Artificial intelligence in recruitment : a qualitative analysis and requirements for promoting fairness

Abstract

Integrating Artificial Intelligence (AI) in recruitment processes is often seen as advantageous. However, despite claims of objectivity, AI systems have demonstrated vulnerabilities to biases and errors that can lead to discrimination. Research on fairness in AI-assisted recruitment (AI recruitment) is currently limited. Additionally, organisations face challenges in selecting AI recruitment tools concerning fairness.Using a design science research methodology, this thesis begins by formulating criteria for a fair recruitment process and identifying potential biases in AI recruitment. The focus is on the phases where AI tools are used, directly influencing the evaluation and selection of suitable candidates. The definition of fairness is examined from two perspectives: the perceived fairness by the candidates and the objective fairness concerning the AI system. Through a systematic literature review of AI principles and guidelines, relevant dimensions and requirements for AI recruitment tools that promote fairness are derived. Legal perspectives, such as the EU AI Act, are also considered. An artefact with guiding questions is developed to help organisations identify potential issues in AI recruitment tools. The correctness and completeness of the artefact are validated through comparative analyses with similar research. To evaluate the practicability and usefulness of the artefact, this thesis conducts qualitative case studies on selected AI recruitment applications and identifies more aspects than those documented in existing literature.This thesis contributes to both academic and practical fields. It provides an overview of AI recruitment, including its associated challenges, thereby raising awareness. It adapts abstract ethical AI guidelines to the context of AI recruitment and promotes the development and adoption of trustworthy AI systems. It supports the understanding of fair AI applications in recruitment. The developed requirements and guiding questions foster transparency and informed decision-making in the selection of fairer AI tools. Additionally, the results offer valuable insights for providers of AI recruitment tools and encourage improvements in alignment with the requirements. Furthermore, this study emphasises the importance of domain-specific AI guidelines and the necessity of making critical ethical AI principles binding and provides recommendations for future enhancements.