Overconfidence, gender, and authority attribution in algorithmic versus human advice: A cognitive perspective on compliance.

Abstract

Authority bias and overconfidence are two common cognitive tendencies that influence how individuals react to expert advice when faced with uncertainty. This study explores the interaction between these biases in decision-making, specifically examining how the source of advice (whether human or AI), perceived epistemic authority, and gender affect compliance with that advice. In a between-subjects experimental design (N = 242), participants predicted the outcomes of football matches under one of three conditions: AI-generated advice, human-generated advice, or no advice (control). Surprisingly, labelling a source as an expert did not significantly increase compliance, suggesting that mere symbolic cues of authority may not consistently lead to authority bias. Additionally, overconfidence was inversely related to advice-taking, particularly among male participants who rated their knowledge highly despite having lower predictive accuracy. Interestingly, female participants outperformed males only in the AI condition, which may indicate a gender difference in receptivity to algorithmic sources. Furthermore, perceptions of epistemic authority did not vary significantly by gender or type of source, nor did they predict compliance behaviour. These results clarify how overconfidence and authority bias interact and influence decision-making under uncertainty. They also highlight overconfidence's potential buffering role and women's relatively higher receptivity to AI-generated advice, within the context.