparency and liability of bias should be regulated to clarify who shall be liable and why shall be liable.