Employees show greater willingness to voice toward algorithmic than human leaders in cognitive tasks through fairness perception and psychological safety

The development of AI technology has enabled algorithms to gradually take on some of the management tasks previously completed by human managers in organizations, such as task allocation, performance review, employee promotion, and so on. As algorithms increasingly assume managerial roles, a novel form of human–machine interaction emerges, characterized by the relationship between algorithmic leaders and human subordinates. Therefore, it is worth exploring how people respond to algorithmic leaders and how they respond differently to human leaders. This study investigates voice behavior, a key form of upward communication and pro-organizational behavior in employee-leader interactions. It explores differences in employees’ willingness to engage in voice behavior when led by algorithmic versus human leaders and examine the underlying mechanisms. We conducted three experimental studies, all using scenario-based materials. Across the three studies, the independent variable was leader type (algorithmic leader vs. human leader), and the dependent variable was willingness to voice to the leader. Participants in all three studies were recruited from the Credamo platform and were based in China. Study 1 examined the differences in human employees’ willingness to engage in voice behavior toward different types of leaders. Study 2 explored the moderating role of task type, investigating how employees’ willingness to engage in voice behavior differed across cognitive and emotional tasks when interacting with different types of leaders. Study 3 focused on the serial mediation mechanism by measuring participants’ fairness perception and psychological safety in the experiment. Study 1 found that employees are more willing to voice to algorithmic leaders than human leaders ( F (1,179) = 7.08, p < 0.01, η 2 p = 0.038). Study 2 found that task type (cognitive vs. emotional) influenced the differences between the two leader types ( F (1, 284) = 6.64, p = 0.010, η 2 p = 0.023), with employees more willing to voice to algorithmic leaders than human leaders on cognitive tasks ( F (1,139) = 10.86, p < 0.001, η 2 p = 0.072). This effect was absent in the emotional tasks ( F (1,139) = 0.001, p = 0.978, η 2 p < 0.001). Study 3 found that individuals had higher fairness perceptions toward algorithmic leaders than toward human leaders, which was associated with higher psychological safety and, in turn, greater willingness to engage in voice behavior. ( b = − 0.047, 95% CI = [− 0.090, − 0.018]). This study reveals people’s willingness to engage in voice behavior to algorithmic leader, which is present in cognitive tasks but not in emotional tasks, and reveals the serial mediation model of fairness perception and psychological safety. This study examines the modest association between algorithmic leadership and human subordinates’ willingness to engage in voice behavior, thereby contributing to the literature on algorithmic leadership, human-machine interaction, and human-AI collaboration. The findings also offer practical insights for deploying and designing algorithmic systems in organizational settings.