Bias in the Machine: Analyzing Gender, Race, and Age Discrimination in LLM‑Based Recruitment
Digitalization and Women's Rights: Women and Sustainability: Volume 2, CRC, Ghent, Belgium , ss.206-221, 2025
- Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
- Basım Tarihi: 2025
- Doi Numarası: 10.1201/9781003542551-20
- Yayınevi: CRC, Ghent, Belgium
- Sayfa Sayıları: ss.206-221
- Boğaziçi Üniversitesi Adresli: Evet
Özet
Large language models (LLMs) are increasingly being used in various industries, including human resources, where they assist in recruitment processes. However, these models often exhibit biases that could negatively impact fairness and inclusivity in hiring. This study examines the possible gender, race, and age biases in LLM-based personnel selection. By presenting two similar resumes to an LLM (Meta-Llama-3.1–8b), with variations in age, race, and gender, the research analyzed the model’s selection behaviors. The findings revealed that the LLM tended to favor female candidates and younger candidates in most job categories and showed racial preferences, particularly for Black candidates in certain positions. Additionally, the model demonstrated a high rate of abstention. In some cases, the LLM provided explicit rationalizations for its biases, such as associating youth with energy and adaptability. Furthermore, inaccuracies in the model’s interpretations of candidate experience were noted, raising concerns about its decision-making reliability. Overall, these results highlight the need for addressing bias in LLMs to ensure fairer outcomes in automated recruitment processes.