Mitigating Bias in AI-Driven Recruitment: A Comparative Study of Machine Learning Algorithms Vs. Human Screeners

Abstract

The integration of Artificial Intelligence (AI) into human resource management offers a promising avenue for enhancing efficiency, yet concerns regarding algorithmic fairness persist. This study investigates the efficacy of Machine Learning (ML) algorithms compared to human screeners in mitigating bias during the initial candidate screening process. The purpose of this research is to determine whether AI systems reduce demographic biases or merely replicate the historical prejudices embedded in training data. Using a controlled experimental design, the study analyzed a dataset of 5,000 anonymized resumes. A supervised learning model (Random Forest) was pitted against a panel of experienced human recruiters to evaluate candidates based on identical job descriptions. Key findings indicate that while human screeners exhibited significant \"affinity bias\"—favoring candidates with similar educational backgrounds—the baseline ML model initially perpetuated gender bias found in historical hiring data. However, after applying algorithmic de-biasing techniques, the AI demonstrated a higher degree of consistency and fairness than the human control group. The study concludes that AI is not a silver bullet; however, when rigorously audited, it serves as a crucial objective counterweight to human subjectivity. These results advocate for a \"human-in-the-loop\" hybrid approach to ensure equitable recruitment practices.

Introduction

The study examines the integration of Artificial Intelligence (AI) in Human Resource Management (HRM), particularly in recruitment, and its implications for bias and fairness. With the digitization of HR and high-volume applicant pools, AI tools like Machine Learning (ML) algorithms and Natural Language Processing (NLP) have become essential for resume parsing, candidate screening, and predictive hiring. While AI increases efficiency, it also amplifies algorithmic bias, reflecting historical patterns of discrimination in hiring data. Human recruiters, meanwhile, are prone to cognitive biases such as affinity bias, halo effect, and gender or racial bias.

Problem and Research Gap

The key challenge is understanding how AI bias compares to human bias, and whether human-in-the-loop approaches mitigate or reinforce discriminatory patterns. Existing studies focus separately on AI or human bias, leaving a gap in direct comparative analysis.

Objectives

1. Identify sources and patterns of bias in AI recruitment systems.

2. Compare AI-driven screening outcomes with human decision-making.

3. Explore mitigation techniques like algorithmic de-biasing and blinded resumes.

Literature Review Highlights

• AI in recruitment enhances efficiency but risks scaling historical biases.

• Human bias persists in gender, racial, and affinity preferences.

• Algorithmic bias arises from training data and proxy variables; notable cases include Amazon’s 2018 AI recruiting tool and LinkedIn ad targeting.

• Mitigation approaches include bias auditing, explainable AI (XAI), and data balancing.

Methodology

The study employs a mixed-methods design, comparing AI algorithms (Logistic Regression, Random Forest, Gradient Boosting) with human screeners using a 2,000-resume anonymized dataset. Quantitative analysis uses Disparate Impact Ratio (DIR), precision, and recall, while qualitative analysis explores recruiter reasoning through thematic coding.

Significance

The research informs HR professionals, developers, and policymakers on fair AI adoption, highlighting strategies to reduce bias, improve diversity, and ensure equitable employment opportunities. It bridges the gap between technological efficiency and ethical recruitment practices, offering empirical evidence for policy and procedural interventions.

Conclusion

This comparative study endeavored to disentangle the complex web of bias inherent in both human cognition and machine learning algorithms within the recruitment domain. The findings lead to three critical conclusions: 1) Humans and Machines Fail Differently: Human screeners exhibited distinct \"Affinity Bias,\" favoring candidates with shared educational backgrounds or social signals, often at the expense of objective skill alignment. Conversely, Baseline AI models functioned as \"historical mirrors,\" accurately replicating and amplifying the systemic gender and ethnic biases present in 10 years of training data. 2) The Mitigation Trade-off: The study demonstrated that algorithmic bias is not immutable. Through adversarial de-biasing and data re-weighting, the Mitigated AI model achieved a Disparate Impact Ratio (DIR) of 0.92, significantly outperforming human fairness levels (0.75). However, this required a deliberate trade-off where \"historical accuracy\" was sacrificed for \"normative fairness.\" 3) Augmentation over Automation: The most effective recruitment model proved to be a hybrid approach. AI excels at high-volume, blind skill matching, while humans are necessary for nuance and ethical oversight—but only when the humans are blinded to demographic markers during the initial review.

References

[1] Barocas, S., & Selbst, A. D. (2016). Big data’s disparate impact. California Law Review, 104(3), 671–732. [2] Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. American Economic Review, 94(4), 991–1013. https://doi.org/10.1257/0002828042002561 [3] Bogen, M., & Rieke, A. (2018). Help wanted: An examination of hiring algorithms, equity, and bias. Upturn. https://www.upturn.org/reports/2018/hiring-algorithms/ [4] Dastin, J. (2018, October 10). Amazon scraps secret AI recruiting tool that showed bias against women. Reuters. https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G [5] Dwork, C., Hardt, M., Pitassi, T., Reingold, O., & Zemel, R. (2012). Fairness through awareness. Proceedings of the 3rd Innovations in Theoretical Computer Science Conference, 214–226. ACM. [6] Kim, P. T. (2017). Data-driven discrimination at work. William & Mary Law Review, 58(3), 857–936. [7] Lambrecht, A., & Tucker, C. (2019). Algorithmic bias? An empirical study of apparent gender-based discrimination in the display of STEM career ads. Management Science, 65(7), 2966–2981. https://doi.org/10.1287/mnsc.2018.3093 [8] Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM Computing Surveys, 54(6), 1–35. [9] Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism. New York University Press. [10] O’Neil, C. (2016). Weapons of math destruction: How big data increases inequality and threatens democracy. Crown. [11] Raghavan, M., Barocas, S., Kleinberg, J., & Levy, K. (2020). Mitigating bias in algorithmic hiring: Evaluating claims and practices. Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, 469–481. ACM. [12] U.S. Equal Employment Opportunity Commission. (1978). Uniform guidelines on employee selection procedures. 29 C.F.R. § 1607.

Copyright

Copyright © 2025 Satyam Bajpai, Siddhant Mishra. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.