Does Data-Driven Recruitment Lead to Less Discrimination? – A Technical Perspective

Abstract

Due to its large cost-saving potential, data-driven recruitment is becoming increasingly popular across various industries. However, several cases were reported where the use of corresponding technologies had caused systematic discrimination against certain candidate groups. While existing approaches to discover and prevent discrimination in data classification mostly perform well within a specific context, it remains unclear to what extent data-driven recruitment can be conducted discrimination-free in real-world business applications, where the respective context-specific assumptions do not necessarily hold. Hence, we first define a generic discrimination model that allows for arbitrary descriptions of job candidate characteristics, before applying two sophisticated discrimination-prevention algorithms on a sample data set generated from our model to evaluate the algorithms' performance. Our analysis shows that the amount of removed discrimination highly depends on the application context and its underlying definitions and assumptions, making it hard to provide a holistic answer to our research question.