"Using AI & ML to automate and enhance recruiting processes" (Anand, 2025)
Used for:
Shaped by cognitive biases, cultural stereotypes & heuristics (Dror, 2020)
Perceived as objective, rule-based & consistent so people trust it
AI inherits & amplifies historical prejudices rather than reducing them (Martínez et al., 2022)
| Theory | Core Idea | In AI Hiring |
|---|---|---|
| Social Role Theory | Historical divisions create stereotypes: men = leadership, women = support | AI trained on past data learns certain roles "belong" to a gender |
| System Justification | We want to see systems as fair. This leading to automation bias | Recruiters defer to AI even against their own gut feeling |
| Stereotype Content Model | Judging happens on two axes: Competence & Warmth | AI sorts men → leadership, women → support roles accordingly |
(Eagly et al., 2026; Nadeem et al., 2021; Jost & Banaji, 1994; Fiske et al., 2002)
Amazon's tool systematically downgraded female résumés, using gendered markers as proxies (nobody noticed for years)
Tools are opaque. Candidates don't know if AI was involved or why they were rejected. Facial data, voice & social media also collected, threatening privacy.
A biased system violates human dignity regardless of outcome (the means matter, not just the ends)
Discriminating against half the world's population can't be outweighed by efficiency gains.
If hiring perpetuates structural inequality, the system simply fails the fairness test.
Findings demonstrated that gender bias within AI hiring systems does not emerge as a singular fault or fluke of technology. Instead emerges from a historical and societal set of prejudices, and is embedded within training data, current organizational practices, and human decision-making processes.