Causal Representation Learning for Bias Detection in AI Hiring Systems

Artificial intelligence is used widely in HR hiring systems for resume screening and ranking, yet models trained on past decisions often carry group bias through hidden paths from protected attributes to hiring outcomes.This study presents a causal representation learning framework that reduces these effects by using structural modeling, adversarial training, and counterfactual simulation.The method is tested on a structured dataset of 225 applicants and the Utrecht Fairness Recruitment Dataset with close to ten thousand records.The framework lowers the demographic parity gap from 19% to 9% and reduces the equal opportunity gap from 22% to 11%.Counterfactual consistency rises from 67.1% to 84.6%, while the Causal Disparity Index drops from 28% to 11%.Predictive performance also improves, reaching 84.3% accuracy, 82.7% precision, 79.4% recall, and an F1 score of 80.9%.Graph reconstruction error decreases from 0.071 to 0.026.These results show that causal representation learning supports fair and reliable HR hiring systems without reducing predictive strength.