When Bias Backfires: The Modulatory Role of Counterfactual Explanations on the Adoption of Algorithmic Bias in XAI-Supported Human Decision-Making

AI bias can subtly reshape human judgment even when people don't consciously detect it: Only 8 out of 294 participants (2.7%) noticed the gender bias in AI recommendations, yet those who weren't given explanations showed altered hiring preferences that persisted even after the AI was removed, suggesting bias can be absorbed unconsciously.

Counterfactual explanations can reverse the harmful effects of biased AI: When participants received counterfactual explanations (showing alternative scenarios), they actually reversed the bias in their independent decisions during the post-interaction phase, whereas those without explanations maintained the biased preferences they had adopted.

People follow biased AI recommendations at high rates regardless of explanation type: Participants followed AI recommendations 70% of the time when candidate qualifications were similar, indicating that explainability alone doesn't prevent over-reliance on AI—the type of explanation matters for preventing lasting bias adoption.

Trust in AI systems doesn't protect against bias adoption: Reported trust levels were similar across all conditions, meaning participants who trusted the system weren't necessarily more protected from absorbing its biases, highlighting that subjective trust is not a reliable safeguard against algorithmic bias.

Explainable AI (XAI) design choices have real consequences for equitable decision-making: The study demonstrates that how AI systems explain their recommendations—not just whether they do—directly influences whether humans internalize algorithmic bias, making XAI design a critical component of responsible AI deployment.