Artificial intelligence, gender and work

Abstract

This thesis explores the use of artificial intelligence (AI) technologies in hiring and their impact on gender inequality in the labor market. While AI has been adopted by firms with the expectation of unbiased decision-making processes, existing research shows that AI often tends to exhibit bias learned from humans, thereby reinforcing gender disparities. The thesis investigates why this is the case, focusing on two widely adopted AI tools in hiring: predictive algorithms and assessment software. Previous studies have primarily focused on predictive algorithms, demonstrating that they can perpetuate human biases due to their reliance on firms’ historical employment choices to make the hiring decision. In contrast, assessment software, which evaluates candidates through resume screening or cognitive tests, has received less attention in the literature. This thesis sheds new light on how assessment software and predictive algorithms differently affect gender inequality in hiring. Chapter 1 introduces the key aspects of the thesis, including its motivation, theoretical framework, and empirical strategy. It discusses the use of AI in hiring and highlights the fresh research perspectives it brings to the study of gender discrimination in the labor market. The chapter aims to (i) provide an overview of the thesis’s contribution to the existing literature on AI and gender discrimination and (ii) explain the primary theoretical and empirical approach employed in the thesis. Chapter 2 presents empirical evidence based on data from Global Fortune 500 firms, employing a difference-in-differences approach. By examining the combined impact of assessment software and predictive algorithms, the chapter shows that the use of AI in hiring increases the proportion of female managers hired by firms and is correlated with a reduction in firms facing gender discrimination lawsuits related to hiring. Chapter 3 delves deeper into the study of AI and explores the heterogeneous effects of assessment software and predictive algorithms on gender inequality when they automate the hiring process. An intervention study conducted in a private company shows that, when granted full autonomy in hiring, assessment software significantly increases the representation of female applicants shortlisted for job interviews. Conversely, predictive algorithms do not differ significantly from human recruiters in promoting gender diversity in the hiring process. Both AI tools ensure, unlike human recruiters, that the selected applicants are highly qualified. Chapter 4 completes the picture by examining the use of assessment software and predictive algorithms as complements to human recruiters in the hiring process. By modeling employers’ hiring choices and conducting an online experiment, the study demonstrates that both assessment software and predictive algorithms enable recruiters to escape information cascades. Additionally, both AI tools improve the overall productivity of selected applicants. Assessment software can also alter employers’ prior beliefs about job candidates and enhance the diversity of hires, particularly when significant productivity differences exist among the job applicants under consideration. The thesis concludes by emphasizing the importance of understanding the implications of assessment software’s autonomy in hiring decisions. Although assessment software can reduce gender inequality in hiring when it automates hiring decisions and supports human recruiters, it may not address existing gender biases effectively when used in tandem with recruiters, similar to predictive algorithms. The thesis suggests that granting full autonomy to assessment software may be a more effective approach to reducing gender inequality in the labor market.